\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: XML ouput for psql\"\n },\n {\n \"msg_contents\": \"\\nDone. Added to 1.15) How can I financially assist PostgreSQL?\\n\\n Also, if you have a success story about PostgreSQL, please submit\\n it to our advocacy site at \\n http://advocacy.postgresql.org.\\n\\n\\n---------------------------------------------------------------------------\\n\\nShridhar Daithankar wrote:\\n> Hi,\\n> \\n> We should probably put out an FAQ saying, if you have success story, please \\n> make a write-up and send to us at http://advocacy.postgresql.org.\\n> \\n> Shridhar\\n> \\n> On Tuesday 15 April 2003 18:42, Ericson Smith wrote:\\n> > As one of the top search engine campaign optimization companies in the\\n> > space, we here at Did-it.com have been using Postgresql for over a year\\n> > now. We had serious locking problems with MySQL and even switching to\\n> > their Innodb handler did not solve all the issues.\\n> >\\n> > As the DB administrator, I recommended we switch when it came time to\\n> > re-write our client platform. That we did, and we have not looked back.\\n> > We have millions of listings, keywords and we perform live visitor\\n> > tracking in our database. We capture on the order of about 1 million\\n> > visitors every day, with each hit making updates, selects and possibly\\n> > inserts.\\n> >\\n> > We could not have done this in mySQL. Basically when I see silly posts\\n> > over on Slashdot about MySQL being as good a sliced bread, you can check\\n> > out the debunking posts that I make as \\\"esconsult1\\\".\\n> >\\n> > Yes, perhaps Postgresql needs a central org that manages press and so\\n> > on, but we know that we dont get the press that MySQL does, but quietly\\n> > in the background Postgresql is handling large important things.\\n> >\\n> > - Ericson Smith\\n> > Web Developer\\n> > Db Admin\\n> > http://www.did-it.com\\n> \\n> \\n> ---------------------------(end of broadcast)---------------------------\\n> TIP 1: subscribe and unsubscribe commands go to majordomo@postgresql.org\\n> \\n\\n-- \\n Bruce Momjian | http://candle.pha.pa.us\\n pgman@candle.pha.pa.us | (610) 359-1001\\n + If your life is a hard drive, | 13 Roberts Road\\n + Christ can be your backup. | Newtown Square, Pennsylvania 19073\\n\",\n \"msg_date\": \"Fri, 30 May 2003 22:26:10 -0400 (EDT)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [GENERAL] Are we losing momentum?\"\n },\n {\n \"msg_contents\": \"\\nWhere do we want to go with this? It is interesting that it maps MySQL\\nSHOW commands on top of our existing SHOW syntax in psql. The patch\\ndoesn't look too big.\\n\\nShould this be applied? Sean, I know you have a newer patch the the URL\\nyou posted isn't good anymore. This also contains GUC change. Sean,\\nwould you show each one you added and we can discuss if any are\\ninappropriate.\\n\\n---------------------------------------------------------------------------\\n\\nSean Chittenden wrote:\\n> > > That's a pretty reasonable thought. I work for a shop that sells\\n> > > Postgres support, and even we install MySQL for the Q&D ticket\\n> > > tracking system we recommend because we can't justify the cost to\\n> > > port it to postgres. If the postgres support were there, we would\\n> > > surely be using it.\\n> > \\n> > > How to fix such a situation, I'm not sure. \\\"MySQL Compatability\\n> > > Mode,\\\" anyone? :-)\\n> > \\n> > What issues are creating a compatibility problem for you?\\n> \\n> I don't think these should be hacked into the backend/libpq, but I\\n> think it'd be a huge win to hack in \\\"show *\\\" support into psql for\\n> MySQL users so they can type:\\n> \\n> SHOW (databases|tables|views|functions|triggers|schemas);\\n> \\n> I have yet to meet a MySQL user who understands the concept of system\\n> catalogs even though it's just the 'mysql' database (this irritates me\\n> enough as is)... gah, f- it: mysql users be damned, I have three\\n> developers that think that postgresql is too hard to use because they\\n> can't remember \\\"\\\\d [table name]\\\" and I'm tired of hearing them bitch\\n> when I push using PostgreSQL instead of MySQL. I have better things\\n> to do with my time than convert their output to PostgreSQL. Here goes\\n> nothing...\\n> \\n> I've tainted psql and added a MySQL command compatibility layer for\\n> the family of SHOW commands (psql [-m | --mysql]).\\n> \\n> \\n> The attached patch does a few things:\\n> \\n> 1) Implements quite a number of SHOW commands (AGGREGATES, CASTS,\\n> CATALOGS, COLUMNS, COMMENTS, CONSTRAINTS, CONVERSIONS, DATABASES,\\n> DOMAINS, FUNCTIONS, HELP, INDEX, LARGEOBJECTS, NAMES, OPERATORS,\\n> PRIVILEGES, PROCESSLIST, SCHEMAS, SEQUENCES, SESSION, STATUS,\\n> TABLES, TRANSACTION, TYPES, USERS, VARIABLES, VIEWS)\\n> \\n> SHOW thing\\n> SHOW thing LIKE pattern\\n> SHOW thing FROM pattern\\n> SHOW HELP ON (topic || ALL);\\n> etc.\\n> \\n> Some of these don't have \\\\ command eqiv's. :( I was tempted to add\\n> them, but opted not to for now, but it'd certainly be a nice to\\n> have.\\n> \\n> 2) Implements the necessary tab completion for the SHOW commands for\\n> the tab happy newbies/folks out there. psql is more friendly than\\n> mysql's CLI now in terms of tab completion for the show commands.\\n> \\n> 3) Few trailing whitespace characters were nuked\\n> \\n> 4) guc.c is now in sync with the list of available variables used for\\n> tab completion\\n> \\n> \\n> Few things to note:\\n> \\n> 1) SHOW INDEXES is the same as SHOW INDEX, I think MySQL is wrong in\\n> this regard and that it should be INDEXES to be plural along with\\n> the rest of the types, but INDEX is preserved for compatibility.\\n> \\n> 2) There are two bugs that I have yet to address\\n> \\n> 1) SHOW VARIABLES doesn't work, but \\\"SHOW [TAB][TAB]y\\\" does\\n> 2) \\\"SHOW [variable_of_choice];\\\" doesn't work, but \\\"SHOW\\n> [variable_of_choice]\\\\n;\\\" does work... not sure where this\\n> problem is coming from\\n> \\n> 3) I think psql is more usable as a result of this more verbose\\n> syntax, but it's not the prettiest thing on the planet (wrote a\\n> small parser outside of the backend or libraries: I don't want to\\n> get those dirty with MySQL's filth).\\n> \\n> 4) In an attempt to wean people over to PostgreSQL's syntax, I\\n> included translation tips on how to use the psql equiv of the SHOW\\n> commands. Going from SHOW foo to \\\\d foo is easy, going from \\\\d foo\\n> to SHOW foo is hard and drives me nuts. This'll help userbase\\n> retention of newbies/converts. :)\\n> \\n> 5) The MySQL mode is just a bounce layer that provides different\\n> syntax wrapping exec_command() so it should provide little in the\\n> way of maintenance headaches. Some of the SHOW commands, however,\\n> don't have \\\\ couterparts, but once they do and that code is\\n> centralized, this feature should come for zero cost.\\n> \\n> 6) As an administrator, I'd be interested in having an environment\\n> variable that I could set that'd turn on MySQL mode for some of my\\n> bozo users that way they don't complain if they forget the -m\\n> switch. Thoughts?\\n> \\n> \\n> I'll try and iron out the last of those two bugs/features, but at this\\n> point, would like to see this patch get wider testing/feedback.\\n> Comments, as always, are welcome.\\n> \\n> PostgreSQL_usability++\\n> \\n> -sc\\n> \\n> -- \\n> Sean Chittenden\\n\\n[ Attachment, skipping... ]\\n\\n> \\n> ---------------------------(end of broadcast)---------------------------\\n> TIP 3: if posting/reading through Usenet, please send an appropriate\\n> subscribe-nomail command to majordomo@postgresql.org so that your\\n> message can get through to the mailing list cleanly\\n\\n-- \\n Bruce Momjian | http://candle.pha.pa.us\\n pgman@candle.pha.pa.us | (610) 359-1001\\n + If your life is a hard drive, | 13 Roberts Road\\n + Christ can be your backup. | Newtown Square, Pennsylvania 19073\\n\",\n \"msg_date\": \"Fri, 30 May 2003 23:07:00 -0400 (EDT)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Are we losing momentum?\"\n },\n {\n \"msg_contents\": \"Hi everyone,\\n\\nAs an additional point of interest, we're still processing all of the \\nCase Study submissions received from the last call around February, and \\nshould be beginning translations within the next week or two.\\n\\nThere is a lot of very good news contained in the submissions, and they \\nwill definitely assist in bringing into the open just how good a \\ndatabase system and Community we truly have.\\n\\nHope everyone will be as amazed as I was at just where PostgreSQL is \\nbeing used already. There are some very clued-in people out there.\\n\\nRegards and best wishes,\\n\\nJustin Clift\\n\\n\\nBruce Momjian wrote:\\n> Done. Added to 1.15) How can I financially assist PostgreSQL?\\n> \\n> Also, if you have a success story about PostgreSQL, please submit\\n> it to our advocacy site at \\n> http://advocacy.postgresql.org.\\n> \\n> \\n> ---------------------------------------------------------------------------\\n> \\n> Shridhar Daithankar wrote:\\n> \\n>>Hi,\\n>>\\n>>We should probably put out an FAQ saying, if you have success story, please \\n>>make a write-up and send to us at http://advocacy.postgresql.org.\\n>>\\n>> Shridhar\\n>>\\n>>On Tuesday 15 April 2003 18:42, Ericson Smith wrote:\\n>>\\n>>>As one of the top search engine campaign optimization companies in the\\n>>>space, we here at Did-it.com have been using Postgresql for over a year\\n>>>now. We had serious locking problems with MySQL and even switching to\\n>>>their Innodb handler did not solve all the issues.\\n>>>\\n>>>As the DB administrator, I recommended we switch when it came time to\\n>>>re-write our client platform. That we did, and we have not looked back.\\n>>>We have millions of listings, keywords and we perform live visitor\\n>>>tracking in our database. We capture on the order of about 1 million\\n>>>visitors every day, with each hit making updates, selects and possibly\\n>>>inserts.\\n>>>\\n>>>We could not have done this in mySQL. Basically when I see silly posts\\n>>>over on Slashdot about MySQL being as good a sliced bread, you can check\\n>>>out the debunking posts that I make as \\\"esconsult1\\\".\\n>>>\\n>>>Yes, perhaps Postgresql needs a central org that manages press and so\\n>>>on, but we know that we dont get the press that MySQL does, but quietly\\n>>>in the background Postgresql is handling large important things.\\n>>>\\n>>>- Ericson Smith\\n>>>Web Developer\\n>>>Db Admin\\n>>>http://www.did-it.com\\n>>\\n>>\\n>>---------------------------(end of broadcast)---------------------------\\n>>TIP 1: subscribe and unsubscribe commands go to majordomo@postgresql.org\\n>>\\n> \\n> \\n\\n\\n\",\n \"msg_date\": \"Sat, 31 May 2003 15:39:34 +0800\",\n \"msg_from\": \"Justin Clift \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [GENERAL] Are we losing momentum?\"\n },\n {\n \"msg_contents\": \"Sean Chittenden kirjutas R, 30.05.2003 kell 23:20:\\n> > > I assume we are not moving in the XML/psql direction, right? We\\n> > > want it int he backend, or the psql HTML converted to XHTML?\\n> > \\n> > I don't think a consensus was ever reached. It would certainly be\\n> > better if this was done on the backend, but that seems to be a long\\n> > time away, and some have argued that it is not the job of the engine\\n> > to do this anyway.\\n> \\n> Few points for the archives regarding XML and databases (spent 9mo\\n> working on this kinda stuff during the .com days):\\n> \\n> *) Use libxml2. MIT Licensed, most complete opensource XML\\n> implementation available, and fast. See the XML benchmarks on\\n> sf.net for details. To avoid library naming conflicts, the library\\n> should likely be renamed to pgxml.so and imported into the src\\n> tree. Mention java in this context and risk being clubbed to death.\\n\\nAgree completely on all points ;)\\n\\n> *) There should be two storage formats for XML data:\\n> \\n> a) DOM-esque storage: broken down xmlNodes. This is necessary for\\n> indexing specific places in documents (ala XPath queries).\\n> Actual datums on the disk should be similar in structure to the\\n> xmlNode struct found in libxml2 (would help with the\\n> serialization in either direction). In database xslt\\n> transformations are also possible with the data stored this way.\\n> \\n> b) SAX-esque storage: basically a single BYTEA/TEXT column. Not\\n> all documents need to be indexed/searchable and SAX processing\\n> of data is generally more efficient if you don't know what\\n> you're looking for. This format is the low hanging fruit\\n> though.\\n\\nI think that Oleg and Todor very recently proposed somethink that could\\nuse b) and still provide indexed access. \\n\\nMost flexible would be some way to define, how much of a tree is kept\\ntogether, as xmlNode/tuple would probably be too much overhead for most\\noperations, whereas xmlFile/tuple would also, just for other ops;)\\n\\n--------------\\nHannu\\n\\n\\n\",\n \"msg_date\": \"31 May 2003 13:35:43 +0300\",\n \"msg_from\": \"Hannu Krosing \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: XML ouput for psql\"\n },\n {\n \"msg_contents\": \"Bruce Momjian writes:\\n\\n> Where do we want to go with this? It is interesting that it maps MySQL\\n> SHOW commands on top of our existing SHOW syntax in psql. The patch\\n> doesn't look too big.\\n\\nThe response to \\\"Are we losing momentum?\\\" isn't to add redundant syntax\\nfor nonstandard features that we have no control over.\\n\\n-- \\nPeter Eisentraut peter_e@gmx.net\\n\\n\",\n \"msg_date\": \"Sat, 31 May 2003 17:55:16 +0200 (CEST)\",\n \"msg_from\": \"Peter Eisentraut \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Are we losing momentum?\"\n },\n {\n \"msg_contents\": \"Peter Eisentraut writes:\\n> Bruce Momjian writes:\\n>> Where do we want to go with this? It is interesting that it maps MySQL\\n>> SHOW commands on top of our existing SHOW syntax in psql. The patch\\n>> doesn't look too big.\\n\\n> The response to \\\"Are we losing momentum?\\\" isn't to add redundant syntax\\n> for nonstandard features that we have no control over.\\n\\nI'm of two minds about it myself. I don't like trying to play follow-the-\\nleader with a moving target. But if you think of it as trying to win\\nover converts from MySQL, it seems a lot more palatable.\\n\\nIt would also be interesting to combine this with Rod's idea of driving\\ndescribe-type queries by table instead of hardwired code. Imagine that\\nthe backend's \\\"show foo\\\" command first looks for \\\"foo\\\" as a GUC\\nvariable, as it does now, but upon failing to find one it looks in a\\nsystem table for a query associated with the name \\\"foo\\\". If it finds\\nsuch a query, it runs it and sends back the result. Now, not only can\\nwe emulate \\\"show tables\\\", but people can easily add application-specific\\n\\\"show whatever\\\" commands, which seems tremendously cool.\\n\\nThere are some safety and protection issues to be solved here (probably\\nonly superusers should be allowed to modify the query table, and we\\nshould restrict the form of the query to be a single SELECT command)\\nbut I can't think of any showstoppers. Now that we've abandoned backend\\nautocommit there are no technical reasons that SHOW shouldn't be allowed\\nto run a SELECT query.\\n\\nI did not care for the original patch (which IIRC made psql recognize\\n\\\"show\\\" commands, rather than doing it in the backend) but something like\\nthe above seems reasonably clean.\\n\\n\\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Sat, 31 May 2003 12:49:00 -0400\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Are we losing momentum? \"\n },\n {\n \"msg_contents\": \"> >> Where do we want to go with this? It is interesting that it maps\\n> >> MySQL SHOW commands on top of our existing SHOW syntax in psql.\\n> >> The patch doesn't look too big.\\n> >\\n> > The response to \\\"Are we losing momentum?\\\" isn't to add redundant\\n> > syntax for nonstandard features that we have no control over.\\n> \\n> I'm of two minds about it myself. I don't like trying to play\\n> follow-the- leader with a moving target. But if you think of it as\\n> trying to win over converts from MySQL, it seems a lot more\\n> palatable.\\n\\nThe _only_ reason I added this was to aid in the conversion of MySQL\\nusers who don't know how to navigate their way through psql and\\nPostgreSQL. If you read through the patch, the extended SHOW commands\\nare only enabled if you turn them on via the --mysql option or by\\nsetting an option (don't recall at the moment) that way you can turn\\nthese on by default in a person's .psqlrc file. With Win32 coming\\ndown the pipe (right??? *cough cough* :-P), converting existing users\\nis going to be important for PostgreSQL's long term success if it\\nseeks to gain market share and momentum.\\n\\n> It would also be interesting to combine this with Rod's idea of\\n> driving describe-type queries by table instead of hardwired code.\\n> Imagine that the backend's \\\"show foo\\\" command first looks for \\\"foo\\\"\\n> as a GUC variable, as it does now, but upon failing to find one it\\n> looks in a system table for a query associated with the name \\\"foo\\\".\\n> If it finds such a query, it runs it and sends back the result.\\n> Now, not only can we emulate \\\"show tables\\\", but people can easily\\n> add application-specific \\\"show whatever\\\" commands, which seems\\n> tremendously cool.\\n\\nI really like the ability to program in queries or syntaxes into the\\nbackend, but as it stands, SHOW foo would have to be pretty smart to\\nhandle the LIKE clauses and other bits. And how would tab completion\\nbe handled in psql? I can't remember if I implemented stuff like,\\nSHOW TABLES IN SCHEMA foo, but 'SHOW TABLES LIKE a%' was quite a hit\\nto the people I have using the patch and I don't see how that'd be\\napplicable with what has been proposed so far. The extra verbosity\\nand tab completion-ability is important for the newbie, it's an\\ninteractive environment that lets them learn and explore in a\\nproactive manner. Having IN SCHEMA vs LIKE handled on the backend\\nwould be a rather complex generic model, but in an ideal world, one\\nthat I'd prefer if done right.\\n\\n> There are some safety and protection issues to be solved here\\n> (probably only superusers should be allowed to modify the query\\n> table, and we should restrict the form of the query to be a single\\n> SELECT command) but I can't think of any showstoppers. Now that\\n> we've abandoned backend autocommit there are no technical reasons\\n> that SHOW shouldn't be allowed to run a SELECT query.\\n> \\n> I did not care for the original patch (which IIRC made psql\\n> recognize \\\"show\\\" commands, rather than doing it in the backend) but\\n> something like the above seems reasonably clean.\\n\\nWell, unless something is done very cleanly and abstractly, and can be\\nextended to handle IN SCHEMA, LIKE, etc., I'm of the opinion that this\\nkinda stuff belongs on the client side of things and _not_ in the\\nbackend. Putting MySQL's brokenness in the backend seems, well, to\\nperpetuate the brokenness and acknowledge that their brokenness isn't\\nreally that broken: not a message I'm fond of. Putting MySQL's\\nbrokenness in psql and hiding it behind the --mysql CLI option,\\nhowever, is tolerably broken, IMHO and of the same use as ora2pg or\\nmysql2pg.\\n\\nBruce wrote:\\n> Where do we want to go with this? It is interesting that it maps\\n> MySQL SHOW commands on top of our existing SHOW syntax in psql. The\\n> patch doesn't look too big.\\n>\\n> Should this be applied? Sean, I know you have a newer patch the the\\n> URL you posted isn't good anymore. This also contains GUC change.\\n> Sean, would you show each one you added and we can discuss if any\\n> are inappropriate.\\n\\nThe GUC change was just flushing out various GUC options that weren't\\navailable as tab completion options... the usefulness of those GUCs\\ncould be debated, but at least there's a complete list in the CLI\\nnow... having this pulled from the server would be wise, IMHO, but I\\ndidn't spend the time to add that at the time. If there's interest, I\\ncan do that.\\n\\nI have the patch someplace, don't worry about that. The patch isn't\\n100% correct as it stands: there is a problem in recognizing the end\\nof a SHOW command so that the query buffer was considered non-empty\\n(wasn't scanning properly for ;'s to emulate the end of a query:\\nsomething that's not necessary with the normal \\\\ syntax). Other than\\nthat, there are no bugs that I'm aware of and the work around is to\\nissue a \\\\r: not an earth shattering problem given psql has a history.\\nBefore I finished, I wanted to get some review/thoughts on the patch\\nbefore I completed the work and sent it off only to have it rejected.\\n\\nI know Tom doesn't like the idea because it's not \\\"clean,\\\" but I can't\\nhelp but feel that adding similar functionality to the backend is some\\nhow contaminating PostgreSQL and acknowledges MySQL's done something\\ncreditable, which is too big of an admission, especially since their\\nproduct is pretty fundamentally flawed in terms of spec compliance.\\nThe last thing we want is to have their brokenness turn into spec and\\nfor that to become the norm. So, as life would have it, I'm an\\nadvocate of adding the SHOW syntax to psql and an opponent of having it\\nin the backend, the core developers would rather have the SHOW syntax\\nin the backend because it's not clean to have it in the front end.\\n*shrug*\\n\\npsql has to have a quasi-parser in the front end as is to handle tab\\ncompletion, so I don't really buy the argument that it's bad to have a\\nparser in the front end (sorry Tom). In fact, I actually think that\\nthe parser in psql needs to be _beefed up_ and made _smarter_ that way\\npsql's more user friendly, even if that means increasing the\\ncomplexity of psql. At the moment, there are short falls with psql's\\nability to perform tab completion of column names on aliased tables,\\netc. Some of it's non-trivial and won't ever see the light of day in\\npsql, but there is room for improvement. Anyway, usability is a\\nproblem for clients, not for servers. MySQL usability belongs in the\\nclient, not the backend.... and hidden behind a CLI flag at that\\n'cause I don't available in my personal development environment.\\n\\nAnd before someone says, \\\"that's fine and dandy, who's going to do the\\nwork on psql,\\\" I'll gladly work on psql and even go so far as to clean\\nup the code now that someone's around to apply the patches. :)\\n\\n*listens and waits before updating the patch and fixing the above\\n bugs* -sc\\n\\n-- \\nSean Chittenden\\n\",\n \"msg_date\": \"Sat, 31 May 2003 18:43:23 -0700\",\n \"msg_from\": \"Sean Chittenden \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Are we losing momentum?\"\n },\n {\n \"msg_contents\": \"> It would also be interesting to combine this with Rod's idea of driving\\n> describe-type queries by table instead of hardwired code. Imagine that\\n> the backend's \\\"show foo\\\" command first looks for \\\"foo\\\" as a GUC\\n> variable, as it does now, but upon failing to find one it looks in a\\n> system table for a query associated with the name \\\"foo\\\". If it finds\\n> such a query, it runs it and sends back the result. Now, not only can\\n> we emulate \\\"show tables\\\", but people can easily add application-specific\\n> \\\"show whatever\\\" commands, which seems tremendously cool.\\n\\nEasy enough to accomplish for the most part. I suppose the most\\ndifficult part is whether we support arbitrary syntax?\\n\\nSHOW INDEXES ON TABLE ;\\nSHOW TABLES;\\nSHOW DATABASES;\\nSHOW COLUMNS FROM ; <-- Long form of DESCRIBE \\n\\nI believe all of the above is valid MySQL syntax, so we would need to be\\nable to work with a list of colId, rather than a single one.\\n\\nThis does not help with tab completion or help for the above items. \\nThough one could certainly argue help on available commands should come\\nfrom the backend, and psql could be taught how to read the table to\\ndetermine how to deal with tab completion.\\n\\nOne significant downside is that describe commands would require an\\ninitdb when updated.\\n\\nOh, if it's the backend doing the work, the queries should probably be\\nfunctions, not free-form queries. Simply makes it easier to inject\\nvariables into the right place. In the initial patch, psql is using a\\nprepared statement for this work with the side benefit that the plan is\\ncached. An SQL function would accomplish the same thing.\\n\\n-- \\nRod Taylor \\n\\nPGP Key: http://www.rbt.ca/rbtpub.asc\",\n \"msg_date\": \"31 May 2003 23:08:49 -0400\",\n \"msg_from\": \"Rod Taylor \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Are we losing momentum?\"\n },\n {\n \"msg_contents\": \"[ moving this thread to a more appropriate place ]\\n\\nSean Chittenden writes:\\n>> It would also be interesting to combine this with Rod's idea of\\n>> driving describe-type queries by table instead of hardwired code.\\n>> Imagine that the backend's \\\"show foo\\\" command first looks for \\\"foo\\\"\\n>> as a GUC variable, as it does now, but upon failing to find one it\\n>> looks in a system table for a query associated with the name \\\"foo\\\".\\n>> If it finds such a query, it runs it and sends back the result.\\n>> Now, not only can we emulate \\\"show tables\\\", but people can easily\\n>> add application-specific \\\"show whatever\\\" commands, which seems\\n>> tremendously cool.\\n\\n> I really like the ability to program in queries or syntaxes into the\\n> backend, but as it stands, SHOW foo would have to be pretty smart to\\n> handle the LIKE clauses and other bits.\\n\\nIt's certainly doable. I thought more about how to handle parameters\\nand such, and came up with this sketch:\\n\\n1. We generalize the SHOW syntax to accept 1 or more identifiers (might\\nas well allow strings too). The existing special cases like SHOW TIME\\nZONE would be taken out of the grammar and checked for at runtime.\\n\\n2. The \\\"key\\\" field of the show_queries table is an array of one or more\\nstrings that can be either keywords or parameter placeholders ($n).\\nThere must be at least one keyword. Then SHOW matches a particular\\ntable entry if there are the right number of words and all the keyword\\nstrings match the corresponding words. The other words become the\\nparameter values.\\n\\n3. The \\\"query\\\" field of the table is a SELECT possibly containing\\nparameter references $n. This can be handled the same way as a\\npreparable statement (we already have mechanisms for resolving the types\\nof the parameters).\\n\\nWhile I haven't studied the MySQL manual to see what-all they allow,\\nthis certainly seems sufficient to support \\\"SHOW TABLE foo\\\" and similar\\nvariants. And the possibility of user-added extensions to the table\\nseems really cool.\\n\\n> And how would tab completion be handled in psql?\\n\\nYou look at the table to see what can come after SHOW. We already have\\ndatabase-driven completion, so this doesn't seem out of reach.\\n\\nOne thing that is doable with psql's current hard-wired approach, but\\ndoesn't seem easy to do with this solution, is automatic localization\\nof strings such as column headings. Rod had looked at that a little\\nin his trial patch to convert psql's \\\\d stuff to table-driven form,\\nbut AFAIR he didn't have a satisfactory answer.\\n\\n\\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Sun, 01 Jun 2003 12:03:08 -0400\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Table-driven SHOW (was Re: Are we losing momentum?)\"\n },\n {\n \"msg_contents\": \"Tom Lane writes:\\n\\n> Now, not only can we emulate \\\"show tables\\\", but people can easily add\\n> application-specific \\\"show whatever\\\" commands, which seems tremendously\\n> cool.\\n\\nWe already have that. They're called views.\\n\\n-- \\nPeter Eisentraut peter_e@gmx.net\\n\\n\",\n \"msg_date\": \"Sun, 1 Jun 2003 23:52:45 +0200 (CEST)\",\n \"msg_from\": \"Peter Eisentraut \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Are we losing momentum? \"\n },\n {\n \"msg_contents\": \"> > Now, not only can we emulate \\\"show tables\\\", but people can easily\\n> > add application-specific \\\"show whatever\\\" commands, which seems\\n> > tremendously cool.\\n> \\n> We already have that. They're called views.\\n\\nUm, I'm interested in aiding in the conversion of users from MySQL to\\nPostgreSQL (how ever it happens, I don't really care). Tom and Rod\\nare interested in an extensible/programmable SHOW syntax (very cool\\nand much more interesting than the hack I put together). Views solve\\nnone of the above problems. -sc\\n\\n-- \\nSean Chittenden\\n\",\n \"msg_date\": \"Sun, 1 Jun 2003 14:57:27 -0700\",\n \"msg_from\": \"Sean Chittenden \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Are we losing momentum?\"\n },\n {\n \"msg_contents\": \"> One thing that is doable with psql's current hard-wired approach, but\\n> doesn't seem easy to do with this solution, is automatic localization\\n> of strings such as column headings. Rod had looked at that a little\\n> in his trial patch to convert psql's \\\\d stuff to table-driven form,\\n> but AFAIR he didn't have a satisfactory answer.\\n\\nI've yet to come up with anything better. Best answer I've come up with\\nis to make the column headings a separate query result set and let the\\nback-end do the translations.\\n\\n-- \\nRod Taylor \\n\\nPGP Key: http://www.rbt.ca/rbtpub.asc\",\n \"msg_date\": \"01 Jun 2003 20:02:45 -0400\",\n \"msg_from\": \"Rod Taylor \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Table-driven SHOW (was Re: Are we losing momentum?)\"\n },\n {\n \"msg_contents\": \"Sean Chittenden writes:\\n\\n> Um, I'm interested in aiding in the conversion of users from MySQL to\\n> PostgreSQL (how ever it happens, I don't really care).\\n\\nYour approach to that reminds me of those\\n\\nalias dir='ls'\\nalias md='mkdir'\\n\\nthings that Linux distributors once stuck (or still stick?) in the default\\nprofile files, presumably to help conversion from DOS. It's pretty\\npointless, because Linux is still very different from DOS, and once you\\nwant to do something besides showing or changing directories, you will\\nneed documentation and training. That is what the MySQL conversion\\nprocess needs as well, otherwise you're not converting, you're emulating,\\nand that is not a game you can win.\\n\\n-- \\nPeter Eisentraut peter_e@gmx.net\\n\\n\",\n \"msg_date\": \"Mon, 2 Jun 2003 21:07:11 +0200 (CEST)\",\n \"msg_from\": \"Peter Eisentraut \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Are we losing momentum?\"\n },\n {\n \"msg_contents\": \"> > Um, I'm interested in aiding in the conversion of users from MySQL\\n> > to PostgreSQL (how ever it happens, I don't really care).\\n> \\n> Your approach to that reminds me of those\\n> \\n> alias dir='ls'\\n> alias md='mkdir'\\n> \\n> things that Linux distributors once stuck (or still stick?) in the\\n> default profile files, presumably to help conversion from DOS. It's\\n> pretty pointless, because Linux is still very different from DOS,\\n> and once you want to do something besides showing or changing\\n> directories, you will need documentation and training.\\n\\nWell, interestingly enough, those commands work for getting people in\\nthe door and to the point that they're able to learn more. The first\\nstep to any kind of adult education or reeducation is to have concepts\\nthat the people are familiar with (in this case MySQL) be translated\\ninto the concepts of the area that they're trying to learn. If you'd\\nhave read the original patch that I'd posted, you'd see that I'd done\\nthat by adding a TIP section to the top of the response.\\n\\n*SNIP*\\nSHOW COLUMNS FROM [tblname];\\n\\n TIP: In psql, \\\"SHOW COLUMNS FROM [tblname]\\\" is natively written as \\\\dt [tblname]\\n\\n[normal output from \\\\dt tblname]\\n*END SNIP*\\n\\nThe point of my patch was to aid the conversion, not to gimp along\\nb0rk3d habits from MySQL.\\n\\n> That is what the MySQL conversion process needs as well, otherwise\\n> you're not converting, you're emulating, and that is not a game you\\n> can win.\\n\\nEmulation within reason. dir->ls and md->mkdir worked for a handful\\nof people that I've transitioned into the UNIX world from Win32 land,\\nin fact, I have one friend from school who's been so successful that\\nhe's converted from using Win32 on his desktop to using Linux, worked\\nwith me on a job where we were hacking mod_perl on a site pushing in\\nexcess of 80Mbps to 25M people a day, but still types dir to this day.\\nNot bad for an aero student who graduated with a 4.0 in his major,\\nexceedingly bright, adaptive, learns fast, etc. My point is that\\nregardless of how bright the person or what the right invocation, aids\\nlike these help get people in the door and if they like what they see\\nonce they're through the door, they'll stay. Once people try and use\\nPostgreSQL, they stay. When people try MySQL, they're left wanting or\\nneeding more and are bound by the limits of the software... that's not\\nreally the case with PostgreSQL.\\n\\nTo get people to try, you play the association or emulation game and\\nit works. Ask anyone in adult education and they will say the same.\\nThe Internet was an \\\"information super highway\\\" because that was\\nsomething that people could grasp, regardless of how flawed it really\\nis. Bandwidth is thought of as pipes in various sizes diameters, a\\nmuch better analogy. To geeks, broadcast is explained as the same as\\nradio and unicast as satellite TV. To communication majors, TCP is\\ndescried as a letter that's been chopped up into a thousand numbered\\npieces and sent to the other side of the US via the postal mail.\\nLeche is milk to English speakers learning Spanish. One way or\\nanother, you have to play the game of working within the understanding\\nof the target audience, in this case, MySQL users who use a SHOW\\nTABLES type syntax.\\n\\n-sc\\n\\n-- \\nSean Chittenden\\n\",\n \"msg_date\": \"Mon, 2 Jun 2003 12:37:55 -0700\",\n \"msg_from\": \"Sean Chittenden \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Are we losing momentum?\"\n },\n {\n \"msg_contents\": \"\\nWow, this is a good argument! I must admit I made 'ps -ef' work on BSD\\nbecause I was just so used to it on mainframe Unix. There is that\\n'fingers type without thinking' thing, and I think that is what he is\\ntalking about.\\n\\nI wonder if we should just support SHOW TABLES or the most common ones.\\n\\nMaybe emulation is the wrong approach --- maybe we just need 'finger\\nthinking' shortcuts.\\n\\n---------------------------------------------------------------------------\\n\\nSean Chittenden wrote:\\n> > > Um, I'm interested in aiding in the conversion of users from MySQL\\n> > > to PostgreSQL (how ever it happens, I don't really care).\\n> > \\n> > Your approach to that reminds me of those\\n> > \\n> > alias dir='ls'\\n> > alias md='mkdir'\\n> > \\n> > things that Linux distributors once stuck (or still stick?) in the\\n> > default profile files, presumably to help conversion from DOS. It's\\n> > pretty pointless, because Linux is still very different from DOS,\\n> > and once you want to do something besides showing or changing\\n> > directories, you will need documentation and training.\\n> \\n> Well, interestingly enough, those commands work for getting people in\\n> the door and to the point that they're able to learn more. The first\\n> step to any kind of adult education or reeducation is to have concepts\\n> that the people are familiar with (in this case MySQL) be translated\\n> into the concepts of the area that they're trying to learn. If you'd\\n> have read the original patch that I'd posted, you'd see that I'd done\\n> that by adding a TIP section to the top of the response.\\n> \\n> *SNIP*\\n> SHOW COLUMNS FROM [tblname];\\n> \\n> TIP: In psql, \\\"SHOW COLUMNS FROM [tblname]\\\" is natively written as \\\\dt [tblname]\\n> \\n> [normal output from \\\\dt tblname]\\n> *END SNIP*\\n> \\n> The point of my patch was to aid the conversion, not to gimp along\\n> b0rk3d habits from MySQL.\\n> \\n> > That is what the MySQL conversion process needs as well, otherwise\\n> > you're not converting, you're emulating, and that is not a game you\\n> > can win.\\n> \\n> Emulation within reason. dir->ls and md->mkdir worked for a handful\\n> of people that I've transitioned into the UNIX world from Win32 land,\\n> in fact, I have one friend from school who's been so successful that\\n> he's converted from using Win32 on his desktop to using Linux, worked\\n> with me on a job where we were hacking mod_perl on a site pushing in\\n> excess of 80Mbps to 25M people a day, but still types dir to this day.\\n> Not bad for an aero student who graduated with a 4.0 in his major,\\n> exceedingly bright, adaptive, learns fast, etc. My point is that\\n> regardless of how bright the person or what the right invocation, aids\\n> like these help get people in the door and if they like what they see\\n> once they're through the door, they'll stay. Once people try and use\\n> PostgreSQL, they stay. When people try MySQL, they're left wanting or\\n> needing more and are bound by the limits of the software... that's not\\n> really the case with PostgreSQL.\\n> \\n> To get people to try, you play the association or emulation game and\\n> it works. Ask anyone in adult education and they will say the same.\\n> The Internet was an \\\"information super highway\\\" because that was\\n> something that people could grasp, regardless of how flawed it really\\n> is. Bandwidth is thought of as pipes in various sizes diameters, a\\n> much better analogy. To geeks, broadcast is explained as the same as\\n> radio and unicast as satellite TV. To communication majors, TCP is\\n> descried as a letter that's been chopped up into a thousand numbered\\n> pieces and sent to the other side of the US via the postal mail.\\n> Leche is milk to English speakers learning Spanish. One way or\\n> another, you have to play the game of working within the understanding\\n> of the target audience, in this case, MySQL users who use a SHOW\\n> TABLES type syntax.\\n> \\n> -sc\\n> \\n> -- \\n> Sean Chittenden\\n> \\n> ---------------------------(end of broadcast)---------------------------\\n> TIP 2: you can get off all lists at once with the unregister command\\n> (send \\\"unregister YourEmailAddressHere\\\" to majordomo@postgresql.org)\\n> \\n\\n-- \\n Bruce Momjian | http://candle.pha.pa.us\\n pgman@candle.pha.pa.us | (610) 359-1001\\n + If your life is a hard drive, | 13 Roberts Road\\n + Christ can be your backup. | Newtown Square, Pennsylvania 19073\\n\",\n \"msg_date\": \"Mon, 2 Jun 2003 15:53:10 -0400 (EDT)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Are we losing momentum?\"\n },\n {\n \"msg_contents\": \"\\nI assume we agreed against adding a MySQL mode --- just verifying.\\n\\n---------------------------------------------------------------------------\\n\\nSean Chittenden wrote:\\n> > > That's a pretty reasonable thought. I work for a shop that sells\\n> > > Postgres support, and even we install MySQL for the Q&D ticket\\n> > > tracking system we recommend because we can't justify the cost to\\n> > > port it to postgres. If the postgres support were there, we would\\n> > > surely be using it.\\n> > \\n> > > How to fix such a situation, I'm not sure. \\\"MySQL Compatability\\n> > > Mode,\\\" anyone? :-)\\n> > \\n> > What issues are creating a compatibility problem for you?\\n> \\n> I don't think these should be hacked into the backend/libpq, but I\\n> think it'd be a huge win to hack in \\\"show *\\\" support into psql for\\n> MySQL users so they can type:\\n> \\n> SHOW (databases|tables|views|functions|triggers|schemas);\\n> \\n> I have yet to meet a MySQL user who understands the concept of system\\n> catalogs even though it's just the 'mysql' database (this irritates me\\n> enough as is)... gah, f- it: mysql users be damned, I have three\\n> developers that think that postgresql is too hard to use because they\\n> can't remember \\\"\\\\d [table name]\\\" and I'm tired of hearing them bitch\\n> when I push using PostgreSQL instead of MySQL. I have better things\\n> to do with my time than convert their output to PostgreSQL. Here goes\\n> nothing...\\n> \\n> I've tainted psql and added a MySQL command compatibility layer for\\n> the family of SHOW commands (psql [-m | --mysql]).\\n> \\n> \\n> The attached patch does a few things:\\n> \\n> 1) Implements quite a number of SHOW commands (AGGREGATES, CASTS,\\n> CATALOGS, COLUMNS, COMMENTS, CONSTRAINTS, CONVERSIONS, DATABASES,\\n> DOMAINS, FUNCTIONS, HELP, INDEX, LARGEOBJECTS, NAMES, OPERATORS,\\n> PRIVILEGES, PROCESSLIST, SCHEMAS, SEQUENCES, SESSION, STATUS,\\n> TABLES, TRANSACTION, TYPES, USERS, VARIABLES, VIEWS)\\n> \\n> SHOW thing\\n> SHOW thing LIKE pattern\\n> SHOW thing FROM pattern\\n> SHOW HELP ON (topic || ALL);\\n> etc.\\n> \\n> Some of these don't have \\\\ command eqiv's. :( I was tempted to add\\n> them, but opted not to for now, but it'd certainly be a nice to\\n> have.\\n> \\n> 2) Implements the necessary tab completion for the SHOW commands for\\n> the tab happy newbies/folks out there. psql is more friendly than\\n> mysql's CLI now in terms of tab completion for the show commands.\\n> \\n> 3) Few trailing whitespace characters were nuked\\n> \\n> 4) guc.c is now in sync with the list of available variables used for\\n> tab completion\\n> \\n> \\n> Few things to note:\\n> \\n> 1) SHOW INDEXES is the same as SHOW INDEX, I think MySQL is wrong in\\n> this regard and that it should be INDEXES to be plural along with\\n> the rest of the types, but INDEX is preserved for compatibility.\\n> \\n> 2) There are two bugs that I have yet to address\\n> \\n> 1) SHOW VARIABLES doesn't work, but \\\"SHOW [TAB][TAB]y\\\" does\\n> 2) \\\"SHOW [variable_of_choice];\\\" doesn't work, but \\\"SHOW\\n> [variable_of_choice]\\\\n;\\\" does work... not sure where this\\n> problem is coming from\\n> \\n> 3) I think psql is more usable as a result of this more verbose\\n> syntax, but it's not the prettiest thing on the planet (wrote a\\n> small parser outside of the backend or libraries: I don't want to\\n> get those dirty with MySQL's filth).\\n> \\n> 4) In an attempt to wean people over to PostgreSQL's syntax, I\\n> included translation tips on how to use the psql equiv of the SHOW\\n> commands. Going from SHOW foo to \\\\d foo is easy, going from \\\\d foo\\n> to SHOW foo is hard and drives me nuts. This'll help userbase\\n> retention of newbies/converts. :)\\n> \\n> 5) The MySQL mode is just a bounce layer that provides different\\n> syntax wrapping exec_command() so it should provide little in the\\n> way of maintenance headaches. Some of the SHOW commands, however,\\n> don't have \\\\ couterparts, but once they do and that code is\\n> centralized, this feature should come for zero cost.\\n> \\n> 6) As an administrator, I'd be interested in having an environment\\n> variable that I could set that'd turn on MySQL mode for some of my\\n> bozo users that way they don't complain if they forget the -m\\n> switch. Thoughts?\\n> \\n> \\n> I'll try and iron out the last of those two bugs/features, but at this\\n> point, would like to see this patch get wider testing/feedback.\\n> Comments, as always, are welcome.\\n> \\n> PostgreSQL_usability++\\n> \\n> -sc\\n> \\n> -- \\n> Sean Chittenden\\n\\n[ Attachment, skipping... ]\\n\\n> \\n> ---------------------------(end of broadcast)---------------------------\\n> TIP 3: if posting/reading through Usenet, please send an appropriate\\n> subscribe-nomail command to majordomo@postgresql.org so that your\\n> message can get through to the mailing list cleanly\\n\\n-- \\n Bruce Momjian | http://candle.pha.pa.us\\n pgman@candle.pha.pa.us | (610) 359-1001\\n + If your life is a hard drive, | 13 Roberts Road\\n + Christ can be your backup. | Newtown Square, Pennsylvania 19073\\n\",\n \"msg_date\": \"Sat, 16 Aug 2003 19:22:47 -0400 (EDT)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Are we losing momentum?\"\n },\n {\n \"msg_contents\": \"Bruce Momjian wrote:\\n\\n> I assume we agreed against adding a MySQL mode --- just verifying.\\n\\nWe agreed that applications that need schema information are much better \\noff using the schema views.\\n\\n\\nJan\\n\\n> \\n> ---------------------------------------------------------------------------\\n> \\n> Sean Chittenden wrote:\\n>> > > That's a pretty reasonable thought. I work for a shop that sells\\n>> > > Postgres support, and even we install MySQL for the Q&D ticket\\n>> > > tracking system we recommend because we can't justify the cost to\\n>> > > port it to postgres. If the postgres support were there, we would\\n>> > > surely be using it.\\n>> > \\n>> > > How to fix such a situation, I'm not sure. \\\"MySQL Compatability\\n>> > > Mode,\\\" anyone? :-)\\n>> > \\n>> > What issues are creating a compatibility problem for you?\\n>> \\n>> I don't think these should be hacked into the backend/libpq, but I\\n>> think it'd be a huge win to hack in \\\"show *\\\" support into psql for\\n>> MySQL users so they can type:\\n>> \\n>> SHOW (databases|tables|views|functions|triggers|schemas);\\n>> \\n>> I have yet to meet a MySQL user who understands the concept of system\\n>> catalogs even though it's just the 'mysql' database (this irritates me\\n>> enough as is)... gah, f- it: mysql users be damned, I have three\\n>> developers that think that postgresql is too hard to use because they\\n>> can't remember \\\"\\\\d [table name]\\\" and I'm tired of hearing them bitch\\n>> when I push using PostgreSQL instead of MySQL. I have better things\\n>> to do with my time than convert their output to PostgreSQL. Here goes\\n>> nothing...\\n>> \\n>> I've tainted psql and added a MySQL command compatibility layer for\\n>> the family of SHOW commands (psql [-m | --mysql]).\\n>> \\n>> \\n>> The attached patch does a few things:\\n>> \\n>> 1) Implements quite a number of SHOW commands (AGGREGATES, CASTS,\\n>> CATALOGS, COLUMNS, COMMENTS, CONSTRAINTS, CONVERSIONS, DATABASES,\\n>> DOMAINS, FUNCTIONS, HELP, INDEX, LARGEOBJECTS, NAMES, OPERATORS,\\n>> PRIVILEGES, PROCESSLIST, SCHEMAS, SEQUENCES, SESSION, STATUS,\\n>> TABLES, TRANSACTION, TYPES, USERS, VARIABLES, VIEWS)\\n>> \\n>> SHOW thing\\n>> SHOW thing LIKE pattern\\n>> SHOW thing FROM pattern\\n>> SHOW HELP ON (topic || ALL);\\n>> etc.\\n>> \\n>> Some of these don't have \\\\ command eqiv's. :( I was tempted to add\\n>> them, but opted not to for now, but it'd certainly be a nice to\\n>> have.\\n>> \\n>> 2) Implements the necessary tab completion for the SHOW commands for\\n>> the tab happy newbies/folks out there. psql is more friendly than\\n>> mysql's CLI now in terms of tab completion for the show commands.\\n>> \\n>> 3) Few trailing whitespace characters were nuked\\n>> \\n>> 4) guc.c is now in sync with the list of available variables used for\\n>> tab completion\\n>> \\n>> \\n>> Few things to note:\\n>> \\n>> 1) SHOW INDEXES is the same as SHOW INDEX, I think MySQL is wrong in\\n>> this regard and that it should be INDEXES to be plural along with\\n>> the rest of the types, but INDEX is preserved for compatibility.\\n>> \\n>> 2) There are two bugs that I have yet to address\\n>> \\n>> 1) SHOW VARIABLES doesn't work, but \\\"SHOW [TAB][TAB]y\\\" does\\n>> 2) \\\"SHOW [variable_of_choice];\\\" doesn't work, but \\\"SHOW\\n>> [variable_of_choice]\\\\n;\\\" does work... not sure where this\\n>> problem is coming from\\n>> \\n>> 3) I think psql is more usable as a result of this more verbose\\n>> syntax, but it's not the prettiest thing on the planet (wrote a\\n>> small parser outside of the backend or libraries: I don't want to\\n>> get those dirty with MySQL's filth).\\n>> \\n>> 4) In an attempt to wean people over to PostgreSQL's syntax, I\\n>> included translation tips on how to use the psql equiv of the SHOW\\n>> commands. Going from SHOW foo to \\\\d foo is easy, going from \\\\d foo\\n>> to SHOW foo is hard and drives me nuts. This'll help userbase\\n>> retention of newbies/converts. :)\\n>> \\n>> 5) The MySQL mode is just a bounce layer that provides different\\n>> syntax wrapping exec_command() so it should provide little in the\\n>> way of maintenance headaches. Some of the SHOW commands, however,\\n>> don't have \\\\ couterparts, but once they do and that code is\\n>> centralized, this feature should come for zero cost.\\n>> \\n>> 6) As an administrator, I'd be interested in having an environment\\n>> variable that I could set that'd turn on MySQL mode for some of my\\n>> bozo users that way they don't complain if they forget the -m\\n>> switch. Thoughts?\\n>> \\n>> \\n>> I'll try and iron out the last of those two bugs/features, but at this\\n>> point, would like to see this patch get wider testing/feedback.\\n>> Comments, as always, are welcome.\\n>> \\n>> PostgreSQL_usability++\\n>> \\n>> -sc\\n>> \\n>> -- \\n>> Sean Chittenden\\n> \\n> [ Attachment, skipping... ]\\n> \\n>> \\n>> ---------------------------(end of broadcast)---------------------------\\n>> TIP 3: if posting/reading through Usenet, please send an appropriate\\n>> subscribe-nomail command to majordomo@postgresql.org so that your\\n>> message can get through to the mailing list cleanly\\n> \\n\\n\\n-- \\n#======================================================================#\\n# It's easier to get forgiveness for being wrong than for being right. #\\n# Let's break this rule - forgive me. #\\n#================================================== JanWieck@Yahoo.com #\\n\\n\",\n \"msg_date\": \"Sat, 16 Aug 2003 22:41:39 -0400\",\n \"msg_from\": \"Jan Wieck \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Are we losing momentum?\"\n },\n {\n \"msg_contents\": \"\\nPersonally, I think adding the SHOW commands would be a good thing ...\\npsql is nice with its \\\\df to get information without having to learn all\\nthe JOINs required ... having that ability easily from any of the\\ninterfaces would definitely be a plus ... to me, its not about MySQL\\ncompatibility, but about a small improvement to ease of use :)\\n\\nOn Sat, 16 Aug 2003, Bruce Momjian wrote:\\n\\n>\\n> I assume we agreed against adding a MySQL mode --- just verifying.\\n>\\n> ---------------------------------------------------------------------------\\n>\\n> Sean Chittenden wrote:\\n> > > > That's a pretty reasonable thought. I work for a shop that sells\\n> > > > Postgres support, and even we install MySQL for the Q&D ticket\\n> > > > tracking system we recommend because we can't justify the cost to\\n> > > > port it to postgres. If the postgres support were there, we would\\n> > > > surely be using it.\\n> > >\\n> > > > How to fix such a situation, I'm not sure. \\\"MySQL Compatability\\n> > > > Mode,\\\" anyone? :-)\\n> > >\\n> > > What issues are creating a compatibility problem for you?\\n> >\\n> > I don't think these should be hacked into the backend/libpq, but I\\n> > think it'd be a huge win to hack in \\\"show *\\\" support into psql for\\n> > MySQL users so they can type:\\n> >\\n> > SHOW (databases|tables|views|functions|triggers|schemas);\\n> >\\n> > I have yet to meet a MySQL user who understands the concept of system\\n> > catalogs even though it's just the 'mysql' database (this irritates me\\n> > enough as is)... gah, f- it: mysql users be damned, I have three\\n> > developers that think that postgresql is too hard to use because they\\n> > can't remember \\\"\\\\d [table name]\\\" and I'm tired of hearing them bitch\\n> > when I push using PostgreSQL instead of MySQL. I have better things\\n> > to do with my time than convert their output to PostgreSQL. Here goes\\n> > nothing...\\n> >\\n> > I've tainted psql and added a MySQL command compatibility layer for\\n> > the family of SHOW commands (psql [-m | --mysql]).\\n> >\\n> >\\n> > The attached patch does a few things:\\n> >\\n> > 1) Implements quite a number of SHOW commands (AGGREGATES, CASTS,\\n> > CATALOGS, COLUMNS, COMMENTS, CONSTRAINTS, CONVERSIONS, DATABASES,\\n> > DOMAINS, FUNCTIONS, HELP, INDEX, LARGEOBJECTS, NAMES, OPERATORS,\\n> > PRIVILEGES, PROCESSLIST, SCHEMAS, SEQUENCES, SESSION, STATUS,\\n> > TABLES, TRANSACTION, TYPES, USERS, VARIABLES, VIEWS)\\n> >\\n> > SHOW thing\\n> > SHOW thing LIKE pattern\\n> > SHOW thing FROM pattern\\n> > SHOW HELP ON (topic || ALL);\\n> > etc.\\n> >\\n> > Some of these don't have \\\\ command eqiv's. :( I was tempted to add\\n> > them, but opted not to for now, but it'd certainly be a nice to\\n> > have.\\n> >\\n> > 2) Implements the necessary tab completion for the SHOW commands for\\n> > the tab happy newbies/folks out there. psql is more friendly than\\n> > mysql's CLI now in terms of tab completion for the show commands.\\n> >\\n> > 3) Few trailing whitespace characters were nuked\\n> >\\n> > 4) guc.c is now in sync with the list of available variables used for\\n> > tab completion\\n> >\\n> >\\n> > Few things to note:\\n> >\\n> > 1) SHOW INDEXES is the same as SHOW INDEX, I think MySQL is wrong in\\n> > this regard and that it should be INDEXES to be plural along with\\n> > the rest of the types, but INDEX is preserved for compatibility.\\n> >\\n> > 2) There are two bugs that I have yet to address\\n> >\\n> > 1) SHOW VARIABLES doesn't work, but \\\"SHOW [TAB][TAB]y\\\" does\\n> > 2) \\\"SHOW [variable_of_choice];\\\" doesn't work, but \\\"SHOW\\n> > [variable_of_choice]\\\\n;\\\" does work... not sure where this\\n> > problem is coming from\\n> >\\n> > 3) I think psql is more usable as a result of this more verbose\\n> > syntax, but it's not the prettiest thing on the planet (wrote a\\n> > small parser outside of the backend or libraries: I don't want to\\n> > get those dirty with MySQL's filth).\\n> >\\n> > 4) In an attempt to wean people over to PostgreSQL's syntax, I\\n> > included translation tips on how to use the psql equiv of the SHOW\\n> > commands. Going from SHOW foo to \\\\d foo is easy, going from \\\\d foo\\n> > to SHOW foo is hard and drives me nuts. This'll help userbase\\n> > retention of newbies/converts. :)\\n> >\\n> > 5) The MySQL mode is just a bounce layer that provides different\\n> > syntax wrapping exec_command() so it should provide little in the\\n> > way of maintenance headaches. Some of the SHOW commands, however,\\n> > don't have \\\\ couterparts, but once they do and that code is\\n> > centralized, this feature should come for zero cost.\\n> >\\n> > 6) As an administrator, I'd be interested in having an environment\\n> > variable that I could set that'd turn on MySQL mode for some of my\\n> > bozo users that way they don't complain if they forget the -m\\n> > switch. Thoughts?\\n> >\\n> >\\n> > I'll try and iron out the last of those two bugs/features, but at this\\n> > point, would like to see this patch get wider testing/feedback.\\n> > Comments, as always, are welcome.\\n> >\\n> > PostgreSQL_usability++\\n> >\\n> > -sc\\n> >\\n> > --\\n> > Sean Chittenden\\n>\\n> [ Attachment, skipping... ]\\n>\\n> >\\n> > ---------------------------(end of broadcast)---------------------------\\n> > TIP 3: if posting/reading through Usenet, please send an appropriate\\n> > subscribe-nomail command to majordomo@postgresql.org so that your\\n> > message can get through to the mailing list cleanly\\n>\\n> --\\n> Bruce Momjian | http://candle.pha.pa.us\\n> pgman@candle.pha.pa.us | (610) 359-1001\\n> + If your life is a hard drive, | 13 Roberts Road\\n> + Christ can be your backup. | Newtown Square, Pennsylvania 19073\\n>\\n> ---------------------------(end of broadcast)---------------------------\\n> TIP 1: subscribe and unsubscribe commands go to majordomo@postgresql.org\\n>\\n\\nMarc G. Fournier ICQ#7615664 IRC Nick: Scrappy\\nSystems Administrator @ hub.org\\nprimary: scrappy@hub.org secondary: scrappy@{freebsd|postgresql}.org\\n\",\n \"msg_date\": \"Sat, 16 Aug 2003 23:48:59 -0300 (ADT)\",\n \"msg_from\": \"The Hermit Hacker \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Are we losing momentum?\"\n },\n {\n \"msg_contents\": \"> Bruce Momjian wrote:\\n> >I assume we agreed against adding a MySQL mode --- just verifying.\\n> \\n> We agreed that applications that need schema information are much better \\n> off using the schema views.\\n> \\n> Jan\\n\\nHeh, I don't think there was any agreement on anything in that thread,\\neveryone had their own view (no pun intended).\\n\\n> From: The Hermit Hacker \\n>\\n> Personally, I think adding the SHOW commands would be a good thing\\n> ... psql is nice with its \\\\df to get information without having to\\n> learn all the JOINs required ... having that ability easily from any\\n> of the interfaces would definitely be a plus ... to me, its not\\n> about MySQL compatibility, but about a small improvement to ease of\\n> use :)\\n\\nWhich goes back to the point about there being little agreement on\\nthis patch or its issues. A handful of folks think it's a _user\\ninterface_ issue (read: psql, phppgadmin, pgadminIII, etc) and would\\nbe good for converting MySQL users to PostgreSQL (or simply because\\nits easy and less obtuse than a \\\\ command), others thought it was a\\nfugly hack to have a parser in the front end and that it should be\\nhandled on the backend by extending SQL to conform to MySQL's\\ninterface (that some argue is incorrect and would unjustly bloat the\\nbackend) that way all clients have the SHOW syntax (thus averting a\\npossible FAQ), and others took a more elitist mindset and simply\\nthought that everyone should just select from the information schemas.\\n\\n*shrug* I tabled working on the patch until there was some kind of\\nagreement from someone with commit privs and am waiting to pick up\\nquashing the remaining parser state bug until after 7.4's out the door\\nor there's renewed interest from non-users.\\n\\n-sc\\n\\n-- \\nSean Chittenden\\n\",\n \"msg_date\": \"Sat, 16 Aug 2003 22:48:16 -0700\",\n \"msg_from\": \"Sean Chittenden \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Are we losing momentum?\"\n },\n {\n \"msg_contents\": \"Short summary:\\n\\n This patch allows ISO 8601 \\\"time intervals\\\" using the \\\"format \\n with time-unit designators\\\" to specify postgresql \\\"intervals\\\".\\n\\n Below I have (A) What these time intervals are, (B) What I\\n modified to support them, (C) Issues with intervals I want\\n to bring up, and (D) a patch supporting them.\\n\\n It's helpful to me. Any feedback is appreciated. If you \\n did want to consider including it, let me know what to clean \\n up. If not, I thought I'd just put it here if anyone else finds\\n it useful too.\\n\\n Thanks for your time,\\n \\n Ron Mayer\\n\\nLonger:\\n\\n(A) What these intervals are.\\n\\n ISO 8601, the standard from which PostgreSQL gets some of it's \\n time syntax, also has a specification for \\\"time-intervals\\\".\\n \\n In particular, section 5.5.4.2 has a \\\"Representation of\\n time-interval by duration only\\\" which I believe maps\\n nicely to ISO intervals.\\n\\n Compared to the ISO 8601 time interval specification, the\\n postgresql interval syntax is quite verbose. For example:\\n\\n Postgresql interval: ISO8601 Interval\\n ---------------------------------------------------\\n '1 year 6 months' 'P1Y6M'\\n '3 hours 25 minutes 42 seconds' 'PT3H25M42S'\\n\\n Yeah, it's uglier, but it sure is short which can make\\n for quicker typing and shorter scripts, and if for some\\n strange reason you had an application using this format\\n it's nice not to have to translate.\\n\\n The syntax is as follows:\\n Basic extended format: PnYnMnDTnHnMnS\\n PnW\\n\\n Where everything before the \\\"T\\\" is a date-part and everything\\n after is a time-part. W is for weeks.\\n In the date-part, Y=Year, M=Month, D=Day\\n In the time-part, H=Hour, M=Minute, S=Second\\n\\n Much more info can be found from the draft standard\\n ftp://ftp.qsl.net/pub/g1smd/154N362_.PDF\\n The final standard's only available for $$$ so I didn't\\n look at it. Some other sites imply that this part didn't\\n change from the last draft to the standard.\\n\\n\\n(B) This change was made by adding two functions to \\\"datetime.c\\\"\\n next to where DecodeInterval parses the normal interval syntax.\\n\\n A total of 313 lines were added, including comments and sgml docs.\\n Of these only 136 are actual code, the rest, comments, whitespace, etc.\\n\\n\\n One new function \\\"DecodeISO8601Interval\\\" follows the style of\\n \\\"DecodeInterval\\\" below it, and trys to strictly follow the ISO\\n syntax. If it doesn't match, it'll return -1 and the old syntax\\n will be checked as before.\\n\\n The first test (first character of the first field must be 'P', \\n and second character must be 'T' or '\\\\0') should be fast so I don't\\n think this will impact performance of existing code.\\n\\n\\n The second function (\\\"adjust_fval\\\") is just a small helper-function\\n to remove some of the cut&paste style that DecodeInterval used.\\n\\n It seems to work.\\n =======================================================================\\n betadb=# select 'P1M15DT12H30M7S'::interval;\\n interval \\n ------------------------\\n 1 mon 15 days 12:30:07\\n (1 row)\\n\\n betadb=# select '1 month 15 days 12 hours 30 minutes 7 seconds'::interval;\\n\\t interval \\n ------------------------\\n 1 mon 15 days 12:30:07\\n (1 row)\\n =====================================================================\\n\\n\\n\\n(C) Open issues with intervals, and questions I'd like to ask.\\n\\n 1. DecodeInterval seems to have a hardcoded '.' for specifying\\n fractional times. ISO 8601 states that both '.' and ',' are\\n ok, but \\\"of these, the comma is the preferred sign\\\".\\n\\n In DecodeISO8601Interval I loosened the test to allow\\n both but left it as it was in DecodeInterval. Should\\n both be changed to make them more consistant?\\n\\n 2. In \\\"DecodeInterval\\\", fractional weeks and fractional months\\n can produce seconds; but fractional years can not (rounded\\n to months). I didn't understand the reasoning for this, so\\n I left it the same, and followed the same convention for\\n ISO intervals. Should I change this?\\n\\n 3. I could save a bunch of copy-paste-lines-of-code from the\\n pre-existing DecodeInterval by calling the adjust_fval helper\\n function. The tradeoff is a few extra function-calls when\\n decoding an interval. However I didn't want to risk changes\\n to the existing part unless you guys encourage me to do so.\\n\\n\\n(D) The patch.\\n\\n\\nIndex: doc/src/sgml/datatype.sgml\\n===================================================================\\nRCS file: /projects/cvsroot/pgsql-server/doc/src/sgml/datatype.sgml,v\\nretrieving revision 1.123\\ndiff -u -1 -0 -r1.123 datatype.sgml\\n--- doc/src/sgml/datatype.sgml\\t31 Aug 2003 17:32:18 -0000\\t1.123\\n+++ doc/src/sgml/datatype.sgml\\t8 Sep 2003 04:04:58 -0000\\n@@ -1735,20 +1735,71 @@\\n Quantities of days, hours, minutes, and seconds can be specified without\\n explicit unit markings. For example, '1 12:59:10' is read\\n the same as '1 day 12 hours 59 min 10 sec'.\\n \\n \\n \\n The optional precision\\n p should be between 0 and 6, and\\n defaults to the precision of the input literal.\\n \\n+\\n+\\n+ \\n+ Alternatively, interval values can be written as \\n+ ISO 8601 time intervals, using the \\\"Format with time-unit designators\\\".\\n+ This format always starts with the character 'P', followed \\n+ by a string of values followed by single character time-unit designators.\\n+ A 'T' separates the date and time parts of the interval.\\n+ \\n+\\n+ \\n+ Format: PnYnMnDTnHnMnS\\n+ \\n+ \\n+ In this format, 'n' gets replaced by a number, and \\n+ Y represents years, \\n+ M (in the date part) months,\\n+ D months,\\n+ H hours,\\n+ M (in the time part) minutes,\\n+ and S seconds.\\n+ \\n+ \\n+\\n+ \\n+\\t Interval Example\\n+\\t \\n+\\t\\t\\n+\\t\\t \\n+\\t\\t Traditional\\n+\\t\\t ISO-8601 time-interval\\n+\\t\\t
\\n+\\t\\t\\n+\\t\\t\\n+\\t\\t \\n+\\t\\t 1 month\\n+\\t\\t P1M\\n+\\t\\t
\\n+\\t\\t \\n+\\t\\t 1 hour 30 minutes\\n+\\t\\t PT1H30M\\n+\\t\\t
\\n+\\t\\t \\n+\\t\\t 2 years 10 months 15 days 10 hours 30 minutes 20 seconds\\n+\\t\\t P2Y10M15DT10H30M20S\\n+\\t\\t
\\n+\\t\\t\\n+\\t \\n+\\t
\\n+\\t \\n+ \\n \\n \\n \\n Special Values\\n \\n \\n time\\n constants\\n \\n \\nIndex: src/backend/utils/adt/datetime.c\\n===================================================================\\nRCS file: /projects/cvsroot/pgsql-server/src/backend/utils/adt/datetime.c,v\\nretrieving revision 1.116\\ndiff -u -1 -0 -r1.116 datetime.c\\n--- src/backend/utils/adt/datetime.c\\t27 Aug 2003 23:29:28 -0000\\t1.116\\n+++ src/backend/utils/adt/datetime.c\\t8 Sep 2003 04:04:59 -0000\\n@@ -30,20 +30,21 @@\\n \\t\\t\\t struct tm * tm, fsec_t *fsec, int *is2digits);\\n static int DecodeNumberField(int len, char *str,\\n \\t\\t\\t\\t int fmask, int *tmask,\\n \\t\\t\\t\\t struct tm * tm, fsec_t *fsec, int *is2digits);\\n static int DecodeTime(char *str, int fmask, int *tmask,\\n \\t\\t struct tm * tm, fsec_t *fsec);\\n static int\\tDecodeTimezone(char *str, int *tzp);\\n static datetkn *datebsearch(char *key, datetkn *base, unsigned int nel);\\n static int\\tDecodeDate(char *str, int fmask, int *tmask, struct tm * tm);\\n static void TrimTrailingZeros(char *str);\\n+static int DecodeISO8601Interval(char **field, int *ftype, int nf, int *dtype, struct tm * tm, fsec_t *fsec);\\n \\n \\n int\\t\\t\\tday_tab[2][13] = {\\n \\t{31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31, 0},\\n {31, 29, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31, 0}};\\n \\n char\\t *months[] = {\\\"Jan\\\", \\\"Feb\\\", \\\"Mar\\\", \\\"Apr\\\", \\\"May\\\", \\\"Jun\\\",\\n \\\"Jul\\\", \\\"Aug\\\", \\\"Sep\\\", \\\"Oct\\\", \\\"Nov\\\", \\\"Dec\\\", NULL};\\n \\n char\\t *days[] = {\\\"Sunday\\\", \\\"Monday\\\", \\\"Tuesday\\\", \\\"Wednesday\\\",\\n@@ -2872,30 +2873,271 @@\\n \\t\\t\\tdefault:\\n \\t\\t\\t\\t*val = tp->value;\\n \\t\\t\\t\\tbreak;\\n \\t\\t}\\n \\t}\\n \\n \\treturn type;\\n }\\n \\n \\n+void adjust_fval(double fval,struct tm * tm, fsec_t *fsec, int scale);\\n+{\\n+\\tint\\tsec;\\n+\\tfval\\t *= scale;\\n+\\tsec\\t\\t = fval;\\n+\\ttm->tm_sec += sec;\\n+#ifdef HAVE_INT64_TIMESTAMP\\n+\\t*fsec\\t += ((fval - sec) * 1000000);\\n+#else\\n+\\t*fsec\\t += (fval - sec);\\n+#endif\\n+}\\n+\\n+\\n+/* DecodeISO8601Interval()\\n+ *\\n+ * Check if it's a ISO 8601 Section 5.5.4.2 \\\"Representation of\\n+ * time-interval by duration only.\\\" \\n+ * Basic extended format: PnYnMnDTnHnMnS\\n+ * PnW\\n+ * For more info.\\n+ * http://www.astroclark.freeserve.co.uk/iso8601/index.html\\n+ * ftp://ftp.qsl.net/pub/g1smd/154N362_.PDF\\n+ *\\n+ * Examples: P1D for 1 day\\n+ * PT1H for 1 hour\\n+ * P2Y6M7DT1H30M for 2 years, 6 months, 7 days 1 hour 30 min\\n+ *\\n+ * The first field is exactly \\\"p\\\" or \\\"pt\\\" it may be of this type.\\n+ *\\n+ * Returns -1 if the field is not of this type.\\n+ *\\n+ * It pretty strictly checks the spec, with the two exceptions\\n+ * that a week field ('W') may coexist with other units, and that\\n+ * this function allows decimals in fields other than the least\\n+ * significant units.\\n+ */\\n+int\\n+DecodeISO8601Interval(char **field, int *ftype, int nf, int *dtype, struct tm * tm, fsec_t *fsec) \\n+{\\n+\\tchar\\t *cp;\\n+\\tint\\t\\t\\tfmask = 0,\\n+\\t\\t\\t\\ttmask;\\n+\\tint\\t\\t\\tval;\\n+\\tdouble\\t\\tfval;\\n+\\tint\\t\\t\\targ;\\n+\\tint\\t\\t\\tdatepart;\\n+\\n+ /*\\n+\\t * An ISO 8601 \\\"time-interval by duration only\\\" must start\\n+\\t * with a 'P'. If it contains a date-part, 'p' will be the\\n+\\t * only character in the field. If it contains no date part\\n+\\t * it will contain exactly to characters 'PT' indicating a\\n+\\t * time part.\\n+\\t * Anything else is illegal and will be treated like a \\n+\\t * traditional postgresql interval.\\n+\\t */\\n+ if (!(field[0][0] == 'p' &&\\n+ ((field[0][1] == 0) || (field[0][1] == 't' && field[0][2] == 0))))\\n+\\t{\\n+\\t return -1;\\n+\\t}\\n+\\n+\\n+ /*\\n+\\t * If the first field is exactly 1 character ('P'), it starts\\n+\\t * with date elements. Otherwise it's two characters ('PT');\\n+\\t * indicating it starts with a time part.\\n+\\t */\\n+\\tdatepart = (field[0][1] == 0);\\n+\\n+\\t/*\\n+\\t * Every value must have a unit, so we require an even\\n+\\t * number of value/unit pairs. Therefore we require an\\n+\\t * odd nubmer of fields, including the prefix 'P'.\\n+\\t */\\n+\\tif ((nf & 1) == 0)\\n+\\t\\treturn -1;\\n+\\n+\\t/*\\n+\\t * Process pairs of fields at a time.\\n+\\t */\\n+\\tfor (arg = 1 ; arg < nf ; arg+=2) \\n+\\t{\\n+\\t\\tchar * value = field[arg ];\\n+\\t\\tchar * units = field[arg+1];\\n+\\n+\\t\\t/*\\n+\\t\\t * The value part must be a number.\\n+\\t\\t */\\n+\\t\\tif (ftype[arg] != DTK_NUMBER) \\n+\\t\\t\\treturn -1;\\n+\\n+\\t\\t/*\\n+\\t\\t * extract the number, almost exactly like the non-ISO interval.\\n+\\t\\t */\\n+\\t\\tval = strtol(value, &cp, 10);\\n+\\n+\\t\\t/*\\n+\\t\\t * One difference from the normal postgresql interval below...\\n+\\t\\t * ISO 8601 states that \\\"Of these, the comma is the preferred \\n+\\t\\t * sign\\\" so I allow it here for locales that support it.\\n+\\t\\t * Note: Perhaps the old-style interval code below should\\n+\\t\\t * allow for this too, but I didn't want to risk backward\\n+\\t\\t * compatability.\\n+\\t\\t */\\n+\\t\\tif (*cp == '.' || *cp == ',') \\n+\\t\\t{\\n+\\t\\t\\tfval = strtod(cp, &cp);\\n+\\t\\t\\tif (*cp != '\\\\0')\\n+\\t\\t\\t\\treturn -1;\\n+\\n+\\t\\t\\tif (val < 0)\\n+\\t\\t\\t\\tfval = -(fval);\\n+\\t\\t}\\n+\\t\\telse if (*cp == '\\\\0')\\n+\\t\\t\\tfval = 0;\\n+\\t\\telse\\n+\\t\\t\\treturn -1;\\n+\\n+\\n+\\t\\tif (datepart)\\n+\\t\\t{\\n+\\t\\t\\t/*\\n+\\t\\t\\t * All the 8601 unit specifiers are 1 character, but may\\n+\\t\\t\\t * be followed by a 'T' character if transitioning between\\n+\\t\\t\\t * the date part and the time part. If it's not either\\n+\\t\\t\\t * one character or two characters with the second being 't'\\n+\\t\\t\\t * it's an error.\\n+\\t\\t\\t */\\n+\\t\\t\\tif (!(units[1] == 0 || (units[1] == 't' && units[2] == 0)))\\n+\\t\\t\\t\\treturn -1;\\n+\\n+\\t\\t\\tif (units[1] == 't')\\n+\\t\\t\\t\\tdatepart = 0;\\n+\\n+\\t\\t\\tswitch (units[0]) /* Y M D W */\\n+\\t\\t\\t{\\n+\\t\\t\\t\\tcase 'd':\\n+\\t\\t\\t\\t\\ttm->tm_mday += val;\\n+\\t\\t\\t\\t\\tif (fval != 0)\\n+\\t\\t\\t\\t\\t adjust_fval(fval,tm,fsec, 86400);\\n+\\t\\t\\t\\t\\ttmask = ((fmask & DTK_M(DAY)) ? 0 : DTK_M(DAY));\\n+\\t\\t\\t\\t\\tbreak;\\n+\\n+\\t\\t\\t\\tcase 'w':\\n+\\t\\t\\t\\t\\ttm->tm_mday += val * 7;\\n+\\t\\t\\t\\t\\tif (fval != 0)\\n+\\t\\t\\t\\t\\t adjust_fval(fval,tm,fsec,7 * 86400);\\n+\\t\\t\\t\\t\\ttmask = ((fmask & DTK_M(DAY)) ? 0 : DTK_M(DAY));\\n+\\t\\t\\t\\t\\tbreak;\\n+\\n+\\t\\t\\t\\tcase 'm':\\n+\\t\\t\\t\\t\\ttm->tm_mon += val;\\n+\\t\\t\\t\\t\\tif (fval != 0)\\n+\\t\\t\\t\\t\\t adjust_fval(fval,tm,fsec,30 * 86400);\\n+\\t\\t\\t\\t\\ttmask = DTK_M(MONTH);\\n+\\t\\t\\t\\t\\tbreak;\\n+\\n+\\t\\t\\t\\tcase 'y':\\n+\\t\\t\\t\\t\\t/*\\n+\\t\\t\\t\\t\\t * Why can fractional months produce seconds,\\n+\\t\\t\\t\\t\\t * but fractional years can't? Well the older\\n+\\t\\t\\t\\t\\t * interval code below has the same property\\n+\\t\\t\\t\\t\\t * so this one follows the other one too.\\n+\\t\\t\\t\\t\\t */\\n+\\t\\t\\t\\t\\ttm->tm_year += val;\\n+\\t\\t\\t\\t\\tif (fval != 0)\\n+\\t\\t\\t\\t\\t\\ttm->tm_mon += (fval * 12);\\n+\\t\\t\\t\\t\\ttmask = ((fmask & DTK_M(YEAR)) ? 0 : DTK_M(YEAR));\\n+\\t\\t\\t\\t\\tbreak;\\n+\\n+\\t\\t\\t\\tdefault:\\n+\\t\\t\\t\\t\\treturn -1; /* invald date unit prefix */\\n+\\t\\t\\t}\\n+\\t\\t}\\n+\\t\\telse\\n+\\t\\t{\\n+\\t\\t\\t/*\\n+\\t\\t\\t * ISO 8601 time part.\\n+\\t\\t\\t * In the time part, only one-character\\n+\\t\\t\\t * unit prefixes are allowed. If it's more\\n+\\t\\t\\t * than one character, it's not a valid ISO 8601\\n+\\t\\t\\t * time interval by duration.\\n+\\t\\t\\t */\\n+\\t\\t\\tif (units[1] != 0)\\n+\\t\\t\\t\\treturn -1;\\n+\\n+\\t\\t\\tswitch (units[0]) /* H M S */\\n+\\t\\t\\t{\\n+\\t\\t\\t\\tcase 's':\\n+\\t\\t\\t\\t\\ttm->tm_sec += val;\\n+#ifdef HAVE_INT64_TIMESTAMP\\n+\\t\\t\\t\\t\\t*fsec += (fval * 1000000);\\n+#else\\n+\\t\\t\\t\\t\\t*fsec += fval;\\n+#endif\\n+\\t\\t\\t\\t\\ttmask = DTK_M(SECOND);\\n+\\t\\t\\t\\t\\tbreak;\\n+\\n+\\t\\t\\t\\tcase 'm':\\n+\\t\\t\\t\\t\\ttm->tm_min += val;\\n+\\t\\t\\t\\t\\tif (fval != 0)\\n+\\t\\t\\t\\t\\t adjust_fval(fval,tm,fsec,60);\\n+\\t\\t\\t\\t\\ttmask = DTK_M(MINUTE);\\n+\\t\\t\\t\\t\\tbreak;\\n+\\n+\\t\\t\\t\\tcase 'h':\\n+\\t\\t\\t\\t\\ttm->tm_hour += val;\\n+\\t\\t\\t\\t\\tif (fval != 0)\\n+\\t\\t\\t\\t\\t adjust_fval(fval,tm,fsec,3600);\\n+\\t\\t\\t\\t\\ttmask = DTK_M(HOUR);\\n+\\t\\t\\t\\t\\tbreak;\\n+\\n+\\t\\t\\t\\tdefault:\\n+\\t\\t\\t\\t\\treturn -1; /* invald time unit prefix */\\n+\\t\\t\\t}\\n+\\t\\t}\\n+\\t\\tfmask |= tmask;\\n+\\t}\\n+\\n+\\tif (*fsec != 0)\\n+\\t{\\n+\\t\\tint\\t\\t\\tsec;\\n+\\n+#ifdef HAVE_INT64_TIMESTAMP\\n+\\t\\tsec = (*fsec / INT64CONST(1000000));\\n+\\t\\t*fsec -= (sec * INT64CONST(1000000));\\n+#else\\n+\\t\\tTMODULO(*fsec, sec, 1e0);\\n+#endif\\n+\\t\\ttm->tm_sec += sec;\\n+\\t}\\n+\\treturn (fmask != 0) ? 0 : -1;\\n+}\\n+\\n+\\n /* DecodeInterval()\\n * Interpret previously parsed fields for general time interval.\\n * Returns 0 if successful, DTERR code if bogus input detected.\\n *\\n * Allow \\\"date\\\" field DTK_DATE since this could be just\\n *\\tan unsigned floating point number. - thomas 1997-11-16\\n *\\n * Allow ISO-style time span, with implicit units on number of days\\n *\\tpreceding an hh:mm:ss field. - thomas 1998-04-30\\n+ * \\n+ * Allow ISO-8601 style \\\"Representation of time-interval by duration only\\\"\\n+ * of the format 'PnYnMnDTnHnMnS' and 'PnW' - ron 2003-08-30\\n */\\n+\\n int\\n DecodeInterval(char **field, int *ftype, int nf, int *dtype, struct tm * tm, fsec_t *fsec)\\n {\\n \\tint\\t\\t\\tis_before = FALSE;\\n \\tchar\\t *cp;\\n \\tint\\t\\t\\tfmask = 0,\\n \\t\\t\\t\\ttmask,\\n \\t\\t\\t\\ttype;\\n \\tint\\t\\t\\ti;\\n \\tint\\t\\t\\tdterr;\\n@@ -2906,20 +3148,37 @@\\n \\n \\ttype = IGNORE_DTF;\\n \\ttm->tm_year = 0;\\n \\ttm->tm_mon = 0;\\n \\ttm->tm_mday = 0;\\n \\ttm->tm_hour = 0;\\n \\ttm->tm_min = 0;\\n \\ttm->tm_sec = 0;\\n \\t*fsec = 0;\\n \\n+\\t/*\\n+\\t * Check if it's a ISO 8601 Section 5.5.4.2 \\\"Representation of\\n+ * time-interval by duration only.\\\" \\n+\\t * Basic extended format: PnYnMnDTnHnMnS\\n+\\t * PnW\\n+\\t * http://www.astroclark.freeserve.co.uk/iso8601/index.html\\n+\\t * ftp://ftp.qsl.net/pub/g1smd/154N362_.PDF\\n+\\t * Examples: P1D for 1 day\\n+\\t * PT1H for 1 hour\\n+\\t * P2Y6M7DT1H30M for 2 years, 6 months, 7 days 1 hour 30 min\\n+\\t *\\n+\\t * The first field is exactly \\\"p\\\" or \\\"pt\\\" it may be of this type.\\n+\\t */\\n+\\tif (DecodeISO8601Interval(field,ftype,nf,dtype,tm,fsec) == 0) {\\n+\\t return 0;\\n+ }\\n+\\n \\t/* read through list backwards to pick up units before values */\\n \\tfor (i = nf - 1; i >= 0; i--)\\n \\t{\\n \\t\\tswitch (ftype[i])\\n \\t\\t{\\n \\t\\t\\tcase DTK_TIME:\\n \\t\\t\\t\\tdterr = DecodeTime(field[i], fmask, &tmask, tm, fsec);\\n \\t\\t\\t\\tif (dterr)\\n \\t\\t\\t\\t\\treturn dterr;\\n \\t\\t\\t\\ttype = DTK_DAY;\\n@@ -2983,20 +3242,21 @@\\n \\t\\t\\t\\t}\\n \\t\\t\\t\\t/* DROP THROUGH */\\n \\n \\t\\t\\tcase DTK_DATE:\\n \\t\\t\\tcase DTK_NUMBER:\\n \\t\\t\\t\\tval = strtol(field[i], &cp, 10);\\n \\n \\t\\t\\t\\tif (type == IGNORE_DTF)\\n \\t\\t\\t\\t\\ttype = DTK_SECOND;\\n \\n+\\t\\t\\t\\t/* should this allow ',' for locales that use it ? */\\n \\t\\t\\t\\tif (*cp == '.')\\n \\t\\t\\t\\t{\\n \\t\\t\\t\\t\\tfval = strtod(cp, &cp);\\n \\t\\t\\t\\t\\tif (*cp != '\\\\0')\\n \\t\\t\\t\\t\\t\\treturn DTERR_BAD_FORMAT;\\n \\n \\t\\t\\t\\t\\tif (val < 0)\\n \\t\\t\\t\\t\\t\\tfval = -(fval);\\n \\t\\t\\t\\t}\\n \\t\\t\\t\\telse if (*cp == '\\\\0')\\n\\n===================================================================\\n\",\n \"msg_date\": \"Sun, 7 Sep 2003 21:50:49 -0700\",\n \"msg_from\": \"\\\"Ron Mayer\\\" \",\n \"msg_from_op\": false,\n \"msg_subject\": \"ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit deignators\\\"\"\n },\n {\n \"msg_contents\": \"\\\"Ron Mayer\\\" writes:\\n> Compared to the ISO 8601 time interval specification, the\\n> postgresql interval syntax is quite verbose. For example:\\n\\n> Postgresql interval: ISO8601 Interval\\n> ---------------------------------------------------\\n> '1 year 6 months' 'P1Y6M'\\n> '3 hours 25 minutes 42 seconds' 'PT3H25M42S'\\n\\nEr, don't we support that already? I know I saw code to support\\nsomething much like that syntax last time I looked into the datetime\\nroutines.\\n\\n\\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Mon, 08 Sep 2003 01:47:17 -0400\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\\n\\tdeignators\\\"\"\n },\n {\n \"msg_contents\": \"\\nIs there a way of producing as well as reading this format? Or did I miss\\nsomething?\\n\\ncheers\\n\\nandrew\\n\\nRon Mayer said:\\n> Short summary:\\n>\\n> This patch allows ISO 8601 \\\"time intervals\\\" using the \\\"format\\n> with time-unit designators\\\" to specify postgresql \\\"intervals\\\".\\n>\\n> Below I have (A) What these time intervals are, (B) What I\\n> modified to support them, (C) Issues with intervals I want\\n> to bring up, and (D) a patch supporting them.\\n>\\n> It's helpful to me. Any feedback is appreciated. If you\\n> did want to consider including it, let me know what to clean\\n> up. If not, I thought I'd just put it here if anyone else finds it\\n> useful too.\\n>\\n> Thanks for your time,\\n>\\n> Ron Mayer\\n>\\n> Longer:\\n>\\n> (A) What these intervals are.\\n>\\n> ISO 8601, the standard from which PostgreSQL gets some of it's time\\n> syntax, also has a specification for \\\"time-intervals\\\".\\n>\\n> In particular, section 5.5.4.2 has a \\\"Representation of\\n> time-interval by duration only\\\" which I believe maps\\n> nicely to ISO intervals.\\n>\\n> Compared to the ISO 8601 time interval specification, the\\n> postgresql interval syntax is quite verbose. For example:\\n>\\n> Postgresql interval: ISO8601 Interval\\n> ---------------------------------------------------\\n> '1 year 6 months' 'P1Y6M'\\n> '3 hours 25 minutes 42 seconds' 'PT3H25M42S'\\n>\\n> Yeah, it's uglier, but it sure is short which can make\\n> for quicker typing and shorter scripts, and if for some\\n> strange reason you had an application using this format\\n> it's nice not to have to translate.\\n>\\n> The syntax is as follows:\\n> Basic extended format: PnYnMnDTnHnMnS\\n> PnW\\n>\\n> Where everything before the \\\"T\\\" is a date-part and everything\\n> after is a time-part. W is for weeks.\\n> In the date-part, Y=Year, M=Month, D=Day\\n> In the time-part, H=Hour, M=Minute, S=Second\\n>\\n> Much more info can be found from the draft standard\\n> ftp://ftp.qsl.net/pub/g1smd/154N362_.PDF\\n> The final standard's only available for $$$ so I didn't\\n> look at it. Some other sites imply that this part didn't\\n> change from the last draft to the standard.\\n>\\n>\\n> (B) This change was made by adding two functions to \\\"datetime.c\\\"\\n> next to where DecodeInterval parses the normal interval syntax.\\n>\\n> A total of 313 lines were added, including comments and sgml docs.\\n> Of these only 136 are actual code, the rest, comments, whitespace,\\n> etc.\\n>\\n>\\n> One new function \\\"DecodeISO8601Interval\\\" follows the style of\\n> \\\"DecodeInterval\\\" below it, and trys to strictly follow the ISO\\n> syntax. If it doesn't match, it'll return -1 and the old syntax\\n> will be checked as before.\\n>\\n> The first test (first character of the first field must be 'P', and\\n> second character must be 'T' or '\\\\0') should be fast so I don't\\n> think this will impact performance of existing code.\\n>\\n>\\n> The second function (\\\"adjust_fval\\\") is just a small helper-function\\n> to remove some of the cut&paste style that DecodeInterval used.\\n>\\n> It seems to work.\\n>\\n=======================================================================\\n> betadb=# select 'P1M15DT12H30M7S'::interval;\\n> interval\\n> ------------------------\\n> 1 mon 15 days 12:30:07\\n> (1 row)\\n>\\n> betadb=# select '1 month 15 days 12 hours 30 minutes 7\\n> seconds'::interval;\\n> \\t interval\\n> ------------------------\\n> 1 mon 15 days 12:30:07\\n> (1 row)\\n> =====================================================================\\n>\\n>\\n>\\n> (C) Open issues with intervals, and questions I'd like to ask.\\n>\\n> 1. DecodeInterval seems to have a hardcoded '.' for specifying\\n> fractional times. ISO 8601 states that both '.' and ',' are ok,\\n> but \\\"of these, the comma is the preferred sign\\\".\\n>\\n> In DecodeISO8601Interval I loosened the test to allow\\n> both but left it as it was in DecodeInterval. Should\\n> both be changed to make them more consistant?\\n>\\n> 2. In \\\"DecodeInterval\\\", fractional weeks and fractional months\\n> can produce seconds; but fractional years can not (rounded to\\n> months). I didn't understand the reasoning for this, so I left\\n> it the same, and followed the same convention for\\n> ISO intervals. Should I change this?\\n>\\n> 3. I could save a bunch of copy-paste-lines-of-code from the\\n> pre-existing DecodeInterval by calling the adjust_fval helper\\n> function. The tradeoff is a few extra function-calls when\\n> decoding an interval. However I didn't want to risk changes to\\n> the existing part unless you guys encourage me to do so.\\n>\\n>\\n> (D) The patch.\\n>\\n>\\n[snip]\\n\\n\\n\",\n \"msg_date\": \"Mon, 8 Sep 2003 02:32:17 -0400 (EDT)\",\n \"msg_from\": \"\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 'Time Intervals' of the 'format with time-unit\\n\\tdeignators'\"\n },\n {\n \"msg_contents\": \"\\nTom wrote:\\n> \\\"Ron Mayer\\\" writes:\\n> > Compared to the ISO 8601 time interval specification, the\\n> > postgresql interval syntax is quite verbose. For example:\\n> \\n> > Postgresql interval: ISO8601 Interval\\n> > ---------------------------------------------------\\n> > '1 year 6 months' 'P1Y6M'\\n> > '3 hours 25 minutes 42 seconds' 'PT3H25M42S'\\n> \\n> Er, don't we support that already? I know I saw code to support\\n> something much like that syntax last time I looked into the datetime\\n> routines.\\n> \\n\\nNope.\\n\\nPostgresql supports a rather bizzare shorthand that has a similar\\nsyntax, but AFAICT, doesn't match ISO 8601 in any way that makes \\nit practical.\\n\\n\\nA disclaimer, I have the \\\"Final Draft\\\" (ISO/TC 154N 362 \\nof 2000-12-19) of the spec; but have not seen the official,\\nexpensive, version. \\nftp://ftp.qsl.net/pub/g1smd/154N362_.PDF\\n\\n\\nFor example, if I read it right, I have differences\\nlike this:\\n\\n Interval ISO Postgres\\n 8601 shorthand\\n -----------------------------------------------------\\n '1 year 1 minute' 'P1YT1M' '1Y1M'\\n '1 year 1 month' 'P1Y1M' N/A\\n\\nThe best part about the postgresql syntax is that\\nthey omit the required 'P', so it's easy to differentiate\\nbetween the two. :-)\\n\\n\\nPerhaps one could argue that the postgres shorthand should \\nfollow the ISO conventions, but I'd not want to break backward\\ncompatability, incase someone out there is using '1H30M' and\\nexpecting minutes instead of months. If we didn't want to\\nsupport two syntaxes, I wouldn't mind eventually depricating\\nthe less-standard one.\\n\\n Ron\\n\\n\\n\",\n \"msg_date\": \"Mon, 8 Sep 2003 11:59:50 -0700\",\n \"msg_from\": \"\\\"Ron Mayer\\\" \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\\n\\tdeignators\\\"\"\n },\n {\n \"msg_contents\": \"andrew@dunslane.net wrote:\\n> \\n> Is there a way of producing as well as reading this format? Or did I miss\\n> something?\\n\\nNot yet, but I'd be happy to add it. \\n\\nMy immediate problem was having some 'P1Y6M' intervals to load. \\nI posted this much largely because it was useful to me so might \\nhelp others, and to see if it was of interest to others and \\nget feedback on what else to change.\\n\\nI'd be happy to make it produce the output, and have some style\\nquestions for doing so.\\n\\nI'd hate to trigger this output on the already-existing 'datestyle' \\nof 'ISO', since that would break backward compatability.\\nI do notice that 8601 has both \\\"basic\\\" and \\\"extended\\\" formats.\\nThe \\\"basic format\\\" is more terse ('19980115' instead of '1998-01-15').\\n\\nWould it be useful if I added a 'datestyle' of 'ISO basic' which\\nwould produce the most terse formats ('19980115' for dates, \\nand 'P1Y1M' for intervals)?\\n\\n Ron\\n\\nPS: What's the best inexpenive way for me to know if this changed \\nat all between the final draft and the published standard? \\n\\n> Ron Mayer said:\\n> > This patch allows ISO 8601 \\\"time intervals\\\" using the \\\"format\\n> > with time-unit designators\\\" to specify postgresql \\\"intervals\\\".\\n> >...\\n> > Much more info can be found from the draft standard\\n> > ftp://ftp.qsl.net/pub/g1smd/154N362_.PDF\\n> > The final standard's only available for $$$ so I didn't\\n> > look at it.\\n\\n\",\n \"msg_date\": \"Mon, 8 Sep 2003 12:16:50 -0700\",\n \"msg_from\": \"\\\"Ron Mayer\\\" \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 'Time Intervals' of the 'format with time-unit\\n\\tdeignators'\"\n },\n {\n \"msg_contents\": \"\\\"Ron Mayer\\\" writes:\\n> Tom wrote:\\n>> Er, don't we support that already?\\n\\n> Postgresql supports a rather bizzare shorthand that has a similar\\n> syntax, but AFAICT, doesn't match ISO 8601 in any way that makes \\n> it practical.\\n\\nWell, it's *supposed* to match ISO, AFAICT (the comments in the code\\ntalk about \\\"ISO dates\\\"). Unless ISO has put out multiple specs that\\ncover this?\\n\\n> Perhaps one could argue that the postgres shorthand should \\n> follow the ISO conventions, but I'd not want to break backward\\n> compatability, incase someone out there is using '1H30M' and\\n> expecting minutes instead of months.\\n\\nI doubt anyone is using it, because it's completely undocumented.\\nIf we're going to support the real ISO spec, I'd suggest ripping\\nout any not-quite-there variant. (Especially so noting that your\\ncode seems a lot cleaner than the ptype stuff.)\\n\\nThe datetime code is kind of a mess right now, because Thomas Lockhart\\nwalked away from the project while only partway through some significant\\nadditions. He left some incomplete features and quite a number of bugs\\nin new-and-untested code. We've been gradually cleaning up the problems,\\nbut if if you find something that doesn't seem to make sense, it's\\nlikely a bug rather than anything we want to preserve. In particular,\\ngiven the knowledge that it doesn't meet the ISO spec, I'd judge that\\nthe existing code for the ISO shorthand was a work-in-progress.\\n\\n\\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Mon, 08 Sep 2003 15:59:20 -0400\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\\n\\tdeignators\\\"\"\n },\n {\n \"msg_contents\": \"\\\"Ron Mayer\\\" writes:\\n> Would it be useful if I added a 'datestyle' of 'ISO basic' which\\n> would produce the most terse formats ('19980115' for dates, \\n> and 'P1Y1M' for intervals)?\\n\\nI don't really care for using that name for it --- for one thing, you\\ncouldn't do\\n\\tset datestyle to iso basic;\\nbecause of syntax limitations. A one-word name is a much better idea.\\n\\nPerhaps call it \\\"compact\\\" or \\\"terse\\\" datestyle?\\n\\n\\n> PS: What's the best inexpenive way for me to know if this changed \\n> at all between the final draft and the published standard? \\n\\nANSI sells PDFs of ISO specs at their online store\\nhttp://webstore.ansi.org/ansidocstore/default.asp\\nalthough it looks like they want $81 for 8601, which is not my idea of\\n\\\"inexpensive\\\".\\n\\nUsually ISO final drafts differ very little from the published specs;\\nI think you could just work from the draft and no one would complain.\\n\\n\\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Mon, 08 Sep 2003 16:17:19 -0400\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 'Time Intervals' of the 'format with time-unit\\n\\tdeignators'\"\n },\n {\n \"msg_contents\": \"Tom Lane wrote:\\n> > Perhaps one could argue that the postgres shorthand should \\n> > follow the ISO conventions, but I'd not want to break backward\\n> > compatability, incase someone out there is using '1H30M' and\\n> > expecting minutes instead of months.\\n> \\n> I doubt anyone is using it, because it's completely undocumented.\\n> If we're going to support the real ISO spec, I'd suggest ripping\\n> out any not-quite-there variant. (Especially so noting that your\\n> code seems a lot cleaner than the ptype stuff.)\\n\\nAgreed. Let me put your code in the queue for 7.5.\\n\\n-- \\n Bruce Momjian | http://candle.pha.pa.us\\n pgman@candle.pha.pa.us | (610) 359-1001\\n + If your life is a hard drive, | 13 Roberts Road\\n + Christ can be your backup. | Newtown Square, Pennsylvania 19073\\n\",\n \"msg_date\": \"Mon, 8 Sep 2003 16:27:16 -0400 (EDT)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\"\n },\n {\n \"msg_contents\": \"\\nThis has been saved for the 7.5 release:\\n\\n\\thttp:/momjian.postgresql.org/cgi-bin/pgpatches2\\n\\nFeel free to submit an updated patch that rips out the old syntax, as\\ndiscussed, or replace this patch with a more comprehensive one.\\n\\n---------------------------------------------------------------------------\\n\\nRon Mayer wrote:\\n> Short summary:\\n> \\n> This patch allows ISO 8601 \\\"time intervals\\\" using the \\\"format \\n> with time-unit designators\\\" to specify postgresql \\\"intervals\\\".\\n> \\n> Below I have (A) What these time intervals are, (B) What I\\n> modified to support them, (C) Issues with intervals I want\\n> to bring up, and (D) a patch supporting them.\\n> \\n> It's helpful to me. Any feedback is appreciated. If you \\n> did want to consider including it, let me know what to clean \\n> up. If not, I thought I'd just put it here if anyone else finds\\n> it useful too.\\n> \\n> Thanks for your time,\\n> \\n> Ron Mayer\\n> \\n> Longer:\\n> \\n> (A) What these intervals are.\\n> \\n> ISO 8601, the standard from which PostgreSQL gets some of it's \\n> time syntax, also has a specification for \\\"time-intervals\\\".\\n> \\n> In particular, section 5.5.4.2 has a \\\"Representation of\\n> time-interval by duration only\\\" which I believe maps\\n> nicely to ISO intervals.\\n> \\n> Compared to the ISO 8601 time interval specification, the\\n> postgresql interval syntax is quite verbose. For example:\\n> \\n> Postgresql interval: ISO8601 Interval\\n> ---------------------------------------------------\\n> '1 year 6 months' 'P1Y6M'\\n> '3 hours 25 minutes 42 seconds' 'PT3H25M42S'\\n> \\n> Yeah, it's uglier, but it sure is short which can make\\n> for quicker typing and shorter scripts, and if for some\\n> strange reason you had an application using this format\\n> it's nice not to have to translate.\\n> \\n> The syntax is as follows:\\n> Basic extended format: PnYnMnDTnHnMnS\\n> PnW\\n> \\n> Where everything before the \\\"T\\\" is a date-part and everything\\n> after is a time-part. W is for weeks.\\n> In the date-part, Y=Year, M=Month, D=Day\\n> In the time-part, H=Hour, M=Minute, S=Second\\n> \\n> Much more info can be found from the draft standard\\n> ftp://ftp.qsl.net/pub/g1smd/154N362_.PDF\\n> The final standard's only available for $$$ so I didn't\\n> look at it. Some other sites imply that this part didn't\\n> change from the last draft to the standard.\\n> \\n> \\n> (B) This change was made by adding two functions to \\\"datetime.c\\\"\\n> next to where DecodeInterval parses the normal interval syntax.\\n> \\n> A total of 313 lines were added, including comments and sgml docs.\\n> Of these only 136 are actual code, the rest, comments, whitespace, etc.\\n> \\n> \\n> One new function \\\"DecodeISO8601Interval\\\" follows the style of\\n> \\\"DecodeInterval\\\" below it, and trys to strictly follow the ISO\\n> syntax. If it doesn't match, it'll return -1 and the old syntax\\n> will be checked as before.\\n> \\n> The first test (first character of the first field must be 'P', \\n> and second character must be 'T' or '\\\\0') should be fast so I don't\\n> think this will impact performance of existing code.\\n> \\n> \\n> The second function (\\\"adjust_fval\\\") is just a small helper-function\\n> to remove some of the cut&paste style that DecodeInterval used.\\n> \\n> It seems to work.\\n> =======================================================================\\n> betadb=# select 'P1M15DT12H30M7S'::interval;\\n> interval \\n> ------------------------\\n> 1 mon 15 days 12:30:07\\n> (1 row)\\n> \\n> betadb=# select '1 month 15 days 12 hours 30 minutes 7 seconds'::interval;\\n> \\t interval \\n> ------------------------\\n> 1 mon 15 days 12:30:07\\n> (1 row)\\n> =====================================================================\\n> \\n> \\n> \\n> (C) Open issues with intervals, and questions I'd like to ask.\\n> \\n> 1. DecodeInterval seems to have a hardcoded '.' for specifying\\n> fractional times. ISO 8601 states that both '.' and ',' are\\n> ok, but \\\"of these, the comma is the preferred sign\\\".\\n> \\n> In DecodeISO8601Interval I loosened the test to allow\\n> both but left it as it was in DecodeInterval. Should\\n> both be changed to make them more consistant?\\n> \\n> 2. In \\\"DecodeInterval\\\", fractional weeks and fractional months\\n> can produce seconds; but fractional years can not (rounded\\n> to months). I didn't understand the reasoning for this, so\\n> I left it the same, and followed the same convention for\\n> ISO intervals. Should I change this?\\n> \\n> 3. I could save a bunch of copy-paste-lines-of-code from the\\n> pre-existing DecodeInterval by calling the adjust_fval helper\\n> function. The tradeoff is a few extra function-calls when\\n> decoding an interval. However I didn't want to risk changes\\n> to the existing part unless you guys encourage me to do so.\\n> \\n> \\n> (D) The patch.\\n> \\n> \\n> Index: doc/src/sgml/datatype.sgml\\n> ===================================================================\\n> RCS file: /projects/cvsroot/pgsql-server/doc/src/sgml/datatype.sgml,v\\n> retrieving revision 1.123\\n> diff -u -1 -0 -r1.123 datatype.sgml\\n> --- doc/src/sgml/datatype.sgml\\t31 Aug 2003 17:32:18 -0000\\t1.123\\n> +++ doc/src/sgml/datatype.sgml\\t8 Sep 2003 04:04:58 -0000\\n> @@ -1735,20 +1735,71 @@\\n> Quantities of days, hours, minutes, and seconds can be specified without\\n> explicit unit markings. For example, '1 12:59:10' is read\\n> the same as '1 day 12 hours 59 min 10 sec'.\\n> \\n> \\n> \\n> The optional precision\\n> p should be between 0 and 6, and\\n> defaults to the precision of the input literal.\\n> \\n> +\\n> +\\n> + \\n> + Alternatively, interval values can be written as \\n> + ISO 8601 time intervals, using the \\\"Format with time-unit designators\\\".\\n> + This format always starts with the character 'P', followed \\n> + by a string of values followed by single character time-unit designators.\\n> + A 'T' separates the date and time parts of the interval.\\n> + \\n> +\\n> + \\n> + Format: PnYnMnDTnHnMnS\\n> + \\n> + \\n> + In this format, 'n' gets replaced by a number, and \\n> + Y represents years, \\n> + M (in the date part) months,\\n> + D months,\\n> + H hours,\\n> + M (in the time part) minutes,\\n> + and S seconds.\\n> + \\n> + \\n> +\\n> + \\n> +\\t Interval Example\\n> +\\t \\n> +\\t\\t\\n> +\\t\\t \\n> +\\t\\t Traditional\\n> +\\t\\t ISO-8601 time-interval\\n> +\\t\\t
\\n> +\\t\\t\\n> +\\t\\t\\n> +\\t\\t \\n> +\\t\\t 1 month\\n> +\\t\\t P1M\\n> +\\t\\t
\\n> +\\t\\t \\n> +\\t\\t 1 hour 30 minutes\\n> +\\t\\t PT1H30M\\n> +\\t\\t
\\n> +\\t\\t \\n> +\\t\\t 2 years 10 months 15 days 10 hours 30 minutes 20 seconds\\n> +\\t\\t P2Y10M15DT10H30M20S\\n> +\\t\\t
\\n> +\\t\\t\\n> +\\t \\n> +\\t
\\n> +\\t \\n> + \\n> \\n> \\n> \\n> Special Values\\n> \\n> \\n> time\\n> constants\\n> \\n> \\n> Index: src/backend/utils/adt/datetime.c\\n> ===================================================================\\n> RCS file: /projects/cvsroot/pgsql-server/src/backend/utils/adt/datetime.c,v\\n> retrieving revision 1.116\\n> diff -u -1 -0 -r1.116 datetime.c\\n> --- src/backend/utils/adt/datetime.c\\t27 Aug 2003 23:29:28 -0000\\t1.116\\n> +++ src/backend/utils/adt/datetime.c\\t8 Sep 2003 04:04:59 -0000\\n> @@ -30,20 +30,21 @@\\n> \\t\\t\\t struct tm * tm, fsec_t *fsec, int *is2digits);\\n> static int DecodeNumberField(int len, char *str,\\n> \\t\\t\\t\\t int fmask, int *tmask,\\n> \\t\\t\\t\\t struct tm * tm, fsec_t *fsec, int *is2digits);\\n> static int DecodeTime(char *str, int fmask, int *tmask,\\n> \\t\\t struct tm * tm, fsec_t *fsec);\\n> static int\\tDecodeTimezone(char *str, int *tzp);\\n> static datetkn *datebsearch(char *key, datetkn *base, unsigned int nel);\\n> static int\\tDecodeDate(char *str, int fmask, int *tmask, struct tm * tm);\\n> static void TrimTrailingZeros(char *str);\\n> +static int DecodeISO8601Interval(char **field, int *ftype, int nf, int *dtype, struct tm * tm, fsec_t *fsec);\\n> \\n> \\n> int\\t\\t\\tday_tab[2][13] = {\\n> \\t{31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31, 0},\\n> {31, 29, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31, 0}};\\n> \\n> char\\t *months[] = {\\\"Jan\\\", \\\"Feb\\\", \\\"Mar\\\", \\\"Apr\\\", \\\"May\\\", \\\"Jun\\\",\\n> \\\"Jul\\\", \\\"Aug\\\", \\\"Sep\\\", \\\"Oct\\\", \\\"Nov\\\", \\\"Dec\\\", NULL};\\n> \\n> char\\t *days[] = {\\\"Sunday\\\", \\\"Monday\\\", \\\"Tuesday\\\", \\\"Wednesday\\\",\\n> @@ -2872,30 +2873,271 @@\\n> \\t\\t\\tdefault:\\n> \\t\\t\\t\\t*val = tp->value;\\n> \\t\\t\\t\\tbreak;\\n> \\t\\t}\\n> \\t}\\n> \\n> \\treturn type;\\n> }\\n> \\n> \\n> +void adjust_fval(double fval,struct tm * tm, fsec_t *fsec, int scale);\\n> +{\\n> +\\tint\\tsec;\\n> +\\tfval\\t *= scale;\\n> +\\tsec\\t\\t = fval;\\n> +\\ttm->tm_sec += sec;\\n> +#ifdef HAVE_INT64_TIMESTAMP\\n> +\\t*fsec\\t += ((fval - sec) * 1000000);\\n> +#else\\n> +\\t*fsec\\t += (fval - sec);\\n> +#endif\\n> +}\\n> +\\n> +\\n> +/* DecodeISO8601Interval()\\n> + *\\n> + * Check if it's a ISO 8601 Section 5.5.4.2 \\\"Representation of\\n> + * time-interval by duration only.\\\" \\n> + * Basic extended format: PnYnMnDTnHnMnS\\n> + * PnW\\n> + * For more info.\\n> + * http://www.astroclark.freeserve.co.uk/iso8601/index.html\\n> + * ftp://ftp.qsl.net/pub/g1smd/154N362_.PDF\\n> + *\\n> + * Examples: P1D for 1 day\\n> + * PT1H for 1 hour\\n> + * P2Y6M7DT1H30M for 2 years, 6 months, 7 days 1 hour 30 min\\n> + *\\n> + * The first field is exactly \\\"p\\\" or \\\"pt\\\" it may be of this type.\\n> + *\\n> + * Returns -1 if the field is not of this type.\\n> + *\\n> + * It pretty strictly checks the spec, with the two exceptions\\n> + * that a week field ('W') may coexist with other units, and that\\n> + * this function allows decimals in fields other than the least\\n> + * significant units.\\n> + */\\n> +int\\n> +DecodeISO8601Interval(char **field, int *ftype, int nf, int *dtype, struct tm * tm, fsec_t *fsec) \\n> +{\\n> +\\tchar\\t *cp;\\n> +\\tint\\t\\t\\tfmask = 0,\\n> +\\t\\t\\t\\ttmask;\\n> +\\tint\\t\\t\\tval;\\n> +\\tdouble\\t\\tfval;\\n> +\\tint\\t\\t\\targ;\\n> +\\tint\\t\\t\\tdatepart;\\n> +\\n> + /*\\n> +\\t * An ISO 8601 \\\"time-interval by duration only\\\" must start\\n> +\\t * with a 'P'. If it contains a date-part, 'p' will be the\\n> +\\t * only character in the field. If it contains no date part\\n> +\\t * it will contain exactly to characters 'PT' indicating a\\n> +\\t * time part.\\n> +\\t * Anything else is illegal and will be treated like a \\n> +\\t * traditional postgresql interval.\\n> +\\t */\\n> + if (!(field[0][0] == 'p' &&\\n> + ((field[0][1] == 0) || (field[0][1] == 't' && field[0][2] == 0))))\\n> +\\t{\\n> +\\t return -1;\\n> +\\t}\\n> +\\n> +\\n> + /*\\n> +\\t * If the first field is exactly 1 character ('P'), it starts\\n> +\\t * with date elements. Otherwise it's two characters ('PT');\\n> +\\t * indicating it starts with a time part.\\n> +\\t */\\n> +\\tdatepart = (field[0][1] == 0);\\n> +\\n> +\\t/*\\n> +\\t * Every value must have a unit, so we require an even\\n> +\\t * number of value/unit pairs. Therefore we require an\\n> +\\t * odd nubmer of fields, including the prefix 'P'.\\n> +\\t */\\n> +\\tif ((nf & 1) == 0)\\n> +\\t\\treturn -1;\\n> +\\n> +\\t/*\\n> +\\t * Process pairs of fields at a time.\\n> +\\t */\\n> +\\tfor (arg = 1 ; arg < nf ; arg+=2) \\n> +\\t{\\n> +\\t\\tchar * value = field[arg ];\\n> +\\t\\tchar * units = field[arg+1];\\n> +\\n> +\\t\\t/*\\n> +\\t\\t * The value part must be a number.\\n> +\\t\\t */\\n> +\\t\\tif (ftype[arg] != DTK_NUMBER) \\n> +\\t\\t\\treturn -1;\\n> +\\n> +\\t\\t/*\\n> +\\t\\t * extract the number, almost exactly like the non-ISO interval.\\n> +\\t\\t */\\n> +\\t\\tval = strtol(value, &cp, 10);\\n> +\\n> +\\t\\t/*\\n> +\\t\\t * One difference from the normal postgresql interval below...\\n> +\\t\\t * ISO 8601 states that \\\"Of these, the comma is the preferred \\n> +\\t\\t * sign\\\" so I allow it here for locales that support it.\\n> +\\t\\t * Note: Perhaps the old-style interval code below should\\n> +\\t\\t * allow for this too, but I didn't want to risk backward\\n> +\\t\\t * compatability.\\n> +\\t\\t */\\n> +\\t\\tif (*cp == '.' || *cp == ',') \\n> +\\t\\t{\\n> +\\t\\t\\tfval = strtod(cp, &cp);\\n> +\\t\\t\\tif (*cp != '\\\\0')\\n> +\\t\\t\\t\\treturn -1;\\n> +\\n> +\\t\\t\\tif (val < 0)\\n> +\\t\\t\\t\\tfval = -(fval);\\n> +\\t\\t}\\n> +\\t\\telse if (*cp == '\\\\0')\\n> +\\t\\t\\tfval = 0;\\n> +\\t\\telse\\n> +\\t\\t\\treturn -1;\\n> +\\n> +\\n> +\\t\\tif (datepart)\\n> +\\t\\t{\\n> +\\t\\t\\t/*\\n> +\\t\\t\\t * All the 8601 unit specifiers are 1 character, but may\\n> +\\t\\t\\t * be followed by a 'T' character if transitioning between\\n> +\\t\\t\\t * the date part and the time part. If it's not either\\n> +\\t\\t\\t * one character or two characters with the second being 't'\\n> +\\t\\t\\t * it's an error.\\n> +\\t\\t\\t */\\n> +\\t\\t\\tif (!(units[1] == 0 || (units[1] == 't' && units[2] == 0)))\\n> +\\t\\t\\t\\treturn -1;\\n> +\\n> +\\t\\t\\tif (units[1] == 't')\\n> +\\t\\t\\t\\tdatepart = 0;\\n> +\\n> +\\t\\t\\tswitch (units[0]) /* Y M D W */\\n> +\\t\\t\\t{\\n> +\\t\\t\\t\\tcase 'd':\\n> +\\t\\t\\t\\t\\ttm->tm_mday += val;\\n> +\\t\\t\\t\\t\\tif (fval != 0)\\n> +\\t\\t\\t\\t\\t adjust_fval(fval,tm,fsec, 86400);\\n> +\\t\\t\\t\\t\\ttmask = ((fmask & DTK_M(DAY)) ? 0 : DTK_M(DAY));\\n> +\\t\\t\\t\\t\\tbreak;\\n> +\\n> +\\t\\t\\t\\tcase 'w':\\n> +\\t\\t\\t\\t\\ttm->tm_mday += val * 7;\\n> +\\t\\t\\t\\t\\tif (fval != 0)\\n> +\\t\\t\\t\\t\\t adjust_fval(fval,tm,fsec,7 * 86400);\\n> +\\t\\t\\t\\t\\ttmask = ((fmask & DTK_M(DAY)) ? 0 : DTK_M(DAY));\\n> +\\t\\t\\t\\t\\tbreak;\\n> +\\n> +\\t\\t\\t\\tcase 'm':\\n> +\\t\\t\\t\\t\\ttm->tm_mon += val;\\n> +\\t\\t\\t\\t\\tif (fval != 0)\\n> +\\t\\t\\t\\t\\t adjust_fval(fval,tm,fsec,30 * 86400);\\n> +\\t\\t\\t\\t\\ttmask = DTK_M(MONTH);\\n> +\\t\\t\\t\\t\\tbreak;\\n> +\\n> +\\t\\t\\t\\tcase 'y':\\n> +\\t\\t\\t\\t\\t/*\\n> +\\t\\t\\t\\t\\t * Why can fractional months produce seconds,\\n> +\\t\\t\\t\\t\\t * but fractional years can't? Well the older\\n> +\\t\\t\\t\\t\\t * interval code below has the same property\\n> +\\t\\t\\t\\t\\t * so this one follows the other one too.\\n> +\\t\\t\\t\\t\\t */\\n> +\\t\\t\\t\\t\\ttm->tm_year += val;\\n> +\\t\\t\\t\\t\\tif (fval != 0)\\n> +\\t\\t\\t\\t\\t\\ttm->tm_mon += (fval * 12);\\n> +\\t\\t\\t\\t\\ttmask = ((fmask & DTK_M(YEAR)) ? 0 : DTK_M(YEAR));\\n> +\\t\\t\\t\\t\\tbreak;\\n> +\\n> +\\t\\t\\t\\tdefault:\\n> +\\t\\t\\t\\t\\treturn -1; /* invald date unit prefix */\\n> +\\t\\t\\t}\\n> +\\t\\t}\\n> +\\t\\telse\\n> +\\t\\t{\\n> +\\t\\t\\t/*\\n> +\\t\\t\\t * ISO 8601 time part.\\n> +\\t\\t\\t * In the time part, only one-character\\n> +\\t\\t\\t * unit prefixes are allowed. If it's more\\n> +\\t\\t\\t * than one character, it's not a valid ISO 8601\\n> +\\t\\t\\t * time interval by duration.\\n> +\\t\\t\\t */\\n> +\\t\\t\\tif (units[1] != 0)\\n> +\\t\\t\\t\\treturn -1;\\n> +\\n> +\\t\\t\\tswitch (units[0]) /* H M S */\\n> +\\t\\t\\t{\\n> +\\t\\t\\t\\tcase 's':\\n> +\\t\\t\\t\\t\\ttm->tm_sec += val;\\n> +#ifdef HAVE_INT64_TIMESTAMP\\n> +\\t\\t\\t\\t\\t*fsec += (fval * 1000000);\\n> +#else\\n> +\\t\\t\\t\\t\\t*fsec += fval;\\n> +#endif\\n> +\\t\\t\\t\\t\\ttmask = DTK_M(SECOND);\\n> +\\t\\t\\t\\t\\tbreak;\\n> +\\n> +\\t\\t\\t\\tcase 'm':\\n> +\\t\\t\\t\\t\\ttm->tm_min += val;\\n> +\\t\\t\\t\\t\\tif (fval != 0)\\n> +\\t\\t\\t\\t\\t adjust_fval(fval,tm,fsec,60);\\n> +\\t\\t\\t\\t\\ttmask = DTK_M(MINUTE);\\n> +\\t\\t\\t\\t\\tbreak;\\n> +\\n> +\\t\\t\\t\\tcase 'h':\\n> +\\t\\t\\t\\t\\ttm->tm_hour += val;\\n> +\\t\\t\\t\\t\\tif (fval != 0)\\n> +\\t\\t\\t\\t\\t adjust_fval(fval,tm,fsec,3600);\\n> +\\t\\t\\t\\t\\ttmask = DTK_M(HOUR);\\n> +\\t\\t\\t\\t\\tbreak;\\n> +\\n> +\\t\\t\\t\\tdefault:\\n> +\\t\\t\\t\\t\\treturn -1; /* invald time unit prefix */\\n> +\\t\\t\\t}\\n> +\\t\\t}\\n> +\\t\\tfmask |= tmask;\\n> +\\t}\\n> +\\n> +\\tif (*fsec != 0)\\n> +\\t{\\n> +\\t\\tint\\t\\t\\tsec;\\n> +\\n> +#ifdef HAVE_INT64_TIMESTAMP\\n> +\\t\\tsec = (*fsec / INT64CONST(1000000));\\n> +\\t\\t*fsec -= (sec * INT64CONST(1000000));\\n> +#else\\n> +\\t\\tTMODULO(*fsec, sec, 1e0);\\n> +#endif\\n> +\\t\\ttm->tm_sec += sec;\\n> +\\t}\\n> +\\treturn (fmask != 0) ? 0 : -1;\\n> +}\\n> +\\n> +\\n> /* DecodeInterval()\\n> * Interpret previously parsed fields for general time interval.\\n> * Returns 0 if successful, DTERR code if bogus input detected.\\n> *\\n> * Allow \\\"date\\\" field DTK_DATE since this could be just\\n> *\\tan unsigned floating point number. - thomas 1997-11-16\\n> *\\n> * Allow ISO-style time span, with implicit units on number of days\\n> *\\tpreceding an hh:mm:ss field. - thomas 1998-04-30\\n> + * \\n> + * Allow ISO-8601 style \\\"Representation of time-interval by duration only\\\"\\n> + * of the format 'PnYnMnDTnHnMnS' and 'PnW' - ron 2003-08-30\\n> */\\n> +\\n> int\\n> DecodeInterval(char **field, int *ftype, int nf, int *dtype, struct tm * tm, fsec_t *fsec)\\n> {\\n> \\tint\\t\\t\\tis_before = FALSE;\\n> \\tchar\\t *cp;\\n> \\tint\\t\\t\\tfmask = 0,\\n> \\t\\t\\t\\ttmask,\\n> \\t\\t\\t\\ttype;\\n> \\tint\\t\\t\\ti;\\n> \\tint\\t\\t\\tdterr;\\n> @@ -2906,20 +3148,37 @@\\n> \\n> \\ttype = IGNORE_DTF;\\n> \\ttm->tm_year = 0;\\n> \\ttm->tm_mon = 0;\\n> \\ttm->tm_mday = 0;\\n> \\ttm->tm_hour = 0;\\n> \\ttm->tm_min = 0;\\n> \\ttm->tm_sec = 0;\\n> \\t*fsec = 0;\\n> \\n> +\\t/*\\n> +\\t * Check if it's a ISO 8601 Section 5.5.4.2 \\\"Representation of\\n> + * time-interval by duration only.\\\" \\n> +\\t * Basic extended format: PnYnMnDTnHnMnS\\n> +\\t * PnW\\n> +\\t * http://www.astroclark.freeserve.co.uk/iso8601/index.html\\n> +\\t * ftp://ftp.qsl.net/pub/g1smd/154N362_.PDF\\n> +\\t * Examples: P1D for 1 day\\n> +\\t * PT1H for 1 hour\\n> +\\t * P2Y6M7DT1H30M for 2 years, 6 months, 7 days 1 hour 30 min\\n> +\\t *\\n> +\\t * The first field is exactly \\\"p\\\" or \\\"pt\\\" it may be of this type.\\n> +\\t */\\n> +\\tif (DecodeISO8601Interval(field,ftype,nf,dtype,tm,fsec) == 0) {\\n> +\\t return 0;\\n> + }\\n> +\\n> \\t/* read through list backwards to pick up units before values */\\n> \\tfor (i = nf - 1; i >= 0; i--)\\n> \\t{\\n> \\t\\tswitch (ftype[i])\\n> \\t\\t{\\n> \\t\\t\\tcase DTK_TIME:\\n> \\t\\t\\t\\tdterr = DecodeTime(field[i], fmask, &tmask, tm, fsec);\\n> \\t\\t\\t\\tif (dterr)\\n> \\t\\t\\t\\t\\treturn dterr;\\n> \\t\\t\\t\\ttype = DTK_DAY;\\n> @@ -2983,20 +3242,21 @@\\n> \\t\\t\\t\\t}\\n> \\t\\t\\t\\t/* DROP THROUGH */\\n> \\n> \\t\\t\\tcase DTK_DATE:\\n> \\t\\t\\tcase DTK_NUMBER:\\n> \\t\\t\\t\\tval = strtol(field[i], &cp, 10);\\n> \\n> \\t\\t\\t\\tif (type == IGNORE_DTF)\\n> \\t\\t\\t\\t\\ttype = DTK_SECOND;\\n> \\n> +\\t\\t\\t\\t/* should this allow ',' for locales that use it ? */\\n> \\t\\t\\t\\tif (*cp == '.')\\n> \\t\\t\\t\\t{\\n> \\t\\t\\t\\t\\tfval = strtod(cp, &cp);\\n> \\t\\t\\t\\t\\tif (*cp != '\\\\0')\\n> \\t\\t\\t\\t\\t\\treturn DTERR_BAD_FORMAT;\\n> \\n> \\t\\t\\t\\t\\tif (val < 0)\\n> \\t\\t\\t\\t\\t\\tfval = -(fval);\\n> \\t\\t\\t\\t}\\n> \\t\\t\\t\\telse if (*cp == '\\\\0')\\n> \\n> ===================================================================\\n> \\n> ---------------------------(end of broadcast)---------------------------\\n> TIP 6: Have you searched our list archives?\\n> \\n> http://archives.postgresql.org\\n> \\n\\n-- \\n Bruce Momjian | http://candle.pha.pa.us\\n pgman@candle.pha.pa.us | (610) 359-1001\\n + If your life is a hard drive, | 13 Roberts Road\\n + Christ can be your backup. | Newtown Square, Pennsylvania 19073\\n\",\n \"msg_date\": \"Mon, 8 Sep 2003 16:29:56 -0400 (EDT)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\"\n },\n {\n \"msg_contents\": \"Tom wrote: \\n> \\\"Ron Mayer\\\" writes:\\n> > Tom wrote:\\n> >> Er, don't we support that already?\\n> > ...AFAICT, doesn't match ISO 8601...\\n> \\n> Well, it's *supposed* to match ISO.... Unless ISO has put out \\n> multiple specs that cover this?\\n\\nAny way to tell if this is the case. \\n8601's the one I see cited the most.\\n\\n\\n> > ...I'd not want to break backward compatability...'1H30M'\\n>\\n> I doubt anyone is using it, because it's completely undocumented.\\n> If we're going to support the real ISO spec, I'd suggest ripping\\n> out any not-quite-there variant.\\n\\nI'm happy to look into it. Rip out completely? Ifdef? \\n\\n> We've been gradually cleaning up the problems, but if if you find \\n> something that doesn't seem to make sense, it's likely a bug rather\\n> than anything we want to preserve. \\n\\nI've seen a few more cases that don't make sense.\\n\\nFor example \\\"why is 0.001 years less than 0.001 months\\\".\\n\\n betadb=# select '0.01 years'::interval\\n interval\\n ----------\\n 00:00:00\\n\\n betadb=# select '0.01 months'::interval\\n interval\\n ----------\\n 07:12:00\\n\\nIf I'm breaking backward compatability anyway, I'd be happy to tweak\\nthings like this one too. Unless, of course someone can give me a \\nreason why we want fractional years rounded to months, but fractional \\nmonths are rounded to fractions of a second.\\n\\n Ron Mayer.\\n\\nPS: mailinglist etiquite question... for discussion, should I\\n more this to hackers, or continue it here.\\n\\n\",\n \"msg_date\": \"Mon, 8 Sep 2003 13:42:11 -0700\",\n \"msg_from\": \"\\\"Ron Mayer\\\" \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\\n\\tdeignators\\\"\"\n },\n {\n \"msg_contents\": \"Ron Mayer wrote:\\n> If I'm breaking backward compatability anyway, I'd be happy to tweak\\n> things like this one too. Unless, of course someone can give me a \\n> reason why we want fractional years rounded to months, but fractional \\n> months are rounded to fractions of a second.\\n> \\n> Ron Mayer.\\n> \\n> PS: mailinglist etiquite question... for discussion, should I\\n> more this to hackers, or continue it here.\\n\\nYour choice, but you get a larger audience on hackers. I usually keep\\nthings on patches when I have lots of code to post, and other times move\\nto hackers.\\n\\n-- \\n Bruce Momjian | http://candle.pha.pa.us\\n pgman@candle.pha.pa.us | (610) 359-1001\\n + If your life is a hard drive, | 13 Roberts Road\\n + Christ can be your backup. | Newtown Square, Pennsylvania 19073\\n\\n\",\n \"msg_date\": \"Mon, 8 Sep 2003 17:05:14 -0400 (EDT)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\"\n },\n {\n \"msg_contents\": \"\\\"Ron Mayer\\\" writes:\\n> For example \\\"why is 0.001 years less than 0.001 months\\\".\\n\\nAnd look at this:\\n\\nregression=# select '0.99 years'::interval;\\n interval\\n----------\\n 11 mons\\n(1 row)\\n\\nregression=# select '0.99 months'::interval;\\n interval\\n------------------\\n 29 days 16:48:00\\n(1 row)\\n\\nIt kinda looks like fractional years are converted to integer months\\n(truncating) while fractional months are moved to the seconds part\\nof the interval. Ick. The handling ought to be consistent if you\\nask me.\\n\\n> If I'm breaking backward compatability anyway, I'd be happy to tweak\\n> things like this one too. Unless, of course someone can give me a \\n> reason why we want fractional years rounded to months, but fractional \\n> months are rounded to fractions of a second.\\n\\nActually, what I'd like to see done with interval is re-implement it as\\na three-field entity, separately storing months, days, and seconds.\\nThe separation between months and smaller units is good because a month\\nisn't a fixed number of any smaller unit, but the same holds true for\\ndays and smaller units (days are not always 24 hours, consider DST\\ntransitions). This would no doubt cause some backwards compatibility\\nproblems, but overall it would fix many more cases than it breaks.\\nWe see complaints about related issues regularly, every spring and fall...\\n\\nI'm unsure whether fractional months or fractional days are sensible\\nto accept, but surely we should accept both or reject both. (This might\\nsuggest that the underlying storage for the month and day fields should\\nbe float not int, btw, but I am not sure about it.)\\n\\n> PS: mailinglist etiquite question... for discussion, should I\\n> more this to hackers, or continue it here.\\n\\nAt this point it should move to pghackers, I think.\\n\\n\\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Mon, 08 Sep 2003 17:19:13 -0400\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\\n\\tdeignators\\\"\"\n },\n {\n \"msg_contents\": \"Tom Lane writes:\\n\\n> \\\"Ron Mayer\\\" writes:\\n> > Would it be useful if I added a 'datestyle' of 'ISO basic' which\\n> > would produce the most terse formats ('19980115' for dates,\\n> > and 'P1Y1M' for intervals)?\\n>\\n> I don't really care for using that name for it --- for one thing, you\\n> couldn't do\\n> \\tset datestyle to iso basic;\\n> because of syntax limitations. A one-word name is a much better idea.\\n\\niso8601\\n\\nKeep in mind that SQL itself is also a kind of ISO, so being more specific\\nis useful.\\n\\n-- \\nPeter Eisentraut peter_e@gmx.net\\n\\n\",\n \"msg_date\": \"Mon, 8 Sep 2003 23:58:09 +0200 (CEST)\",\n \"msg_from\": \"Peter Eisentraut \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 'Time Intervals' of the 'format with time-unit\"\n },\n {\n \"msg_contents\": \"Peter Eisentraut writes:\\n> Tom Lane writes:\\n>> I don't really care for using that name for it ---\\n\\n> iso8601\\n\\n> Keep in mind that SQL itself is also a kind of ISO, so being more specific\\n> is useful.\\n\\nYes, but by the same token \\\"iso8601\\\" isn't specific enough either.\\nSeveral of the other input formats we support have at least as good a\\nclaim on that name.\\n\\n\\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Mon, 08 Sep 2003 18:06:11 -0400\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 'Time Intervals' of the 'format with time-unit\\n\\tdeignators'\"\n },\n {\n \"msg_contents\": \"Tom Lane writes:\\n\\n> Yes, but by the same token \\\"iso8601\\\" isn't specific enough either.\\n> Several of the other input formats we support have at least as good a\\n> claim on that name.\\n\\nThe only input formats we support are along the lines of\\n\\n@ 1 year 2 mons 3 days 4 hours 5 mins 6 secs\\n@ 1 year 2 mons 3 days 04:05:06\\n\\nThese are also the supported output formats (the first you get for 'sql',\\n'postgres', and 'german' formats; the second is 'iso'). A quick check of\\nISO 8601 shows, however, that neither of these are close to anything\\nspecified in that standard.\\n\\n-- \\nPeter Eisentraut peter_e@gmx.net\\n\\n\",\n \"msg_date\": \"Tue, 9 Sep 2003 00:39:44 +0200 (CEST)\",\n \"msg_from\": \"Peter Eisentraut \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 'Time Intervals' of the 'format with time-unit\"\n },\n {\n \"msg_contents\": \"Peter Eisentraut writes:\\n> Tom Lane writes:\\n>> Yes, but by the same token \\\"iso8601\\\" isn't specific enough either.\\n>> Several of the other input formats we support have at least as good a\\n>> claim on that name.\\n\\n> The only input formats we support are along the lines of\\n\\n> @ 1 year 2 mons 3 days 4 hours 5 mins 6 secs\\n> @ 1 year 2 mons 3 days 04:05:06\\n\\nSorry, I was thinking of timestamp formats not intervals.\\nYou're right that we don't have anything else particularly ISO-standard\\nfor intervals, but my understanding is that formats like '2003-09-08\\n18:43:31.046283-04' are ISO8601 compatible for timestamps. (Possibly\\nyou need to put a T in there for strict compatibility, not sure.)\\n\\n\\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Mon, 08 Sep 2003 18:45:03 -0400\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 'Time Intervals' of the 'format with time-unit\\n\\tdeignators'\"\n },\n {\n \"msg_contents\": \"Tom wrote...\\n> At this point it should move to pghackers, I think.\\n\\nBackground for pghackers first, open issues below...\\n\\n Over on pgpatches we've been discussing ISO syntax for\\n �time intervals� of the �format with time-unit designators�.\\n http://archives.postgresql.org/pgsql-patches/2003-09/msg00103.php\\n A short summary is that I�ve submitted a patch that\\n accepts intervals of this format..\\n Postgresql interval: ISO8601 Interval\\n ---------------------------------------------------\\n '1 year 6 months' 'P1Y6M'\\n '3 hours 25 minutes 42 seconds' 'PT3H25M42S'\\n The final draft is here\\n ftp://ftp.qsl.net/pub/g1smd/154N362_.PDF\\n\\n This patch was backward-compatable, but further improvements\\n discussed on patches may break compatability so I wanted to\\n discuss them here before implementing them. I�ll also\\n be submitting a new datestyle �iso8601� to output these intervals.\\n\\nOpen issues:\\n\\n1. Postgresql supported a shorthand for intervals that had\\n a similar, but not compatable syntax:\\n Interval ISO Existing postgres\\n 8601 shorthand\\n -----------------------------------------------------\\n '1 year 1 minute' 'P1YT1M' '1Y1M'\\n '1 year 1 month' 'P1Y1M' N/A\\n\\n The current thinking of the thread in pgpatches is to remove\\n the existing (undocumented) syntax.\\n\\n Removing this will break backward compatability if anyone\\n used this feature. Let me know if you needed it.\\n\\n2. Some of the parsing for intervals is inconsistant and\\n confusing. For example, note that �0.01 years� is\\n less than �0.01 months�.\\n\\n betadb=# select '0.01 month'::interval as hundredth_of_month,\\n betadb-# '0.01 year'::interval as hundredth_of_year;\\n hundredth_of_month | hundredth_of_year\\n --------------------+-------------------\\n 07:12:00 | 00:00:00\\n\\n This occurs because the current interval parsing rounds\\n fractional years to the month, but fractional months\\n to the fraction of a second.\\n\\n The current thinking on the thread in patches is\\n at the very least to make these consistant, but with\\n some open-issues because months aren�t a fixed number\\n of days, and days aren�t a fixed number of seconds.\\n\\n The easiest and most minimal change would be to assume\\n that any fractional part automatically gets turned\\n into seconds, assuming things like 30 seconds/month,\\n 24 hrs/day. Since all units except years work that way\\n today, it�d would have the least impact on existing code.\\n\\n A probably better way that Tom recommended would remember\\n fractional months and fractional days. This has the\\n advantage that unlike today,\\n �.5 months�::interval + �.5 months�::interval\\n would then equal 1 month.\\n\\n So what should �.5 years� be?\\n\\n Today, it�s �6 mons�. But I could just as easily\\n argue that it should be 365.2425/2 days, or 4382.91\\n seconds. Each of these will be different (the last\\n two are different durring daylight savings).\\n\\n3. This all is based on the final draft standard of\\n ISO 8601, but I haven�t seen the actual expensive\\n standard. If anyone has it handy...\\n\\n Also, I�m curious to know what if anything the SQL\\n spec says about intervals and units. Any pointers.\\n\\n Ron\\n\\nAny other interval annoyances I should hit at the same time?\\n\\n\",\n \"msg_date\": \"Mon, 8 Sep 2003 16:05:38 -0700\",\n \"msg_from\": \"\\\"Ron Mayer\\\" \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [PATCHES] ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\\n\\tdesignators\\\"\"\n },\n {\n \"msg_contents\": \"\\nTom wrote:\\n> Peter Eisentraut writes:\\n> > Tom Lane writes:\\n> >> Yes, but by the same token \\\"iso8601\\\" isn't specific enough either.\\n\\nISO 8601 gives more specific names.\\n\\n ISO 8601 Basic Format: P2Y10M15DT10H20M30S\\n ISO 8601 Alternative Format: P00021015T102030\\n ISO 8601 Extended Format: P0002-10-15T10:20:30\\n\\nIn a way, the Extended Format is kinda nice, since it�s\\nalmost human readable.\\n\\nI could put in both the basic and extended ones, and\\ncall the dateformats �iso8601basic� and �iso8601extended�.\\nThe negative is that to do �iso8601basic� right, I�d also\\nhave to tweak the �date� and �time� parts of the code too.\\n\\n\",\n \"msg_date\": \"Mon, 8 Sep 2003 16:46:13 -0700\",\n \"msg_from\": \"\\\"Ron Mayer\\\" \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 'Time Intervals' of the 'format with time-unit\\n\\tdeignators'\"\n },\n {\n \"msg_contents\": \"[ backtracking a little ]\\n\\n\\\"Ron Mayer\\\" writes:\\n> Tom wrote: \\n>> I doubt anyone is using it, because it's completely undocumented.\\n>> If we're going to support the real ISO spec, I'd suggest ripping\\n>> out any not-quite-there variant.\\n\\n> I'm happy to look into it. Rip out completely? Ifdef? \\n\\n\\\"Rip\\\" was what I had in mind --- the idea is to simplify the code,\\nwhich you will surely agree is too complicated as it stands. ifdefs\\nwon't simplify the code or make it more understandable, rather the\\nreverse.\\n\\nI have no problem with complex code when it's needed, but in this case\\nthe ptype implementation of almost-ISO notation seems to me to affect\\nmuch more of the code than it has any right to on a usefulness basis.\\n\\n\\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Tue, 09 Sep 2003 00:19:48 -0400\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\\n\\tdeignators\\\"\"\n },\n {\n \"msg_contents\": \"\\nWhere did we leave this?\\n\\n---------------------------------------------------------------------------\\n\\nRon Mayer wrote:\\n> \\n> Tom wrote:\\n> > Peter Eisentraut writes:\\n> > > Tom Lane writes:\\n> > >> Yes, but by the same token \\\"iso8601\\\" isn't specific enough either.\\n> \\n> ISO 8601 gives more specific names.\\n> \\n> ISO 8601 Basic Format: P2Y10M15DT10H20M30S\\n> ISO 8601 Alternative Format: P00021015T102030\\n> ISO 8601 Extended Format: P0002-10-15T10:20:30\\n> \\n> In a way, the Extended Format is kinda nice, since it?s\\n> almost human readable.\\n> \\n> I could put in both the basic and extended ones, and\\n> call the dateformats ?iso8601basic? and ?iso8601extended?.\\n> The negative is that to do ?iso8601basic? right, I?d also\\n> have to tweak the ?date? and ?time? parts of the code too.\\n> \\n> \\n> ---------------------------(end of broadcast)---------------------------\\n> TIP 4: Don't 'kill -9' the postmaster\\n> \\n\\n-- \\n Bruce Momjian | http://candle.pha.pa.us\\n pgman@candle.pha.pa.us | (610) 359-1001\\n + If your life is a hard drive, | 13 Roberts Road\\n + Christ can be your backup. | Newtown Square, Pennsylvania 19073\\n\",\n \"msg_date\": \"Fri, 26 Sep 2003 18:49:08 -0400 (EDT)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 'Time Intervals' of the 'format with time-unit\"\n },\n {\n \"msg_contents\": \"Bruce Momjian writes:\\n> Where did we leave this?\\n\\nI thought it was proposed work for 7.5.\\n\\n\\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Fri, 26 Sep 2003 18:54:09 -0400\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 'Time Intervals' of the 'format with time-unit\\n\\tdeignators'\"\n },\n {\n \"msg_contents\": \"So far,\\n\\n I have submitted the input-part.\\n\\n I have a working output-part (attached below, but I'm\\n still cleaning up the documentation so I'll submit another\\n one later). The output is chosen by setting\\n the datestyle to 'iso8601basic'.\\n\\n Those two changes don't break backward compatability\\n but don't fix too much odd behavior except ISO time interval I/O.\\n\\n\\n I was encouraged to look into changing the way timestamp\\n math is done (keeping month, and day, and second separate\\n until the end). This is a bigger change and I don't have\\n a stable version yet, and it breaks backward compatability. \\n I hope to submit a proposal for changes early enough in the\\n 7.5 timeframe to submit fixes then as well. At the very least\\n I will fully document the existing interval-math as part of\\n this proposal so the docs can be updated even if the proposal\\n gets rejected.\\n\\n Ron\\n\\n> -----Original Message-----\\n> From: Tom Lane [mailto:tgl@sss.pgh.pa.us]\\n> Sent: Friday, September 26, 2003 3:54 PM\\n> To: Bruce Momjian\\n> Cc: Ron Mayer; Peter Eisentraut; andrew@dunslane.net;\\n> pgsql-patches@postgresql.org\\n> Subject: Re: [PATCHES] ISO 8601 'Time Intervals' of the 'format with\\n> time-unit deignators' \\n> \\n> \\n> Bruce Momjian writes:\\n> > Where did we leave this?\\n> \\n> I thought it was proposed work for 7.5.\\n> \\n> \\t\\t\\tregards, tom lane\\n> \\n>\",\n \"msg_date\": \"Fri, 26 Sep 2003 15:54:58 -0700\",\n \"msg_from\": \"\\\"Ron Mayer\\\" \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 'Time Intervals' of the 'format with time-unit\\n\\tdeignators'\"\n },\n {\n \"msg_contents\": \"> I have a working output-part (attached below, but I'm\\n> still cleaning up the documentation so I'll submit another\\n> one later)\\n\\nUgh. Something in this pc quoted some characters in the attachment.\\nRather than trying to apply it, wait a couple days and I'll submit\\nan update where the docs match.\\n\\n Ron\\n\\n\\n\",\n \"msg_date\": \"Fri, 26 Sep 2003 16:04:37 -0700\",\n \"msg_from\": \"\\\"Ron Mayer\\\" \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 'Time Intervals' of the 'format with time-unit\\n\\tdeignators'\"\n },\n {\n \"msg_contents\": \"\\nOh... and I'm pretty sure noone expects this before 7.5. :-)\\nLooks like I'm the only one with input files that have these \\nwierd-but-iso8601 'P1Y6M' and 'PT30M' inputs; so I can just \\nuse my own patch. :-)\\n\\n\\n> \\n> So far,\\n> \\n> I have submitted the input-part.\\n> \\n> I have a working output-part (attached below, but I'm\\n> still cleaning up the documentation so I'll submit another\\n> one later). The output is chosen by setting\\n> the datestyle to 'iso8601basic'.\\n> \\n> Those two changes don't break backward compatability\\n> but don't fix too much odd behavior except ISO time interval I/O.\\n> \\n> \\n> I was encouraged to look into changing the way timestamp\\n> math is done (keeping month, and day, and second separate\\n> until the end). This is a bigger change and I don't have\\n> a stable version yet, and it breaks backward compatability. \\n> I hope to submit a proposal for changes early enough in the\\n> 7.5 timeframe to submit fixes then as well. At the very least\\n> I will fully document the existing interval-math as part of\\n> this proposal so the docs can be updated even if the proposal\\n> gets rejected.\\n> \\n> Ron\\n> \\n> > -----Original Message-----\\n> > From: Tom Lane [mailto:tgl@sss.pgh.pa.us]\\n> > Sent: Friday, September 26, 2003 3:54 PM\\n> > To: Bruce Momjian\\n> > Cc: Ron Mayer; Peter Eisentraut; andrew@dunslane.net;\\n> > pgsql-patches@postgresql.org\\n> > Subject: Re: [PATCHES] ISO 8601 'Time Intervals' of the 'format with\\n> > time-unit deignators' \\n> > \\n> > \\n> > Bruce Momjian writes:\\n> > > Where did we leave this?\\n> > \\n> > I thought it was proposed work for 7.5.\\n> > \\n> > \\t\\t\\tregards, tom lane\\n> > \\n> > \\n\\n\",\n \"msg_date\": \"Fri, 26 Sep 2003 16:04:37 -0700\",\n \"msg_from\": \"\\\"Ron Mayer\\\" \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 'Time Intervals' of the 'format with time-unit\\n\\tdeignators'\"\n },\n {\n \"msg_contents\": \"\\nIs this ready for application? It looks good to me. However, there is\\nan \\\"Open issues\\\" section.\\n\\n---------------------------------------------------------------------------\\n\\nRon Mayer wrote:\\n> Short summary:\\n> \\n> This patch allows ISO 8601 \\\"time intervals\\\" using the \\\"format \\n> with time-unit designators\\\" to specify postgresql \\\"intervals\\\".\\n> \\n> Below I have (A) What these time intervals are, (B) What I\\n> modified to support them, (C) Issues with intervals I want\\n> to bring up, and (D) a patch supporting them.\\n> \\n> It's helpful to me. Any feedback is appreciated. If you \\n> did want to consider including it, let me know what to clean \\n> up. If not, I thought I'd just put it here if anyone else finds\\n> it useful too.\\n> \\n> Thanks for your time,\\n> \\n> Ron Mayer\\n> \\n> Longer:\\n> \\n> (A) What these intervals are.\\n> \\n> ISO 8601, the standard from which PostgreSQL gets some of it's \\n> time syntax, also has a specification for \\\"time-intervals\\\".\\n> \\n> In particular, section 5.5.4.2 has a \\\"Representation of\\n> time-interval by duration only\\\" which I believe maps\\n> nicely to ISO intervals.\\n> \\n> Compared to the ISO 8601 time interval specification, the\\n> postgresql interval syntax is quite verbose. For example:\\n> \\n> Postgresql interval: ISO8601 Interval\\n> ---------------------------------------------------\\n> '1 year 6 months' 'P1Y6M'\\n> '3 hours 25 minutes 42 seconds' 'PT3H25M42S'\\n> \\n> Yeah, it's uglier, but it sure is short which can make\\n> for quicker typing and shorter scripts, and if for some\\n> strange reason you had an application using this format\\n> it's nice not to have to translate.\\n> \\n> The syntax is as follows:\\n> Basic extended format: PnYnMnDTnHnMnS\\n> PnW\\n> \\n> Where everything before the \\\"T\\\" is a date-part and everything\\n> after is a time-part. W is for weeks.\\n> In the date-part, Y=Year, M=Month, D=Day\\n> In the time-part, H=Hour, M=Minute, S=Second\\n> \\n> Much more info can be found from the draft standard\\n> ftp://ftp.qsl.net/pub/g1smd/154N362_.PDF\\n> The final standard's only available for $$$ so I didn't\\n> look at it. Some other sites imply that this part didn't\\n> change from the last draft to the standard.\\n> \\n> \\n> (B) This change was made by adding two functions to \\\"datetime.c\\\"\\n> next to where DecodeInterval parses the normal interval syntax.\\n> \\n> A total of 313 lines were added, including comments and sgml docs.\\n> Of these only 136 are actual code, the rest, comments, whitespace, etc.\\n> \\n> \\n> One new function \\\"DecodeISO8601Interval\\\" follows the style of\\n> \\\"DecodeInterval\\\" below it, and trys to strictly follow the ISO\\n> syntax. If it doesn't match, it'll return -1 and the old syntax\\n> will be checked as before.\\n> \\n> The first test (first character of the first field must be 'P', \\n> and second character must be 'T' or '\\\\0') should be fast so I don't\\n> think this will impact performance of existing code.\\n> \\n> \\n> The second function (\\\"adjust_fval\\\") is just a small helper-function\\n> to remove some of the cut&paste style that DecodeInterval used.\\n> \\n> It seems to work.\\n> =======================================================================\\n> betadb=# select 'P1M15DT12H30M7S'::interval;\\n> interval \\n> ------------------------\\n> 1 mon 15 days 12:30:07\\n> (1 row)\\n> \\n> betadb=# select '1 month 15 days 12 hours 30 minutes 7 seconds'::interval;\\n> \\t interval \\n> ------------------------\\n> 1 mon 15 days 12:30:07\\n> (1 row)\\n> =====================================================================\\n> \\n> \\n> \\n> (C) Open issues with intervals, and questions I'd like to ask.\\n> \\n> 1. DecodeInterval seems to have a hardcoded '.' for specifying\\n> fractional times. ISO 8601 states that both '.' and ',' are\\n> ok, but \\\"of these, the comma is the preferred sign\\\".\\n> \\n> In DecodeISO8601Interval I loosened the test to allow\\n> both but left it as it was in DecodeInterval. Should\\n> both be changed to make them more consistant?\\n> \\n> 2. In \\\"DecodeInterval\\\", fractional weeks and fractional months\\n> can produce seconds; but fractional years can not (rounded\\n> to months). I didn't understand the reasoning for this, so\\n> I left it the same, and followed the same convention for\\n> ISO intervals. Should I change this?\\n> \\n> 3. I could save a bunch of copy-paste-lines-of-code from the\\n> pre-existing DecodeInterval by calling the adjust_fval helper\\n> function. The tradeoff is a few extra function-calls when\\n> decoding an interval. However I didn't want to risk changes\\n> to the existing part unless you guys encourage me to do so.\\n> \\n> \\n> (D) The patch.\\n> \\n> \\n> Index: doc/src/sgml/datatype.sgml\\n> ===================================================================\\n> RCS file: /projects/cvsroot/pgsql-server/doc/src/sgml/datatype.sgml,v\\n> retrieving revision 1.123\\n> diff -u -1 -0 -r1.123 datatype.sgml\\n> --- doc/src/sgml/datatype.sgml\\t31 Aug 2003 17:32:18 -0000\\t1.123\\n> +++ doc/src/sgml/datatype.sgml\\t8 Sep 2003 04:04:58 -0000\\n> @@ -1735,20 +1735,71 @@\\n> Quantities of days, hours, minutes, and seconds can be specified without\\n> explicit unit markings. For example, '1 12:59:10' is read\\n> the same as '1 day 12 hours 59 min 10 sec'.\\n> \\n> \\n> \\n> The optional precision\\n> p should be between 0 and 6, and\\n> defaults to the precision of the input literal.\\n> \\n> +\\n> +\\n> + \\n> + Alternatively, interval values can be written as \\n> + ISO 8601 time intervals, using the \\\"Format with time-unit designators\\\".\\n> + This format always starts with the character 'P', followed \\n> + by a string of values followed by single character time-unit designators.\\n> + A 'T' separates the date and time parts of the interval.\\n> + \\n> +\\n> + \\n> + Format: PnYnMnDTnHnMnS\\n> + \\n> + \\n> + In this format, 'n' gets replaced by a number, and \\n> + Y represents years, \\n> + M (in the date part) months,\\n> + D months,\\n> + H hours,\\n> + M (in the time part) minutes,\\n> + and S seconds.\\n> + \\n> + \\n> +\\n> + \\n> +\\t Interval Example\\n> +\\t \\n> +\\t\\t\\n> +\\t\\t \\n> +\\t\\t Traditional\\n> +\\t\\t ISO-8601 time-interval\\n> +\\t\\t
\\n> +\\t\\t\\n> +\\t\\t\\n> +\\t\\t \\n> +\\t\\t 1 month\\n> +\\t\\t P1M\\n> +\\t\\t
\\n> +\\t\\t \\n> +\\t\\t 1 hour 30 minutes\\n> +\\t\\t PT1H30M\\n> +\\t\\t
\\n> +\\t\\t \\n> +\\t\\t 2 years 10 months 15 days 10 hours 30 minutes 20 seconds\\n> +\\t\\t P2Y10M15DT10H30M20S\\n> +\\t\\t
\\n> +\\t\\t\\n> +\\t \\n> +\\t
\\n> +\\t \\n> + \\n> \\n> \\n> \\n> Special Values\\n> \\n> \\n> time\\n> constants\\n> \\n> \\n> Index: src/backend/utils/adt/datetime.c\\n> ===================================================================\\n> RCS file: /projects/cvsroot/pgsql-server/src/backend/utils/adt/datetime.c,v\\n> retrieving revision 1.116\\n> diff -u -1 -0 -r1.116 datetime.c\\n> --- src/backend/utils/adt/datetime.c\\t27 Aug 2003 23:29:28 -0000\\t1.116\\n> +++ src/backend/utils/adt/datetime.c\\t8 Sep 2003 04:04:59 -0000\\n> @@ -30,20 +30,21 @@\\n> \\t\\t\\t struct tm * tm, fsec_t *fsec, int *is2digits);\\n> static int DecodeNumberField(int len, char *str,\\n> \\t\\t\\t\\t int fmask, int *tmask,\\n> \\t\\t\\t\\t struct tm * tm, fsec_t *fsec, int *is2digits);\\n> static int DecodeTime(char *str, int fmask, int *tmask,\\n> \\t\\t struct tm * tm, fsec_t *fsec);\\n> static int\\tDecodeTimezone(char *str, int *tzp);\\n> static datetkn *datebsearch(char *key, datetkn *base, unsigned int nel);\\n> static int\\tDecodeDate(char *str, int fmask, int *tmask, struct tm * tm);\\n> static void TrimTrailingZeros(char *str);\\n> +static int DecodeISO8601Interval(char **field, int *ftype, int nf, int *dtype, struct tm * tm, fsec_t *fsec);\\n> \\n> \\n> int\\t\\t\\tday_tab[2][13] = {\\n> \\t{31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31, 0},\\n> {31, 29, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31, 0}};\\n> \\n> char\\t *months[] = {\\\"Jan\\\", \\\"Feb\\\", \\\"Mar\\\", \\\"Apr\\\", \\\"May\\\", \\\"Jun\\\",\\n> \\\"Jul\\\", \\\"Aug\\\", \\\"Sep\\\", \\\"Oct\\\", \\\"Nov\\\", \\\"Dec\\\", NULL};\\n> \\n> char\\t *days[] = {\\\"Sunday\\\", \\\"Monday\\\", \\\"Tuesday\\\", \\\"Wednesday\\\",\\n> @@ -2872,30 +2873,271 @@\\n> \\t\\t\\tdefault:\\n> \\t\\t\\t\\t*val = tp->value;\\n> \\t\\t\\t\\tbreak;\\n> \\t\\t}\\n> \\t}\\n> \\n> \\treturn type;\\n> }\\n> \\n> \\n> +void adjust_fval(double fval,struct tm * tm, fsec_t *fsec, int scale);\\n> +{\\n> +\\tint\\tsec;\\n> +\\tfval\\t *= scale;\\n> +\\tsec\\t\\t = fval;\\n> +\\ttm->tm_sec += sec;\\n> +#ifdef HAVE_INT64_TIMESTAMP\\n> +\\t*fsec\\t += ((fval - sec) * 1000000);\\n> +#else\\n> +\\t*fsec\\t += (fval - sec);\\n> +#endif\\n> +}\\n> +\\n> +\\n> +/* DecodeISO8601Interval()\\n> + *\\n> + * Check if it's a ISO 8601 Section 5.5.4.2 \\\"Representation of\\n> + * time-interval by duration only.\\\" \\n> + * Basic extended format: PnYnMnDTnHnMnS\\n> + * PnW\\n> + * For more info.\\n> + * http://www.astroclark.freeserve.co.uk/iso8601/index.html\\n> + * ftp://ftp.qsl.net/pub/g1smd/154N362_.PDF\\n> + *\\n> + * Examples: P1D for 1 day\\n> + * PT1H for 1 hour\\n> + * P2Y6M7DT1H30M for 2 years, 6 months, 7 days 1 hour 30 min\\n> + *\\n> + * The first field is exactly \\\"p\\\" or \\\"pt\\\" it may be of this type.\\n> + *\\n> + * Returns -1 if the field is not of this type.\\n> + *\\n> + * It pretty strictly checks the spec, with the two exceptions\\n> + * that a week field ('W') may coexist with other units, and that\\n> + * this function allows decimals in fields other than the least\\n> + * significant units.\\n> + */\\n> +int\\n> +DecodeISO8601Interval(char **field, int *ftype, int nf, int *dtype, struct tm * tm, fsec_t *fsec) \\n> +{\\n> +\\tchar\\t *cp;\\n> +\\tint\\t\\t\\tfmask = 0,\\n> +\\t\\t\\t\\ttmask;\\n> +\\tint\\t\\t\\tval;\\n> +\\tdouble\\t\\tfval;\\n> +\\tint\\t\\t\\targ;\\n> +\\tint\\t\\t\\tdatepart;\\n> +\\n> + /*\\n> +\\t * An ISO 8601 \\\"time-interval by duration only\\\" must start\\n> +\\t * with a 'P'. If it contains a date-part, 'p' will be the\\n> +\\t * only character in the field. If it contains no date part\\n> +\\t * it will contain exactly to characters 'PT' indicating a\\n> +\\t * time part.\\n> +\\t * Anything else is illegal and will be treated like a \\n> +\\t * traditional postgresql interval.\\n> +\\t */\\n> + if (!(field[0][0] == 'p' &&\\n> + ((field[0][1] == 0) || (field[0][1] == 't' && field[0][2] == 0))))\\n> +\\t{\\n> +\\t return -1;\\n> +\\t}\\n> +\\n> +\\n> + /*\\n> +\\t * If the first field is exactly 1 character ('P'), it starts\\n> +\\t * with date elements. Otherwise it's two characters ('PT');\\n> +\\t * indicating it starts with a time part.\\n> +\\t */\\n> +\\tdatepart = (field[0][1] == 0);\\n> +\\n> +\\t/*\\n> +\\t * Every value must have a unit, so we require an even\\n> +\\t * number of value/unit pairs. Therefore we require an\\n> +\\t * odd nubmer of fields, including the prefix 'P'.\\n> +\\t */\\n> +\\tif ((nf & 1) == 0)\\n> +\\t\\treturn -1;\\n> +\\n> +\\t/*\\n> +\\t * Process pairs of fields at a time.\\n> +\\t */\\n> +\\tfor (arg = 1 ; arg < nf ; arg+=2) \\n> +\\t{\\n> +\\t\\tchar * value = field[arg ];\\n> +\\t\\tchar * units = field[arg+1];\\n> +\\n> +\\t\\t/*\\n> +\\t\\t * The value part must be a number.\\n> +\\t\\t */\\n> +\\t\\tif (ftype[arg] != DTK_NUMBER) \\n> +\\t\\t\\treturn -1;\\n> +\\n> +\\t\\t/*\\n> +\\t\\t * extract the number, almost exactly like the non-ISO interval.\\n> +\\t\\t */\\n> +\\t\\tval = strtol(value, &cp, 10);\\n> +\\n> +\\t\\t/*\\n> +\\t\\t * One difference from the normal postgresql interval below...\\n> +\\t\\t * ISO 8601 states that \\\"Of these, the comma is the preferred \\n> +\\t\\t * sign\\\" so I allow it here for locales that support it.\\n> +\\t\\t * Note: Perhaps the old-style interval code below should\\n> +\\t\\t * allow for this too, but I didn't want to risk backward\\n> +\\t\\t * compatability.\\n> +\\t\\t */\\n> +\\t\\tif (*cp == '.' || *cp == ',') \\n> +\\t\\t{\\n> +\\t\\t\\tfval = strtod(cp, &cp);\\n> +\\t\\t\\tif (*cp != '\\\\0')\\n> +\\t\\t\\t\\treturn -1;\\n> +\\n> +\\t\\t\\tif (val < 0)\\n> +\\t\\t\\t\\tfval = -(fval);\\n> +\\t\\t}\\n> +\\t\\telse if (*cp == '\\\\0')\\n> +\\t\\t\\tfval = 0;\\n> +\\t\\telse\\n> +\\t\\t\\treturn -1;\\n> +\\n> +\\n> +\\t\\tif (datepart)\\n> +\\t\\t{\\n> +\\t\\t\\t/*\\n> +\\t\\t\\t * All the 8601 unit specifiers are 1 character, but may\\n> +\\t\\t\\t * be followed by a 'T' character if transitioning between\\n> +\\t\\t\\t * the date part and the time part. If it's not either\\n> +\\t\\t\\t * one character or two characters with the second being 't'\\n> +\\t\\t\\t * it's an error.\\n> +\\t\\t\\t */\\n> +\\t\\t\\tif (!(units[1] == 0 || (units[1] == 't' && units[2] == 0)))\\n> +\\t\\t\\t\\treturn -1;\\n> +\\n> +\\t\\t\\tif (units[1] == 't')\\n> +\\t\\t\\t\\tdatepart = 0;\\n> +\\n> +\\t\\t\\tswitch (units[0]) /* Y M D W */\\n> +\\t\\t\\t{\\n> +\\t\\t\\t\\tcase 'd':\\n> +\\t\\t\\t\\t\\ttm->tm_mday += val;\\n> +\\t\\t\\t\\t\\tif (fval != 0)\\n> +\\t\\t\\t\\t\\t adjust_fval(fval,tm,fsec, 86400);\\n> +\\t\\t\\t\\t\\ttmask = ((fmask & DTK_M(DAY)) ? 0 : DTK_M(DAY));\\n> +\\t\\t\\t\\t\\tbreak;\\n> +\\n> +\\t\\t\\t\\tcase 'w':\\n> +\\t\\t\\t\\t\\ttm->tm_mday += val * 7;\\n> +\\t\\t\\t\\t\\tif (fval != 0)\\n> +\\t\\t\\t\\t\\t adjust_fval(fval,tm,fsec,7 * 86400);\\n> +\\t\\t\\t\\t\\ttmask = ((fmask & DTK_M(DAY)) ? 0 : DTK_M(DAY));\\n> +\\t\\t\\t\\t\\tbreak;\\n> +\\n> +\\t\\t\\t\\tcase 'm':\\n> +\\t\\t\\t\\t\\ttm->tm_mon += val;\\n> +\\t\\t\\t\\t\\tif (fval != 0)\\n> +\\t\\t\\t\\t\\t adjust_fval(fval,tm,fsec,30 * 86400);\\n> +\\t\\t\\t\\t\\ttmask = DTK_M(MONTH);\\n> +\\t\\t\\t\\t\\tbreak;\\n> +\\n> +\\t\\t\\t\\tcase 'y':\\n> +\\t\\t\\t\\t\\t/*\\n> +\\t\\t\\t\\t\\t * Why can fractional months produce seconds,\\n> +\\t\\t\\t\\t\\t * but fractional years can't? Well the older\\n> +\\t\\t\\t\\t\\t * interval code below has the same property\\n> +\\t\\t\\t\\t\\t * so this one follows the other one too.\\n> +\\t\\t\\t\\t\\t */\\n> +\\t\\t\\t\\t\\ttm->tm_year += val;\\n> +\\t\\t\\t\\t\\tif (fval != 0)\\n> +\\t\\t\\t\\t\\t\\ttm->tm_mon += (fval * 12);\\n> +\\t\\t\\t\\t\\ttmask = ((fmask & DTK_M(YEAR)) ? 0 : DTK_M(YEAR));\\n> +\\t\\t\\t\\t\\tbreak;\\n> +\\n> +\\t\\t\\t\\tdefault:\\n> +\\t\\t\\t\\t\\treturn -1; /* invald date unit prefix */\\n> +\\t\\t\\t}\\n> +\\t\\t}\\n> +\\t\\telse\\n> +\\t\\t{\\n> +\\t\\t\\t/*\\n> +\\t\\t\\t * ISO 8601 time part.\\n> +\\t\\t\\t * In the time part, only one-character\\n> +\\t\\t\\t * unit prefixes are allowed. If it's more\\n> +\\t\\t\\t * than one character, it's not a valid ISO 8601\\n> +\\t\\t\\t * time interval by duration.\\n> +\\t\\t\\t */\\n> +\\t\\t\\tif (units[1] != 0)\\n> +\\t\\t\\t\\treturn -1;\\n> +\\n> +\\t\\t\\tswitch (units[0]) /* H M S */\\n> +\\t\\t\\t{\\n> +\\t\\t\\t\\tcase 's':\\n> +\\t\\t\\t\\t\\ttm->tm_sec += val;\\n> +#ifdef HAVE_INT64_TIMESTAMP\\n> +\\t\\t\\t\\t\\t*fsec += (fval * 1000000);\\n> +#else\\n> +\\t\\t\\t\\t\\t*fsec += fval;\\n> +#endif\\n> +\\t\\t\\t\\t\\ttmask = DTK_M(SECOND);\\n> +\\t\\t\\t\\t\\tbreak;\\n> +\\n> +\\t\\t\\t\\tcase 'm':\\n> +\\t\\t\\t\\t\\ttm->tm_min += val;\\n> +\\t\\t\\t\\t\\tif (fval != 0)\\n> +\\t\\t\\t\\t\\t adjust_fval(fval,tm,fsec,60);\\n> +\\t\\t\\t\\t\\ttmask = DTK_M(MINUTE);\\n> +\\t\\t\\t\\t\\tbreak;\\n> +\\n> +\\t\\t\\t\\tcase 'h':\\n> +\\t\\t\\t\\t\\ttm->tm_hour += val;\\n> +\\t\\t\\t\\t\\tif (fval != 0)\\n> +\\t\\t\\t\\t\\t adjust_fval(fval,tm,fsec,3600);\\n> +\\t\\t\\t\\t\\ttmask = DTK_M(HOUR);\\n> +\\t\\t\\t\\t\\tbreak;\\n> +\\n> +\\t\\t\\t\\tdefault:\\n> +\\t\\t\\t\\t\\treturn -1; /* invald time unit prefix */\\n> +\\t\\t\\t}\\n> +\\t\\t}\\n> +\\t\\tfmask |= tmask;\\n> +\\t}\\n> +\\n> +\\tif (*fsec != 0)\\n> +\\t{\\n> +\\t\\tint\\t\\t\\tsec;\\n> +\\n> +#ifdef HAVE_INT64_TIMESTAMP\\n> +\\t\\tsec = (*fsec / INT64CONST(1000000));\\n> +\\t\\t*fsec -= (sec * INT64CONST(1000000));\\n> +#else\\n> +\\t\\tTMODULO(*fsec, sec, 1e0);\\n> +#endif\\n> +\\t\\ttm->tm_sec += sec;\\n> +\\t}\\n> +\\treturn (fmask != 0) ? 0 : -1;\\n> +}\\n> +\\n> +\\n> /* DecodeInterval()\\n> * Interpret previously parsed fields for general time interval.\\n> * Returns 0 if successful, DTERR code if bogus input detected.\\n> *\\n> * Allow \\\"date\\\" field DTK_DATE since this could be just\\n> *\\tan unsigned floating point number. - thomas 1997-11-16\\n> *\\n> * Allow ISO-style time span, with implicit units on number of days\\n> *\\tpreceding an hh:mm:ss field. - thomas 1998-04-30\\n> + * \\n> + * Allow ISO-8601 style \\\"Representation of time-interval by duration only\\\"\\n> + * of the format 'PnYnMnDTnHnMnS' and 'PnW' - ron 2003-08-30\\n> */\\n> +\\n> int\\n> DecodeInterval(char **field, int *ftype, int nf, int *dtype, struct tm * tm, fsec_t *fsec)\\n> {\\n> \\tint\\t\\t\\tis_before = FALSE;\\n> \\tchar\\t *cp;\\n> \\tint\\t\\t\\tfmask = 0,\\n> \\t\\t\\t\\ttmask,\\n> \\t\\t\\t\\ttype;\\n> \\tint\\t\\t\\ti;\\n> \\tint\\t\\t\\tdterr;\\n> @@ -2906,20 +3148,37 @@\\n> \\n> \\ttype = IGNORE_DTF;\\n> \\ttm->tm_year = 0;\\n> \\ttm->tm_mon = 0;\\n> \\ttm->tm_mday = 0;\\n> \\ttm->tm_hour = 0;\\n> \\ttm->tm_min = 0;\\n> \\ttm->tm_sec = 0;\\n> \\t*fsec = 0;\\n> \\n> +\\t/*\\n> +\\t * Check if it's a ISO 8601 Section 5.5.4.2 \\\"Representation of\\n> + * time-interval by duration only.\\\" \\n> +\\t * Basic extended format: PnYnMnDTnHnMnS\\n> +\\t * PnW\\n> +\\t * http://www.astroclark.freeserve.co.uk/iso8601/index.html\\n> +\\t * ftp://ftp.qsl.net/pub/g1smd/154N362_.PDF\\n> +\\t * Examples: P1D for 1 day\\n> +\\t * PT1H for 1 hour\\n> +\\t * P2Y6M7DT1H30M for 2 years, 6 months, 7 days 1 hour 30 min\\n> +\\t *\\n> +\\t * The first field is exactly \\\"p\\\" or \\\"pt\\\" it may be of this type.\\n> +\\t */\\n> +\\tif (DecodeISO8601Interval(field,ftype,nf,dtype,tm,fsec) == 0) {\\n> +\\t return 0;\\n> + }\\n> +\\n> \\t/* read through list backwards to pick up units before values */\\n> \\tfor (i = nf - 1; i >= 0; i--)\\n> \\t{\\n> \\t\\tswitch (ftype[i])\\n> \\t\\t{\\n> \\t\\t\\tcase DTK_TIME:\\n> \\t\\t\\t\\tdterr = DecodeTime(field[i], fmask, &tmask, tm, fsec);\\n> \\t\\t\\t\\tif (dterr)\\n> \\t\\t\\t\\t\\treturn dterr;\\n> \\t\\t\\t\\ttype = DTK_DAY;\\n> @@ -2983,20 +3242,21 @@\\n> \\t\\t\\t\\t}\\n> \\t\\t\\t\\t/* DROP THROUGH */\\n> \\n> \\t\\t\\tcase DTK_DATE:\\n> \\t\\t\\tcase DTK_NUMBER:\\n> \\t\\t\\t\\tval = strtol(field[i], &cp, 10);\\n> \\n> \\t\\t\\t\\tif (type == IGNORE_DTF)\\n> \\t\\t\\t\\t\\ttype = DTK_SECOND;\\n> \\n> +\\t\\t\\t\\t/* should this allow ',' for locales that use it ? */\\n> \\t\\t\\t\\tif (*cp == '.')\\n> \\t\\t\\t\\t{\\n> \\t\\t\\t\\t\\tfval = strtod(cp, &cp);\\n> \\t\\t\\t\\t\\tif (*cp != '\\\\0')\\n> \\t\\t\\t\\t\\t\\treturn DTERR_BAD_FORMAT;\\n> \\n> \\t\\t\\t\\t\\tif (val < 0)\\n> \\t\\t\\t\\t\\t\\tfval = -(fval);\\n> \\t\\t\\t\\t}\\n> \\t\\t\\t\\telse if (*cp == '\\\\0')\\n> \\n> ===================================================================\\n> \\n> ---------------------------(end of broadcast)---------------------------\\n> TIP 6: Have you searched our list archives?\\n> \\n> http://archives.postgresql.org\\n> \\n\\n-- \\n Bruce Momjian | http://candle.pha.pa.us\\n pgman@candle.pha.pa.us | (610) 359-1001\\n + If your life is a hard drive, | 13 Roberts Road\\n + Christ can be your backup. | Newtown Square, Pennsylvania 19073\\n\",\n \"msg_date\": \"Sun, 30 Nov 2003 23:52:10 -0500 (EST)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\"\n },\n {\n \"msg_contents\": \"\\nHere is an email on the open issues.\\n\\n---------------------------------------------------------------------------\\n\\nRon Mayer wrote:\\n> Tom wrote...\\n> > At this point it should move to pghackers, I think.\\n> \\n> Background for pghackers first, open issues below...\\n> \\n> Over on pgpatches we've been discussing ISO syntax for\\n> ?time intervals? of the ?format with time-unit designators?.\\n> http://archives.postgresql.org/pgsql-patches/2003-09/msg00103.php\\n> A short summary is that I?ve submitted a patch that\\n> accepts intervals of this format..\\n> Postgresql interval: ISO8601 Interval\\n> ---------------------------------------------------\\n> '1 year 6 months' 'P1Y6M'\\n> '3 hours 25 minutes 42 seconds' 'PT3H25M42S'\\n> The final draft is here\\n> ftp://ftp.qsl.net/pub/g1smd/154N362_.PDF\\n> \\n> This patch was backward-compatable, but further improvements\\n> discussed on patches may break compatability so I wanted to\\n> discuss them here before implementing them. I?ll also\\n> be submitting a new datestyle ?iso8601? to output these intervals.\\n> \\n> Open issues:\\n> \\n> 1. Postgresql supported a shorthand for intervals that had\\n> a similar, but not compatable syntax:\\n> Interval ISO Existing postgres\\n> 8601 shorthand\\n> -----------------------------------------------------\\n> '1 year 1 minute' 'P1YT1M' '1Y1M'\\n> '1 year 1 month' 'P1Y1M' N/A\\n> \\n> The current thinking of the thread in pgpatches is to remove\\n> the existing (undocumented) syntax.\\n> \\n> Removing this will break backward compatability if anyone\\n> used this feature. Let me know if you needed it.\\n> \\n> 2. Some of the parsing for intervals is inconsistant and\\n> confusing. For example, note that ?0.01 years? is\\n> less than ?0.01 months?.\\n> \\n> betadb=# select '0.01 month'::interval as hundredth_of_month,\\n> betadb-# '0.01 year'::interval as hundredth_of_year;\\n> hundredth_of_month | hundredth_of_year\\n> --------------------+-------------------\\n> 07:12:00 | 00:00:00\\n> \\n> This occurs because the current interval parsing rounds\\n> fractional years to the month, but fractional months\\n> to the fraction of a second.\\n> \\n> The current thinking on the thread in patches is\\n> at the very least to make these consistant, but with\\n> some open-issues because months aren?t a fixed number\\n> of days, and days aren?t a fixed number of seconds.\\n> \\n> The easiest and most minimal change would be to assume\\n> that any fractional part automatically gets turned\\n> into seconds, assuming things like 30 seconds/month,\\n> 24 hrs/day. Since all units except years work that way\\n> today, it?d would have the least impact on existing code.\\n> \\n> A probably better way that Tom recommended would remember\\n> fractional months and fractional days. This has the\\n> advantage that unlike today,\\n> ?.5 months?::interval + ?.5 months?::interval\\n> would then equal 1 month.\\n> \\n> So what should ?.5 years? be?\\n> \\n> Today, it?s ?6 mons?. But I could just as easily\\n> argue that it should be 365.2425/2 days, or 4382.91\\n> seconds. Each of these will be different (the last\\n> two are different durring daylight savings).\\n> \\n> 3. This all is based on the final draft standard of\\n> ISO 8601, but I haven?t seen the actual expensive\\n> standard. If anyone has it handy...\\n> \\n> Also, I?m curious to know what if anything the SQL\\n> spec says about intervals and units. Any pointers.\\n> \\n> Ron\\n> \\n> Any other interval annoyances I should hit at the same time?\\n> \\n> \\n> ---------------------------(end of broadcast)---------------------------\\n> TIP 3: if posting/reading through Usenet, please send an appropriate\\n> subscribe-nomail command to majordomo@postgresql.org so that your\\n> message can get through to the mailing list cleanly\\n> \\n\\n-- \\n Bruce Momjian | http://candle.pha.pa.us\\n pgman@candle.pha.pa.us | (610) 359-1001\\n + If your life is a hard drive, | 13 Roberts Road\\n + Christ can be your backup. | Newtown Square, Pennsylvania 19073\\n\",\n \"msg_date\": \"Sun, 30 Nov 2003 23:53:13 -0500 (EST)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [PATCHES] ISO 8601 \\\"Time Intervals\\\" of the \\\"format with\"\n },\n {\n \"msg_contents\": \"\\nAnd another open issues email.\\n\\n---------------------------------------------------------------------------\\n\\nRon Mayer wrote:\\n> \\n> Tom wrote:\\n> > Peter Eisentraut writes:\\n> > > Tom Lane writes:\\n> > >> Yes, but by the same token \\\"iso8601\\\" isn't specific enough either.\\n> \\n> ISO 8601 gives more specific names.\\n> \\n> ISO 8601 Basic Format: P2Y10M15DT10H20M30S\\n> ISO 8601 Alternative Format: P00021015T102030\\n> ISO 8601 Extended Format: P0002-10-15T10:20:30\\n> \\n> In a way, the Extended Format is kinda nice, since it?s\\n> almost human readable.\\n> \\n> I could put in both the basic and extended ones, and\\n> call the dateformats ?iso8601basic? and ?iso8601extended?.\\n> The negative is that to do ?iso8601basic? right, I?d also\\n> have to tweak the ?date? and ?time? parts of the code too.\\n> \\n> \\n> ---------------------------(end of broadcast)---------------------------\\n> TIP 4: Don't 'kill -9' the postmaster\\n> \\n\\n-- \\n Bruce Momjian | http://candle.pha.pa.us\\n pgman@candle.pha.pa.us | (610) 359-1001\\n + If your life is a hard drive, | 13 Roberts Road\\n + Christ can be your backup. | Newtown Square, Pennsylvania 19073\\n\",\n \"msg_date\": \"Sun, 30 Nov 2003 23:53:22 -0500 (EST)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 'Time Intervals' of the 'format with time-unit\"\n },\n {\n \"msg_contents\": \"Bruce Momjian writes:\\n\\n> Is this ready for application? It looks good to me. However, there is\\n> an \\\"Open issues\\\" section.\\n\\nIt would be more useful to implement the SQL standard for intervals first\\ninstead of inventing more nonstandard formats for it.\\n\\n>\\n> ---------------------------------------------------------------------------\\n>\\n> Ron Mayer wrote:\\n> > Short summary:\\n> >\\n> > This patch allows ISO 8601 \\\"time intervals\\\" using the \\\"format\\n> > with time-unit designators\\\" to specify postgresql \\\"intervals\\\".\\n> >\\n> > Below I have (A) What these time intervals are, (B) What I\\n> > modified to support them, (C) Issues with intervals I want\\n> > to bring up, and (D) a patch supporting them.\\n> >\\n> > It's helpful to me. Any feedback is appreciated. If you\\n> > did want to consider including it, let me know what to clean\\n> > up. If not, I thought I'd just put it here if anyone else finds\\n> > it useful too.\\n> >\\n> > Thanks for your time,\\n> >\\n> > Ron Mayer\\n> >\\n> > Longer:\\n> >\\n> > (A) What these intervals are.\\n> >\\n> > ISO 8601, the standard from which PostgreSQL gets some of it's\\n> > time syntax, also has a specification for \\\"time-intervals\\\".\\n> >\\n> > In particular, section 5.5.4.2 has a \\\"Representation of\\n> > time-interval by duration only\\\" which I believe maps\\n> > nicely to ISO intervals.\\n> >\\n> > Compared to the ISO 8601 time interval specification, the\\n> > postgresql interval syntax is quite verbose. For example:\\n> >\\n> > Postgresql interval: ISO8601 Interval\\n> > ---------------------------------------------------\\n> > '1 year 6 months' 'P1Y6M'\\n> > '3 hours 25 minutes 42 seconds' 'PT3H25M42S'\\n> >\\n> > Yeah, it's uglier, but it sure is short which can make\\n> > for quicker typing and shorter scripts, and if for some\\n> > strange reason you had an application using this format\\n> > it's nice not to have to translate.\\n> >\\n> > The syntax is as follows:\\n> > Basic extended format: PnYnMnDTnHnMnS\\n> > PnW\\n> >\\n> > Where everything before the \\\"T\\\" is a date-part and everything\\n> > after is a time-part. W is for weeks.\\n> > In the date-part, Y=Year, M=Month, D=Day\\n> > In the time-part, H=Hour, M=Minute, S=Second\\n> >\\n> > Much more info can be found from the draft standard\\n> > ftp://ftp.qsl.net/pub/g1smd/154N362_.PDF\\n> > The final standard's only available for $$$ so I didn't\\n> > look at it. Some other sites imply that this part didn't\\n> > change from the last draft to the standard.\\n> >\\n> >\\n> > (B) This change was made by adding two functions to \\\"datetime.c\\\"\\n> > next to where DecodeInterval parses the normal interval syntax.\\n> >\\n> > A total of 313 lines were added, including comments and sgml docs.\\n> > Of these only 136 are actual code, the rest, comments, whitespace, etc.\\n> >\\n> >\\n> > One new function \\\"DecodeISO8601Interval\\\" follows the style of\\n> > \\\"DecodeInterval\\\" below it, and trys to strictly follow the ISO\\n> > syntax. If it doesn't match, it'll return -1 and the old syntax\\n> > will be checked as before.\\n> >\\n> > The first test (first character of the first field must be 'P',\\n> > and second character must be 'T' or '\\\\0') should be fast so I don't\\n> > think this will impact performance of existing code.\\n> >\\n> >\\n> > The second function (\\\"adjust_fval\\\") is just a small helper-function\\n> > to remove some of the cut&paste style that DecodeInterval used.\\n> >\\n> > It seems to work.\\n> > =======================================================================\\n> > betadb=# select 'P1M15DT12H30M7S'::interval;\\n> > interval\\n> > ------------------------\\n> > 1 mon 15 days 12:30:07\\n> > (1 row)\\n> >\\n> > betadb=# select '1 month 15 days 12 hours 30 minutes 7 seconds'::interval;\\n> > \\t interval\\n> > ------------------------\\n> > 1 mon 15 days 12:30:07\\n> > (1 row)\\n> > =====================================================================\\n> >\\n> >\\n> >\\n> > (C) Open issues with intervals, and questions I'd like to ask.\\n> >\\n> > 1. DecodeInterval seems to have a hardcoded '.' for specifying\\n> > fractional times. ISO 8601 states that both '.' and ',' are\\n> > ok, but \\\"of these, the comma is the preferred sign\\\".\\n> >\\n> > In DecodeISO8601Interval I loosened the test to allow\\n> > both but left it as it was in DecodeInterval. Should\\n> > both be changed to make them more consistant?\\n> >\\n> > 2. In \\\"DecodeInterval\\\", fractional weeks and fractional months\\n> > can produce seconds; but fractional years can not (rounded\\n> > to months). I didn't understand the reasoning for this, so\\n> > I left it the same, and followed the same convention for\\n> > ISO intervals. Should I change this?\\n> >\\n> > 3. I could save a bunch of copy-paste-lines-of-code from the\\n> > pre-existing DecodeInterval by calling the adjust_fval helper\\n> > function. The tradeoff is a few extra function-calls when\\n> > decoding an interval. However I didn't want to risk changes\\n> > to the existing part unless you guys encourage me to do so.\\n> >\\n> >\\n> > (D) The patch.\\n> >\\n> >\\n> > Index: doc/src/sgml/datatype.sgml\\n> > ===================================================================\\n> > RCS file: /projects/cvsroot/pgsql-server/doc/src/sgml/datatype.sgml,v\\n> > retrieving revision 1.123\\n> > diff -u -1 -0 -r1.123 datatype.sgml\\n> > --- doc/src/sgml/datatype.sgml\\t31 Aug 2003 17:32:18 -0000\\t1.123\\n> > +++ doc/src/sgml/datatype.sgml\\t8 Sep 2003 04:04:58 -0000\\n> > @@ -1735,20 +1735,71 @@\\n> > Quantities of days, hours, minutes, and seconds can be specified without\\n> > explicit unit markings. For example, '1 12:59:10' is read\\n> > the same as '1 day 12 hours 59 min 10 sec'.\\n> > \\n> >\\n> > \\n> > The optional precision\\n> > p should be between 0 and 6, and\\n> > defaults to the precision of the input literal.\\n> > \\n> > +\\n> > +\\n> > + \\n> > + Alternatively, interval values can be written as\\n> > + ISO 8601 time intervals, using the \\\"Format with time-unit designators\\\".\\n> > + This format always starts with the character 'P', followed\\n> > + by a string of values followed by single character time-unit designators.\\n> > + A 'T' separates the date and time parts of the interval.\\n> > + \\n> > +\\n> > + \\n> > + Format: PnYnMnDTnHnMnS\\n> > + \\n> > + \\n> > + In this format, 'n' gets replaced by a number, and\\n> > + Y represents years,\\n> > + M (in the date part) months,\\n> > + D months,\\n> > + H hours,\\n> > + M (in the time part) minutes,\\n> > + and S seconds.\\n> > + \\n> > +\\n> > +\\n> > + \\n> > +\\t Interval Example\\n> > +\\t \\n> > +\\t\\t\\n> > +\\t\\t \\n> > +\\t\\t Traditional\\n> > +\\t\\t ISO-8601 time-interval\\n> > +\\t\\t
\\n> > +\\t\\t\\n> > +\\t\\t\\n> > +\\t\\t \\n> > +\\t\\t 1 month\\n> > +\\t\\t P1M\\n> > +\\t\\t
\\n> > +\\t\\t \\n> > +\\t\\t 1 hour 30 minutes\\n> > +\\t\\t PT1H30M\\n> > +\\t\\t
\\n> > +\\t\\t \\n> > +\\t\\t 2 years 10 months 15 days 10 hours 30 minutes 20 seconds\\n> > +\\t\\t P2Y10M15DT10H30M20S\\n> > +\\t\\t
\\n> > +\\t\\t\\n> > +\\t \\n> > +\\t
\\n> > +\\n> > + \\n> > \\n> >\\n> > \\n> > Special Values\\n> >\\n> > \\n> > time\\n> > constants\\n> > \\n> >\\n> > Index: src/backend/utils/adt/datetime.c\\n> > ===================================================================\\n> > RCS file: /projects/cvsroot/pgsql-server/src/backend/utils/adt/datetime.c,v\\n> > retrieving revision 1.116\\n> > diff -u -1 -0 -r1.116 datetime.c\\n> > --- src/backend/utils/adt/datetime.c\\t27 Aug 2003 23:29:28 -0000\\t1.116\\n> > +++ src/backend/utils/adt/datetime.c\\t8 Sep 2003 04:04:59 -0000\\n> > @@ -30,20 +30,21 @@\\n> > \\t\\t\\t struct tm * tm, fsec_t *fsec, int *is2digits);\\n> > static int DecodeNumberField(int len, char *str,\\n> > \\t\\t\\t\\t int fmask, int *tmask,\\n> > \\t\\t\\t\\t struct tm * tm, fsec_t *fsec, int *is2digits);\\n> > static int DecodeTime(char *str, int fmask, int *tmask,\\n> > \\t\\t struct tm * tm, fsec_t *fsec);\\n> > static int\\tDecodeTimezone(char *str, int *tzp);\\n> > static datetkn *datebsearch(char *key, datetkn *base, unsigned int nel);\\n> > static int\\tDecodeDate(char *str, int fmask, int *tmask, struct tm * tm);\\n> > static void TrimTrailingZeros(char *str);\\n> > +static int DecodeISO8601Interval(char **field, int *ftype, int nf, int *dtype, struct tm * tm, fsec_t *fsec);\\n> >\\n> >\\n> > int\\t\\t\\tday_tab[2][13] = {\\n> > \\t{31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31, 0},\\n> > {31, 29, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31, 0}};\\n> >\\n> > char\\t *months[] = {\\\"Jan\\\", \\\"Feb\\\", \\\"Mar\\\", \\\"Apr\\\", \\\"May\\\", \\\"Jun\\\",\\n> > \\\"Jul\\\", \\\"Aug\\\", \\\"Sep\\\", \\\"Oct\\\", \\\"Nov\\\", \\\"Dec\\\", NULL};\\n> >\\n> > char\\t *days[] = {\\\"Sunday\\\", \\\"Monday\\\", \\\"Tuesday\\\", \\\"Wednesday\\\",\\n> > @@ -2872,30 +2873,271 @@\\n> > \\t\\t\\tdefault:\\n> > \\t\\t\\t\\t*val = tp->value;\\n> > \\t\\t\\t\\tbreak;\\n> > \\t\\t}\\n> > \\t}\\n> >\\n> > \\treturn type;\\n> > }\\n> >\\n> >\\n> > +void adjust_fval(double fval,struct tm * tm, fsec_t *fsec, int scale);\\n> > +{\\n> > +\\tint\\tsec;\\n> > +\\tfval\\t *= scale;\\n> > +\\tsec\\t\\t = fval;\\n> > +\\ttm->tm_sec += sec;\\n> > +#ifdef HAVE_INT64_TIMESTAMP\\n> > +\\t*fsec\\t += ((fval - sec) * 1000000);\\n> > +#else\\n> > +\\t*fsec\\t += (fval - sec);\\n> > +#endif\\n> > +}\\n> > +\\n> > +\\n> > +/* DecodeISO8601Interval()\\n> > + *\\n> > + * Check if it's a ISO 8601 Section 5.5.4.2 \\\"Representation of\\n> > + * time-interval by duration only.\\\"\\n> > + * Basic extended format: PnYnMnDTnHnMnS\\n> > + * PnW\\n> > + * For more info.\\n> > + * http://www.astroclark.freeserve.co.uk/iso8601/index.html\\n> > + * ftp://ftp.qsl.net/pub/g1smd/154N362_.PDF\\n> > + *\\n> > + * Examples: P1D for 1 day\\n> > + * PT1H for 1 hour\\n> > + * P2Y6M7DT1H30M for 2 years, 6 months, 7 days 1 hour 30 min\\n> > + *\\n> > + * The first field is exactly \\\"p\\\" or \\\"pt\\\" it may be of this type.\\n> > + *\\n> > + * Returns -1 if the field is not of this type.\\n> > + *\\n> > + * It pretty strictly checks the spec, with the two exceptions\\n> > + * that a week field ('W') may coexist with other units, and that\\n> > + * this function allows decimals in fields other than the least\\n> > + * significant units.\\n> > + */\\n> > +int\\n> > +DecodeISO8601Interval(char **field, int *ftype, int nf, int *dtype, struct tm * tm, fsec_t *fsec)\\n> > +{\\n> > +\\tchar\\t *cp;\\n> > +\\tint\\t\\t\\tfmask = 0,\\n> > +\\t\\t\\t\\ttmask;\\n> > +\\tint\\t\\t\\tval;\\n> > +\\tdouble\\t\\tfval;\\n> > +\\tint\\t\\t\\targ;\\n> > +\\tint\\t\\t\\tdatepart;\\n> > +\\n> > + /*\\n> > +\\t * An ISO 8601 \\\"time-interval by duration only\\\" must start\\n> > +\\t * with a 'P'. If it contains a date-part, 'p' will be the\\n> > +\\t * only character in the field. If it contains no date part\\n> > +\\t * it will contain exactly to characters 'PT' indicating a\\n> > +\\t * time part.\\n> > +\\t * Anything else is illegal and will be treated like a\\n> > +\\t * traditional postgresql interval.\\n> > +\\t */\\n> > + if (!(field[0][0] == 'p' &&\\n> > + ((field[0][1] == 0) || (field[0][1] == 't' && field[0][2] == 0))))\\n> > +\\t{\\n> > +\\t return -1;\\n> > +\\t}\\n> > +\\n> > +\\n> > + /*\\n> > +\\t * If the first field is exactly 1 character ('P'), it starts\\n> > +\\t * with date elements. Otherwise it's two characters ('PT');\\n> > +\\t * indicating it starts with a time part.\\n> > +\\t */\\n> > +\\tdatepart = (field[0][1] == 0);\\n> > +\\n> > +\\t/*\\n> > +\\t * Every value must have a unit, so we require an even\\n> > +\\t * number of value/unit pairs. Therefore we require an\\n> > +\\t * odd nubmer of fields, including the prefix 'P'.\\n> > +\\t */\\n> > +\\tif ((nf & 1) == 0)\\n> > +\\t\\treturn -1;\\n> > +\\n> > +\\t/*\\n> > +\\t * Process pairs of fields at a time.\\n> > +\\t */\\n> > +\\tfor (arg = 1 ; arg < nf ; arg+=2)\\n> > +\\t{\\n> > +\\t\\tchar * value = field[arg ];\\n> > +\\t\\tchar * units = field[arg+1];\\n> > +\\n> > +\\t\\t/*\\n> > +\\t\\t * The value part must be a number.\\n> > +\\t\\t */\\n> > +\\t\\tif (ftype[arg] != DTK_NUMBER)\\n> > +\\t\\t\\treturn -1;\\n> > +\\n> > +\\t\\t/*\\n> > +\\t\\t * extract the number, almost exactly like the non-ISO interval.\\n> > +\\t\\t */\\n> > +\\t\\tval = strtol(value, &cp, 10);\\n> > +\\n> > +\\t\\t/*\\n> > +\\t\\t * One difference from the normal postgresql interval below...\\n> > +\\t\\t * ISO 8601 states that \\\"Of these, the comma is the preferred\\n> > +\\t\\t * sign\\\" so I allow it here for locales that support it.\\n> > +\\t\\t * Note: Perhaps the old-style interval code below should\\n> > +\\t\\t * allow for this too, but I didn't want to risk backward\\n> > +\\t\\t * compatability.\\n> > +\\t\\t */\\n> > +\\t\\tif (*cp == '.' || *cp == ',')\\n> > +\\t\\t{\\n> > +\\t\\t\\tfval = strtod(cp, &cp);\\n> > +\\t\\t\\tif (*cp != '\\\\0')\\n> > +\\t\\t\\t\\treturn -1;\\n> > +\\n> > +\\t\\t\\tif (val < 0)\\n> > +\\t\\t\\t\\tfval = -(fval);\\n> > +\\t\\t}\\n> > +\\t\\telse if (*cp == '\\\\0')\\n> > +\\t\\t\\tfval = 0;\\n> > +\\t\\telse\\n> > +\\t\\t\\treturn -1;\\n> > +\\n> > +\\n> > +\\t\\tif (datepart)\\n> > +\\t\\t{\\n> > +\\t\\t\\t/*\\n> > +\\t\\t\\t * All the 8601 unit specifiers are 1 character, but may\\n> > +\\t\\t\\t * be followed by a 'T' character if transitioning between\\n> > +\\t\\t\\t * the date part and the time part. If it's not either\\n> > +\\t\\t\\t * one character or two characters with the second being 't'\\n> > +\\t\\t\\t * it's an error.\\n> > +\\t\\t\\t */\\n> > +\\t\\t\\tif (!(units[1] == 0 || (units[1] == 't' && units[2] == 0)))\\n> > +\\t\\t\\t\\treturn -1;\\n> > +\\n> > +\\t\\t\\tif (units[1] == 't')\\n> > +\\t\\t\\t\\tdatepart = 0;\\n> > +\\n> > +\\t\\t\\tswitch (units[0]) /* Y M D W */\\n> > +\\t\\t\\t{\\n> > +\\t\\t\\t\\tcase 'd':\\n> > +\\t\\t\\t\\t\\ttm->tm_mday += val;\\n> > +\\t\\t\\t\\t\\tif (fval != 0)\\n> > +\\t\\t\\t\\t\\t adjust_fval(fval,tm,fsec, 86400);\\n> > +\\t\\t\\t\\t\\ttmask = ((fmask & DTK_M(DAY)) ? 0 : DTK_M(DAY));\\n> > +\\t\\t\\t\\t\\tbreak;\\n> > +\\n> > +\\t\\t\\t\\tcase 'w':\\n> > +\\t\\t\\t\\t\\ttm->tm_mday += val * 7;\\n> > +\\t\\t\\t\\t\\tif (fval != 0)\\n> > +\\t\\t\\t\\t\\t adjust_fval(fval,tm,fsec,7 * 86400);\\n> > +\\t\\t\\t\\t\\ttmask = ((fmask & DTK_M(DAY)) ? 0 : DTK_M(DAY));\\n> > +\\t\\t\\t\\t\\tbreak;\\n> > +\\n> > +\\t\\t\\t\\tcase 'm':\\n> > +\\t\\t\\t\\t\\ttm->tm_mon += val;\\n> > +\\t\\t\\t\\t\\tif (fval != 0)\\n> > +\\t\\t\\t\\t\\t adjust_fval(fval,tm,fsec,30 * 86400);\\n> > +\\t\\t\\t\\t\\ttmask = DTK_M(MONTH);\\n> > +\\t\\t\\t\\t\\tbreak;\\n> > +\\n> > +\\t\\t\\t\\tcase 'y':\\n> > +\\t\\t\\t\\t\\t/*\\n> > +\\t\\t\\t\\t\\t * Why can fractional months produce seconds,\\n> > +\\t\\t\\t\\t\\t * but fractional years can't? Well the older\\n> > +\\t\\t\\t\\t\\t * interval code below has the same property\\n> > +\\t\\t\\t\\t\\t * so this one follows the other one too.\\n> > +\\t\\t\\t\\t\\t */\\n> > +\\t\\t\\t\\t\\ttm->tm_year += val;\\n> > +\\t\\t\\t\\t\\tif (fval != 0)\\n> > +\\t\\t\\t\\t\\t\\ttm->tm_mon += (fval * 12);\\n> > +\\t\\t\\t\\t\\ttmask = ((fmask & DTK_M(YEAR)) ? 0 : DTK_M(YEAR));\\n> > +\\t\\t\\t\\t\\tbreak;\\n> > +\\n> > +\\t\\t\\t\\tdefault:\\n> > +\\t\\t\\t\\t\\treturn -1; /* invald date unit prefix */\\n> > +\\t\\t\\t}\\n> > +\\t\\t}\\n> > +\\t\\telse\\n> > +\\t\\t{\\n> > +\\t\\t\\t/*\\n> > +\\t\\t\\t * ISO 8601 time part.\\n> > +\\t\\t\\t * In the time part, only one-character\\n> > +\\t\\t\\t * unit prefixes are allowed. If it's more\\n> > +\\t\\t\\t * than one character, it's not a valid ISO 8601\\n> > +\\t\\t\\t * time interval by duration.\\n> > +\\t\\t\\t */\\n> > +\\t\\t\\tif (units[1] != 0)\\n> > +\\t\\t\\t\\treturn -1;\\n> > +\\n> > +\\t\\t\\tswitch (units[0]) /* H M S */\\n> > +\\t\\t\\t{\\n> > +\\t\\t\\t\\tcase 's':\\n> > +\\t\\t\\t\\t\\ttm->tm_sec += val;\\n> > +#ifdef HAVE_INT64_TIMESTAMP\\n> > +\\t\\t\\t\\t\\t*fsec += (fval * 1000000);\\n> > +#else\\n> > +\\t\\t\\t\\t\\t*fsec += fval;\\n> > +#endif\\n> > +\\t\\t\\t\\t\\ttmask = DTK_M(SECOND);\\n> > +\\t\\t\\t\\t\\tbreak;\\n> > +\\n> > +\\t\\t\\t\\tcase 'm':\\n> > +\\t\\t\\t\\t\\ttm->tm_min += val;\\n> > +\\t\\t\\t\\t\\tif (fval != 0)\\n> > +\\t\\t\\t\\t\\t adjust_fval(fval,tm,fsec,60);\\n> > +\\t\\t\\t\\t\\ttmask = DTK_M(MINUTE);\\n> > +\\t\\t\\t\\t\\tbreak;\\n> > +\\n> > +\\t\\t\\t\\tcase 'h':\\n> > +\\t\\t\\t\\t\\ttm->tm_hour += val;\\n> > +\\t\\t\\t\\t\\tif (fval != 0)\\n> > +\\t\\t\\t\\t\\t adjust_fval(fval,tm,fsec,3600);\\n> > +\\t\\t\\t\\t\\ttmask = DTK_M(HOUR);\\n> > +\\t\\t\\t\\t\\tbreak;\\n> > +\\n> > +\\t\\t\\t\\tdefault:\\n> > +\\t\\t\\t\\t\\treturn -1; /* invald time unit prefix */\\n> > +\\t\\t\\t}\\n> > +\\t\\t}\\n> > +\\t\\tfmask |= tmask;\\n> > +\\t}\\n> > +\\n> > +\\tif (*fsec != 0)\\n> > +\\t{\\n> > +\\t\\tint\\t\\t\\tsec;\\n> > +\\n> > +#ifdef HAVE_INT64_TIMESTAMP\\n> > +\\t\\tsec = (*fsec / INT64CONST(1000000));\\n> > +\\t\\t*fsec -= (sec * INT64CONST(1000000));\\n> > +#else\\n> > +\\t\\tTMODULO(*fsec, sec, 1e0);\\n> > +#endif\\n> > +\\t\\ttm->tm_sec += sec;\\n> > +\\t}\\n> > +\\treturn (fmask != 0) ? 0 : -1;\\n> > +}\\n> > +\\n> > +\\n> > /* DecodeInterval()\\n> > * Interpret previously parsed fields for general time interval.\\n> > * Returns 0 if successful, DTERR code if bogus input detected.\\n> > *\\n> > * Allow \\\"date\\\" field DTK_DATE since this could be just\\n> > *\\tan unsigned floating point number. - thomas 1997-11-16\\n> > *\\n> > * Allow ISO-style time span, with implicit units on number of days\\n> > *\\tpreceding an hh:mm:ss field. - thomas 1998-04-30\\n> > + *\\n> > + * Allow ISO-8601 style \\\"Representation of time-interval by duration only\\\"\\n> > + * of the format 'PnYnMnDTnHnMnS' and 'PnW' - ron 2003-08-30\\n> > */\\n> > +\\n> > int\\n> > DecodeInterval(char **field, int *ftype, int nf, int *dtype, struct tm * tm, fsec_t *fsec)\\n> > {\\n> > \\tint\\t\\t\\tis_before = FALSE;\\n> > \\tchar\\t *cp;\\n> > \\tint\\t\\t\\tfmask = 0,\\n> > \\t\\t\\t\\ttmask,\\n> > \\t\\t\\t\\ttype;\\n> > \\tint\\t\\t\\ti;\\n> > \\tint\\t\\t\\tdterr;\\n> > @@ -2906,20 +3148,37 @@\\n> >\\n> > \\ttype = IGNORE_DTF;\\n> > \\ttm->tm_year = 0;\\n> > \\ttm->tm_mon = 0;\\n> > \\ttm->tm_mday = 0;\\n> > \\ttm->tm_hour = 0;\\n> > \\ttm->tm_min = 0;\\n> > \\ttm->tm_sec = 0;\\n> > \\t*fsec = 0;\\n> >\\n> > +\\t/*\\n> > +\\t * Check if it's a ISO 8601 Section 5.5.4.2 \\\"Representation of\\n> > + * time-interval by duration only.\\\"\\n> > +\\t * Basic extended format: PnYnMnDTnHnMnS\\n> > +\\t * PnW\\n> > +\\t * http://www.astroclark.freeserve.co.uk/iso8601/index.html\\n> > +\\t * ftp://ftp.qsl.net/pub/g1smd/154N362_.PDF\\n> > +\\t * Examples: P1D for 1 day\\n> > +\\t * PT1H for 1 hour\\n> > +\\t * P2Y6M7DT1H30M for 2 years, 6 months, 7 days 1 hour 30 min\\n> > +\\t *\\n> > +\\t * The first field is exactly \\\"p\\\" or \\\"pt\\\" it may be of this type.\\n> > +\\t */\\n> > +\\tif (DecodeISO8601Interval(field,ftype,nf,dtype,tm,fsec) == 0) {\\n> > +\\t return 0;\\n> > + }\\n> > +\\n> > \\t/* read through list backwards to pick up units before values */\\n> > \\tfor (i = nf - 1; i >= 0; i--)\\n> > \\t{\\n> > \\t\\tswitch (ftype[i])\\n> > \\t\\t{\\n> > \\t\\t\\tcase DTK_TIME:\\n> > \\t\\t\\t\\tdterr = DecodeTime(field[i], fmask, &tmask, tm, fsec);\\n> > \\t\\t\\t\\tif (dterr)\\n> > \\t\\t\\t\\t\\treturn dterr;\\n> > \\t\\t\\t\\ttype = DTK_DAY;\\n> > @@ -2983,20 +3242,21 @@\\n> > \\t\\t\\t\\t}\\n> > \\t\\t\\t\\t/* DROP THROUGH */\\n> >\\n> > \\t\\t\\tcase DTK_DATE:\\n> > \\t\\t\\tcase DTK_NUMBER:\\n> > \\t\\t\\t\\tval = strtol(field[i], &cp, 10);\\n> >\\n> > \\t\\t\\t\\tif (type == IGNORE_DTF)\\n> > \\t\\t\\t\\t\\ttype = DTK_SECOND;\\n> >\\n> > +\\t\\t\\t\\t/* should this allow ',' for locales that use it ? */\\n> > \\t\\t\\t\\tif (*cp == '.')\\n> > \\t\\t\\t\\t{\\n> > \\t\\t\\t\\t\\tfval = strtod(cp, &cp);\\n> > \\t\\t\\t\\t\\tif (*cp != '\\\\0')\\n> > \\t\\t\\t\\t\\t\\treturn DTERR_BAD_FORMAT;\\n> >\\n> > \\t\\t\\t\\t\\tif (val < 0)\\n> > \\t\\t\\t\\t\\t\\tfval = -(fval);\\n> > \\t\\t\\t\\t}\\n> > \\t\\t\\t\\telse if (*cp == '\\\\0')\\n> >\\n> > ===================================================================\\n> >\\n> > ---------------------------(end of broadcast)---------------------------\\n> > TIP 6: Have you searched our list archives?\\n> >\\n> > http://archives.postgresql.org\\n> >\\n>\\n>\\n\\n-- \\nPeter Eisentraut peter_e@gmx.net\\n\\n\",\n \"msg_date\": \"Mon, 1 Dec 2003 07:20:21 +0100 (CET)\",\n \"msg_from\": \"Peter Eisentraut \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\"\n },\n {\n \"msg_contents\": \"Peter Eisentraut wrote:\\n> Bruce Momjian writes:\\n> \\n> > Is this ready for application? It looks good to me. However, there is\\n> > an \\\"Open issues\\\" section.\\n> \\n> It would be more useful to implement the SQL standard for intervals first\\n> instead of inventing more nonstandard formats for it.\\n\\nOK, patch removed from queue.\\n\\n-- \\n Bruce Momjian | http://candle.pha.pa.us\\n pgman@candle.pha.pa.us | (610) 359-1001\\n + If your life is a hard drive, | 13 Roberts Road\\n + Christ can be your backup. | Newtown Square, Pennsylvania 19073\\n\",\n \"msg_date\": \"Mon, 1 Dec 2003 09:12:35 -0500 (EST)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\"\n },\n {\n \"msg_contents\": \"> -----Original Message-----\\n> \\n> Is this ready for application? It looks good to me. However, there is\\n> an \\\"Open issues\\\" section.\\n\\nIn my mind there were two categories of open issues\\n a) ones that are 100% backward (such as the comment about \\n outputting this format)\\nand\\n b) ones that aren't (such as deprecating the current\\n postgresql shorthand of \\n '1Y1M'::interval = 1 year 1 minute\\n in favor of the ISO-8601\\n 'P1Y1M'::interval = 1 year 1 month.\\n\\nAttached is a patch that addressed all the discussed issues that\\ndid not break backward compatability, including the ability to\\noutput ISO-8601 compliant intervals by setting datestyle to\\niso8601basic.\\n\\n Ron\",\n \"msg_date\": \"Mon, 1 Dec 2003 12:50:47 -0800\",\n \"msg_from\": \"\\\"Ron Mayer\\\" \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\\n\\tdeignators\\\"\"\n },\n {\n \"msg_contents\": \"Peter wrote:\\n> \\n> It would be more useful to implement the SQL standard for intervals first\\n> instead of inventing more nonstandard formats for it.\\n\\nMuch of the postgresql docs talks about ISO-8601 formats, so I would think\\nof the patch more as a standards-based improvemnt for the current interval \\nshortand. \\n\\nFor example, where today postgresql accepts \\\"1Y1M\\\" as '1 year 1 minute',\\nwith the patch, the ISO-8601-standard \\\"P1Y1M\\\" would mean '1 year 1 month'.\\n\\nI would be happy to implement the SQL-standard-intervals as well if \\nsomeone can point me to that spec. It's just that the system I worked\\nwith happend to exchange data with ISO8601 time intervals.\\n\\n Ron Mayer\\n\\n[Moderators of psql-patches... I first replied to Peter from a\\nseparate (personal) account. I think this reply is now in the \\nmoderation queue). In retrospect I thought it might be better if \\nthe archives had the whole thread from the original email address. \\nIf it's not too late, could you reject my post in the moderation queue?\\nIf not, sorry for the spam.]\\n\\n\\n\",\n \"msg_date\": \"Mon, 1 Dec 2003 13:18:20 -0800\",\n \"msg_from\": \"\\\"Ron Mayer\\\" \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\"\n },\n {\n \"msg_contents\": \"[sNip]\\n> ISO 8601 gives more specific names.\\n> \\n> ISO 8601 Basic Format: P2Y10M15DT10H20M30S\\n> ISO 8601 Alternative Format: P00021015T102030\\n> ISO 8601 Extended Format: P0002-10-15T10:20:30\\n> \\n> In a way, the Extended Format is kinda nice, since it�s\\n> almost human readable.\\n> \\n> I could put in both the basic and extended ones, and\\n> call the dateformats �iso8601basic� and �iso8601extended�.\\n> The negative is that to do �iso8601basic� right, I�d also\\n> have to tweak the �date� and �time� parts of the code too.\\n\\n \\tPerhaps all three formats should be supported, and if the following \\nnames were all valid things could be simplified further too:\\n\\n \\t \\tiso8601basic\\n \\t \\tiso8601bas\\n \\t \\tiso8601alternative\\n \\t \\tiso8601alt\\n \\t \\tiso8601extended\\n \\t \\tiso8601ext\\n\\n \\tThe reason for allowing shorter names is to simplify database \\nmanagement for anyone who may need to store the format name in a column for \\nsome reason (I can't think of one now, but I get a feeling that someone \\nwill want to do this type of thing in the future).\\n\\n \\tFor that matter, the first letter could be used instead of the first \\nthree for the short versions. Any thoughts on this?\\n\\n-- \\nRandolf Richardson - rr@8x.ca\\nVancouver, British Columbia, Canada\\n\\nPlease do not eMail me directly when responding\\nto my postings in the newsgroups.\\n\",\n \"msg_date\": \"Wed, 3 Dec 2003 10:20:23 +0000 (UTC)\",\n \"msg_from\": \"Randolf Richardson \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 'Time Intervals' of the 'format with time-unit\\n\\tdeignators'\"\n },\n {\n \"msg_contents\": \"Randolf Richardson wrote:\\n> [sNip]\\n> > ISO 8601 gives more specific names.\\n> > \\n> > ISO 8601 Basic Format: P2Y10M15DT10H20M30S\\n> > ISO 8601 Alternative Format: P00021015T102030\\n> > ISO 8601 Extended Format: P0002-10-15T10:20:30\\n> > \\n> > In a way, the Extended Format is kinda nice, since it�s\\n> > almost human readable.\\n> > \\n> > I could put in both the basic and extended ones, and\\n> > call the dateformats �iso8601basic� and �iso8601extended�.\\n> > The negative is that to do �iso8601basic� right, I�d also\\n> > have to tweak the �date� and �time� parts of the code too.\\n> \\n> \\tPerhaps all three formats should be supported, and if the following \\n> names were all valid things could be simplified further too:\\n> \\n> \\t \\tiso8601basic\\n> \\t \\tiso8601bas\\n> \\t \\tiso8601alternative\\n> \\t \\tiso8601alt\\n> \\t \\tiso8601extended\\n> \\t \\tiso8601ext\\n> \\n> \\tThe reason for allowing shorter names is to simplify database \\n> management for anyone who may need to store the format name in a column for \\n> some reason (I can't think of one now, but I get a feeling that someone \\n> will want to do this type of thing in the future).\\n> \\n> \\tFor that matter, the first letter could be used instead of the first \\n> three for the short versions. Any thoughts on this?\\n\\nJust go with the full spellings.\\n\\n\\n-- \\n Bruce Momjian | http://candle.pha.pa.us\\n pgman@candle.pha.pa.us | (610) 359-1001\\n + If your life is a hard drive, | 13 Roberts Road\\n + Christ can be your backup. | Newtown Square, Pennsylvania 19073\\n\",\n \"msg_date\": \"Tue, 9 Dec 2003 18:28:20 -0500 (EST)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 'Time Intervals' of the 'format with time-unit\"\n },\n {\n \"msg_contents\": \"\\nYour patch has been added to the PostgreSQL unapplied patches list at:\\n\\n\\thttp://momjian.postgresql.org/cgi-bin/pgpatches\\n\\nI will try to apply it within the next 48 hours.\\n\\n---------------------------------------------------------------------------\\n\\n\\nRon Mayer wrote:\\n> > -----Original Message-----\\n> > \\n> > Is this ready for application? It looks good to me. However, there is\\n> > an \\\"Open issues\\\" section.\\n> \\n> In my mind there were two categories of open issues\\n> a) ones that are 100% backward (such as the comment about \\n> outputting this format)\\n> and\\n> b) ones that aren't (such as deprecating the current\\n> postgresql shorthand of \\n> '1Y1M'::interval = 1 year 1 minute\\n> in favor of the ISO-8601\\n> 'P1Y1M'::interval = 1 year 1 month.\\n> \\n> Attached is a patch that addressed all the discussed issues that\\n> did not break backward compatability, including the ability to\\n> output ISO-8601 compliant intervals by setting datestyle to\\n> iso8601basic.\\n> \\n> Ron\\n\\n[ Attachment, skipping... ]\\n\\n> \\n> ---------------------------(end of broadcast)---------------------------\\n> TIP 8: explain analyze is your friend\\n\\n-- \\n Bruce Momjian | http://candle.pha.pa.us\\n pgman@candle.pha.pa.us | (610) 359-1001\\n + If your life is a hard drive, | 13 Roberts Road\\n + Christ can be your backup. | Newtown Square, Pennsylvania 19073\\n\",\n \"msg_date\": \"Mon, 15 Dec 2003 18:30:09 -0500 (EST)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\"\n },\n {\n \"msg_contents\": \"Bruce Momjian wrote:\\n> Your patch has been added to the PostgreSQL unapplied patches list\\n> at:\\n>\\n> \\thttp://momjian.postgresql.org/cgi-bin/pgpatches\\n>\\n> I will try to apply it within the next 48 hours.\\n\\nI keep reading about open issues, and deprecating certain things, and \\npatch removed, and patch readded. What is going on?\\n\\n\",\n \"msg_date\": \"Tue, 16 Dec 2003 01:06:32 +0100\",\n \"msg_from\": \"Peter Eisentraut \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\"\n },\n {\n \"msg_contents\": \"Peter Eisentraut wrote:\\n> Bruce Momjian wrote:\\n> > Your patch has been added to the PostgreSQL unapplied patches list\\n> > at:\\n> >\\n> > \\thttp://momjian.postgresql.org/cgi-bin/pgpatches\\n> >\\n> > I will try to apply it within the next 48 hours.\\n> \\n> I keep reading about open issues, and deprecating certain things, and \\n> patch removed, and patch readded. What is going on?\\n\\nI think the patch just added is OK, no?\\n\\n-- \\n Bruce Momjian | http://candle.pha.pa.us\\n pgman@candle.pha.pa.us | (610) 359-1001\\n + If your life is a hard drive, | 13 Roberts Road\\n + Christ can be your backup. | Newtown Square, Pennsylvania 19073\\n\",\n \"msg_date\": \"Mon, 15 Dec 2003 19:09:15 -0500 (EST)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\"\n },\n {\n \"msg_contents\": \"Bruce Momjian wrote:\\n> > I keep reading about open issues, and deprecating certain things,\\n> > and patch removed, and patch readded. What is going on?\\n>\\n> I think the patch just added is OK, no?\\n\\nI don't know, but earlier the identical patch was rejected by you.\\n\\n\",\n \"msg_date\": \"Tue, 16 Dec 2003 01:37:50 +0100\",\n \"msg_from\": \"Peter Eisentraut \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\"\n },\n {\n \"msg_contents\": \"Peter Eisentraut wrote:\\n> Bruce Momjian wrote:\\n> > > I keep reading about open issues, and deprecating certain things,\\n> > > and patch removed, and patch readded. What is going on?\\n> >\\n> > I think the patch just added is OK, no?\\n> \\n> I don't know, but earlier the identical patch was rejected by you.\\n\\nI thought he made an adjustment so no backward compatibility was broken.\\n\\n\\n-- \\n Bruce Momjian | http://candle.pha.pa.us\\n pgman@candle.pha.pa.us | (610) 359-1001\\n + If your life is a hard drive, | 13 Roberts Road\\n + Christ can be your backup. | Newtown Square, Pennsylvania 19073\\n\",\n \"msg_date\": \"Mon, 15 Dec 2003 19:39:00 -0500 (EST)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\"\n },\n {\n \"msg_contents\": \"Bruce Momjian wrote:\\n> Peter Eisentraut wrote:\\n> > Bruce Momjian wrote:\\n> > > > I keep reading about open issues, and deprecating certain\\n> > > > things, and patch removed, and patch readded. What is going\\n> > > > on?\\n> > >\\n> > > I think the patch just added is OK, no?\\n> >\\n> > I don't know, but earlier the identical patch was rejected by you.\\n>\\n> I thought he made an adjustment so no backward compatibility was\\n> broken.\\n\\nThen I wouldn't have said \\\"identical\\\".\\n\\n\",\n \"msg_date\": \"Tue, 16 Dec 2003 01:42:09 +0100\",\n \"msg_from\": \"Peter Eisentraut \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\"\n },\n {\n \"msg_contents\": \"Peter Eisentraut wrote:\\n> Bruce Momjian wrote:\\n> > Peter Eisentraut wrote:\\n> > > Bruce Momjian wrote:\\n> > > > > I keep reading about open issues, and deprecating certain\\n> > > > > things, and patch removed, and patch readded. What is going\\n> > > > > on?\\n> > > >\\n> > > > I think the patch just added is OK, no?\\n> > >\\n> > > I don't know, but earlier the identical patch was rejected by you.\\n> >\\n> > I thought he made an adjustment so no backward compatibility was\\n> > broken.\\n> \\n> Then I wouldn't have said \\\"identical\\\".\\n\\nOK, can anyone raise an objection to the patch. The new description\\nmeans to me that he addressed our concerns and that my original\\nhesitation was unwarranted.\\n\\n-- \\n Bruce Momjian | http://candle.pha.pa.us\\n pgman@candle.pha.pa.us | (610) 359-1001\\n + If your life is a hard drive, | 13 Roberts Road\\n + Christ can be your backup. | Newtown Square, Pennsylvania 19073\\n\",\n \"msg_date\": \"Mon, 15 Dec 2003 19:44:11 -0500 (EST)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\"\n },\n {\n \"msg_contents\": \"\\nPatch applied. Thanks.\\n\\n---------------------------------------------------------------------------\\n\\n\\nRon Mayer wrote:\\n> > -----Original Message-----\\n> > \\n> > Is this ready for application? It looks good to me. However, there is\\n> > an \\\"Open issues\\\" section.\\n> \\n> In my mind there were two categories of open issues\\n> a) ones that are 100% backward (such as the comment about \\n> outputting this format)\\n> and\\n> b) ones that aren't (such as deprecating the current\\n> postgresql shorthand of \\n> '1Y1M'::interval = 1 year 1 minute\\n> in favor of the ISO-8601\\n> 'P1Y1M'::interval = 1 year 1 month.\\n> \\n> Attached is a patch that addressed all the discussed issues that\\n> did not break backward compatability, including the ability to\\n> output ISO-8601 compliant intervals by setting datestyle to\\n> iso8601basic.\\n> \\n> Ron\\n\\n[ Attachment, skipping... ]\\n\\n> \\n> ---------------------------(end of broadcast)---------------------------\\n> TIP 8: explain analyze is your friend\\n\\n-- \\n Bruce Momjian | http://candle.pha.pa.us\\n pgman@candle.pha.pa.us | (610) 359-1001\\n + If your life is a hard drive, | 13 Roberts Road\\n + Christ can be your backup. | Newtown Square, Pennsylvania 19073\\n\",\n \"msg_date\": \"Sat, 20 Dec 2003 10:32:45 -0500 (EST)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: ISO 8601 \\\"Time Intervals\\\" of the \\\"format with time-unit\"\n },\n {\n \"msg_contents\": \"Hello,\\n\\nIs there ANY chance to recover data from a database system that suffered disk \\ncrash, and is not missing the data/global directory?\\n\\nVersion is 7.2.4. Database files seem to be intact as well as pg_clog and \\npg_xlog directories.\\n\\nThanks in advance for any ideas.\\n\\nDaniel\\n\\n\",\n \"msg_date\": \"Mon, 23 Aug 2004 16:00:01 +0300\",\n \"msg_from\": \"Daniel Kalchev \",\n \"msg_from_op\": false,\n \"msg_subject\": \"missing data/global\"\n },\n {\n \"msg_contents\": \"Daniel Kalchev writes:\\n> Is there ANY chance to recover data from a database system that suffered disk\\n> crash, and is not missing the data/global directory?\\n> Version is 7.2.4. Database files seem to be intact as well as pg_clog and \\n> pg_xlog directories.\\n\\nThe hard part I think would be reconstructing pg_database, because you'd\\nneed to get the database OIDs right. I can't think of any way to do\\nthat that doesn't involve poking at the file with a hex editor.\\n\\nHere's a sketch of how I'd proceed:\\n\\n1. Make a tar backup of what you have! That way you can start over\\nafter you screw up ;-)\\n\\n2. I assume you know the names and properties of your databases, users,\\nand groups if any; also the SYSID numbers for the users and groups.\\nA recent pg_dumpall script would be a good place to get this info.\\n\\n3. You're also going to need to figure out the OIDs of your databases\\n(the OIDs are the same as the names of their subdirectories under\\n$PGDATA/base). Possibly you can do this just from directory/file sizes.\\nNote that template1 should be OID 1, and template0 will have the next\\nlowest number (probably 16555, in 7.2).\\n\\n4. Initdb a scratch database in some other place (or move aside your\\nexisting files, if that seems safer). In this scratch DB, create\\ndatabases, users, and groups to match your old setup. You should be\\nable to duplicate everything except the database OIDs using standard\\nSQL commands.\\n\\n5. Shut down scratch postmaster, then hex-edit pg_database to insert the\\ncorrect OIDs. Use pg_filedump or a similar tool to verify that you did\\nthis properly.\\n\\n6. Restart scratch postmaster, and VACUUM FREEZE pg_database, pg_shadow,\\nand pg_group (from any database). This will make the next step safe.\\n\\n7. Stop scratch postmaster, and then copy over its $PGDATA/global\\ndirectory into the old DB.\\n\\n8. Cross your fingers and start postmaster ...\\n\\nThis will probably *not* work if you had been doing anything to\\npg_database, pg_shadow, or pg_group between your last checkpoint and the\\ncrash, because the reconstructed tables are not going to be physically\\nidentical to what they were before, so any actions replayed from WAL\\nagainst those tables will be wrong. Hopefully you won't have that\\nproblem. If you do, it might work to shut down the postmaster and again\\ncopy the scratch $PGDATA/global directory into the old DB, thereby\\noverwriting what the WAL replay did. This is getting into the realm of\\nspeculation though.\\n\\n\\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Mon, 23 Aug 2004 14:46:17 -0400\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: missing data/global \"\n },\n {\n \"msg_contents\": \"Tom,\\n\\nThis is basically what I had in mind, but you described it better than I ever \\ncould :)\\n\\nWhat I need from this database system is just one database and probably not \\nall of the tables anyway (but some do seem valuable). This database happens to \\nbe second in the pg_dumpall script. The next databases are rather big (and I \\nactually have more recent backup and could eventually recreate the data) The \\nvaluable database hasn't had significant structure changes since the backup.\\n\\nLooking at the files, I am confident which is the proper database oid - if \\nthis cannot be properly fixed, is there .. reasonable way to dump data from \\nthe (heap) files?\\n\\nHere is what I have:\\n\\nsu-2.02# du\\n1747 ./base/1\\n1693 ./base/16555\\n1 ./base/77573557/pgsql_tmp\\n127036 ./base/77573557\\n1 ./base/13255137/pgsql_tmp\\n1379190 ./base/13255137\\n11246 ./base/95521309\\n1781 ./base/96388007\\n1 ./base/133512058/pgsql_tmp\\n11933861 ./base/133512058\\n13456555 ./base\\n98209 ./pg_xlog\\n41315 ./pg_clog\\n13596100 .\\n\\nMy database should be with oid 77573557, template0 is apparently 16555\\n\\nLet's see how all this works.\\n\\nDaniel\\n\\n>>>Tom Lane said:\\n > Daniel Kalchev writes:\\n > > Is there ANY chance to recover data from a database system that suffered d\\n isk\\n > > crash, and is not missing the data/global directory?\\n > > Version is 7.2.4. Database files seem to be intact as well as pg_clog and \\n > > pg_xlog directories.\\n > \\n > The hard part I think would be reconstructing pg_database, because you'd\\n > need to get the database OIDs right. I can't think of any way to do\\n > that that doesn't involve poking at the file with a hex editor.\\n > \\n > Here's a sketch of how I'd proceed:\\n > \\n > 1. Make a tar backup of what you have! That way you can start over\\n > after you screw up ;-)\\n > \\n > 2. I assume you know the names and properties of your databases, users,\\n > and groups if any; also the SYSID numbers for the users and groups.\\n > A recent pg_dumpall script would be a good place to get this info.\\n > \\n > 3. You're also going to need to figure out the OIDs of your databases\\n > (the OIDs are the same as the names of their subdirectories under\\n > $PGDATA/base). Possibly you can do this just from directory/file sizes.\\n > Note that template1 should be OID 1, and template0 will have the next\\n > lowest number (probably 16555, in 7.2).\\n > \\n > 4. Initdb a scratch database in some other place (or move aside your\\n > existing files, if that seems safer). In this scratch DB, create\\n > databases, users, and groups to match your old setup. You should be\\n > able to duplicate everything except the database OIDs using standard\\n > SQL commands.\\n > \\n > 5. Shut down scratch postmaster, then hex-edit pg_database to insert the\\n > correct OIDs. Use pg_filedump or a similar tool to verify that you did\\n > this properly.\\n > \\n > 6. Restart scratch postmaster, and VACUUM FREEZE pg_database, pg_shadow,\\n > and pg_group (from any database). This will make the next step safe.\\n > \\n > 7. Stop scratch postmaster, and then copy over its $PGDATA/global\\n > directory into the old DB.\\n > \\n > 8. Cross your fingers and start postmaster ...\\n > \\n > This will probably *not* work if you had been doing anything to\\n > pg_database, pg_shadow, or pg_group between your last checkpoint and the\\n > crash, because the reconstructed tables are not going to be physically\\n > identical to what they were before, so any actions replayed from WAL\\n > against those tables will be wrong. Hopefully you won't have that\\n > problem. If you do, it might work to shut down the postmaster and again\\n > copy the scratch $PGDATA/global directory into the old DB, thereby\\n > overwriting what the WAL replay did. This is getting into the realm of\\n > speculation though.\\n > \\n > \\t\\t\\tregards, tom lane\\n\\n\\n\",\n \"msg_date\": \"Mon, 23 Aug 2004 23:02:49 +0300\",\n \"msg_from\": \"Daniel Kalchev \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: missing data/global \"\n },\n {\n \"msg_contents\": \"If you're not missing your data dir, clog or xlog then what's the problem?\\n\\nDaniel Kalchev wrote:\\n> Hello,\\n> \\n> Is there ANY chance to recover data from a database system that suffered disk \\n> crash, and is not missing the data/global directory?\\n> \\n> Version is 7.2.4. Database files seem to be intact as well as pg_clog and \\n> pg_xlog directories.\\n> \\n> Thanks in advance for any ideas.\\n> \\n> Daniel\\n> \\n> \\n> ---------------------------(end of broadcast)---------------------------\\n> TIP 1: subscribe and unsubscribe commands go to majordomo@postgresql.org\\n\",\n \"msg_date\": \"Tue, 24 Aug 2004 09:11:26 +0800\",\n \"msg_from\": \"Christopher Kings-Lynne \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: missing data/global\"\n },\n {\n \"msg_contents\": \"data/base/global is missing and this is where postgres gets all it's startup \\ndata from (database oids, next oid, transaction id etc).\\n\\nLet's see how easy to recover from this it will turn to be.\\n\\nDaniel\\n\\n>>>Christopher Kings-Lynne said:\\n > If you're not missing your data dir, clog or xlog then what's the problem?\\n > \\n > Daniel Kalchev wrote:\\n > > Hello,\\n > > \\n > > Is there ANY chance to recover data from a database system that suffered d\\n isk \\n > > crash, and is not missing the data/global directory?\\n > > \\n > > Version is 7.2.4. Database files seem to be intact as well as pg_clog and \\n > > pg_xlog directories.\\n > > \\n > > Thanks in advance for any ideas.\\n > > \\n > > Daniel\\n > > \\n > > \\n > > ---------------------------(end of broadcast)---------------------------\\n > > TIP 1: subscribe and unsubscribe commands go to majordomo@postgresql.org\\n\\n\\n\",\n \"msg_date\": \"Tue, 24 Aug 2004 11:30:20 +0300\",\n \"msg_from\": \"Daniel Kalchev \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: missing data/global \"\n },\n {\n \"msg_contents\": \"Ah, you said 'is NOT missing'.\\n\\nChris\\n\\nDaniel Kalchev wrote:\\n\\n> data/base/global is missing and this is where postgres gets all it's startup \\n> data from (database oids, next oid, transaction id etc).\\n> \\n> Let's see how easy to recover from this it will turn to be.\\n> \\n> Daniel\\n> \\n> \\n>>>>Christopher Kings-Lynne said:\\n> \\n> > If you're not missing your data dir, clog or xlog then what's the problem?\\n> > \\n> > Daniel Kalchev wrote:\\n> > > Hello,\\n> > > \\n> > > Is there ANY chance to recover data from a database system that suffered d\\n> isk \\n> > > crash, and is not missing the data/global directory?\\n> > > \\n> > > Version is 7.2.4. Database files seem to be intact as well as pg_clog and \\n> > > pg_xlog directories.\\n> > > \\n> > > Thanks in advance for any ideas.\\n> > > \\n> > > Daniel\\n> > > \\n> > > \\n> > > ---------------------------(end of broadcast)---------------------------\\n> > > TIP 1: subscribe and unsubscribe commands go to majordomo@postgresql.org\\n> \\n\",\n \"msg_date\": \"Tue, 24 Aug 2004 17:07:22 +0800\",\n \"msg_from\": \"Christopher Kings-Lynne \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: missing data/global\"\n },\n {\n \"msg_contents\": \"Tom I did the following:\\n\\n(found out 7.2.3 does not have pg_database)\\n\\n1. saved old data etc.\\n\\n2. created new database, and the database. database oid was 16556;\\n\\n3. moved data/global to the old data directory.\\n\\n4. though, that postmaster would actually use the database oid to locate the \\ndirectory, then load everything from there.. old database oid was 77573557, so \\nI just linked this to 16556 in the data/base direcotry. (this might be the \\nfirst possible error)\\n\\nNow I can connect to the 'old' database, but get the error \\n\\nFATAL 1: Index pg_operator_oid_index is not a btree\\n\\n(if I run postmaster with -P I get not errors, but no tables as well).\\n\\nBy the way, I had to copy over the 'new' files from pg_clog and pg_xlog (this \\nis the second possible error) to get the postmaster running. Perhaps better \\nwould be to use pg_resetxlog or similar?\\n\\nDaniel\\n\\n>>>Tom Lane said:\\n > Daniel Kalchev writes:\\n > > Is there ANY chance to recover data from a database system that suffered d\\n isk\\n > > crash, and is not missing the data/global directory?\\n > > Version is 7.2.4. Database files seem to be intact as well as pg_clog and \\n > > pg_xlog directories.\\n > \\n > The hard part I think would be reconstructing pg_database, because you'd\\n > need to get the database OIDs right. I can't think of any way to do\\n > that that doesn't involve poking at the file with a hex editor.\\n > \\n > Here's a sketch of how I'd proceed:\\n > \\n > 1. Make a tar backup of what you have! That way you can start over\\n > after you screw up ;-)\\n > \\n > 2. I assume you know the names and properties of your databases, users,\\n > and groups if any; also the SYSID numbers for the users and groups.\\n > A recent pg_dumpall script would be a good place to get this info.\\n > \\n > 3. You're also going to need to figure out the OIDs of your databases\\n > (the OIDs are the same as the names of their subdirectories under\\n > $PGDATA/base). Possibly you can do this just from directory/file sizes.\\n > Note that template1 should be OID 1, and template0 will have the next\\n > lowest number (probably 16555, in 7.2).\\n > \\n > 4. Initdb a scratch database in some other place (or move aside your\\n > existing files, if that seems safer). In this scratch DB, create\\n > databases, users, and groups to match your old setup. You should be\\n > able to duplicate everything except the database OIDs using standard\\n > SQL commands.\\n > \\n > 5. Shut down scratch postmaster, then hex-edit pg_database to insert the\\n > correct OIDs. Use pg_filedump or a similar tool to verify that you did\\n > this properly.\\n > \\n > 6. Restart scratch postmaster, and VACUUM FREEZE pg_database, pg_shadow,\\n > and pg_group (from any database). This will make the next step safe.\\n > \\n > 7. Stop scratch postmaster, and then copy over its $PGDATA/global\\n > directory into the old DB.\\n > \\n > 8. Cross your fingers and start postmaster ...\\n > \\n > This will probably *not* work if you had been doing anything to\\n > pg_database, pg_shadow, or pg_group between your last checkpoint and the\\n > crash, because the reconstructed tables are not going to be physically\\n > identical to what they were before, so any actions replayed from WAL\\n > against those tables will be wrong. Hopefully you won't have that\\n > problem. If you do, it might work to shut down the postmaster and again\\n > copy the scratch $PGDATA/global directory into the old DB, thereby\\n > overwriting what the WAL replay did. This is getting into the realm of\\n > speculation though.\\n > \\n > \\t\\t\\tregards, tom lane\\n > \\n > ---------------------------(end of broadcast)---------------------------\\n > TIP 2: you can get off all lists at once with the unregister command\\n > (send \\\"unregister YourEmailAddressHere\\\" to majordomo@postgresql.org)\\n\\n\\n\",\n \"msg_date\": \"Tue, 24 Aug 2004 12:54:26 +0300\",\n \"msg_from\": \"Daniel Kalchev \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: missing data/global \"\n },\n {\n \"msg_contents\": \"Daniel Kalchev writes:\\n> (found out 7.2.3 does not have pg_database)\\n\\nYou think not?\\n\\n> By the way, I had to copy over the 'new' files from pg_clog and pg_xlog (this\\n> is the second possible error) to get the postmaster running.\\n\\nThat was *not* part of the recipe, and is guaranteed *not* to work.\\n\\nIt seems likely though that you are wasting your time --- the index\\nfailure suggests strongly that you have more corruption than just the\\nloss of the /global subdirectory :-(\\n\\n\\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Tue, 24 Aug 2004 10:43:41 -0400\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: missing data/global \"\n },\n {\n \"msg_contents\": \">>>Tom Lane said:\\n > Daniel Kalchev writes:\\n > > (found out 7.2.3 does not have pg_database)\\n > \\n > You think not?\\n\\nNot as a file similar to pg_control. pg_database is indeed table in the system \\ncatalog.\\n\\n > > By the way, I had to copy over the 'new' files from pg_clog and pg_xlog (t\\n his\\n > > is the second possible error) to get the postmaster running.\\n > \\n > That was *not* part of the recipe, and is guaranteed *not* to work.\\n\\nI know that, but wondered if it would help in any way.. By the way, what would \\nbe the solution to sync WAL with the pg_control contents?\\n\\n > \\n > It seems likely though that you are wasting your time --- the index\\n > failure suggests strongly that you have more corruption than just the\\n > loss of the /global subdirectory :-(\\n\\nAfter spending some time to find possible ways to adjust pointers (could \\neventually save part of the data), I decided to move to plan B, which is to \\nhave few people manually re-enter the data - would have been more effective to \\nwaste my time anyway - but not if it will take days and the result be not \\nguaranteed to be consistent.\\n\\nDoes such toll exist, that could dump data (records?) from the heap files \\ngiven the table structure?\\n\\nRegards,\\nDaniel\\n\\n\",\n \"msg_date\": \"Wed, 25 Aug 2004 19:07:23 +0300\",\n \"msg_from\": \"Daniel Kalchev \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: missing data/global \"\n },\n {\n \"msg_contents\": \"On Wed, Aug 25, 2004 at 07:07:23PM +0300, Daniel Kalchev wrote:\\n\\n> Does such toll exist, that could dump data (records?) from the heap files \\n> given the table structure?\\n\\nYou may want to check pg_filedump (from http://sources.redhat.com/rhdb\\nIIRC).\\n\\n(What happened to pg_fsck BTW?)\\n\\n-- \\nAlvaro Herrera ()\\n\\n\",\n \"msg_date\": \"Wed, 25 Aug 2004 12:34:34 -0400\",\n \"msg_from\": \"Alvaro Herrera \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: missing data/global\"\n },\n {\n \"msg_contents\": \"Coincidentally I JUST NOW built 8.0 on Solaris 9, and ran into the same\\nproblem. As they say, \\\"this used to work\\\"..... \\n\\nWe build databases as part of the build of our product, and I'm looking\\ninto what we need to do to upgrade from 7.4.5, and this was the first\\nthing I ran into. I hadn't gotten as far as truss yet, so thanks Kenneth\\nfor that extra info.\\n\\nDid initdb previously just assume the -D path existed, and now it is\\ntrying to create the whole path, if necessary?\\n\\n- DAP\\n\\n>-----Original Message-----\\n>From: pgsql-hackers-owner@postgresql.org \\n>[mailto:pgsql-hackers-owner@postgresql.org] On Behalf Of Kenneth Lareau\\n>Sent: Thursday, January 27, 2005 5:23 PM\\n>To: pgsql-hackers@postgresql.org\\n>Subject: [HACKERS] Strange issue with initdb on 8.0 and \\n>Solaris automounts\\n>\\n>Folks,\\n>\\n>I ran into an interesting issue when installing PostgreSQL 8.0 \\n>that I'm not sure how to resolve correctly. My system is a \\n>Sun machine (Blade\\n>1000) running Solaris 9, with relatively recent patches. After \\n>install- ing 8.0, I went to run the 'initdb' command and was \\n>greeted with the\\n>following:\\n>\\n>[delirium:postgres] ~\\n>(11) initdb -D /software/postgresql-8.0.0/data The files \\n>belonging to this database system will be owned by user \\\"postgres\\\".\\n>This user must also own the server process.\\n>\\n>The database cluster will be initialized with locale C.\\n>\\n>creating directory /software/postgresql-8.0.0/data ... initdb: \\n>could not create directory \\\"/software/postgresql-8.0.0\\\": \\n>Operation not applicable\\n>\\n>\\n>The error message was a bit confusing, so I decided to run a \\n>truss on the process to see what might be happening, and this \\n>is what I came\\n>across:\\n>\\n>[...]\\n>8802/1: write(1, \\\" c r e a t i n g d i r\\\".., 62) = 62\\n>8802/1: umask(0) = 077\\n>8802/1: umask(077) = 0\\n>8802/1: mkdir(\\\"/software\\\", 0777) \\n> Err#17 EEXIST\\n>8802/1: stat64(\\\"/software\\\", 0xFFBFC858) = 0\\n>8802/1: mkdir(\\\"/software/postgresql-8.0.0\\\", 0777) \\n> Err#89 ENOSYS\\n>[...]\\n>\\n>\\n>The last error in that section, ENOSYS, is very strange, as \\n>the Solaris manpage for 'mkdir' does not mention it as a \\n>possible error. One thing to note in this, however, is that \\n>'/software/postgresql-8.0.0' is not a regular directory, but \\n>an automount point (which in this case is just a local \\n>loopback mount). So the indication is that Solaris seems to \\n>have a bug not in mkdir, but deeper in their VFS code that's \\n>causing this seemingly strange issue.\\n>\\n>Two workarounds for this problem have been found: running \\n>'initdb' with a directory that's *not* an automount point and \\n>then moving the 'data'\\n>directory to its final destination worked fine, along with a \\n>suggestion from Andrew Dunstan (on the #postgresql IRC \\n>channel) with using a rela- tive path for the data directory. \\n>Both were successful in avoiding the issue, but I decided to \\n>mention this here in case someone felt it might be worth \\n>looking into to see if the Sun problem can be avoided; I am \\n>going to notify Sun of their bug, just don't know how long it \\n>will take them to actually resolve it (if they ever do).\\n>\\n>While I can fully understand that a code change here may not \\n>be desire- able, might some notes in the documentation be \\n>useful for those who might stumble across the problem as well? \\n> Just a suggestion...\\n>\\n>I hope I gave sufficient information on the problem, though \\n>I'm always willing to give any clarification needed. Thank \\n>you for your time.\\n>\\n>\\n>Ken Lareau\\n>elessar@numenor.org\\n>\\n>---------------------------(end of \\n>broadcast)---------------------------\\n>TIP 6: Have you searched our list archives?\\n>\\n> http://archives.postgresql.org\\n>\\n\",\n \"msg_date\": \"Thu, 27 Jan 2005 17:49:37 -0500\",\n \"msg_from\": \"\\\"David Parker\\\" \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Strange issue with initdb on 8.0 and Solaris automounts\"\n },\n {\n \"msg_contents\": \"\\\"David Parker\\\" writes:\\n> Did initdb previously just assume the -D path existed, and now it is\\n> trying to create the whole path, if necessary?\\n\\nPre-8.0 it was using mkdir(1), which might possibly contain some weird\\nworkaround for this case on Solaris.\\n\\nI suppose that manually creating the data directory before running\\ninitdb would also avoid this issue, since the mkdir(2) loop is only\\nentered if we don't find the directory in existence.\\n\\n\\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Thu, 27 Jan 2005 18:22:18 -0500\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Strange issue with initdb on 8.0 and Solaris automounts \"\n },\n {\n \"msg_contents\": \"In message <21723.1106868138@sss.pgh.pa.us>, Tom Lane writes:\\n>\\\"David Parker\\\" writes:\\n>> Did initdb previously just assume the -D path existed, and now it is\\n>> trying to create the whole path, if necessary?\\n>\\n>Pre-8.0 it was using mkdir(1), which might possibly contain some weird\\n>workaround for this case on Solaris.\\n>\\n>I suppose that manually creating the data directory before running\\n>initdb would also avoid this issue, since the mkdir(2) loop is only\\n>entered if we don't find the directory in existence.\\n>\\n>\\t\\t\\tregards, tom lane\\n>\\n\\nActually, creating the 'data' directory first doesn't work either:\\n\\n[delirium:postgres] ~\\n(17) mkdir data\\n[delirium:postgres] ~\\n(18) initdb -D /software/postgresql-8.0.0/data\\nThe files belonging to this database system will be owned by user \\\"postgres\\\".\\nThis user must also own the server process.\\n\\nThe database cluster will be initialized with locale C.\\n\\nfixing permissions on existing directory /software/postgresql-8.0.0/data ... ok\\ncreating directory /software/postgresql-8.0.0/data/global ... initdb: could not create directory \\\"/software/postgresql-8.0.0\\\": Operation not applicable\\ninitdb: removing contents of data directory \\\"/software/postgresql-8.0.0/data\\\"\\n\\n\\nSince there's subdirectories that need to be created, it still runs into\\nthe problem. I don't know why the command 'mkdir' doesn't exhibit the\\nsame problem as the function 'mkdir', but running:\\n\\n mkdir /software/postgresql-8.0.0\\n\\nproduces the correct error \\\"File exists\\\" on my system. I suspect the\\n'mkdir' command probably checks to see if the directory exists first\\nbefore trying to create it, which avoids the problem.\\n\\n\\nKen Lareau\\nelessar@numenor.org\\n\",\n \"msg_date\": \"Thu, 27 Jan 2005 15:35:41 -0800\",\n \"msg_from\": \"Kenneth Lareau \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Strange issue with initdb on 8.0 and Solaris automounts\"\n },\n {\n \"msg_contents\": \"Kenneth Lareau writes:\\n> In message <21723.1106868138@sss.pgh.pa.us>, Tom Lane writes:\\n>> I suppose that manually creating the data directory before running\\n>> initdb would also avoid this issue, since the mkdir(2) loop is only\\n>> entered if we don't find the directory in existence.\\n\\n> Actually, creating the 'data' directory first doesn't work either:\\n\\nGood point.\\n\\n> I don't know why the command 'mkdir' doesn't exhibit the\\n> same problem as the function 'mkdir', but running:\\n\\n> mkdir /software/postgresql-8.0.0\\n\\n> produces the correct error \\\"File exists\\\" on my system.\\n\\nCould you truss that and see what it does? It would be a simple change\\nin initdb to make it stat before mkdir instead of after, but I'm not\\ntotally convinced that would fix the problem. If mkdir returns a funny\\nerror code then stat might as well ...\\n\\n\\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Thu, 27 Jan 2005 18:50:48 -0500\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Strange issue with initdb on 8.0 and Solaris automounts \"\n },\n {\n \"msg_contents\": \"In message <22095.1106869848@sss.pgh.pa.us>, Tom Lane writes:\\n>Kenneth Lareau writes:\\n>> In message <21723.1106868138@sss.pgh.pa.us>, Tom Lane writes:\\n>>> I suppose that manually creating the data directory before running\\n>>> initdb would also avoid this issue, since the mkdir(2) loop is only\\n>>> entered if we don't find the directory in existence.\\n>\\n>> Actually, creating the 'data' directory first doesn't work either:\\n>\\n>Good point.\\n>\\n>> I don't know why the command 'mkdir' doesn't exhibit the\\n>> same problem as the function 'mkdir', but running:\\n>\\n>> mkdir /software/postgresql-8.0.0\\n>\\n>> produces the correct error \\\"File exists\\\" on my system.\\n>\\n>Could you truss that and see what it does? It would be a simple change\\n>in initdb to make it stat before mkdir instead of after, but I'm not\\n>totally convinced that would fix the problem. If mkdir returns a funny\\n>error code then stat might as well ...\\n>\\n>\\t\\t\\tregards, tom lane\\n>\\n\\nHere's the relevant truss output from 'mkdir /software/postgresql-8.0.0'\\non my Solaris 9 system:\\n\\n10832: umask(0) = 077\\n10832: umask(077) = 0\\n10832: mkdir(\\\"/software/postgresql-8.0.0\\\", 0777) Err#89 ENOSYS\\n10832: stat64(\\\"/software/postgresql-8.0.0\\\", 0xFFBFFA38) = 0\\n10832: fstat64(2, 0xFFBFEB78) = 0\\n10832: write(2, \\\" m k d i r\\\", 5) = 5\\n10832: write(2, \\\" : \\\", 2) = 2\\n10832: write(2, \\\" c a n n o t c r e a t\\\".., 24) = 24\\n10832: write(2, \\\" ` / s o f t w a r e / p\\\".., 28) = 28\\n10832: write(2, \\\" : \\\", 2) = 2\\n10832: write(2, \\\" F i l e e x i s t s\\\", 11) = 11\\n10832: write(2, \\\"\\\\n\\\", 1) = 1\\n10832: _exit(1)\\n\\n\\nIt's doing the stat after the mkdir attempt it seems, and coming back\\nwith the correct response. Hmm, maybe I should look at the Solaris 8\\ncode for the mkdir command...\\n\\n\\nKen Lareau\\nelessar@numenor.org\\n\",\n \"msg_date\": \"Thu, 27 Jan 2005 16:18:36 -0800\",\n \"msg_from\": \"Kenneth Lareau \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Strange issue with initdb on 8.0 and Solaris automounts\"\n },\n {\n \"msg_contents\": \"\\n\\nTom Lane wrote:\\n\\n>>I don't know why the command 'mkdir' doesn't exhibit the\\n>>same problem as the function 'mkdir', but running:\\n>> \\n>>\\n>\\n> \\n>\\n>> mkdir /software/postgresql-8.0.0\\n>> \\n>>\\n>\\n> \\n>\\n>>produces the correct error \\\"File exists\\\" on my system.\\n>> \\n>>\\n>\\n>Could you truss that and see what it does? It would be a simple change\\n>in initdb to make it stat before mkdir instead of after, but I'm not\\n>totally convinced that would fix the problem. If mkdir returns a funny\\n>error code then stat might as well ...\\n>\\n>\\n> \\n>\\n\\nThere's also a tiny race condition, which I guess isn't worth worrying \\nabout.\\n\\nReturning ENOSYS is pretty bogus ...\\n\\ncheers\\n\\nandrew\\n\",\n \"msg_date\": \"Thu, 27 Jan 2005 19:28:25 -0500\",\n \"msg_from\": \"Andrew Dunstan \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Strange issue with initdb on 8.0 and Solaris automounts\"\n },\n {\n \"msg_contents\": \"Kenneth Lareau writes:\\n> In message <22095.1106869848@sss.pgh.pa.us>, Tom Lane writes:\\n>> Could you truss that and see what it does?\\n\\n> Here's the relevant truss output from 'mkdir /software/postgresql-8.0.0'\\n> on my Solaris 9 system:\\n\\n> 10832: mkdir(\\\"/software/postgresql-8.0.0\\\", 0777) Err#89 ENOSYS\\n> 10832: stat64(\\\"/software/postgresql-8.0.0\\\", 0xFFBFFA38) = 0\\n\\n> It's doing the stat after the mkdir attempt it seems, and coming back\\n> with the correct response. Hmm, maybe I should look at the Solaris 8\\n> code for the mkdir command...\\n\\nWell, the important point is that the stat does succeed. I'm not going\\nto put in anything as specific as a check for ENOSYS, but it seems\\nreasonable to try the stat first and mkdir only if stat fails.\\nI've applied the attached patch.\\n\\n\\t\\t\\tregards, tom lane\\n\\n*** src/bin/initdb/initdb.c.orig\\tSat Jan 8 17:51:12 2005\\n--- src/bin/initdb/initdb.c\\tThu Jan 27 19:23:49 2005\\n***************\\n*** 476,481 ****\\n--- 476,484 ----\\n * this tries to build all the elements of a path to a directory a la mkdir -p\\n * we assume the path is in canonical form, i.e. uses / as the separator\\n * we also assume it isn't null.\\n+ *\\n+ * note that on failure, the path arg has been modified to show the particular\\n+ * directory level we had problems with.\\n */\\n static int\\n mkdir_p(char *path, mode_t omode)\\n***************\\n*** 544,573 ****\\n \\t\\t}\\n \\t\\tif (last)\\n \\t\\t\\t(void) umask(oumask);\\n! \\t\\tif (mkdir(path, last ? omode : S_IRWXU | S_IRWXG | S_IRWXO) < 0)\\n \\t\\t{\\n! \\t\\t\\tif (errno == EEXIST || errno == EISDIR)\\n! \\t\\t\\t{\\n! \\t\\t\\t\\tif (stat(path, &sb) < 0)\\n! \\t\\t\\t\\t{\\n! \\t\\t\\t\\t\\tretval = 1;\\n! \\t\\t\\t\\t\\tbreak;\\n! \\t\\t\\t\\t}\\n! \\t\\t\\t\\telse if (!S_ISDIR(sb.st_mode))\\n! \\t\\t\\t\\t{\\n! \\t\\t\\t\\t\\tif (last)\\n! \\t\\t\\t\\t\\t\\terrno = EEXIST;\\n! \\t\\t\\t\\t\\telse\\n! \\t\\t\\t\\t\\t\\terrno = ENOTDIR;\\n! \\t\\t\\t\\t\\tretval = 1;\\n! \\t\\t\\t\\t\\tbreak;\\n! \\t\\t\\t\\t}\\n! \\t\\t\\t}\\n! \\t\\t\\telse\\n \\t\\t\\t{\\n \\t\\t\\t\\tretval = 1;\\n \\t\\t\\t\\tbreak;\\n \\t\\t\\t}\\n \\t\\t}\\n \\t\\tif (!last)\\n \\t\\t\\t*p = '/';\\n--- 547,570 ----\\n \\t\\t}\\n \\t\\tif (last)\\n \\t\\t\\t(void) umask(oumask);\\n! \\n! \\t\\t/* check for pre-existing directory; ok if it's a parent */\\n! \\t\\tif (stat(path, &sb) == 0)\\n \\t\\t{\\n! \\t\\t\\tif (!S_ISDIR(sb.st_mode))\\n \\t\\t\\t{\\n+ \\t\\t\\t\\tif (last)\\n+ \\t\\t\\t\\t\\terrno = EEXIST;\\n+ \\t\\t\\t\\telse\\n+ \\t\\t\\t\\t\\terrno = ENOTDIR;\\n \\t\\t\\t\\tretval = 1;\\n \\t\\t\\t\\tbreak;\\n \\t\\t\\t}\\n+ \\t\\t}\\n+ \\t\\telse if (mkdir(path, last ? omode : S_IRWXU | S_IRWXG | S_IRWXO) < 0)\\n+ \\t\\t{\\n+ \\t\\t\\tretval = 1;\\n+ \\t\\t\\tbreak;\\n \\t\\t}\\n \\t\\tif (!last)\\n \\t\\t\\t*p = '/';\\n\",\n \"msg_date\": \"Thu, 27 Jan 2005 19:37:33 -0500\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Strange issue with initdb on 8.0 and Solaris automounts \"\n },\n {\n \"msg_contents\": \"Andrew Dunstan writes:\\n> There's also a tiny race condition, which I guess isn't worth worrying \\n> about.\\n\\nConsidering that we're not checking ownership or permissions of the\\nparent directories, I'd say not.\\n\\n\\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Thu, 27 Jan 2005 19:39:59 -0500\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Strange issue with initdb on 8.0 and Solaris automounts \"\n },\n {\n \"msg_contents\": \"In message <22687.1106872653@sss.pgh.pa.us>, Tom Lane writes:\\n>Kenneth Lareau writes:\\n>> In message <22095.1106869848@sss.pgh.pa.us>, Tom Lane writes:\\n>>> Could you truss that and see what it does?\\n>\\n>> Here's the relevant truss output from 'mkdir /software/postgresql-8.0.0'\\n>> on my Solaris 9 system:\\n>\\n>> 10832: mkdir(\\\"/software/postgresql-8.0.0\\\", 0777) Err#89 ENOSYS\\n>> 10832: stat64(\\\"/software/postgresql-8.0.0\\\", 0xFFBFFA38) = 0\\n>\\n>> It's doing the stat after the mkdir attempt it seems, and coming back\\n>> with the correct response. Hmm, maybe I should look at the Solaris 8\\n>> code for the mkdir command...\\n>\\n>Well, the important point is that the stat does succeed. I'm not going\\n>to put in anything as specific as a check for ENOSYS, but it seems\\n>reasonable to try the stat first and mkdir only if stat fails.\\n>I've applied the attached patch.\\n>\\n>\\t\\t\\tregards, tom lane\\n\\n\\nTom, thank you very much for the patch, it worked like a charm.\\n\\n\\nKen Lareau\\nelessar@numenor.org\\n\",\n \"msg_date\": \"Thu, 27 Jan 2005 17:10:04 -0800\",\n \"msg_from\": \"Kenneth Lareau \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Strange issue with initdb on 8.0 and Solaris automounts\"\n },\n {\n \"msg_contents\": \"> My opinion is that this is a very bogus shortcut in the \\n> network datatype code. There are no cases outside the \\n> inet/cidr group where an operator doesn't exactly match its \\n> underlying function. (The whole business of inet and cidr \\n> being almost but not quite the same type is maldesigned\\n> anyway...)\\n> \\n> The right solution for you is to declare two SQL functions. \\n> Whether you make them point at the same underlying C code is \\n> up to you.\\n\\nRight,...\\n\\nIn that case may I suggest fixing the catalog so network_* functions exists for both datatypes!\\nAnything less I'd consider inconsistent...\\n\\nKind regards,\\n\\nJohn\\n\",\n \"msg_date\": \"Sun, 30 Jan 2005 13:46:34 +1100\",\n \"msg_from\": \"\\\"John Hansen\\\" \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Bug in create operator and/or initdb \"\n },\n {\n \"msg_contents\": \"\\\"John Hansen\\\" writes:\\n> In that case may I suggest fixing the catalog so network_* functions exists for both datatypes!\\n\\nRedesigning the inet/cidr distinction is on the to-do list (though I'm\\nafraid not very high on the list). ISTM it should either be one type\\nwith a distinguishing bit in the runtime representation, or two types\\nwith no such bit needed. Having both is a schizophrenic design. It's\\nled directly to bugs in the past, and I think there are still some\\ncorner cases that act oddly (see the archives).\\n\\n\\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Sat, 29 Jan 2005 22:07:30 -0500\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: Bug in create operator and/or initdb \"\n },\n {\n \"msg_contents\": \"On Sat, Jan 29, 2005 at 10:07:30PM -0500, Tom Lane wrote:\\n> \\\"John Hansen\\\" writes:\\n> > In that case may I suggest fixing the catalog so network_* functions exists for both datatypes!\\n> \\n> Redesigning the inet/cidr distinction is on the to-do list (though I'm\\n> afraid not very high on the list). ISTM it should either be one type\\n> with a distinguishing bit in the runtime representation, or two types\\n> with no such bit needed. Having both is a schizophrenic design. It's\\n> led directly to bugs in the past, and I think there are still some\\n> corner cases that act oddly (see the archives).\\n\\n From a network engineering point of view the inet type is utterly\\nbogus. I'm not aware of data of that type being needed or used in\\nany real application. Given that, the complexity that it causes\\nsimply by existing seems too high a cost.\\n\\nI suspect that the right thing to do is to kill the inet type\\nentirely, and replace it with a special case of cidr. (And possibly\\nthen to kill cidr and replace it with something that can be indexed\\nmore effectively.)\\n\\nFor a replacement type, how important is it that it be completely\\ncompatible with the existing inet/cidr types? Is anyone actually using\\ninet types with a non-cidr mask?\\n\\nCheers,\\n Steve\\n\",\n \"msg_date\": \"Sat, 29 Jan 2005 20:20:10 -0800\",\n \"msg_from\": \"Steve Atkins \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [BUGS] Bug in create operator and/or initdb\"\n },\n {\n \"msg_contents\": \"Steve Atkins writes:\\n> For a replacement type, how important is it that it be completely\\n> compatible with the existing inet/cidr types? Is anyone actually using\\n> inet types with a non-cidr mask?\\n\\nIf you check the archives you'll discover that our current inet/cidr\\ntypes were largely designed and implemented by Paul Vixie (yes, that\\nVixie). I'm disinclined to second-guess Paul about the external\\ndefinition of these types; I just want to rationalize the internal\\nrepresentation a bit. In particular we've got some issues about\\nconversions between the two types ...\\n\\n\\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Sun, 30 Jan 2005 01:46:19 -0500\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [BUGS] Bug in create operator and/or initdb \"\n },\n {\n \"msg_contents\": \"On Sun, 30 Jan 2005, Tom Lane wrote:\\n\\n> Steve Atkins writes:\\n>> For a replacement type, how important is it that it be completely\\n>> compatible with the existing inet/cidr types? Is anyone actually using\\n>> inet types with a non-cidr mask?\\n>\\n> If you check the archives you'll discover that our current inet/cidr\\n> types were largely designed and implemented by Paul Vixie (yes, that\\n> Vixie). I'm disinclined to second-guess Paul about the external\\n> definition of these types; I just want to rationalize the internal\\n> representation a bit. In particular we've got some issues about\\n> conversions between the two types ...\\nPlease do **NOT** break the external representations. We had enough fights\\nabout that 2-3 releases ago, and I personally don't want to revisit them.\\n\\nYes, we do flakey things with inet on the masking stuff.\\n\\nLER\\n\\n-- \\nLarry Rosenman http://www.lerctr.org/~ler\\nPhone: +1 972-414-9812 E-Mail: ler@lerctr.org\\nUS Mail: 1905 Steamboat Springs Drive, Garland, TX 75044-6749\\n\",\n \"msg_date\": \"Sun, 30 Jan 2005 21:49:43 -0600 (CST)\",\n \"msg_from\": \"Larry Rosenman \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [BUGS] Bug in create operator and/or initdb\"\n },\n {\n \"msg_contents\": \"On Sun, Jan 30, 2005 at 09:49:43PM -0600, Larry Rosenman wrote:\\n> On Sun, 30 Jan 2005, Tom Lane wrote:\\n> \\n> >Steve Atkins writes:\\n> >>For a replacement type, how important is it that it be completely\\n> >>compatible with the existing inet/cidr types? Is anyone actually using\\n> >>inet types with a non-cidr mask?\\n> >\\n> >If you check the archives you'll discover that our current inet/cidr\\n> >types were largely designed and implemented by Paul Vixie (yes, that\\n> >Vixie). I'm disinclined to second-guess Paul about the external\\n> >definition of these types; I just want to rationalize the internal\\n> >representation a bit. In particular we've got some issues about\\n> >conversions between the two types ...\\n>\\n> Please do **NOT** break the external representations. We had enough fights\\n> about that 2-3 releases ago, and I personally don't want to revisit them.\\n\\n> Yes, we do flakey things with inet on the masking stuff.\\n\\nWell, if you want the ability to store both a host address and a\\nnetmask in the same datatype the inet masking stuff makes sense.\\nThat's not really a useful datatype for any actual use, but it's\\nfairly well-defined. The problem is that when someone looks at the\\ndocs they'll see inet as the obvious datatype to use to store IP\\naddresses, and it isn't very good for that.\\n\\nBut that's not all that's flakey, unfortunately.\\n\\nThe CIDR input format is documented to be classful, which in itself is\\nhorribly obsolete and completely useless in this decades internet (and\\nwas when the current code was written in '98).\\n\\nBut the implementation isn't either classful or classless, and the\\nbehaviour disagrees with documented behaviour, and the behaviour you'd\\nreasonably expect, in many cases.\\n\\n-- Class A - documented to be 10.0.0.0/8\\n steve=# select '10.0.0.0'::cidr;\\n cidr \\n -------------\\n 10.0.0.0/32\\n\\n-- Class B - documented to be 128.0.0.0/16\\n steve=# select '128.0.0.0'::cidr;\\n cidr \\n --------------\\n 128.0.0.0/32\\n\\n-- Class C - documented to be 223.10.0.0/24\\n steve=# select '223.10.0.0'::cidr;\\n cidr \\n ---------------\\n 223.10.0.0/32\\n\\n-- Class D\\n steve=# select '224.10.0.0'::cidr;\\n ERROR: invalid cidr value: \\\"224.10.0.0\\\"\\n DETAIL: Value has bits set to right of mask.\\n\\n steve=# select '224.0.0.0'::cidr;\\n cidr \\n -------------\\n 224.0.0.0/4\\n\\n-- Class E\\n steve=# select '240.10.0.0'::cidr;\\n cidr \\n ---------------\\n 240.10.0.0/32\\n\\nI use postgresql for network-related applications and for IP address\\nrelated data mining, so I'm dealing with IP addresses in postgresql\\non a daily basis.\\n\\nThe cidr type, including it's external interface, is simply broken.\\nThere is no way to fix it that doesn't change that external interface.\\n\\nI know of at least two independant implementations of function IP\\naddress types that have been put together for specific projects to\\nimplement a working IP datatype. The ability to use gist indexes on\\nthem to accelerate range-based lookups is a bonus.\\n\\nIf it's not possible (for backwards compatibility reasons) to fix\\ninet+cidr, would migrating them out to contrib be a possibility?\\nData types in the core tend to be widely used, even if they're\\nbroken and there are better datatypes implemented as external\\nmodules.\\n\\nCheers,\\n Steve\\n\",\n \"msg_date\": \"Mon, 31 Jan 2005 07:23:05 -0800\",\n \"msg_from\": \"Steve Atkins \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [BUGS] Bug in create operator and/or initdb\"\n },\n {\n \"msg_contents\": \"Steve Atkins writes:\\n> The cidr type, including it's external interface, is simply broken.\\n\\nThat is a large claim that I don't think you have demonstrated.\\nThe only one of your examples that seems to me to contradict the\\ndocumentation is this one:\\n\\n steve=# select '224.0.0.0'::cidr;\\n cidr \\n -------------\\n 224.0.0.0/4\\n\\nwhich should be /32 according to what the docs say:\\n\\n: If y is omitted, it is calculated using assumptions from the older\\n: classful network numbering system, except that it will be at least large\\n: enough to include all of the octets written in the input.\\n\\nThe bogus netmask is in turn responsible for this case:\\n\\n steve=# select '224.10.0.0'::cidr;\\n ERROR: invalid cidr value: \\\"224.10.0.0\\\"\\n DETAIL: Value has bits set to right of mask.\\n\\n\\nLooking at the source code, there seems to be a special case for \\\"class D\\\"\\nnetwork numbers that causes the code not to extend y to cover the\\nsupplied inputs:\\n\\n /* If no CIDR spec was given, infer width from net class. */\\n if (bits == -1)\\n {\\n if (*odst >= 240) /* Class E */\\n bits = 32;\\n else if (*odst >= 224) /* Class D */\\n bits = 4;\\n else if (*odst >= 192) /* Class C */\\n bits = 24;\\n else if (*odst >= 128) /* Class B */\\n bits = 16;\\n else /* Class A */\\n bits = 8;\\n /* If imputed mask is narrower than specified octets, widen. */\\n if (bits >= 8 && bits < ((dst - odst) * 8))\\n ^^^^^^^^^\\n bits = (dst - odst) * 8;\\n }\\n\\nI think the test for \\\"bits >= 8\\\" should be removed. Does anyone know\\nwhy it's there?\\n\\n\\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Mon, 31 Jan 2005 12:16:26 -0500\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [BUGS] Bug in create operator and/or initdb \"\n },\n {\n \"msg_contents\": \"On Mon, Jan 31, 2005 at 12:16:26PM -0500, Tom Lane wrote:\\n> Steve Atkins writes:\\n> > The cidr type, including it's external interface, is simply broken.\\n> \\n> That is a large claim that I don't think you have demonstrated.\\n> The only one of your examples that seems to me to contradict the\\n> documentation is this one:\\n> \\n> steve=# select '224.0.0.0'::cidr;\\n> cidr \\n> -------------\\n> 224.0.0.0/4\\n> \\n> which should be /32 according to what the docs say:\\n\\nOK. If this sort of thing is considered a bug, rather than part\\nof the external interface that shouldn't be changed, then I'd\\nagree that cidr isn't entirely broken and it may well be possible\\nto improve it without changing the interface.\\n\\n/me goes grovelling through the IPv6 inet code...\\n\\nCheers,\\n Steve\\n\\n\",\n \"msg_date\": \"Mon, 31 Jan 2005 09:54:24 -0800\",\n \"msg_from\": \"Steve Atkins \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [BUGS] Bug in create operator and/or initdb\"\n },\n {\n \"msg_contents\": \"when my cidr datatype was integrated into pgsql, the decision was made to\\nincorporate a copy of bind's inet_net_pton.c rather than add a link-time\\ndependence to libbind.a (libbind.so). thus, when this bug was fixed in\\n2003:\\n\\n----------------------------\\nrevision 1.14\\ndate: 2003/08/20 02:21:08; author: marka; state: Exp; lines: +10 -4\\n1580. [bug] inet_net_pton() didn't fully handle implicit\\n multicast IPv4 network addresses.\\n\\nthe pgsql \\\"fork\\\" of this code did not benefit from the fix. the patch was:\\n\\nIndex: inet_net_pton.c\\n===================================================================\\nRCS file: /proj/cvs/prod/bind8/src/lib/inet/inet_net_pton.c,v\\nretrieving revision 1.13\\nretrieving revision 1.14\\ndiff -u -r1.13 -r1.14\\n--- inet_net_pton.c 27 Sep 2001 15:08:38 -0000 1.13\\n+++ inet_net_pton.c 20 Aug 2003 02:21:08 -0000 1.14\\n@@ -16,7 +16,7 @@\\n */\\n \\n #if defined(LIBC_SCCS) && !defined(lint)\\n-static const char rcsid[] = \\\"$Id: inet_net_pton.c,v 1.13 2001/09/27 15:08:38 marka Exp $\\\";\\n+static const char rcsid[] = \\\"$Id: inet_net_pton.c,v 1.14 2003/08/20 02:21:08 marka Exp $\\\";\\n #endif\\n \\n #include \\\"port_before.h\\\"\\n@@ -59,7 +59,7 @@\\n * Paul Vixie (ISC), June 1996\\n */\\n static int\\n-inet_net_pton_ipv4( const char *src, u_char *dst, size_t size) {\\n+inet_net_pton_ipv4(const char *src, u_char *dst, size_t size) {\\n static const char xdigits[] = \\\"0123456789abcdef\\\";\\n static const char digits[] = \\\"0123456789\\\";\\n int n, ch, tmp = 0, dirty, bits;\\n@@ -152,7 +152,7 @@\\n if (*odst >= 240) /* Class E */\\n bits = 32;\\n else if (*odst >= 224) /* Class D */\\n- bits = 4;\\n+ bits = 8;\\n else if (*odst >= 192) /* Class C */\\n bits = 24;\\n else if (*odst >= 128) /* Class B */\\n@@ -160,8 +160,14 @@\\n else /* Class A */\\n bits = 8;\\n /* If imputed mask is narrower than specified octets, widen. */\\n- if (bits >= 8 && bits < ((dst - odst) * 8))\\n+ if (bits < ((dst - odst) * 8))\\n bits = (dst - odst) * 8;\\n+ /*\\n+ * If there are no additional bits specified for a class D\\n+ * address adjust bits to 4.\\n+ */\\n+ if (bits == 8 && *odst == 224)\\n+ bits = 4;\\n }\\n /* Extend network to cover the actual mask. */\\n while (bits > ((dst - odst) * 8)) {\\n\\nre:\\n\\n> To: Steve Atkins \\n> Cc: pgsql-hackers , paul@vix.com\\n> Subject: Re: [HACKERS] [BUGS] Bug in create operator and/or initdb \\n> Comments: In-reply-to Steve Atkins \\n> \\tmessage dated \\\"Mon, 31 Jan 2005 07:23:05 -0800\\\"\\n> Date: Mon, 31 Jan 2005 12:16:26 -0500\\n> From: Tom Lane \\n> \\n> Steve Atkins writes:\\n> > The cidr type, including it's external interface, is simply broken.\\n> \\n> That is a large claim that I don't think you have demonstrated.\\n> The only one of your examples that seems to me to contradict the\\n> documentation is this one:\\n> \\n> steve=# select '224.0.0.0'::cidr;\\n> cidr \\n> -------------\\n> 224.0.0.0/4\\n> \\n> which should be /32 according to what the docs say:\\n> \\n> : If y is omitted, it is calculated using assumptions from the older\\n> : classful network numbering system, except that it will be at least large\\n> : enough to include all of the octets written in the input.\\n> \\n> The bogus netmask is in turn responsible for this case:\\n> \\n> steve=# select '224.10.0.0'::cidr;\\n> ERROR: invalid cidr value: \\\"224.10.0.0\\\"\\n> DETAIL: Value has bits set to right of mask.\\n> \\n> \\n> Looking at the source code, there seems to be a special case for \\\"class D\\\"\\n> network numbers that causes the code not to extend y to cover the\\n> supplied inputs:\\n> \\n> /* If no CIDR spec was given, infer width from net class. */\\n> if (bits == -1)\\n> {\\n> if (*odst >= 240) /* Class E */\\n> bits = 32;\\n> else if (*odst >= 224) /* Class D */\\n> bits = 4;\\n> else if (*odst >= 192) /* Class C */\\n> bits = 24;\\n> else if (*odst >= 128) /* Class B */\\n> bits = 16;\\n> else /* Class A */\\n> bits = 8;\\n> /* If imputed mask is narrower than specified octets, widen. */\\n> if (bits >= 8 && bits < ((dst - odst) * 8))\\n> ^^^^^^^^^\\n> bits = (dst - odst) * 8;\\n> }\\n> \\n> I think the test for \\\"bits >= 8\\\" should be removed. Does anyone know\\n> why it's there?\\n> \\n> \\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Mon, 31 Jan 2005 19:28:28 +0000\",\n \"msg_from\": \"Paul Vixie \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [BUGS] Bug in create operator and/or initdb \"\n },\n {\n \"msg_contents\": \"Tom Lane writes:\\n\\n> steve=# select '224.0.0.0'::cidr;\\n> cidr \\n> -------------\\n> 224.0.0.0/4\\n> \\n> which should be /32 according to what the docs say:\\n\\n224-239 are multicast addresses. Making it /4 makes the entire multicast\\naddress space one network block which is about as reasonable an answer as\\nanything else.\\n\\n> if (bits >= 8 && bits < ((dst - odst) * 8))\\n> ^^^^^^^^^\\n> bits = (dst - odst) * 8;\\n> }\\n> \\n> I think the test for \\\"bits >= 8\\\" should be removed. Does anyone know\\n> why it's there?\\n\\nI guess Vixie figured network blocks subdividing multicast address space\\nweren't a sensible concept? It's a bit of a strange constraint to hard code\\ninto the C code though.\\n\\nIncidentally, how can that code possibly work? It treats odst as a pointer in\\nsome places but then calculates bits using arithmetic on it directly without\\ndereferencing?\\n\\n-- \\ngreg\\n\\n\",\n \"msg_date\": \"31 Jan 2005 14:35:58 -0500\",\n \"msg_from\": \"Greg Stark \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [BUGS] Bug in create operator and/or initdb\"\n },\n {\n \"msg_contents\": \"Paul Vixie writes:\\n> when my cidr datatype was integrated into pgsql, the decision was made to\\n> incorporate a copy of bind's inet_net_pton.c rather than add a link-time\\n> dependence to libbind.a (libbind.so).\\n\\nWe didn't really want to assume that all platforms are using libbind :-(\\n\\n> thus, when this bug was fixed in 2003:\\n\\n> ----------------------------\\n> revision 1.14\\n> date: 2003/08/20 02:21:08; author: marka; state: Exp; lines: +10 -4\\n> 1580. [bug] inet_net_pton() didn't fully handle implicit\\n> multicast IPv4 network addresses.\\n\\n> the pgsql \\\"fork\\\" of this code did not benefit from the fix. the patch was:\\n\\nAh-hah. Many thanks for supplying the patch --- will integrate it.\\n\\n\\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Mon, 31 Jan 2005 14:36:33 -0500\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [BUGS] Bug in create operator and/or initdb \"\n },\n {\n \"msg_contents\": \"> We didn't really want to assume that all platforms are using libbind :-(\\n\\ni think you could have, at the time, since windows wasn't even a gleam in\\npgsql's eye. even now, libbind would be a dependable universal dependency,\\nsince we publish windows binaries.\\n\\n> > the pgsql \\\"fork\\\" of this code did not benefit from the fix. the patch was:\\n> \\n> Ah-hah. Many thanks for supplying the patch --- will integrate it.\\n\\ni have two suggestions. first, look at the rest of the current source file,\\nin case there are other fixes. second, track changes this source file during\\nyour release engineering process for each new pgsql version.\\n\",\n \"msg_date\": \"Mon, 31 Jan 2005 22:44:38 +0000\",\n \"msg_from\": \"Paul Vixie \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [BUGS] Bug in create operator and/or initdb \"\n },\n {\n \"msg_contents\": \"Paul Vixie writes:\\n> i have two suggestions. first, look at the rest of the current source file,\\n> in case there are other fixes.\\n\\nRight, I already grabbed the latest.\\n\\n> second, track changes this source file during\\n> your release engineering process for each new pgsql version.\\n\\nBruce, do you think this is worth adding to RELEASE_CHANGES?\\ninet_net_ntop.c and inet_net_pton.c are both extracted from the BIND\\ndistribution. But they're hardly the only files we took from elsewhere.\\n\\n\\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Mon, 31 Jan 2005 17:52:40 -0500\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [BUGS] Bug in create operator and/or initdb \"\n },\n {\n \"msg_contents\": \"Tom Lane wrote:\\n> Paul Vixie writes:\\n> > i have two suggestions. first, look at the rest of the current source file,\\n> > in case there are other fixes.\\n> \\n> Right, I already grabbed the latest.\\n> \\n> > second, track changes this source file during\\n> > your release engineering process for each new pgsql version.\\n> \\n> Bruce, do you think this is worth adding to RELEASE_CHANGES?\\n> inet_net_ntop.c and inet_net_pton.c are both extracted from the BIND\\n> distribution. But they're hardly the only files we took from elsewhere.\\n\\nYes, I do. Most of the stuff we pull from other OS projects has\\nclearly-defined behavior, while inet/cidr seem to be still in flux a\\nlittle.\\n\\nAdded to release checklist:\\n\\n\\t* Update inet/cidr data types with newest Bind patches\\n\\n-- \\n Bruce Momjian | http://candle.pha.pa.us\\n pgman@candle.pha.pa.us | (610) 359-1001\\n + If your life is a hard drive, | 13 Roberts Road\\n + Christ can be your backup. | Newtown Square, Pennsylvania 19073\\n\",\n \"msg_date\": \"Wed, 2 Feb 2005 12:00:05 -0500 (EST)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [BUGS] Bug in create operator and/or initdb\"\n },\n {\n \"msg_contents\": \"Bruce Momjian writes:\\n> Added to release checklist:\\n> \\t* Update inet/cidr data types with newest Bind patches\\n\\nYou should also add \\\"check for zic database updates\\\".\\n\\n\\t\\t\\tregards, tom lane\\n\",\n \"msg_date\": \"Wed, 02 Feb 2005 12:02:58 -0500\",\n \"msg_from\": \"Tom Lane \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [BUGS] Bug in create operator and/or initdb \"\n },\n {\n \"msg_contents\": \"Tom Lane wrote:\\n> Bruce Momjian writes:\\n> > Added to release checklist:\\n> > \\t* Update inet/cidr data types with newest Bind patches\\n> \\n> You should also add \\\"check for zic database updates\\\".\\n\\nUh, we already have:\\n\\n\\t* Update timezone data to match latest zic database (see\\n\\t src/timezone/README)\\n\\n-- \\n Bruce Momjian | http://candle.pha.pa.us\\n pgman@candle.pha.pa.us | (610) 359-1001\\n + If your life is a hard drive, | 13 Roberts Road\\n + Christ can be your backup. | Newtown Square, Pennsylvania 19073\\n\",\n \"msg_date\": \"Wed, 2 Feb 2005 12:38:04 -0500 (EST)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [BUGS] Bug in create operator and/or initdb\"\n }\n]"}}},{"rowIdx":129,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"\nHiya...\n\n\tWould send email directly, but only have the address at home :(\n\n\tD'Arcy...you've been working on a Python interface, correct? Do\nwe want to include that as part of src/interfaces? *raised eyebrows*\n\n\n","msg_date":"Thu, 15 Jan 1998 11:57:17 -0500 (EST)","msg_from":"The Hermit Hacker ","msg_from_op":true,"msg_subject":"Python...?"},{"msg_contents":"Thus spake The Hermit Hacker\n> \tWould send email directly, but only have the address at home :(\n> \n> \tD'Arcy...you've been working on a Python interface, correct? Do\n> we want to include that as part of src/interfaces? *raised eyebrows*\n\nSure. I'm currently working on a few enhancements and I hope to make\nit compliant with the Python standard database API. The current version\nworks fine though so go right ahead.\n\n-- \nD'Arcy J.M. Cain | Democracy is three wolves\nhttp://www.druid.net/darcy/ | and a sheep voting on\n+1 416 424 2871 (DoD#0082) (eNTP) | what's for dinner.\n","msg_date":"Thu, 15 Jan 1998 13:37:43 -0500 (EST)","msg_from":"darcy@druid.net (D'Arcy J.M. Cain)","msg_from_op":false,"msg_subject":"Re: [HACKERS] Python...?"}],"string":"[\n {\n \"msg_contents\": \"\\nHiya...\\n\\n\\tWould send email directly, but only have the address at home :(\\n\\n\\tD'Arcy...you've been working on a Python interface, correct? Do\\nwe want to include that as part of src/interfaces? *raised eyebrows*\\n\\n\\n\",\n \"msg_date\": \"Thu, 15 Jan 1998 11:57:17 -0500 (EST)\",\n \"msg_from\": \"The Hermit Hacker \",\n \"msg_from_op\": true,\n \"msg_subject\": \"Python...?\"\n },\n {\n \"msg_contents\": \"Thus spake The Hermit Hacker\\n> \\tWould send email directly, but only have the address at home :(\\n> \\n> \\tD'Arcy...you've been working on a Python interface, correct? Do\\n> we want to include that as part of src/interfaces? *raised eyebrows*\\n\\nSure. I'm currently working on a few enhancements and I hope to make\\nit compliant with the Python standard database API. The current version\\nworks fine though so go right ahead.\\n\\n-- \\nD'Arcy J.M. Cain | Democracy is three wolves\\nhttp://www.druid.net/darcy/ | and a sheep voting on\\n+1 416 424 2871 (DoD#0082) (eNTP) | what's for dinner.\\n\",\n \"msg_date\": \"Thu, 15 Jan 1998 13:37:43 -0500 (EST)\",\n \"msg_from\": \"darcy@druid.net (D'Arcy J.M. Cain)\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Python...?\"\n }\n]"}}},{"rowIdx":130,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"Hi,\n\nI have installed PostgreSQL6.2.1 from my own user account (I was just trying\nto test out something first). It seems to be running too slow. For example,\nthe self join of a table with roughly 300K records takes 2-3 hours. There is\nan index on the join attribute.\nI am running it on a 8 processor (each is a 248 MHz SUNW,UltraSPARC-II)\n machine with a total of 2.0 GB main memory.\nBut, since it is not parallelized, it may get the power of only one processor.\n\nDoes it make a difference in performance since I have installed it from my\nuser account, and not the root. Also, I compiled it using gcc on a \nsolaris 2.5 machine and the machine I ran it is a solaris 2.6 machine. I did\nthat since the solaris 2.6 machine is the fastest we have here and the \nexecutable\ncompiled on 2.5 was running properly also. Can that have an impact on the\nperformance?\n\nThanks\n--shiby\n\n","msg_date":"Thu, 15 Jan 1998 13:01:37 -0500","msg_from":"Shiby Thomas ","msg_from_op":true,"msg_subject":"postgres performance"},{"msg_contents":"On Thu, 15 Jan 1998, Shiby Thomas wrote:\n\n> Hi,\n> \n> I have installed PostgreSQL6.2.1 from my own user account (I was just trying\n> to test out something first). It seems to be running too slow. For example,\n> the self join of a table with roughly 300K records takes 2-3 hours. There is\n> an index on the join attribute.\n> I am running it on a 8 processor (each is a 248 MHz SUNW,UltraSPARC-II)\n> machine with a total of 2.0 GB main memory.\n> But, since it is not parallelized, it may get the power of only one processor.\n> \n> Does it make a difference in performance since I have installed it from my\n> user account, and not the root. Also, I compiled it using gcc on a \n> solaris 2.5 machine and the machine I ran it is a solaris 2.6 machine. I did\n> that since the solaris 2.6 machine is the fastest we have here and the \n> executable\n> compiled on 2.5 was running properly also. Can that have an impact on the\n> performance?\n\n\tThere may be optimizations in the 2.6 libraries that would improve\nperformance, but I wouldn't suspect that it would make *that* big of a\ndifference. What is your SQL/join statemnt? How are you running\npostmaster? What does 'explain' show?\n\n\n","msg_date":"Thu, 15 Jan 1998 13:15:59 -0500 (EST)","msg_from":"The Hermit Hacker ","msg_from_op":false,"msg_subject":"Re: [HACKERS] postgres performance"},{"msg_contents":"> \n> Hi,\n> \n> I have installed PostgreSQL6.2.1 from my own user account (I was just trying\n> to test out something first). It seems to be running too slow. For example,\n> the self join of a table with roughly 300K records takes 2-3 hours. There is\n> an index on the join attribute.\n> I am running it on a 8 processor (each is a 248 MHz SUNW,UltraSPARC-II)\n> machine with a total of 2.0 GB main memory.\n> But, since it is not parallelized, it may get the power of only one processor.\n\nYou said a self-join. I think we have a performance problem there. \nVadim?\n\n\n> \n> Does it make a difference in performance since I have installed it from my\n> user account, and not the root. Also, I compiled it using gcc on a \n> solaris 2.5 machine and the machine I ran it is a solaris 2.6 machine. I did\n> that since the solaris 2.6 machine is the fastest we have here and the \n> executable\n> compiled on 2.5 was running properly also. Can that have an impact on the\n> performance?\n> \n> Thanks\n> --shiby\n> \n> \n> \n\n\n-- \nBruce Momjian\nmaillist@candle.pha.pa.us\n","msg_date":"Thu, 15 Jan 1998 13:23:07 -0500 (EST)","msg_from":"Bruce Momjian ","msg_from_op":false,"msg_subject":"Re: [HACKERS] postgres performance"},{"msg_contents":"\n=> \tThere may be optimizations in the 2.6 libraries that would improve\n=> performance, but I wouldn't suspect that it would make *that* big of a\n=> difference. What is your SQL/join statemnt? How are you running\n=> postmaster? What does 'explain' show?\n=> \nThe complete query is this:\n\nselect item1, item2, count(t1.tid) into table f2_temp from data t1, data t2, \nc2\nwhere t1.item = c2.item1 and t2.item = c2.item2 and t1.tid = t2.tid group by \nite\nm1, item2\n\ndata is a table with 2 integer columns (tid, item) and it has ~300K records\nc2 is a table (item1, item2), both integers and has ~1.5K records.\n\nI was directly running postgres with the -B and -S flags to give more buffers\nand sortMem. I also tried several join plans by the -f flags. Hash join works\nthe best and that itself is too slow (perhaps due to the self join)\n\n--shiby\n\n\n\n","msg_date":"Thu, 15 Jan 1998 13:33:08 -0500","msg_from":"Shiby Thomas ","msg_from_op":true,"msg_subject":"Re: [HACKERS] postgres performance "},{"msg_contents":"> \n> \n> => \tThere may be optimizations in the 2.6 libraries that would improve\n> => performance, but I wouldn't suspect that it would make *that* big of a\n> => difference. What is your SQL/join statemnt? How are you running\n> => postmaster? What does 'explain' show?\n> => \n> The complete query is this:\n> \n> select item1, item2, count(t1.tid) into table f2_temp from data t1, data t2, \n> c2\n> where t1.item = c2.item1 and t2.item = c2.item2 and t1.tid = t2.tid group by \n> ite\n> m1, item2\n> \n> data is a table with 2 integer columns (tid, item) and it has ~300K records\n> c2 is a table (item1, item2), both integers and has ~1.5K records.\n> \n> I was directly running postgres with the -B and -S flags to give more buffers\n> and sortMem. I also tried several join plans by the -f flags. Hash join works\n> the best and that itself is too slow (perhaps due to the self join)\n> \n\nI have a possible workaround. Turn GEQO on:\n\n\tSET GEQO ON=1\n\nand try it. Let us know.\n\n-- \nBruce Momjian\nmaillist@candle.pha.pa.us\n","msg_date":"Thu, 15 Jan 1998 19:26:45 -0500 (EST)","msg_from":"Bruce Momjian ","msg_from_op":false,"msg_subject":"Re: [HACKERS] postgres performance"},{"msg_contents":"Shiby Thomas wrote:\n> \n> => There may be optimizations in the 2.6 libraries that would improve\n> => performance, but I wouldn't suspect that it would make *that* big of a\n> => difference. What is your SQL/join statemnt? How are you running\n> => postmaster? What does 'explain' show?\n> =>\n> The complete query is this:\n> \n> select item1, item2, count(t1.tid) into table f2_temp from data t1, data t2,\n> c2\n> where t1.item = c2.item1 and t2.item = c2.item2 and t1.tid = t2.tid group by\n> ite\n> m1, item2\n> \n> data is a table with 2 integer columns (tid, item) and it has ~300K records\n> c2 is a table (item1, item2), both integers and has ~1.5K records.\n> \n> I was directly running postgres with the -B and -S flags to give more buffers\n> and sortMem. I also tried several join plans by the -f flags. Hash join works\n> the best and that itself is too slow (perhaps due to the self join)\n\nIndices ?\nEXPLAIN ?\n\nVadim\n","msg_date":"Fri, 16 Jan 1998 10:30:22 +0700","msg_from":"\"Vadim B. Mikheev\" ","msg_from_op":false,"msg_subject":"Re: [HACKERS] postgres performance"}],"string":"[\n {\n \"msg_contents\": \"Hi,\\n\\nI have installed PostgreSQL6.2.1 from my own user account (I was just trying\\nto test out something first). It seems to be running too slow. For example,\\nthe self join of a table with roughly 300K records takes 2-3 hours. There is\\nan index on the join attribute.\\nI am running it on a 8 processor (each is a 248 MHz SUNW,UltraSPARC-II)\\n machine with a total of 2.0 GB main memory.\\nBut, since it is not parallelized, it may get the power of only one processor.\\n\\nDoes it make a difference in performance since I have installed it from my\\nuser account, and not the root. Also, I compiled it using gcc on a \\nsolaris 2.5 machine and the machine I ran it is a solaris 2.6 machine. I did\\nthat since the solaris 2.6 machine is the fastest we have here and the \\nexecutable\\ncompiled on 2.5 was running properly also. Can that have an impact on the\\nperformance?\\n\\nThanks\\n--shiby\\n\\n\",\n \"msg_date\": \"Thu, 15 Jan 1998 13:01:37 -0500\",\n \"msg_from\": \"Shiby Thomas \",\n \"msg_from_op\": true,\n \"msg_subject\": \"postgres performance\"\n },\n {\n \"msg_contents\": \"On Thu, 15 Jan 1998, Shiby Thomas wrote:\\n\\n> Hi,\\n> \\n> I have installed PostgreSQL6.2.1 from my own user account (I was just trying\\n> to test out something first). It seems to be running too slow. For example,\\n> the self join of a table with roughly 300K records takes 2-3 hours. There is\\n> an index on the join attribute.\\n> I am running it on a 8 processor (each is a 248 MHz SUNW,UltraSPARC-II)\\n> machine with a total of 2.0 GB main memory.\\n> But, since it is not parallelized, it may get the power of only one processor.\\n> \\n> Does it make a difference in performance since I have installed it from my\\n> user account, and not the root. Also, I compiled it using gcc on a \\n> solaris 2.5 machine and the machine I ran it is a solaris 2.6 machine. I did\\n> that since the solaris 2.6 machine is the fastest we have here and the \\n> executable\\n> compiled on 2.5 was running properly also. Can that have an impact on the\\n> performance?\\n\\n\\tThere may be optimizations in the 2.6 libraries that would improve\\nperformance, but I wouldn't suspect that it would make *that* big of a\\ndifference. What is your SQL/join statemnt? How are you running\\npostmaster? What does 'explain' show?\\n\\n\\n\",\n \"msg_date\": \"Thu, 15 Jan 1998 13:15:59 -0500 (EST)\",\n \"msg_from\": \"The Hermit Hacker \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] postgres performance\"\n },\n {\n \"msg_contents\": \"> \\n> Hi,\\n> \\n> I have installed PostgreSQL6.2.1 from my own user account (I was just trying\\n> to test out something first). It seems to be running too slow. For example,\\n> the self join of a table with roughly 300K records takes 2-3 hours. There is\\n> an index on the join attribute.\\n> I am running it on a 8 processor (each is a 248 MHz SUNW,UltraSPARC-II)\\n> machine with a total of 2.0 GB main memory.\\n> But, since it is not parallelized, it may get the power of only one processor.\\n\\nYou said a self-join. I think we have a performance problem there. \\nVadim?\\n\\n\\n> \\n> Does it make a difference in performance since I have installed it from my\\n> user account, and not the root. Also, I compiled it using gcc on a \\n> solaris 2.5 machine and the machine I ran it is a solaris 2.6 machine. I did\\n> that since the solaris 2.6 machine is the fastest we have here and the \\n> executable\\n> compiled on 2.5 was running properly also. Can that have an impact on the\\n> performance?\\n> \\n> Thanks\\n> --shiby\\n> \\n> \\n> \\n\\n\\n-- \\nBruce Momjian\\nmaillist@candle.pha.pa.us\\n\",\n \"msg_date\": \"Thu, 15 Jan 1998 13:23:07 -0500 (EST)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] postgres performance\"\n },\n {\n \"msg_contents\": \"\\n=> \\tThere may be optimizations in the 2.6 libraries that would improve\\n=> performance, but I wouldn't suspect that it would make *that* big of a\\n=> difference. What is your SQL/join statemnt? How are you running\\n=> postmaster? What does 'explain' show?\\n=> \\nThe complete query is this:\\n\\nselect item1, item2, count(t1.tid) into table f2_temp from data t1, data t2, \\nc2\\nwhere t1.item = c2.item1 and t2.item = c2.item2 and t1.tid = t2.tid group by \\nite\\nm1, item2\\n\\ndata is a table with 2 integer columns (tid, item) and it has ~300K records\\nc2 is a table (item1, item2), both integers and has ~1.5K records.\\n\\nI was directly running postgres with the -B and -S flags to give more buffers\\nand sortMem. I also tried several join plans by the -f flags. Hash join works\\nthe best and that itself is too slow (perhaps due to the self join)\\n\\n--shiby\\n\\n\\n\\n\",\n \"msg_date\": \"Thu, 15 Jan 1998 13:33:08 -0500\",\n \"msg_from\": \"Shiby Thomas \",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: [HACKERS] postgres performance \"\n },\n {\n \"msg_contents\": \"> \\n> \\n> => \\tThere may be optimizations in the 2.6 libraries that would improve\\n> => performance, but I wouldn't suspect that it would make *that* big of a\\n> => difference. What is your SQL/join statemnt? How are you running\\n> => postmaster? What does 'explain' show?\\n> => \\n> The complete query is this:\\n> \\n> select item1, item2, count(t1.tid) into table f2_temp from data t1, data t2, \\n> c2\\n> where t1.item = c2.item1 and t2.item = c2.item2 and t1.tid = t2.tid group by \\n> ite\\n> m1, item2\\n> \\n> data is a table with 2 integer columns (tid, item) and it has ~300K records\\n> c2 is a table (item1, item2), both integers and has ~1.5K records.\\n> \\n> I was directly running postgres with the -B and -S flags to give more buffers\\n> and sortMem. I also tried several join plans by the -f flags. Hash join works\\n> the best and that itself is too slow (perhaps due to the self join)\\n> \\n\\nI have a possible workaround. Turn GEQO on:\\n\\n\\tSET GEQO ON=1\\n\\nand try it. Let us know.\\n\\n-- \\nBruce Momjian\\nmaillist@candle.pha.pa.us\\n\",\n \"msg_date\": \"Thu, 15 Jan 1998 19:26:45 -0500 (EST)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] postgres performance\"\n },\n {\n \"msg_contents\": \"Shiby Thomas wrote:\\n> \\n> => There may be optimizations in the 2.6 libraries that would improve\\n> => performance, but I wouldn't suspect that it would make *that* big of a\\n> => difference. What is your SQL/join statemnt? How are you running\\n> => postmaster? What does 'explain' show?\\n> =>\\n> The complete query is this:\\n> \\n> select item1, item2, count(t1.tid) into table f2_temp from data t1, data t2,\\n> c2\\n> where t1.item = c2.item1 and t2.item = c2.item2 and t1.tid = t2.tid group by\\n> ite\\n> m1, item2\\n> \\n> data is a table with 2 integer columns (tid, item) and it has ~300K records\\n> c2 is a table (item1, item2), both integers and has ~1.5K records.\\n> \\n> I was directly running postgres with the -B and -S flags to give more buffers\\n> and sortMem. I also tried several join plans by the -f flags. Hash join works\\n> the best and that itself is too slow (perhaps due to the self join)\\n\\nIndices ?\\nEXPLAIN ?\\n\\nVadim\\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 10:30:22 +0700\",\n \"msg_from\": \"\\\"Vadim B. Mikheev\\\" \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] postgres performance\"\n }\n]"}}},{"rowIdx":131,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"\n> OIDs are a bastardization of the relational model. If you have to keep\n> them, then do so, but their use should be SEVERELY discouraged.\n\nExplain yourself, please.\n\nIn my opinion, I view the OID in the same way as I view the SERIAL datatype\nin Informix. It is usually a primary key field in a table. On an insert,\nthe DBMS will increment the current serial-maximum (for that table) and insert\nthe new serial value into that field; thus creating a unique identifier.\n\nThere are differences between OID and SERIAL. The main difference is that\nthe OID field (always called 'oid') is always present whereas a DB designer\nexplicitly creates 'id' fields (of SERIAL type). Thus, postgresql treats\nevery table as an object (which is not always the case).\n\nIs the SERIAL datatype part of the SQL-92 standard? Does PostgreSQL plan\nto support SERIAL in the future. This would be an acceptable replacement\nfor the OID.\n\n-Bryan Basham\n","msg_date":"Thu, 15 Jan 1998 11:39:16 -0700 (MST)","msg_from":"Bryan Basham ","msg_from_op":true,"msg_subject":"Re: [QUESTIONS] Arrays (inserting and removing)"},{"msg_contents":"On Thu, 15 Jan 1998, Bryan Basham wrote:\n\n> \n> > OIDs are a bastardization of the relational model. If you have to keep\n> > them, then do so, but their use should be SEVERELY discouraged.\n> \n> Explain yourself, please.\n> \n> In my opinion, I view the OID in the same way as I view the SERIAL datatype\n> in Informix. It is usually a primary key field in a table. On an insert,\n> the DBMS will increment the current serial-maximum (for that table) and insert\n> the new serial value into that field; thus creating a unique identifier.\n> \n> There are differences between OID and SERIAL. The main difference is that\n> the OID field (always called 'oid') is always present whereas a DB designer\n> explicitly creates 'id' fields (of SERIAL type). Thus, postgresql treats\n> every table as an object (which is not always the case).\n\n\tMajor problem with OID: OIDs are sequenced across the database,\nnot the table. ie. tableA inserts with OID #1, tableB inserts with OID\n#2, tableA inserts next record with OID #3, tableC then gets #4, etc...\n\n\tAnd...# of OIDs is finite...so if you have a lot of tables with\nalot of data in each...you run the risk of running out.\n\n\tIn this sense, sequences are the better alternative, but again,\nthey are a newer feature to PostgreSQL then the code that I wrote using\nOIDs\n\n\n","msg_date":"Thu, 15 Jan 1998 13:48:24 -0500 (EST)","msg_from":"The Hermit Hacker ","msg_from_op":false,"msg_subject":"Re: [HACKERS] Re: [QUESTIONS] Arrays (inserting and removing)"},{"msg_contents":"On Thu, 15 Jan 1998, The Hermit Hacker wrote:\n\n> \tMajor problem with OID: OIDs are sequenced across the database,\n> not the table. ie. tableA inserts with OID #1, tableB inserts with OID\n> #2, tableA inserts next record with OID #3, tableC then gets #4, etc...\n\nin my oo world (i.e., systems i develop), i use oid's to determine the\ntype of object and due to the limitation of postgresql's oids, i use a\nseparate field for the system's oid - kinda redundant but gotta live with\nit. nonetheless, postgresql's oids can still be improved.\n\n> \tAnd...# of OIDs is finite...so if you have a lot of tables with\n> alot of data in each...you run the risk of running out.\n\nfinitely limited :)\n\n[---]\nNeil D. Quiogue \nIPhil Communications Network, Inc.\nOther: neil@postgresql.org\n\n","msg_date":"Fri, 16 Jan 1998 09:04:10 +0800 (HKT)","msg_from":"\"neil d. quiogue\" ","msg_from_op":false,"msg_subject":"Re: [HACKERS] Re: [QUESTIONS] Arrays (inserting and removing)"}],"string":"[\n {\n \"msg_contents\": \"\\n> OIDs are a bastardization of the relational model. If you have to keep\\n> them, then do so, but their use should be SEVERELY discouraged.\\n\\nExplain yourself, please.\\n\\nIn my opinion, I view the OID in the same way as I view the SERIAL datatype\\nin Informix. It is usually a primary key field in a table. On an insert,\\nthe DBMS will increment the current serial-maximum (for that table) and insert\\nthe new serial value into that field; thus creating a unique identifier.\\n\\nThere are differences between OID and SERIAL. The main difference is that\\nthe OID field (always called 'oid') is always present whereas a DB designer\\nexplicitly creates 'id' fields (of SERIAL type). Thus, postgresql treats\\nevery table as an object (which is not always the case).\\n\\nIs the SERIAL datatype part of the SQL-92 standard? Does PostgreSQL plan\\nto support SERIAL in the future. This would be an acceptable replacement\\nfor the OID.\\n\\n-Bryan Basham\\n\",\n \"msg_date\": \"Thu, 15 Jan 1998 11:39:16 -0700 (MST)\",\n \"msg_from\": \"Bryan Basham \",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: [QUESTIONS] Arrays (inserting and removing)\"\n },\n {\n \"msg_contents\": \"On Thu, 15 Jan 1998, Bryan Basham wrote:\\n\\n> \\n> > OIDs are a bastardization of the relational model. If you have to keep\\n> > them, then do so, but their use should be SEVERELY discouraged.\\n> \\n> Explain yourself, please.\\n> \\n> In my opinion, I view the OID in the same way as I view the SERIAL datatype\\n> in Informix. It is usually a primary key field in a table. On an insert,\\n> the DBMS will increment the current serial-maximum (for that table) and insert\\n> the new serial value into that field; thus creating a unique identifier.\\n> \\n> There are differences between OID and SERIAL. The main difference is that\\n> the OID field (always called 'oid') is always present whereas a DB designer\\n> explicitly creates 'id' fields (of SERIAL type). Thus, postgresql treats\\n> every table as an object (which is not always the case).\\n\\n\\tMajor problem with OID: OIDs are sequenced across the database,\\nnot the table. ie. tableA inserts with OID #1, tableB inserts with OID\\n#2, tableA inserts next record with OID #3, tableC then gets #4, etc...\\n\\n\\tAnd...# of OIDs is finite...so if you have a lot of tables with\\nalot of data in each...you run the risk of running out.\\n\\n\\tIn this sense, sequences are the better alternative, but again,\\nthey are a newer feature to PostgreSQL then the code that I wrote using\\nOIDs\\n\\n\\n\",\n \"msg_date\": \"Thu, 15 Jan 1998 13:48:24 -0500 (EST)\",\n \"msg_from\": \"The Hermit Hacker \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Re: [QUESTIONS] Arrays (inserting and removing)\"\n },\n {\n \"msg_contents\": \"On Thu, 15 Jan 1998, The Hermit Hacker wrote:\\n\\n> \\tMajor problem with OID: OIDs are sequenced across the database,\\n> not the table. ie. tableA inserts with OID #1, tableB inserts with OID\\n> #2, tableA inserts next record with OID #3, tableC then gets #4, etc...\\n\\nin my oo world (i.e., systems i develop), i use oid's to determine the\\ntype of object and due to the limitation of postgresql's oids, i use a\\nseparate field for the system's oid - kinda redundant but gotta live with\\nit. nonetheless, postgresql's oids can still be improved.\\n\\n> \\tAnd...# of OIDs is finite...so if you have a lot of tables with\\n> alot of data in each...you run the risk of running out.\\n\\nfinitely limited :)\\n\\n[---]\\nNeil D. Quiogue \\nIPhil Communications Network, Inc.\\nOther: neil@postgresql.org\\n\\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 09:04:10 +0800 (HKT)\",\n \"msg_from\": \"\\\"neil d. quiogue\\\" \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Re: [QUESTIONS] Arrays (inserting and removing)\"\n }\n]"}}},{"rowIdx":132,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"\nHi...\n\n\tI've just installed (massaged in?) a major major patch for PL(?)\nthat was submitted in Nov...snapshot will be created in about 18hrs or\nso...please test and report back any problems...haven't had a chance to\ntest compile it here yet :(\n\n\n\n","msg_date":"Thu, 15 Jan 1998 14:44:20 -0500 (EST)","msg_from":"The Hermit Hacker ","msg_from_op":true,"msg_subject":"*Major* Patch for PL "},{"msg_contents":"> Hi...\n>\n> I've just installed (massaged in?) a major major patch for PL(?)\n> that was submitted in Nov...snapshot will be created in about 18hrs or\n> so...please test and report back any problems...haven't had a chance to\n> test compile it here yet :(\n\n Waited for that so long - thanks. I'll take a look at it and\n run my tests asap.\n\n\nUntil later, Jan\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== jwieck@debis.com (Jan Wieck) #\n\n\n","msg_date":"Mon, 19 Jan 1998 12:12:11 +0100 (MET)","msg_from":"jwieck@debis.com (Jan Wieck)","msg_from_op":false,"msg_subject":"Re: [HACKERS] *Major* Patch for PL"},{"msg_contents":">\n> > Hi...\n> >\n> > I've just installed (massaged in?) a major major patch for PL(?)\n> > that was submitted in Nov...snapshot will be created in about 18hrs or\n> > so...please test and report back any problems...haven't had a chance to\n> > test compile it here yet :(\n>\n> Waited for that so long - thanks. I'll take a look at it and\n> run my tests asap.\n>\n\n Looks O.K. - tests passed through.\n\n Now anything is prepared for PL support in all places\n (aggregates, operators, etc.). At least PL/Tcl ran in all\n that areas.\n\n Found a little bug in interfaces/libpq/fe-connect.c during\n tests. In line 403 conn->pgpass is set to DefaultPassword and\n later in line 665 this might be free()'d. Just wrapping a\n strdup() around DefaultPassword fixed it.\n\n Did anybody worked on other languages for PL modules since\n this new interface was designed? Who would like to dive into\n (perl/phyton/...)? Who would like to co-work on a pure\n PL/pgSQL?\n\n\nUntil later, Jan\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== jwieck@debis.com (Jan Wieck) #\n\n\n","msg_date":"Mon, 19 Jan 1998 14:18:01 +0100 (MET)","msg_from":"jwieck@debis.com (Jan Wieck)","msg_from_op":false,"msg_subject":"Re: [HACKERS] *Major* Patch for PL"}],"string":"[\n {\n \"msg_contents\": \"\\nHi...\\n\\n\\tI've just installed (massaged in?) a major major patch for PL(?)\\nthat was submitted in Nov...snapshot will be created in about 18hrs or\\nso...please test and report back any problems...haven't had a chance to\\ntest compile it here yet :(\\n\\n\\n\\n\",\n \"msg_date\": \"Thu, 15 Jan 1998 14:44:20 -0500 (EST)\",\n \"msg_from\": \"The Hermit Hacker \",\n \"msg_from_op\": true,\n \"msg_subject\": \"*Major* Patch for PL \"\n },\n {\n \"msg_contents\": \"> Hi...\\n>\\n> I've just installed (massaged in?) a major major patch for PL(?)\\n> that was submitted in Nov...snapshot will be created in about 18hrs or\\n> so...please test and report back any problems...haven't had a chance to\\n> test compile it here yet :(\\n\\n Waited for that so long - thanks. I'll take a look at it and\\n run my tests asap.\\n\\n\\nUntil later, Jan\\n\\n--\\n\\n#======================================================================#\\n# It's easier to get forgiveness for being wrong than for being right. #\\n# Let's break this rule - forgive me. #\\n#======================================== jwieck@debis.com (Jan Wieck) #\\n\\n\\n\",\n \"msg_date\": \"Mon, 19 Jan 1998 12:12:11 +0100 (MET)\",\n \"msg_from\": \"jwieck@debis.com (Jan Wieck)\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] *Major* Patch for PL\"\n },\n {\n \"msg_contents\": \">\\n> > Hi...\\n> >\\n> > I've just installed (massaged in?) a major major patch for PL(?)\\n> > that was submitted in Nov...snapshot will be created in about 18hrs or\\n> > so...please test and report back any problems...haven't had a chance to\\n> > test compile it here yet :(\\n>\\n> Waited for that so long - thanks. I'll take a look at it and\\n> run my tests asap.\\n>\\n\\n Looks O.K. - tests passed through.\\n\\n Now anything is prepared for PL support in all places\\n (aggregates, operators, etc.). At least PL/Tcl ran in all\\n that areas.\\n\\n Found a little bug in interfaces/libpq/fe-connect.c during\\n tests. In line 403 conn->pgpass is set to DefaultPassword and\\n later in line 665 this might be free()'d. Just wrapping a\\n strdup() around DefaultPassword fixed it.\\n\\n Did anybody worked on other languages for PL modules since\\n this new interface was designed? Who would like to dive into\\n (perl/phyton/...)? Who would like to co-work on a pure\\n PL/pgSQL?\\n\\n\\nUntil later, Jan\\n\\n--\\n\\n#======================================================================#\\n# It's easier to get forgiveness for being wrong than for being right. #\\n# Let's break this rule - forgive me. #\\n#======================================== jwieck@debis.com (Jan Wieck) #\\n\\n\\n\",\n \"msg_date\": \"Mon, 19 Jan 1998 14:18:01 +0100 (MET)\",\n \"msg_from\": \"jwieck@debis.com (Jan Wieck)\",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] *Major* Patch for PL\"\n }\n]"}}},{"rowIdx":133,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"I keep getting:\n\nToo Large Allocation Request(\"!(0 < (size) && (size) <= \n(0xfffffff)):size=-3 [0xfffffffd]\", File: \"mcxt.c\", Line: 232)\n!(0 < (size) && (size) <= (0xfffffff)) (0) [No such file or directory]\n\nfrom a simple select:\n\nSELECT * from provinces;\n\nincidentally, \"SELECT oid from provinces;\" works...\n(as does any select on one component)... but checking...\nNope - countrycode fails. Prints out garbage (8bit garbage at that)\nAs do all fields after that - though they're just bad data.\n\nFor comparison, I have another database with \"varchar(50), integer,\ntimestamp, timestamp\" - and the only noncorrupted field is the first\nvarchar() entry. Perhaps it's in the string-handling?\n\nThe contents of the database:\n---\nCREATE TABLE provinces\n (\n code\t\tchar(4),\t-- ...\n name\t\tvarchar(50),\t-- ...\n countrycode\tchar(2),\t-- ...\n country\tvarchar(50),\t-- ...\n\n -- creation and modification dates of this record...\n creationdate\ttimestamp default now(),\n modifydate\ttimestamp default now()\n );\n\nINSERT INTO provinces (code, name, countrycode, country)\n\tVALUES ('bc','British Columbia','ca','Canada');\nINSERT INTO provinces (code, name, countrycode, country)\n\tVALUES ('ab','Alberta', 'ca','Canada');\nINSERT INTO provinces (code, name, countrycode, country)\n\tVALUES ('alta','Alberta', 'ca','Canada');\n-- one gets tired of databases referring to \"state\"... so time to turn the\n-- tables *grin*\nINSERT INTO provinces (code, name, countrycode, country)\n\tVALUES ('wa', 'Washington',\t'us','United States');\n---\n\nIncidentally, this worked perfectly up 'till christmas-CVS. I had some\nother problems so I upgraded... and now it core-dumps 'psql' and sends\nthis to the backend.\n\nOn debugging psql I found the error in \"UP()\", but as I don't know how to\nuse gdb properly I can't look any deeper (I've always preferred logfiles -\nespecially when working on videodriver code :)\n[mayhaps psql needs better recovery code... will have to look at that\nsometime]\n\nJDBC also fails FWIW - with these messages: (same query)\njava.sql.SQLException: Error reading from backend: java.io.IOException: EOF\n at postgresql.PG_Stream.ReceiveInteger(PG_Stream.java:184)\n at postgresql.PG_Stream.ReceiveTuple(PG_Stream.java:256)\n at postgresql.Connection.ExecSQL(Connection.java:626)\n at postgresql.Statement.execute(Statement.java:259)\n at postgresql.Statement.executeQuery(Statement.java:46)\n at Mauve.Main.readProvinces(Main.java:234)\n at Mauve.Main.main(Main.java:363)\njava.sql.SQLException: I/O Error: java.io.IOException: Broken pipe\n at postgresql.Connection.ExecSQL(Connection.java:583)\n at postgresql.Statement.execute(Statement.java:259)\n at postgresql.Statement.executeQuery(Statement.java:46)\n at Mauve.Main.readPaymentForms(Main.java:252)\n at Mauve.Main.main(Main.java:366)\n\nI suspect a bug somewhere - but don't know how to find it. Or if my\ndata's incorrect.... many things have changed in postgres.\n\nPlatform:\n\tlinux 2.0.33 (got tired of being flamed for using 2.1 kernels)\n\tglibc-2.0.5c (standard! ref: redhat-5.0)\n\tgcc 2.7.2.1 (I also have egc-1.0 but I don't trust it yet)\n\tCyrix 6x86, 40M ram, ~5G drivespace\n\nThough it's nice to finally see large objects working *grin*.\n\nG'day, eh? :)\n\t- Teunis\n\n","msg_date":"Thu, 15 Jan 1998 12:54:13 -0700 (MST)","msg_from":"teunis ","msg_from_op":true,"msg_subject":"a small problem with current CVS - wondering how to debug"}],"string":"[\n {\n \"msg_contents\": \"I keep getting:\\n\\nToo Large Allocation Request(\\\"!(0 < (size) && (size) <= \\n(0xfffffff)):size=-3 [0xfffffffd]\\\", File: \\\"mcxt.c\\\", Line: 232)\\n!(0 < (size) && (size) <= (0xfffffff)) (0) [No such file or directory]\\n\\nfrom a simple select:\\n\\nSELECT * from provinces;\\n\\nincidentally, \\\"SELECT oid from provinces;\\\" works...\\n(as does any select on one component)... but checking...\\nNope - countrycode fails. Prints out garbage (8bit garbage at that)\\nAs do all fields after that - though they're just bad data.\\n\\nFor comparison, I have another database with \\\"varchar(50), integer,\\ntimestamp, timestamp\\\" - and the only noncorrupted field is the first\\nvarchar() entry. Perhaps it's in the string-handling?\\n\\nThe contents of the database:\\n---\\nCREATE TABLE provinces\\n (\\n code\\t\\tchar(4),\\t-- ...\\n name\\t\\tvarchar(50),\\t-- ...\\n countrycode\\tchar(2),\\t-- ...\\n country\\tvarchar(50),\\t-- ...\\n\\n -- creation and modification dates of this record...\\n creationdate\\ttimestamp default now(),\\n modifydate\\ttimestamp default now()\\n );\\n\\nINSERT INTO provinces (code, name, countrycode, country)\\n\\tVALUES ('bc','British Columbia','ca','Canada');\\nINSERT INTO provinces (code, name, countrycode, country)\\n\\tVALUES ('ab','Alberta', 'ca','Canada');\\nINSERT INTO provinces (code, name, countrycode, country)\\n\\tVALUES ('alta','Alberta', 'ca','Canada');\\n-- one gets tired of databases referring to \\\"state\\\"... so time to turn the\\n-- tables *grin*\\nINSERT INTO provinces (code, name, countrycode, country)\\n\\tVALUES ('wa', 'Washington',\\t'us','United States');\\n---\\n\\nIncidentally, this worked perfectly up 'till christmas-CVS. I had some\\nother problems so I upgraded... and now it core-dumps 'psql' and sends\\nthis to the backend.\\n\\nOn debugging psql I found the error in \\\"UP()\\\", but as I don't know how to\\nuse gdb properly I can't look any deeper (I've always preferred logfiles -\\nespecially when working on videodriver code :)\\n[mayhaps psql needs better recovery code... will have to look at that\\nsometime]\\n\\nJDBC also fails FWIW - with these messages: (same query)\\njava.sql.SQLException: Error reading from backend: java.io.IOException: EOF\\n at postgresql.PG_Stream.ReceiveInteger(PG_Stream.java:184)\\n at postgresql.PG_Stream.ReceiveTuple(PG_Stream.java:256)\\n at postgresql.Connection.ExecSQL(Connection.java:626)\\n at postgresql.Statement.execute(Statement.java:259)\\n at postgresql.Statement.executeQuery(Statement.java:46)\\n at Mauve.Main.readProvinces(Main.java:234)\\n at Mauve.Main.main(Main.java:363)\\njava.sql.SQLException: I/O Error: java.io.IOException: Broken pipe\\n at postgresql.Connection.ExecSQL(Connection.java:583)\\n at postgresql.Statement.execute(Statement.java:259)\\n at postgresql.Statement.executeQuery(Statement.java:46)\\n at Mauve.Main.readPaymentForms(Main.java:252)\\n at Mauve.Main.main(Main.java:366)\\n\\nI suspect a bug somewhere - but don't know how to find it. Or if my\\ndata's incorrect.... many things have changed in postgres.\\n\\nPlatform:\\n\\tlinux 2.0.33 (got tired of being flamed for using 2.1 kernels)\\n\\tglibc-2.0.5c (standard! ref: redhat-5.0)\\n\\tgcc 2.7.2.1 (I also have egc-1.0 but I don't trust it yet)\\n\\tCyrix 6x86, 40M ram, ~5G drivespace\\n\\nThough it's nice to finally see large objects working *grin*.\\n\\nG'day, eh? :)\\n\\t- Teunis\\n\\n\",\n \"msg_date\": \"Thu, 15 Jan 1998 12:54:13 -0700 (MST)\",\n \"msg_from\": \"teunis \",\n \"msg_from_op\": true,\n \"msg_subject\": \"a small problem with current CVS - wondering how to debug\"\n }\n]"}}},{"rowIdx":134,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"\nI wish ppl would quite dropping off CC's :( Especially when this stuff\n*should* be being discussed in pgsql-hackers@postgresql.org...\n\nAs for the pg_privileges relation and whatnot...is this something we want\nfor v6.3 (can it be done in 2weeks?) or wait until after v6.3 is released?\n\n\n\nOn Thu, 15 Jan 1998, todd brandys wrote:\n\n> > wouldn't that pose a problem with setting up someone other the\n> > the superuser to create databases? *Raised eyebrows*\n> \n> Again I feel that PostgreSQL should have a pg_privileges relation which would\n> handle the CREATE DATABASE privilege. There would be no problem then.\n> \n> Todd A. Brandys\n> brandys@eng3.hep.uiuc.edu\n> \n\n","msg_date":"Thu, 15 Jan 1998 15:27:21 -0500 (EST)","msg_from":"The Hermit Hacker ","msg_from_op":true,"msg_subject":"Re: New pg_pwd patch and stuff"},{"msg_contents":"> \n> \n> I wish ppl would quite dropping off CC's :( Especially when this stuff\n> *should* be being discussed in pgsql-hackers@postgresql.org...\n> \n> As for the pg_privileges relation and whatnot...is this something we want\n> for v6.3 (can it be done in 2weeks?) or wait until after v6.3 is released?\n> \n\nWell, I can create the table quite easily. The issue is what type of\nflack we will get by haveing pg_user non-readable, and removing the user\nname from psql's \\d output.\n\n> \n> \n> On Thu, 15 Jan 1998, todd brandys wrote:\n> \n> > > wouldn't that pose a problem with setting up someone other the\n> > > the superuser to create databases? *Raised eyebrows*\n> > \n> > Again I feel that PostgreSQL should have a pg_privileges relation which would\n> > handle the CREATE DATABASE privilege. There would be no problem then.\n> > \n> > Todd A. Brandys\n> > brandys@eng3.hep.uiuc.edu\n> > \n> \n> \n> \n\n\n-- \nBruce Momjian\nmaillist@candle.pha.pa.us\n","msg_date":"Thu, 15 Jan 1998 15:51:34 -0500 (EST)","msg_from":"Bruce Momjian ","msg_from_op":false,"msg_subject":"Re: [HACKERS] Re: New pg_pwd patch and stuff"}],"string":"[\n {\n \"msg_contents\": \"\\nI wish ppl would quite dropping off CC's :( Especially when this stuff\\n*should* be being discussed in pgsql-hackers@postgresql.org...\\n\\nAs for the pg_privileges relation and whatnot...is this something we want\\nfor v6.3 (can it be done in 2weeks?) or wait until after v6.3 is released?\\n\\n\\n\\nOn Thu, 15 Jan 1998, todd brandys wrote:\\n\\n> > wouldn't that pose a problem with setting up someone other the\\n> > the superuser to create databases? *Raised eyebrows*\\n> \\n> Again I feel that PostgreSQL should have a pg_privileges relation which would\\n> handle the CREATE DATABASE privilege. There would be no problem then.\\n> \\n> Todd A. Brandys\\n> brandys@eng3.hep.uiuc.edu\\n> \\n\\n\",\n \"msg_date\": \"Thu, 15 Jan 1998 15:27:21 -0500 (EST)\",\n \"msg_from\": \"The Hermit Hacker \",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: New pg_pwd patch and stuff\"\n },\n {\n \"msg_contents\": \"> \\n> \\n> I wish ppl would quite dropping off CC's :( Especially when this stuff\\n> *should* be being discussed in pgsql-hackers@postgresql.org...\\n> \\n> As for the pg_privileges relation and whatnot...is this something we want\\n> for v6.3 (can it be done in 2weeks?) or wait until after v6.3 is released?\\n> \\n\\nWell, I can create the table quite easily. The issue is what type of\\nflack we will get by haveing pg_user non-readable, and removing the user\\nname from psql's \\\\d output.\\n\\n> \\n> \\n> On Thu, 15 Jan 1998, todd brandys wrote:\\n> \\n> > > wouldn't that pose a problem with setting up someone other the\\n> > > the superuser to create databases? *Raised eyebrows*\\n> > \\n> > Again I feel that PostgreSQL should have a pg_privileges relation which would\\n> > handle the CREATE DATABASE privilege. There would be no problem then.\\n> > \\n> > Todd A. Brandys\\n> > brandys@eng3.hep.uiuc.edu\\n> > \\n> \\n> \\n> \\n\\n\\n-- \\nBruce Momjian\\nmaillist@candle.pha.pa.us\\n\",\n \"msg_date\": \"Thu, 15 Jan 1998 15:51:34 -0500 (EST)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Re: New pg_pwd patch and stuff\"\n }\n]"}}},{"rowIdx":135,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"> Fork off the postgres process first, then authenticate inside of\n> there...which would get rid of the problem with pg_user itself being a\n> text file vs a relation...no?\n\nYes, yes, yes. This is how authentication should be done (for HBA, etc.)\nFurthermore, we could reduce the footprint of the postmaster drastically. It\nwould only need to accept a socket connection and fork the backend. This\nscenario would also allow the postmaster to be run as the root user. Good\nthings could only come of this method.\n\nThe only reason I put my authentication scheme where it is, is that all the\nother authentication schemes take place in the postmaster, and to work things\nproperly, use of my scheme (checking to see if there is a password or not) must\ncome first.\n\nTodd A. Brandys\nbrandys@eng3.hep.uiuc.edu\n\n","msg_date":"Thu, 15 Jan 1998 14:31:37 -0600","msg_from":"todd brandys ","msg_from_op":true,"msg_subject":"Re: New pg_pwd patch and stuff"},{"msg_contents":"todd brandys wrote:\n> \n> > Fork off the postgres process first, then authenticate inside of\n> > there...which would get rid of the problem with pg_user itself being a\n> > text file vs a relation...no?\n> \n> Yes, yes, yes. This is how authentication should be done (for HBA, etc.)\n\nNo, no, no! For security reasons, you can't fork (and exec)\nunauthenticated processes. Especially HBA authentication should be done\nto consume as low resources as possbile. Otherwise you open a giant door\nfor so infamously called Denial of Service attacks. Afterwards, every\nhacker will know that to bring your system running postgres to it's\nknees he just have to try to connect to 5432 port very frequently. \"OK\",\nyou might say, \"I have this firewall\". \"OK\", I say, \"so what's that HBA\nfor?\".\n\nSo it's the postmaster's role to deny as much connections as possible.\nUnless we speak of non-execing postgres childs?\n\nMike\n\n-- \nWWW: http://www.lodz.pdi.net/~mimo tel: Int. Acc. Code + 48 42 148340\nadd: Michal Mosiewicz * Bugaj 66 m.54 * 95-200 Pabianice * POLAND\n","msg_date":"Fri, 16 Jan 1998 02:35:38 +0000","msg_from":"\"Micha��� Mosiewicz\" ","msg_from_op":false,"msg_subject":"Re: [HACKERS] Re: New pg_pwd patch and stuff"},{"msg_contents":"On Fri, 16 Jan 1998, Micha3 Mosiewicz wrote:\n\n> No, no, no! For security reasons, you can't fork (and exec)\n> unauthenticated processes. Especially HBA authentication should be done\n> to consume as low resources as possbile. Otherwise you open a giant door\n> for so infamously called Denial of Service attacks. Afterwards, every\n> hacker will know that to bring your system running postgres to it's\n> knees he just have to try to connect to 5432 port very frequently. \"OK\",\n> you might say, \"I have this firewall\". \"OK\", I say, \"so what's that HBA\n> for?\".\n> \n> So it's the postmaster's role to deny as much connections as possible.\n> Unless we speak of non-execing postgres childs?\n\n\tHrmmmm...i don't quite agree with this. postmaster can handle one \nconnection at a time, and then has to pass it off to the postgres backend\nprocess...DoS attacks are easier now then by forking before HBA. I just have\nto continuously open a connection to port 5432...so, while postmaster is\nhandling that connection, checking HBA, checking a password...no other new \nconnections can happen. Can't think of a stronger DoS then that...? :)\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: scrappy@hub.org secondary: scrappy@{freebsd|postgresql}.org \n\n","msg_date":"Fri, 16 Jan 1998 00:11:46 -0400 (AST)","msg_from":"The Hermit Hacker ","msg_from_op":false,"msg_subject":"Re: [HACKERS] Re: New pg_pwd patch and stuff"},{"msg_contents":"The Hermit Hacker wrote:\n> \n> On Fri, 16 Jan 1998, Micha3 Mosiewicz wrote:\n> \n> > No, no, no! For security reasons, you can't fork (and exec)\n> > unauthenticated processes. Especially HBA authentication should be done\n> > to consume as low resources as possbile. Otherwise you open a giant door\n> > for so infamously called Denial of Service attacks. Afterwards, every\n> > hacker will know that to bring your system running postgres to it's\n> > knees he just have to try to connect to 5432 port very frequently. \"OK\",\n> > you might say, \"I have this firewall\". \"OK\", I say, \"so what's that HBA\n> > for?\".\n> >\n> > So it's the postmaster's role to deny as much connections as possible.\n> > Unless we speak of non-execing postgres childs?\n> \n> Hrmmmm...i don't quite agree with this. postmaster can handle one\n> connection at a time, and then has to pass it off to the postgres backend\n> process...DoS attacks are easier now then by forking before HBA. I just have\n> to continuously open a connection to port 5432...so, while postmaster is\n> handling that connection, checking HBA, checking a password...no other new\n> connections can happen. Can't think of a stronger DoS then that...? :)\n\nI think Micha is right. The postmaster can handle multiple connections\nas the read of the startup packet is done a fragment at a time without\nblocking so there is no DoS problem until the postmaster runs out of\nsockets. I think this is less of a problem than loads of\nunauthenticated, resource hungry backends forked by the postmaster.\n\nIn changing the authentication methods for 6.3 I've had to add the\nability for the postmaster to do non-blocking writes as well as reads so\nthat a two-way (non-blocking) dialog can take place between frontend and\npostmaster.\n\nHaving said that, I won't fix (for 6.3 anyway) other parts of the\npostmaster that do blocked I/O - the ident lookup in particular. \nHowever, it is at least under the control of the DBA whether or not\nident is used.\n\nPhil\n","msg_date":"Fri, 16 Jan 1998 19:45:14 +0000","msg_from":"Phil Thompson ","msg_from_op":false,"msg_subject":"Re: [HACKERS] Re: New pg_pwd patch and stuff"},{"msg_contents":"The Hermit Hacker wrote:\n\n> Hrmmmm...i don't quite agree with this. postmaster can handle one\n> connection at a time, and then has to pass it off to the postgres backend\n> process...DoS attacks are easier now then by forking before HBA. I just have\n\nForking is not so bad... but isn't there any exec also? And of course\nit's a difference if your machine is overloaded by processes or if it's\nonly one service that doesn't respond becouse the access-controling code\nis disabled.\n\nSecond question... if we speak only about forking postmaster, or it's\nabout forking-execing-opening files-reading-etc stuff? If it's only\nfork, I would totally agree with you, otherwise I'm not sure which is\nworse... \n\nMike\n\n-- \nWWW: http://www.lodz.pdi.net/~mimo tel: Int. Acc. Code + 48 42 148340\nadd: Michal Mosiewicz * Bugaj 66 m.54 * 95-200 Pabianice * POLAND\n\n","msg_date":"Sun, 18 Jan 1998 03:39:45 +0000","msg_from":"\"Micha3 Mosiewicz\" ","msg_from_op":false,"msg_subject":"Re: [HACKERS] Re: New pg_pwd patch and stuff"},{"msg_contents":"\n\nOn Fri, 16 Jan 1998, Phil Thompson wrote:\n\n> The Hermit Hacker wrote:\n> > \n> > On Fri, 16 Jan 1998, Micha3 Mosiewicz wrote:\n> > \n> > > No, no, no! For security reasons, you can't fork (and exec)\n> > > unauthenticated processes. Especially HBA authentication should be done\n> > > to consume as low resources as possbile. Otherwise you open a giant door\n> > > for so infamously called Denial of Service attacks. Afterwards, every\n> > > hacker will know that to bring your system running postgres to it's\n> > > knees he just have to try to connect to 5432 port very frequently. \"OK\",\n> > > you might say, \"I have this firewall\". \"OK\", I say, \"so what's that HBA\n> > > for?\".\n> > >\n> > > So it's the postmaster's role to deny as much connections as possible.\n> > > Unless we speak of non-execing postgres childs?\n> > \n> > Hrmmmm...i don't quite agree with this. postmaster can handle one\n> > connection at a time, and then has to pass it off to the postgres backend\n> > process...DoS attacks are easier now then by forking before HBA. I just have\n> > to continuously open a connection to port 5432...so, while postmaster is\n> > handling that connection, checking HBA, checking a password...no other new\n> > connections can happen. Can't think of a stronger DoS then that...? :)\n> \n> I think Micha is right. The postmaster can handle multiple connections\n> as the read of the startup packet is done a fragment at a time without\n> blocking so there is no DoS problem until the postmaster runs out of\n> sockets. I think this is less of a problem than loads of\n> unauthenticated, resource hungry backends forked by the postmaster.\n> \n> In changing the authentication methods for 6.3 I've had to add the\n> ability for the postmaster to do non-blocking writes as well as reads so\n> that a two-way (non-blocking) dialog can take place between frontend and\n> postmaster.\n> \n> Having said that, I won't fix (for 6.3 anyway) other parts of the\n> postmaster that do blocked I/O - the ident lookup in particular. \n> However, it is at least under the control of the DBA whether or not\n> ident is used.\n> \n> Phil\n> \n\nOne way or another PostgreSQL is subject to attack. If we do continue on \nwith postmaster doing authentication, then the postmaster should be made \ncapable of performing queries against the system catalog. This would \nallow HBA to be incorporated into the system catalog (pg_user or \nwhatever). Why would we do this? It will make PostgreSQL easier to \nadminister. The user will not have to edit files to make a change. HBA's\ncharacteristics could be partially handled by adding a:\n\nFROM ( | ) ... [, ( | )]\n\nclause to the 'CREATE USER' statement. Right now the only way this could be\nachieved would be to pass authentication down to the backend processes, \nas the postmaster can not execute SQL statements.\n\nTodd A. Brandys\nbrandys@eng3.hep.uiuc.edu\n\n","msg_date":"Sun, 18 Jan 1998 15:47:30 -0600 (CST)","msg_from":"todd brandys ","msg_from_op":true,"msg_subject":"Re: [HACKERS] Re: New pg_pwd patch and stuff"},{"msg_contents":" So, if postmaster can't access the system tables for authentication\npurposes, why not have one child forked, connected to the main \npostmaster process via some IPC mechanism, whose only purpose\nin life it is to answer SQL-queries for postmaster.\n\n\tmjl\n\n","msg_date":"Mon, 19 Jan 1998 04:18:34 +0100 (MET)","msg_from":"\"Martin J. Laubach\" ","msg_from_op":false,"msg_subject":"Re: [HACKERS] Re: New pg_pwd patch and stuff"},{"msg_contents":"\n\nOn Mon, 19 Jan 1998, Martin J. Laubach wrote:\n\n> So, if postmaster can't access the system tables for authentication\n> purposes, why not have one child forked, connected to the main \n> postmaster process via some IPC mechanism, whose only purpose\n> in life it is to answer SQL-queries for postmaster.\n> \n> \tmjl\n> \n> \n\nI would have to say \"footprint\". All the code for the backend has been \nincorporated into the postmaster. The postmaster is nothing more than a\nbackend run with some special command line options. The postgres executable\nis over 1 MB large and is a static executable. This is quite a bit of\n(virtual) memory. I realize that memory is cheap these days, but to \nhave all the backend code included in with the postmaster and not have it \nuseable is a waste of resources. To then have a second instance running \nwould take more resources.\n\nIf we must maintain user authorization in the postmaster, then I believe \nit would be best to strip this code from the postgres executable to \ncreate a postmaster with a smaller footprint. Furthermore, the postmaster\nshould be made capable of performing heapscans, so that it could at least\nview the data in the system catalog. Then all the data needed for each \nauthentication method (or a single method if we incorporated the best \naspects of each method into one super authentication method), could be \nstored in the system catalog. This information could then be managed by the\nCREATE/ALTER USER statements, which would alleviate the need to edit flat \nfiles for system configuration.\n\nTodd A. Brandys\nbrandys@eng3.hep.uiuc.edu\n\n","msg_date":"Tue, 20 Jan 1998 12:10:23 -0600 (CST)","msg_from":"todd brandys ","msg_from_op":true,"msg_subject":"Re: [HACKERS] Re: New pg_pwd patch and stuff"}],"string":"[\n {\n \"msg_contents\": \"> Fork off the postgres process first, then authenticate inside of\\n> there...which would get rid of the problem with pg_user itself being a\\n> text file vs a relation...no?\\n\\nYes, yes, yes. This is how authentication should be done (for HBA, etc.)\\nFurthermore, we could reduce the footprint of the postmaster drastically. It\\nwould only need to accept a socket connection and fork the backend. This\\nscenario would also allow the postmaster to be run as the root user. Good\\nthings could only come of this method.\\n\\nThe only reason I put my authentication scheme where it is, is that all the\\nother authentication schemes take place in the postmaster, and to work things\\nproperly, use of my scheme (checking to see if there is a password or not) must\\ncome first.\\n\\nTodd A. Brandys\\nbrandys@eng3.hep.uiuc.edu\\n\\n\",\n \"msg_date\": \"Thu, 15 Jan 1998 14:31:37 -0600\",\n \"msg_from\": \"todd brandys \",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: New pg_pwd patch and stuff\"\n },\n {\n \"msg_contents\": \"todd brandys wrote:\\n> \\n> > Fork off the postgres process first, then authenticate inside of\\n> > there...which would get rid of the problem with pg_user itself being a\\n> > text file vs a relation...no?\\n> \\n> Yes, yes, yes. This is how authentication should be done (for HBA, etc.)\\n\\nNo, no, no! For security reasons, you can't fork (and exec)\\nunauthenticated processes. Especially HBA authentication should be done\\nto consume as low resources as possbile. Otherwise you open a giant door\\nfor so infamously called Denial of Service attacks. Afterwards, every\\nhacker will know that to bring your system running postgres to it's\\nknees he just have to try to connect to 5432 port very frequently. \\\"OK\\\",\\nyou might say, \\\"I have this firewall\\\". \\\"OK\\\", I say, \\\"so what's that HBA\\nfor?\\\".\\n\\nSo it's the postmaster's role to deny as much connections as possible.\\nUnless we speak of non-execing postgres childs?\\n\\nMike\\n\\n-- \\nWWW: http://www.lodz.pdi.net/~mimo tel: Int. Acc. Code + 48 42 148340\\nadd: Michal Mosiewicz * Bugaj 66 m.54 * 95-200 Pabianice * POLAND\\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 02:35:38 +0000\",\n \"msg_from\": \"\\\"Micha��� Mosiewicz\\\" \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Re: New pg_pwd patch and stuff\"\n },\n {\n \"msg_contents\": \"On Fri, 16 Jan 1998, Micha3 Mosiewicz wrote:\\n\\n> No, no, no! For security reasons, you can't fork (and exec)\\n> unauthenticated processes. Especially HBA authentication should be done\\n> to consume as low resources as possbile. Otherwise you open a giant door\\n> for so infamously called Denial of Service attacks. Afterwards, every\\n> hacker will know that to bring your system running postgres to it's\\n> knees he just have to try to connect to 5432 port very frequently. \\\"OK\\\",\\n> you might say, \\\"I have this firewall\\\". \\\"OK\\\", I say, \\\"so what's that HBA\\n> for?\\\".\\n> \\n> So it's the postmaster's role to deny as much connections as possible.\\n> Unless we speak of non-execing postgres childs?\\n\\n\\tHrmmmm...i don't quite agree with this. postmaster can handle one \\nconnection at a time, and then has to pass it off to the postgres backend\\nprocess...DoS attacks are easier now then by forking before HBA. I just have\\nto continuously open a connection to port 5432...so, while postmaster is\\nhandling that connection, checking HBA, checking a password...no other new \\nconnections can happen. Can't think of a stronger DoS then that...? :)\\n\\nMarc G. Fournier \\nSystems Administrator @ hub.org \\nprimary: scrappy@hub.org secondary: scrappy@{freebsd|postgresql}.org \\n\\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 00:11:46 -0400 (AST)\",\n \"msg_from\": \"The Hermit Hacker \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Re: New pg_pwd patch and stuff\"\n },\n {\n \"msg_contents\": \"The Hermit Hacker wrote:\\n> \\n> On Fri, 16 Jan 1998, Micha3 Mosiewicz wrote:\\n> \\n> > No, no, no! For security reasons, you can't fork (and exec)\\n> > unauthenticated processes. Especially HBA authentication should be done\\n> > to consume as low resources as possbile. Otherwise you open a giant door\\n> > for so infamously called Denial of Service attacks. Afterwards, every\\n> > hacker will know that to bring your system running postgres to it's\\n> > knees he just have to try to connect to 5432 port very frequently. \\\"OK\\\",\\n> > you might say, \\\"I have this firewall\\\". \\\"OK\\\", I say, \\\"so what's that HBA\\n> > for?\\\".\\n> >\\n> > So it's the postmaster's role to deny as much connections as possible.\\n> > Unless we speak of non-execing postgres childs?\\n> \\n> Hrmmmm...i don't quite agree with this. postmaster can handle one\\n> connection at a time, and then has to pass it off to the postgres backend\\n> process...DoS attacks are easier now then by forking before HBA. I just have\\n> to continuously open a connection to port 5432...so, while postmaster is\\n> handling that connection, checking HBA, checking a password...no other new\\n> connections can happen. Can't think of a stronger DoS then that...? :)\\n\\nI think Micha is right. The postmaster can handle multiple connections\\nas the read of the startup packet is done a fragment at a time without\\nblocking so there is no DoS problem until the postmaster runs out of\\nsockets. I think this is less of a problem than loads of\\nunauthenticated, resource hungry backends forked by the postmaster.\\n\\nIn changing the authentication methods for 6.3 I've had to add the\\nability for the postmaster to do non-blocking writes as well as reads so\\nthat a two-way (non-blocking) dialog can take place between frontend and\\npostmaster.\\n\\nHaving said that, I won't fix (for 6.3 anyway) other parts of the\\npostmaster that do blocked I/O - the ident lookup in particular. \\nHowever, it is at least under the control of the DBA whether or not\\nident is used.\\n\\nPhil\\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 19:45:14 +0000\",\n \"msg_from\": \"Phil Thompson \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Re: New pg_pwd patch and stuff\"\n },\n {\n \"msg_contents\": \"The Hermit Hacker wrote:\\n\\n> Hrmmmm...i don't quite agree with this. postmaster can handle one\\n> connection at a time, and then has to pass it off to the postgres backend\\n> process...DoS attacks are easier now then by forking before HBA. I just have\\n\\nForking is not so bad... but isn't there any exec also? And of course\\nit's a difference if your machine is overloaded by processes or if it's\\nonly one service that doesn't respond becouse the access-controling code\\nis disabled.\\n\\nSecond question... if we speak only about forking postmaster, or it's\\nabout forking-execing-opening files-reading-etc stuff? If it's only\\nfork, I would totally agree with you, otherwise I'm not sure which is\\nworse... \\n\\nMike\\n\\n-- \\nWWW: http://www.lodz.pdi.net/~mimo tel: Int. Acc. Code + 48 42 148340\\nadd: Michal Mosiewicz * Bugaj 66 m.54 * 95-200 Pabianice * POLAND\\n\\n\",\n \"msg_date\": \"Sun, 18 Jan 1998 03:39:45 +0000\",\n \"msg_from\": \"\\\"Micha3 Mosiewicz\\\" \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Re: New pg_pwd patch and stuff\"\n },\n {\n \"msg_contents\": \"\\n\\nOn Fri, 16 Jan 1998, Phil Thompson wrote:\\n\\n> The Hermit Hacker wrote:\\n> > \\n> > On Fri, 16 Jan 1998, Micha3 Mosiewicz wrote:\\n> > \\n> > > No, no, no! For security reasons, you can't fork (and exec)\\n> > > unauthenticated processes. Especially HBA authentication should be done\\n> > > to consume as low resources as possbile. Otherwise you open a giant door\\n> > > for so infamously called Denial of Service attacks. Afterwards, every\\n> > > hacker will know that to bring your system running postgres to it's\\n> > > knees he just have to try to connect to 5432 port very frequently. \\\"OK\\\",\\n> > > you might say, \\\"I have this firewall\\\". \\\"OK\\\", I say, \\\"so what's that HBA\\n> > > for?\\\".\\n> > >\\n> > > So it's the postmaster's role to deny as much connections as possible.\\n> > > Unless we speak of non-execing postgres childs?\\n> > \\n> > Hrmmmm...i don't quite agree with this. postmaster can handle one\\n> > connection at a time, and then has to pass it off to the postgres backend\\n> > process...DoS attacks are easier now then by forking before HBA. I just have\\n> > to continuously open a connection to port 5432...so, while postmaster is\\n> > handling that connection, checking HBA, checking a password...no other new\\n> > connections can happen. Can't think of a stronger DoS then that...? :)\\n> \\n> I think Micha is right. The postmaster can handle multiple connections\\n> as the read of the startup packet is done a fragment at a time without\\n> blocking so there is no DoS problem until the postmaster runs out of\\n> sockets. I think this is less of a problem than loads of\\n> unauthenticated, resource hungry backends forked by the postmaster.\\n> \\n> In changing the authentication methods for 6.3 I've had to add the\\n> ability for the postmaster to do non-blocking writes as well as reads so\\n> that a two-way (non-blocking) dialog can take place between frontend and\\n> postmaster.\\n> \\n> Having said that, I won't fix (for 6.3 anyway) other parts of the\\n> postmaster that do blocked I/O - the ident lookup in particular. \\n> However, it is at least under the control of the DBA whether or not\\n> ident is used.\\n> \\n> Phil\\n> \\n\\nOne way or another PostgreSQL is subject to attack. If we do continue on \\nwith postmaster doing authentication, then the postmaster should be made \\ncapable of performing queries against the system catalog. This would \\nallow HBA to be incorporated into the system catalog (pg_user or \\nwhatever). Why would we do this? It will make PostgreSQL easier to \\nadminister. The user will not have to edit files to make a change. HBA's\\ncharacteristics could be partially handled by adding a:\\n\\nFROM ( | ) ... [, ( | )]\\n\\nclause to the 'CREATE USER' statement. Right now the only way this could be\\nachieved would be to pass authentication down to the backend processes, \\nas the postmaster can not execute SQL statements.\\n\\nTodd A. Brandys\\nbrandys@eng3.hep.uiuc.edu\\n\\n\",\n \"msg_date\": \"Sun, 18 Jan 1998 15:47:30 -0600 (CST)\",\n \"msg_from\": \"todd brandys \",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: [HACKERS] Re: New pg_pwd patch and stuff\"\n },\n {\n \"msg_contents\": \" So, if postmaster can't access the system tables for authentication\\npurposes, why not have one child forked, connected to the main \\npostmaster process via some IPC mechanism, whose only purpose\\nin life it is to answer SQL-queries for postmaster.\\n\\n\\tmjl\\n\\n\",\n \"msg_date\": \"Mon, 19 Jan 1998 04:18:34 +0100 (MET)\",\n \"msg_from\": \"\\\"Martin J. Laubach\\\" \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Re: New pg_pwd patch and stuff\"\n },\n {\n \"msg_contents\": \"\\n\\nOn Mon, 19 Jan 1998, Martin J. Laubach wrote:\\n\\n> So, if postmaster can't access the system tables for authentication\\n> purposes, why not have one child forked, connected to the main \\n> postmaster process via some IPC mechanism, whose only purpose\\n> in life it is to answer SQL-queries for postmaster.\\n> \\n> \\tmjl\\n> \\n> \\n\\nI would have to say \\\"footprint\\\". All the code for the backend has been \\nincorporated into the postmaster. The postmaster is nothing more than a\\nbackend run with some special command line options. The postgres executable\\nis over 1 MB large and is a static executable. This is quite a bit of\\n(virtual) memory. I realize that memory is cheap these days, but to \\nhave all the backend code included in with the postmaster and not have it \\nuseable is a waste of resources. To then have a second instance running \\nwould take more resources.\\n\\nIf we must maintain user authorization in the postmaster, then I believe \\nit would be best to strip this code from the postgres executable to \\ncreate a postmaster with a smaller footprint. Furthermore, the postmaster\\nshould be made capable of performing heapscans, so that it could at least\\nview the data in the system catalog. Then all the data needed for each \\nauthentication method (or a single method if we incorporated the best \\naspects of each method into one super authentication method), could be \\nstored in the system catalog. This information could then be managed by the\\nCREATE/ALTER USER statements, which would alleviate the need to edit flat \\nfiles for system configuration.\\n\\nTodd A. Brandys\\nbrandys@eng3.hep.uiuc.edu\\n\\n\",\n \"msg_date\": \"Tue, 20 Jan 1998 12:10:23 -0600 (CST)\",\n \"msg_from\": \"todd brandys \",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: [HACKERS] Re: New pg_pwd patch and stuff\"\n }\n]"}}},{"rowIdx":136,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"> As for the pg_privileges relation and whatnot...is this something we want\n> for v6.3 (can it be done in 2weeks?) or wait until after v6.3 is released?\n\nI don't think (realistically) that such a task could be done in two weeks. No.\nRather, we should wait until after release 6.3, and then maybe spend some time\ndebating on what the pg_privileges table should look like. After the table is\ncreated (the easy part), then it becomes a hunt to find all the places where\nprivileges are checked and change to the code in these spot (not too bad really).\nFinally, we have to develop the code for governing column permissions (this is\nthe most difficult part of the project). The query processor must be changed\nto check the permissions of each column in all commands.\n\nThis would be a tall order for two weeks. Especially, to be certain that we\nhad a consensus among the hacker community that what was being done was being\ndone in the best way possible.\n\nTodd A. Brandys\nbrandys@eng3.hep.uiuc.edu\n","msg_date":"Thu, 15 Jan 1998 14:41:51 -0600","msg_from":"todd brandys ","msg_from_op":true,"msg_subject":"Re: New pg_pwd patch and stuff"},{"msg_contents":"> \n> > As for the pg_privileges relation and whatnot...is this something we want\n> > for v6.3 (can it be done in 2weeks?) or wait until after v6.3 is released?\n> \n> I don't think (realistically) that such a task could be done in two weeks. No.\n> Rather, we should wait until after release 6.3, and then maybe spend some time\n> debating on what the pg_privileges table should look like. After the table is\n> created (the easy part), then it becomes a hunt to find all the places where\n> privileges are checked and change to the code in these spot (not too bad really).\n> Finally, we have to develop the code for governing column permissions (this is\n> the most difficult part of the project). The query processor must be changed\n> to check the permissions of each column in all commands.\n> \n> This would be a tall order for two weeks. Especially, to be certain that we\n> had a consensus among the hacker community that what was being done was being\n> done in the best way possible.\n\n\nI believe doing permissions on VIEWS would be much simpler than\ncolumn-level permissions. That way, you create the view with the\ncolumns you need, and give that to the user.\n\n-- \nBruce Momjian\nmaillist@candle.pha.pa.us\n","msg_date":"Thu, 15 Jan 1998 16:05:54 -0500 (EST)","msg_from":"Bruce Momjian ","msg_from_op":false,"msg_subject":"Re: [HACKERS] Re: New pg_pwd patch and stuff"}],"string":"[\n {\n \"msg_contents\": \"> As for the pg_privileges relation and whatnot...is this something we want\\n> for v6.3 (can it be done in 2weeks?) or wait until after v6.3 is released?\\n\\nI don't think (realistically) that such a task could be done in two weeks. No.\\nRather, we should wait until after release 6.3, and then maybe spend some time\\ndebating on what the pg_privileges table should look like. After the table is\\ncreated (the easy part), then it becomes a hunt to find all the places where\\nprivileges are checked and change to the code in these spot (not too bad really).\\nFinally, we have to develop the code for governing column permissions (this is\\nthe most difficult part of the project). The query processor must be changed\\nto check the permissions of each column in all commands.\\n\\nThis would be a tall order for two weeks. Especially, to be certain that we\\nhad a consensus among the hacker community that what was being done was being\\ndone in the best way possible.\\n\\nTodd A. Brandys\\nbrandys@eng3.hep.uiuc.edu\\n\",\n \"msg_date\": \"Thu, 15 Jan 1998 14:41:51 -0600\",\n \"msg_from\": \"todd brandys \",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: New pg_pwd patch and stuff\"\n },\n {\n \"msg_contents\": \"> \\n> > As for the pg_privileges relation and whatnot...is this something we want\\n> > for v6.3 (can it be done in 2weeks?) or wait until after v6.3 is released?\\n> \\n> I don't think (realistically) that such a task could be done in two weeks. No.\\n> Rather, we should wait until after release 6.3, and then maybe spend some time\\n> debating on what the pg_privileges table should look like. After the table is\\n> created (the easy part), then it becomes a hunt to find all the places where\\n> privileges are checked and change to the code in these spot (not too bad really).\\n> Finally, we have to develop the code for governing column permissions (this is\\n> the most difficult part of the project). The query processor must be changed\\n> to check the permissions of each column in all commands.\\n> \\n> This would be a tall order for two weeks. Especially, to be certain that we\\n> had a consensus among the hacker community that what was being done was being\\n> done in the best way possible.\\n\\n\\nI believe doing permissions on VIEWS would be much simpler than\\ncolumn-level permissions. That way, you create the view with the\\ncolumns you need, and give that to the user.\\n\\n-- \\nBruce Momjian\\nmaillist@candle.pha.pa.us\\n\",\n \"msg_date\": \"Thu, 15 Jan 1998 16:05:54 -0500 (EST)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Re: New pg_pwd patch and stuff\"\n }\n]"}}},{"rowIdx":137,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"subscribe\n","msg_date":"Thu, 15 Jan 1998 14:42:31 -0600","msg_from":"todd brandys ","msg_from_op":true,"msg_subject":"subscribe"}],"string":"[\n {\n \"msg_contents\": \"subscribe\\n\",\n \"msg_date\": \"Thu, 15 Jan 1998 14:42:31 -0600\",\n \"msg_from\": \"todd brandys \",\n \"msg_from_op\": true,\n \"msg_subject\": \"subscribe\"\n }\n]"}}},{"rowIdx":138,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"Forwarded message:\n> > Why attlen should be -1 ?\n> > attlen in pg_attribute for v in table t is 84, why run-time attlen\n> > should be -1 ? How else maxlen constraint could be checked ?\n> > IMHO, you have to change heap_getattr() to check is atttype == VARCHAROID\n> > and use vl_len if yes. Also, other places where attlen is used must be \n> > changed too - e.g. ExecEvalVar():\n> > \n> > {\n> > len = tuple_type->attrs[attnum - 1]->attlen;\n> > ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n> > byval = tuple_type->attrs[attnum - 1]->attbyval ? true : false;\n> > }\n> > \n> > execConstByVal = byval;\n> > execConstLen = len;\n> > ^^^^^^^^^^^^^^^^^^ - used in nodeHash.c\n> > \n> \n> The major problem is that TupleDesc comes from several places, and\n> attlen means several things.\n> \n> There are some cases where TupleDesc (int numatt, Attrs[]) is created\n> on-the-fly (tupdesc.c), and the attlen is the length of the type. In\n> other cases, we get attlen from opening the relation, heap_open(), and\n> in these cases it is the length as defined for the particular attribute.\n> \n> Certainly a bad situation. I am not sure about a fix.\n\nOK, here is a temporary fix to the problem. It does the heap_open(),\nthen replaces the attrs for VARCHAR with attlen of -1. You can't just\nchange the field, because the data is in a cache and you are just\nreturned a pointer.\n\nCan I add an 'attdeflen' for \"attributed defined length\" field to\npg_attribute, and change the attlen references needed to the new field?\nThis is the only proper way to fix it.\n\n\n---------------------------------------------------------------------------\n\n*** ./backend/executor/execAmi.c.orig\tThu Jan 15 22:42:13 1998\n--- ./backend/executor/execAmi.c\tThu Jan 15 23:54:37 1998\n***************\n*** 42,47 ****\n--- 42,48 ----\n #include \"access/genam.h\"\n #include \"access/heapam.h\"\n #include \"catalog/heap.h\"\n+ #include \"catalog/pg_type.h\"\n \n static Pointer\n ExecBeginScan(Relation relation, int nkeys, ScanKey skeys,\n***************\n*** 124,129 ****\n--- 125,155 ----\n \tif (relation == NULL)\n \t\telog(DEBUG, \"ExecOpenR: relation == NULL, heap_open failed.\");\n \n+ \t{\n+ \t\tint i;\n+ \t\tRelation trel = palloc(sizeof(RelationData));\n+ \t\tTupleDesc tdesc = palloc(sizeof(struct tupleDesc));\n+ \t\tAttributeTupleForm *tatt =\n+ \t\t\t\tpalloc(sizeof(AttributeTupleForm*)*relation->rd_att->natts);\n+ \n+ \t\tmemcpy(trel, relation, sizeof(RelationData));\n+ \t\tmemcpy(tdesc, relation->rd_att, sizeof(struct tupleDesc));\n+ \t\ttrel->rd_att = tdesc;\n+ \t\ttdesc->attrs = tatt;\n+ \t\t\n+ \t\tfor (i = 0; i < relation->rd_att->natts; i++)\n+ \t\t{\n+ \t\t\tif (relation->rd_att->attrs[i]->atttypid != VARCHAROID)\n+ \t\t\t\ttdesc->attrs[i] = relation->rd_att->attrs[i];\n+ \t\t\telse\n+ \t\t\t{\n+ \t\t\t\ttdesc->attrs[i] = palloc(sizeof(FormData_pg_attribute));\n+ \t\t\t\tmemcpy(tdesc->attrs[i], relation->rd_att->attrs[i],\n+ \t\t\t\t\t\t\t\t\t\t\tsizeof(FormData_pg_attribute));\n+ \t\t\t\ttdesc->attrs[i]->attlen = -1;\n+ \t\t\t}\n+ \t\t}\n+ \t}\n \treturn relation;\n }\n \n\n\n-- \nBruce Momjian\nmaillist@candle.pha.pa.us\n","msg_date":"Fri, 16 Jan 1998 00:05:03 -0500 (EST)","msg_from":"Bruce Momjian ","msg_from_op":true,"msg_subject":"Re: [HACKERS] Re: varchar() troubles (fwd)"},{"msg_contents":"> OK, here is a temporary fix to the problem. It does the heap_open(),\n> then replaces the attrs for VARCHAR with attlen of -1. You can't just\n> change the field, because the data is in a cache and you are just\n> returned a pointer.\n>\n> Can I add an 'attdeflen' for \"attributed defined length\" field to\n> pg_attribute, and change the attlen references needed to the new field?\n> This is the only proper way to fix it.\n\nBruce, does your \"temporary fix\" seem to repair all known problems with varchar()? If so, would you be interested in\nholding off on a \"proper fix\" and coming back to it after v6.3 is released? At that time, we can try solving the general\nproblem of retaining column-specific attributes, such as your max len for varchar, declared dimensions for arrays, and\nnumeric() and decimal() types. Or, if you have time to try a solution now _and_ come back to it later...\n\n - Tom\n\n","msg_date":"Fri, 16 Jan 1998 05:26:08 +0000","msg_from":"\"Thomas G. Lockhart\" ","msg_from_op":false,"msg_subject":"Re: [HACKERS] Re: varchar() troubles (fwd)"},{"msg_contents":"> \n> > OK, here is a temporary fix to the problem. It does the heap_open(),\n> > then replaces the attrs for VARCHAR with attlen of -1. You can't just\n> > change the field, because the data is in a cache and you are just\n> > returned a pointer.\n> >\n> > Can I add an 'attdeflen' for \"attributed defined length\" field to\n> > pg_attribute, and change the attlen references needed to the new field?\n> > This is the only proper way to fix it.\n> \n\n> Bruce, does your \"temporary fix\" seem to repair all known problems\nwith varchar()? If so, would you be interested in > holding off on a\n\"proper fix\" and coming back to it after v6.3 is released? At that time,\nwe can try solving the general > problem of retaining column-specific\nattributes, such as your max len for varchar, declared dimensions for\narrays, and > numeric() and decimal() types. Or, if you have time to try\na solution now _and_ come back to it later... > > \n\n[Those wide post really are difficult.]\n\nI don't think my solution is perfect or complete. I only caught one\ngroup of heap_open calls used in the executor. I could funnel all of\nthem through this patched function, but I can imagine there would be\nones I would miss. Once the structure is gotten from the cache, it\nseems to fly around the executor code quite freely, and it is hard to\nknow when a tuple descriptor is being created, if it is being used for\ndata creation or data reference. attlen references are much clearer in\ntheir intent.\n\nIf I add a new field type to FormData_pg_attribute, I can then check\neach attlen reference, and check if it is trying to move through the\non-disk storage (attlen/typlen) or create a new/modify an entry\n(attdeflen).\n\nHow much time I have depends on what Vadim needs me to do for\nsubselects.\n\n\n-- \nBruce Momjian\nmaillist@candle.pha.pa.us\n","msg_date":"Fri, 16 Jan 1998 00:44:11 -0500 (EST)","msg_from":"Bruce Momjian ","msg_from_op":true,"msg_subject":"Re: [HACKERS] Re: varchar() troubles (fwd)"},{"msg_contents":"> \n> > OK, here is a temporary fix to the problem. It does the heap_open(),\n> > then replaces the attrs for VARCHAR with attlen of -1. You can't just\n> > change the field, because the data is in a cache and you are just\n> > returned a pointer.\n> >\n> > Can I add an 'attdeflen' for \"attributed defined length\" field to\n> > pg_attribute, and change the attlen references needed to the new field?\n> > This is the only proper way to fix it.\n> \n> Bruce, does your \"temporary fix\" seem to repair all known problems with varchar()? If so, would you be interested in\n> holding off on a \"proper fix\" and coming back to it after v6.3 is released? At that time, we can try solving the general\n> problem of retaining column-specific attributes, such as your max len for varchar, declared dimensions for arrays, and\n> numeric() and decimal() types. Or, if you have time to try a solution now _and_ come back to it later...\n> \n> - Tom\n> \n> \n\nIn fact, I am inclined to leave attlen unchanged, and add atttyplen that\nis a copy of the length of the type. That way, the attlen for varchar()\nreally contains the defined length, and atttyplen is used for disk\nreferences, and it is very clear what it means.\n\n\n-- \nBruce Momjian\nmaillist@candle.pha.pa.us\n","msg_date":"Fri, 16 Jan 1998 00:49:02 -0500 (EST)","msg_from":"Bruce Momjian ","msg_from_op":true,"msg_subject":"Re: [HACKERS] Re: varchar() troubles (fwd)"},{"msg_contents":"Bruce Momjian wrote:\n> \n> > >\n> > > Can I add an 'attdeflen' for \"attributed defined length\" field to\n> > > pg_attribute, and change the attlen references needed to the new field?\n> > > This is the only proper way to fix it.\n> >\n> > Bruce, does your \"temporary fix\" seem to repair all known problems with varchar()? If so, would you be interested in\n> > holding off on a \"proper fix\" and coming back to it after v6.3 is released? At that time, we can try solving the general\n> > problem of retaining column-specific attributes, such as your max len for varchar, declared dimensions for arrays, and\n> > numeric() and decimal() types. Or, if you have time to try a solution now _and_ come back to it later...\n> >\n> \n> In fact, I am inclined to leave attlen unchanged, and add atttyplen that\n> is a copy of the length of the type. That way, the attlen for varchar()\n> really contains the defined length, and atttyplen is used for disk\n> references, and it is very clear what it means.\n\npg_attribute.h:\n\n int2 attlen; \n\n /*\n * attlen is a copy of the typlen field from pg_type for this\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n * attribute... \n ^^^^^^^^^\n: I would suggest to don't change this to preserve the same meaning\nfor all data types. attlen = -1 means that attribute is varlena.\n\nWe certainly need in new field in pg_attribute! I don't like names like\n\"attdeflen\" or \"atttyplen\" - bad names for NUMERIC etc. Something like\natttspec (ATTribute Type SPECification) is better, imho.\n\nFor the varchar(N) we'll have attlen = -1 and atttspec = N (or N + 4 - what's\nbetter). For the text: attlen = -1 and atttspec = -1. And so on.\n\nOf 'course, it's not so much matter where to put maxlen of varchar.\nattlen = -1 for varchar just seems more clear to me.\n\nBut in any case we need in new field and, imho, this should be added\nin 6.3\n\nVadim\n","msg_date":"Fri, 16 Jan 1998 16:31:45 +0700","msg_from":"\"Vadim B. Mikheev\" ","msg_from_op":false,"msg_subject":"Re: [HACKERS] Re: varchar() troubles (fwd)"},{"msg_contents":"> \n> Bruce Momjian wrote:\n> > \n> > > >\n> > > > Can I add an 'attdeflen' for \"attributed defined length\" field to\n> > > > pg_attribute, and change the attlen references needed to the new field?\n> > > > This is the only proper way to fix it.\n> > >\n> > > Bruce, does your \"temporary fix\" seem to repair all known problems with varchar()? If so, would you be interested in\n> > > holding off on a \"proper fix\" and coming back to it after v6.3 is released? At that time, we can try solving the general\n> > > problem of retaining column-specific attributes, such as your max len for varchar, declared dimensions for arrays, and\n> > > numeric() and decimal() types. Or, if you have time to try a solution now _and_ come back to it later...\n> > >\n> > \n> > In fact, I am inclined to leave attlen unchanged, and add atttyplen that\n> > is a copy of the length of the type. That way, the attlen for varchar()\n> > really contains the defined length, and atttyplen is used for disk\n> > references, and it is very clear what it means.\n> \n> pg_attribute.h:\n> \n> int2 attlen; \n> \n> /*\n> * attlen is a copy of the typlen field from pg_type for this\n> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n> * attribute... \n> ^^^^^^^^^\n> : I would suggest to don't change this to preserve the same meaning\n> for all data types. attlen = -1 means that attribute is varlena.\n> \n> We certainly need in new field in pg_attribute! I don't like names like\n> \"attdeflen\" or \"atttyplen\" - bad names for NUMERIC etc. Something like\n> atttspec (ATTribute Type SPECification) is better, imho.\n> \n> For the varchar(N) we'll have attlen = -1 and atttspec = N (or N + 4 - what's\n> better). For the text: attlen = -1 and atttspec = -1. And so on.\n> \n> Of 'course, it's not so much matter where to put maxlen of varchar.\n> attlen = -1 for varchar just seems more clear to me.\n> \n> But in any case we need in new field and, imho, this should be added\n> in 6.3\n\nOK, we have a new pg_attribute column called 'atttypmod' for\n'type-specific modifier'. Currently, it is only used to hold the char\nand varchar length, but I am sure will be used soon for other types.\n\nHere is the test:\n\t\n\ttest=> insert into test values ('asdfasdfasdfasdfasdfadsfasdf11',3);\n\tINSERT 18282 1\n\ttest=> select * from test;\n\tx |y\n\t--------------------+-\n\tasdfasdfasdfasdfasdf|3\n\t(1 row)\n\n'attlen' was certainly a confusing double-used field that I am glad to\nreturn to single-use status.\n\nI will be installing the patch soon, and will then start on subselects\nin the parser. It will probably take me until Monday to finish that.\n\n-- \nBruce Momjian\nmaillist@candle.pha.pa.us\n","msg_date":"Fri, 16 Jan 1998 17:48:24 -0500 (EST)","msg_from":"Bruce Momjian ","msg_from_op":true,"msg_subject":"Re: [HACKERS] Re: varchar() troubles (fwd)"},{"msg_contents":"Bruce Momjian wrote:\n> \n> OK, we have a new pg_attribute column called 'atttypmod' for\n> 'type-specific modifier'. Currently, it is only used to hold the char\n> and varchar length, but I am sure will be used soon for other types.\n\nNice!\n\n> \n> Here is the test:\n> \n> test=> insert into test values ('asdfasdfasdfasdfasdfadsfasdf11',3);\n> INSERT 18282 1\n> test=> select * from test;\n> x |y\n> --------------------+-\n> asdfasdfasdfasdfasdf|3\n> (1 row)\n> \n> 'attlen' was certainly a confusing double-used field that I am glad to\n> return to single-use status.\n> \n> I will be installing the patch soon, and will then start on subselects\n> in the parser. It will probably take me until Monday to finish that.\n\nOk.\n\nVadim\n","msg_date":"Sun, 18 Jan 1998 15:30:11 +0700","msg_from":"\"Vadim B. Mikheev\" ","msg_from_op":false,"msg_subject":"Re: [HACKERS] Re: varchar() troubles (fwd)"}],"string":"[\n {\n \"msg_contents\": \"Forwarded message:\\n> > Why attlen should be -1 ?\\n> > attlen in pg_attribute for v in table t is 84, why run-time attlen\\n> > should be -1 ? How else maxlen constraint could be checked ?\\n> > IMHO, you have to change heap_getattr() to check is atttype == VARCHAROID\\n> > and use vl_len if yes. Also, other places where attlen is used must be \\n> > changed too - e.g. ExecEvalVar():\\n> > \\n> > {\\n> > len = tuple_type->attrs[attnum - 1]->attlen;\\n> > ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\\n> > byval = tuple_type->attrs[attnum - 1]->attbyval ? true : false;\\n> > }\\n> > \\n> > execConstByVal = byval;\\n> > execConstLen = len;\\n> > ^^^^^^^^^^^^^^^^^^ - used in nodeHash.c\\n> > \\n> \\n> The major problem is that TupleDesc comes from several places, and\\n> attlen means several things.\\n> \\n> There are some cases where TupleDesc (int numatt, Attrs[]) is created\\n> on-the-fly (tupdesc.c), and the attlen is the length of the type. In\\n> other cases, we get attlen from opening the relation, heap_open(), and\\n> in these cases it is the length as defined for the particular attribute.\\n> \\n> Certainly a bad situation. I am not sure about a fix.\\n\\nOK, here is a temporary fix to the problem. It does the heap_open(),\\nthen replaces the attrs for VARCHAR with attlen of -1. You can't just\\nchange the field, because the data is in a cache and you are just\\nreturned a pointer.\\n\\nCan I add an 'attdeflen' for \\\"attributed defined length\\\" field to\\npg_attribute, and change the attlen references needed to the new field?\\nThis is the only proper way to fix it.\\n\\n\\n---------------------------------------------------------------------------\\n\\n*** ./backend/executor/execAmi.c.orig\\tThu Jan 15 22:42:13 1998\\n--- ./backend/executor/execAmi.c\\tThu Jan 15 23:54:37 1998\\n***************\\n*** 42,47 ****\\n--- 42,48 ----\\n #include \\\"access/genam.h\\\"\\n #include \\\"access/heapam.h\\\"\\n #include \\\"catalog/heap.h\\\"\\n+ #include \\\"catalog/pg_type.h\\\"\\n \\n static Pointer\\n ExecBeginScan(Relation relation, int nkeys, ScanKey skeys,\\n***************\\n*** 124,129 ****\\n--- 125,155 ----\\n \\tif (relation == NULL)\\n \\t\\telog(DEBUG, \\\"ExecOpenR: relation == NULL, heap_open failed.\\\");\\n \\n+ \\t{\\n+ \\t\\tint i;\\n+ \\t\\tRelation trel = palloc(sizeof(RelationData));\\n+ \\t\\tTupleDesc tdesc = palloc(sizeof(struct tupleDesc));\\n+ \\t\\tAttributeTupleForm *tatt =\\n+ \\t\\t\\t\\tpalloc(sizeof(AttributeTupleForm*)*relation->rd_att->natts);\\n+ \\n+ \\t\\tmemcpy(trel, relation, sizeof(RelationData));\\n+ \\t\\tmemcpy(tdesc, relation->rd_att, sizeof(struct tupleDesc));\\n+ \\t\\ttrel->rd_att = tdesc;\\n+ \\t\\ttdesc->attrs = tatt;\\n+ \\t\\t\\n+ \\t\\tfor (i = 0; i < relation->rd_att->natts; i++)\\n+ \\t\\t{\\n+ \\t\\t\\tif (relation->rd_att->attrs[i]->atttypid != VARCHAROID)\\n+ \\t\\t\\t\\ttdesc->attrs[i] = relation->rd_att->attrs[i];\\n+ \\t\\t\\telse\\n+ \\t\\t\\t{\\n+ \\t\\t\\t\\ttdesc->attrs[i] = palloc(sizeof(FormData_pg_attribute));\\n+ \\t\\t\\t\\tmemcpy(tdesc->attrs[i], relation->rd_att->attrs[i],\\n+ \\t\\t\\t\\t\\t\\t\\t\\t\\t\\t\\tsizeof(FormData_pg_attribute));\\n+ \\t\\t\\t\\ttdesc->attrs[i]->attlen = -1;\\n+ \\t\\t\\t}\\n+ \\t\\t}\\n+ \\t}\\n \\treturn relation;\\n }\\n \\n\\n\\n-- \\nBruce Momjian\\nmaillist@candle.pha.pa.us\\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 00:05:03 -0500 (EST)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: [HACKERS] Re: varchar() troubles (fwd)\"\n },\n {\n \"msg_contents\": \"> OK, here is a temporary fix to the problem. It does the heap_open(),\\n> then replaces the attrs for VARCHAR with attlen of -1. You can't just\\n> change the field, because the data is in a cache and you are just\\n> returned a pointer.\\n>\\n> Can I add an 'attdeflen' for \\\"attributed defined length\\\" field to\\n> pg_attribute, and change the attlen references needed to the new field?\\n> This is the only proper way to fix it.\\n\\nBruce, does your \\\"temporary fix\\\" seem to repair all known problems with varchar()? If so, would you be interested in\\nholding off on a \\\"proper fix\\\" and coming back to it after v6.3 is released? At that time, we can try solving the general\\nproblem of retaining column-specific attributes, such as your max len for varchar, declared dimensions for arrays, and\\nnumeric() and decimal() types. Or, if you have time to try a solution now _and_ come back to it later...\\n\\n - Tom\\n\\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 05:26:08 +0000\",\n \"msg_from\": \"\\\"Thomas G. Lockhart\\\" \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Re: varchar() troubles (fwd)\"\n },\n {\n \"msg_contents\": \"> \\n> > OK, here is a temporary fix to the problem. It does the heap_open(),\\n> > then replaces the attrs for VARCHAR with attlen of -1. You can't just\\n> > change the field, because the data is in a cache and you are just\\n> > returned a pointer.\\n> >\\n> > Can I add an 'attdeflen' for \\\"attributed defined length\\\" field to\\n> > pg_attribute, and change the attlen references needed to the new field?\\n> > This is the only proper way to fix it.\\n> \\n\\n> Bruce, does your \\\"temporary fix\\\" seem to repair all known problems\\nwith varchar()? If so, would you be interested in > holding off on a\\n\\\"proper fix\\\" and coming back to it after v6.3 is released? At that time,\\nwe can try solving the general > problem of retaining column-specific\\nattributes, such as your max len for varchar, declared dimensions for\\narrays, and > numeric() and decimal() types. Or, if you have time to try\\na solution now _and_ come back to it later... > > \\n\\n[Those wide post really are difficult.]\\n\\nI don't think my solution is perfect or complete. I only caught one\\ngroup of heap_open calls used in the executor. I could funnel all of\\nthem through this patched function, but I can imagine there would be\\nones I would miss. Once the structure is gotten from the cache, it\\nseems to fly around the executor code quite freely, and it is hard to\\nknow when a tuple descriptor is being created, if it is being used for\\ndata creation or data reference. attlen references are much clearer in\\ntheir intent.\\n\\nIf I add a new field type to FormData_pg_attribute, I can then check\\neach attlen reference, and check if it is trying to move through the\\non-disk storage (attlen/typlen) or create a new/modify an entry\\n(attdeflen).\\n\\nHow much time I have depends on what Vadim needs me to do for\\nsubselects.\\n\\n\\n-- \\nBruce Momjian\\nmaillist@candle.pha.pa.us\\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 00:44:11 -0500 (EST)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: [HACKERS] Re: varchar() troubles (fwd)\"\n },\n {\n \"msg_contents\": \"> \\n> > OK, here is a temporary fix to the problem. It does the heap_open(),\\n> > then replaces the attrs for VARCHAR with attlen of -1. You can't just\\n> > change the field, because the data is in a cache and you are just\\n> > returned a pointer.\\n> >\\n> > Can I add an 'attdeflen' for \\\"attributed defined length\\\" field to\\n> > pg_attribute, and change the attlen references needed to the new field?\\n> > This is the only proper way to fix it.\\n> \\n> Bruce, does your \\\"temporary fix\\\" seem to repair all known problems with varchar()? If so, would you be interested in\\n> holding off on a \\\"proper fix\\\" and coming back to it after v6.3 is released? At that time, we can try solving the general\\n> problem of retaining column-specific attributes, such as your max len for varchar, declared dimensions for arrays, and\\n> numeric() and decimal() types. Or, if you have time to try a solution now _and_ come back to it later...\\n> \\n> - Tom\\n> \\n> \\n\\nIn fact, I am inclined to leave attlen unchanged, and add atttyplen that\\nis a copy of the length of the type. That way, the attlen for varchar()\\nreally contains the defined length, and atttyplen is used for disk\\nreferences, and it is very clear what it means.\\n\\n\\n-- \\nBruce Momjian\\nmaillist@candle.pha.pa.us\\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 00:49:02 -0500 (EST)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: [HACKERS] Re: varchar() troubles (fwd)\"\n },\n {\n \"msg_contents\": \"Bruce Momjian wrote:\\n> \\n> > >\\n> > > Can I add an 'attdeflen' for \\\"attributed defined length\\\" field to\\n> > > pg_attribute, and change the attlen references needed to the new field?\\n> > > This is the only proper way to fix it.\\n> >\\n> > Bruce, does your \\\"temporary fix\\\" seem to repair all known problems with varchar()? If so, would you be interested in\\n> > holding off on a \\\"proper fix\\\" and coming back to it after v6.3 is released? At that time, we can try solving the general\\n> > problem of retaining column-specific attributes, such as your max len for varchar, declared dimensions for arrays, and\\n> > numeric() and decimal() types. Or, if you have time to try a solution now _and_ come back to it later...\\n> >\\n> \\n> In fact, I am inclined to leave attlen unchanged, and add atttyplen that\\n> is a copy of the length of the type. That way, the attlen for varchar()\\n> really contains the defined length, and atttyplen is used for disk\\n> references, and it is very clear what it means.\\n\\npg_attribute.h:\\n\\n int2 attlen; \\n\\n /*\\n * attlen is a copy of the typlen field from pg_type for this\\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\\n * attribute... \\n ^^^^^^^^^\\n: I would suggest to don't change this to preserve the same meaning\\nfor all data types. attlen = -1 means that attribute is varlena.\\n\\nWe certainly need in new field in pg_attribute! I don't like names like\\n\\\"attdeflen\\\" or \\\"atttyplen\\\" - bad names for NUMERIC etc. Something like\\natttspec (ATTribute Type SPECification) is better, imho.\\n\\nFor the varchar(N) we'll have attlen = -1 and atttspec = N (or N + 4 - what's\\nbetter). For the text: attlen = -1 and atttspec = -1. And so on.\\n\\nOf 'course, it's not so much matter where to put maxlen of varchar.\\nattlen = -1 for varchar just seems more clear to me.\\n\\nBut in any case we need in new field and, imho, this should be added\\nin 6.3\\n\\nVadim\\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 16:31:45 +0700\",\n \"msg_from\": \"\\\"Vadim B. Mikheev\\\" \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Re: varchar() troubles (fwd)\"\n },\n {\n \"msg_contents\": \"> \\n> Bruce Momjian wrote:\\n> > \\n> > > >\\n> > > > Can I add an 'attdeflen' for \\\"attributed defined length\\\" field to\\n> > > > pg_attribute, and change the attlen references needed to the new field?\\n> > > > This is the only proper way to fix it.\\n> > >\\n> > > Bruce, does your \\\"temporary fix\\\" seem to repair all known problems with varchar()? If so, would you be interested in\\n> > > holding off on a \\\"proper fix\\\" and coming back to it after v6.3 is released? At that time, we can try solving the general\\n> > > problem of retaining column-specific attributes, such as your max len for varchar, declared dimensions for arrays, and\\n> > > numeric() and decimal() types. Or, if you have time to try a solution now _and_ come back to it later...\\n> > >\\n> > \\n> > In fact, I am inclined to leave attlen unchanged, and add atttyplen that\\n> > is a copy of the length of the type. That way, the attlen for varchar()\\n> > really contains the defined length, and atttyplen is used for disk\\n> > references, and it is very clear what it means.\\n> \\n> pg_attribute.h:\\n> \\n> int2 attlen; \\n> \\n> /*\\n> * attlen is a copy of the typlen field from pg_type for this\\n> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\\n> * attribute... \\n> ^^^^^^^^^\\n> : I would suggest to don't change this to preserve the same meaning\\n> for all data types. attlen = -1 means that attribute is varlena.\\n> \\n> We certainly need in new field in pg_attribute! I don't like names like\\n> \\\"attdeflen\\\" or \\\"atttyplen\\\" - bad names for NUMERIC etc. Something like\\n> atttspec (ATTribute Type SPECification) is better, imho.\\n> \\n> For the varchar(N) we'll have attlen = -1 and atttspec = N (or N + 4 - what's\\n> better). For the text: attlen = -1 and atttspec = -1. And so on.\\n> \\n> Of 'course, it's not so much matter where to put maxlen of varchar.\\n> attlen = -1 for varchar just seems more clear to me.\\n> \\n> But in any case we need in new field and, imho, this should be added\\n> in 6.3\\n\\nOK, we have a new pg_attribute column called 'atttypmod' for\\n'type-specific modifier'. Currently, it is only used to hold the char\\nand varchar length, but I am sure will be used soon for other types.\\n\\nHere is the test:\\n\\t\\n\\ttest=> insert into test values ('asdfasdfasdfasdfasdfadsfasdf11',3);\\n\\tINSERT 18282 1\\n\\ttest=> select * from test;\\n\\tx |y\\n\\t--------------------+-\\n\\tasdfasdfasdfasdfasdf|3\\n\\t(1 row)\\n\\n'attlen' was certainly a confusing double-used field that I am glad to\\nreturn to single-use status.\\n\\nI will be installing the patch soon, and will then start on subselects\\nin the parser. It will probably take me until Monday to finish that.\\n\\n-- \\nBruce Momjian\\nmaillist@candle.pha.pa.us\\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 17:48:24 -0500 (EST)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: [HACKERS] Re: varchar() troubles (fwd)\"\n },\n {\n \"msg_contents\": \"Bruce Momjian wrote:\\n> \\n> OK, we have a new pg_attribute column called 'atttypmod' for\\n> 'type-specific modifier'. Currently, it is only used to hold the char\\n> and varchar length, but I am sure will be used soon for other types.\\n\\nNice!\\n\\n> \\n> Here is the test:\\n> \\n> test=> insert into test values ('asdfasdfasdfasdfasdfadsfasdf11',3);\\n> INSERT 18282 1\\n> test=> select * from test;\\n> x |y\\n> --------------------+-\\n> asdfasdfasdfasdfasdf|3\\n> (1 row)\\n> \\n> 'attlen' was certainly a confusing double-used field that I am glad to\\n> return to single-use status.\\n> \\n> I will be installing the patch soon, and will then start on subselects\\n> in the parser. It will probably take me until Monday to finish that.\\n\\nOk.\\n\\nVadim\\n\",\n \"msg_date\": \"Sun, 18 Jan 1998 15:30:11 +0700\",\n \"msg_from\": \"\\\"Vadim B. Mikheev\\\" \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Re: varchar() troubles (fwd)\"\n }\n]"}}},{"rowIdx":139,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"> scan.l: In function `yylex':\n> scan.l:202: `ABORT' undeclared (first use this function)\n\nSince the snapshot is intended for experienced (or wanting to become so)\npostgresql'ers\nthat would like to help the team and don't have CVS (like me on AIX),\nmaybe the snapshot shoud be moved to a development subdir ?\n\nThat said, it's not a bad build, only there seem to be some files\nincluded\nthat should not, like scan.c (that is outdated) .\nDo a 'make clean' before the make and everything is fine.\n\nAndreas\n","msg_date":"Fri, 16 Jan 1998 09:23:53 +0100","msg_from":"Zeugswetter Andreas DBT ","msg_from_op":true,"msg_subject":"Nope: Cannot build recent snapshot"},{"msg_contents":"> \n> > scan.l: In function `yylex':\n> > scan.l:202: `ABORT' undeclared (first use this function)\n> \n> Since the snapshot is intended for experienced (or wanting to become so)\n> postgresql'ers\n> that would like to help the team and don't have CVS (like me on AIX),\n> maybe the snapshot shoud be moved to a development subdir ?\n> \n> That said, it's not a bad build, only there seem to be some files\n> included\n> that should not, like scan.c (that is outdated) .\n> Do a 'make clean' before the make and everything is fine.\n\nWe include scan.c because some lex versions can't handle scan.l.\n\n-- \nBruce Momjian\nmaillist@candle.pha.pa.us\n","msg_date":"Fri, 16 Jan 1998 09:20:20 -0500 (EST)","msg_from":"Bruce Momjian ","msg_from_op":false,"msg_subject":"Re: [HACKERS] Nope: Cannot build recent snapshot"}],"string":"[\n {\n \"msg_contents\": \"> scan.l: In function `yylex':\\n> scan.l:202: `ABORT' undeclared (first use this function)\\n\\nSince the snapshot is intended for experienced (or wanting to become so)\\npostgresql'ers\\nthat would like to help the team and don't have CVS (like me on AIX),\\nmaybe the snapshot shoud be moved to a development subdir ?\\n\\nThat said, it's not a bad build, only there seem to be some files\\nincluded\\nthat should not, like scan.c (that is outdated) .\\nDo a 'make clean' before the make and everything is fine.\\n\\nAndreas\\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 09:23:53 +0100\",\n \"msg_from\": \"Zeugswetter Andreas DBT \",\n \"msg_from_op\": true,\n \"msg_subject\": \"Nope: Cannot build recent snapshot\"\n },\n {\n \"msg_contents\": \"> \\n> > scan.l: In function `yylex':\\n> > scan.l:202: `ABORT' undeclared (first use this function)\\n> \\n> Since the snapshot is intended for experienced (or wanting to become so)\\n> postgresql'ers\\n> that would like to help the team and don't have CVS (like me on AIX),\\n> maybe the snapshot shoud be moved to a development subdir ?\\n> \\n> That said, it's not a bad build, only there seem to be some files\\n> included\\n> that should not, like scan.c (that is outdated) .\\n> Do a 'make clean' before the make and everything is fine.\\n\\nWe include scan.c because some lex versions can't handle scan.l.\\n\\n-- \\nBruce Momjian\\nmaillist@candle.pha.pa.us\\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 09:20:20 -0500 (EST)\",\n \"msg_from\": \"Bruce Momjian \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Nope: Cannot build recent snapshot\"\n }\n]"}}},{"rowIdx":140,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":">> OIDs are a bastardization of the relational model. If you have to\nkeep\n>> them, then do so, but their use should be SEVERELY discouraged.\n>\n>\tActually, I use them quite extensively...I have several\nWWW-based\n>search directories that are searched with:\n>\n>select oid, from where ;\n>\n>\tThat display minimal data to the browser, and then if someone\n>wants more detailed information, I just do:\n>\n>select * from where oid = '';\n>\n>\tIts also great if you mess up the original coding for a table\nand\n>want to remove 1 of many duplicates that you've accidently let pass\n>through :(\n\nSince OID is not a rowid (physical address) it needs an extra index,\nthat has to be built,\nfor what you state ROWID will be a lot better (since it don't need an\nindex)\nThis is on the TODO (something like add ctid to where clause)\n\nAndreas \n","msg_date":"Fri, 16 Jan 1998 09:34:29 +0100","msg_from":"Zeugswetter Andreas DBT ","msg_from_op":true,"msg_subject":"Re: [QUESTIONS] Arrays (inserting and removing)"}],"string":"[\n {\n \"msg_contents\": \">> OIDs are a bastardization of the relational model. If you have to\\nkeep\\n>> them, then do so, but their use should be SEVERELY discouraged.\\n>\\n>\\tActually, I use them quite extensively...I have several\\nWWW-based\\n>search directories that are searched with:\\n>\\n>select oid, from where ;\\n>\\n>\\tThat display minimal data to the browser, and then if someone\\n>wants more detailed information, I just do:\\n>\\n>select * from where oid = '';\\n>\\n>\\tIts also great if you mess up the original coding for a table\\nand\\n>want to remove 1 of many duplicates that you've accidently let pass\\n>through :(\\n\\nSince OID is not a rowid (physical address) it needs an extra index,\\nthat has to be built,\\nfor what you state ROWID will be a lot better (since it don't need an\\nindex)\\nThis is on the TODO (something like add ctid to where clause)\\n\\nAndreas \\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 09:34:29 +0100\",\n \"msg_from\": \"Zeugswetter Andreas DBT \",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: [QUESTIONS] Arrays (inserting and removing)\"\n }\n]"}}},{"rowIdx":141,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"unsubscribe b.aerts@syntegra.net\n","msg_date":"Fri, 16 Jan 1998 10:29:07 +0100","msg_from":"bas aerts ","msg_from_op":true,"msg_subject":"(no subject)"}],"string":"[\n {\n \"msg_contents\": \"unsubscribe b.aerts@syntegra.net\\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 10:29:07 +0100\",\n \"msg_from\": \"bas aerts \",\n \"msg_from_op\": true,\n \"msg_subject\": \"(no subject)\"\n }\n]"}}},{"rowIdx":142,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"unsubscribe pgsql-hackers\n","msg_date":"Fri, 16 Jan 1998 10:29:31 +0100","msg_from":"bas aerts ","msg_from_op":true,"msg_subject":"(no subject)"}],"string":"[\n {\n \"msg_contents\": \"unsubscribe pgsql-hackers\\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 10:29:31 +0100\",\n \"msg_from\": \"bas aerts \",\n \"msg_from_op\": true,\n \"msg_subject\": \"(no subject)\"\n }\n]"}}},{"rowIdx":143,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"> > \n> > > Fork off the postgres process first, then authenticate\ninside of\n> > > there...which would get rid of the problem with pg_user itself\nbeing a\n> > > text file vs a relation...no?\n> > \n> > Yes, yes, yes. This is how authentication should be done (for HBA,\netc.)\n> \n> No, no, no! For security reasons, you can't fork (and exec)\n> unauthenticated processes. Especially HBA authentication should be\ndone\n> to consume as low resources as possbile.\n\nStartup time for a valid connect client is now < 0.16 secs, so is this\nreally a threat ?\nI would say might leave hba to postmaster (since postgres don't need to\nknow about it)\nthen fork off postgres and do the rest of the authentication. \n\nRunning postgres as root though is a **very** bad idea.\nRemember that we have user defined Functions !\n\nno, yes, yes \nAndreas\n","msg_date":"Fri, 16 Jan 1998 10:33:00 +0100","msg_from":"Zeugswetter Andreas DBT ","msg_from_op":true,"msg_subject":"Re: [HACKERS] Re: New pg_pwd patch and stuff"},{"msg_contents":"On Fri, 16 Jan 1998, Zeugswetter Andreas DBT wrote:\n \n> Running postgres as root though is a **very** bad idea.\n> Remember that we have user defined Functions !\n\n\tpostmaster nor postgres will run as root, so I'm not sure where\nyou are coming up with a \"Running postgres as root...\" problem :(\n\n\n","msg_date":"Fri, 16 Jan 1998 08:08:02 -0500 (EST)","msg_from":"The Hermit Hacker ","msg_from_op":false,"msg_subject":"Re: [HACKERS] Re: New pg_pwd patch and stuff"}],"string":"[\n {\n \"msg_contents\": \"> > \\n> > > Fork off the postgres process first, then authenticate\\ninside of\\n> > > there...which would get rid of the problem with pg_user itself\\nbeing a\\n> > > text file vs a relation...no?\\n> > \\n> > Yes, yes, yes. This is how authentication should be done (for HBA,\\netc.)\\n> \\n> No, no, no! For security reasons, you can't fork (and exec)\\n> unauthenticated processes. Especially HBA authentication should be\\ndone\\n> to consume as low resources as possbile.\\n\\nStartup time for a valid connect client is now < 0.16 secs, so is this\\nreally a threat ?\\nI would say might leave hba to postmaster (since postgres don't need to\\nknow about it)\\nthen fork off postgres and do the rest of the authentication. \\n\\nRunning postgres as root though is a **very** bad idea.\\nRemember that we have user defined Functions !\\n\\nno, yes, yes \\nAndreas\\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 10:33:00 +0100\",\n \"msg_from\": \"Zeugswetter Andreas DBT \",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: [HACKERS] Re: New pg_pwd patch and stuff\"\n },\n {\n \"msg_contents\": \"On Fri, 16 Jan 1998, Zeugswetter Andreas DBT wrote:\\n \\n> Running postgres as root though is a **very** bad idea.\\n> Remember that we have user defined Functions !\\n\\n\\tpostmaster nor postgres will run as root, so I'm not sure where\\nyou are coming up with a \\\"Running postgres as root...\\\" problem :(\\n\\n\\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 08:08:02 -0500 (EST)\",\n \"msg_from\": \"The Hermit Hacker \",\n \"msg_from_op\": false,\n \"msg_subject\": \"Re: [HACKERS] Re: New pg_pwd patch and stuff\"\n }\n]"}}},{"rowIdx":144,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"\nAnother *cool* project would be a LDAPd front-end to PgSQL,\nor perhaps rather a PgSQL-backend to SLAPD.\n\nI am planning to look into this when I got the neccesary time,\nbut feel free to start ahead of me if you like the idea.\n\n happy hacking,\n-- \n---------------------------------------------\nG�ran Thyni, sysadm, JMS Bildbasen, Kiruna\n\n","msg_date":"16 Jan 1998 13:21:08 -0000","msg_from":"Goran Thyni ","msg_from_op":true,"msg_subject":"LDAP (was: CBAC ...)"}],"string":"[\n {\n \"msg_contents\": \"\\nAnother *cool* project would be a LDAPd front-end to PgSQL,\\nor perhaps rather a PgSQL-backend to SLAPD.\\n\\nI am planning to look into this when I got the neccesary time,\\nbut feel free to start ahead of me if you like the idea.\\n\\n happy hacking,\\n-- \\n---------------------------------------------\\nG�ran Thyni, sysadm, JMS Bildbasen, Kiruna\\n\\n\",\n \"msg_date\": \"16 Jan 1998 13:21:08 -0000\",\n \"msg_from\": \"Goran Thyni \",\n \"msg_from_op\": true,\n \"msg_subject\": \"LDAP (was: CBAC ...)\"\n }\n]"}}},{"rowIdx":145,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"> On 12-Jan-98 The Hermit Hacker wrote:\n> > Does anyone here *understand* the LGPL? If we put the ODBC\n> >drivers *under* src/interfaces, does that risk contaminating the rest of\n> >the code *in any way*? Anyone here done a reasonably thorough study of\n> >the LGPL and can comment on it?\n> \n> My understanding from Stallman's statements on the matter are: Distribution of\n> GPL'd source with non-GPL'd source is fine, as long as it is simple to figure\n> out which is which. By definition, GPL'd sources can be distributed freely.\n> For binaries which fall under the GPL, again, mixing them with other stuff is\n> OK, as long as GPL'd stuff is identified as such. Sources must be available,\n> of course.\n> \n> LGPL is completely different. LGPL is what you use when you link your\n> non-GPL'd sources against a library built with GPL'd sources. In that case,\n> you are legal IFF you stuff can be re-linked against a different, non-GPL'd\n> library without recompilation. Actually, there's a bit of confusion on my\n> part about how much recompilation is permitted.\n> \n> Companies like DG/Sequent/Sun/etc wouldn't be able to include FSF software on\n> the distributions if the above were not the case.\n> \n> ObCaveat: I'm not a lawyer. I don't look like a lawyer, I don't smell like a\n> lawyer, and I don't lie like a lawyer.\n> \n> \nMy understanding is pretty much the same. Originally there was only GPL. This\nreally says that anything you link with GPL code must be distributed under\nGPL - effectively your source becomes part of the original GPL'd product.\n\nClearly this is ridiculous when you are linking against, say, the GNU\nC-library, so Stallman introduced LGPL which effectively says that any\nmodifications or additions you make to the library fall under the LGPL,\nbut anything which calls the LGPL'd library can have whatever copyright\nyou want. Thus it is possible to produce commercial products which use\nthe GNU C-library, etc., etc.\n\n\nAndrew\n\n(Same caveat as above :-)\n\n\n----------------------------------------------------------------------------\nDr. Andrew C.R. Martin University College London\nEMAIL: (Work) martin@biochem.ucl.ac.uk (Home) andrew@stagleys.demon.co.uk\nURL: http://www.biochem.ucl.ac.uk/~martin\nTel: (Work) +44(0)171 419 3890 (Home) +44(0)1372 275775\n","msg_date":"Fri, 16 Jan 1998 15:06:12 GMT","msg_from":"Andrew Martin ","msg_from_op":true,"msg_subject":"RE: [HACKERS] ODBC & LGPL license..."},{"msg_contents":"On Fri, 16 Jan 1998, Andrew Martin wrote:\n\n> > On 12-Jan-98 The Hermit Hacker wrote:\n> > > Does anyone here *understand* the LGPL? If we put the ODBC\n> > >drivers *under* src/interfaces, does that risk contaminating the rest of\n> > >the code *in any way*? Anyone here done a reasonably thorough study of\n> > >the LGPL and can comment on it?\n> > \n> > My understanding from Stallman's statements on the matter are: Distribution of\n> > GPL'd source with non-GPL'd source is fine, as long as it is simple to figure\n> > out which is which. By definition, GPL'd sources can be distributed freely.\n> > For binaries which fall under the GPL, again, mixing them with other stuff is\n> > OK, as long as GPL'd stuff is identified as such. Sources must be available,\n> > of course.\n> > \n> > LGPL is completely different. LGPL is what you use when you link your\n> > non-GPL'd sources against a library built with GPL'd sources. In that case,\n> > you are legal IFF you stuff can be re-linked against a different, non-GPL'd\n> > library without recompilation. Actually, there's a bit of confusion on my\n> > part about how much recompilation is permitted.\n> > \n> > Companies like DG/Sequent/Sun/etc wouldn't be able to include FSF software on\n> > the distributions if the above were not the case.\n> > \n> > ObCaveat: I'm not a lawyer. I don't look like a lawyer, I don't smell like a\n> > lawyer, and I don't lie like a lawyer.\n> > \n> > \n> My understanding is pretty much the same. Originally there was only GPL. This\n> really says that anything you link with GPL code must be distributed under\n> GPL - effectively your source becomes part of the original GPL'd product.\n> \n> Clearly this is ridiculous when you are linking against, say, the GNU\n> C-library, so Stallman introduced LGPL which effectively says that any\n> modifications or additions you make to the library fall under the LGPL,\n> but anything which calls the LGPL'd library can have whatever copyright\n> you want. Thus it is possible to produce commercial products which use\n> the GNU C-library, etc., etc.\n\n\tOkay, then going back to the original...the PostODBC drivers that\nI'd like to include as part of the src/interfaces directory falls under\nLGPL...if we did include it, then we wouldn't/shouldn't be contaminating\nthe source tree in any way?\n\n\n\n","msg_date":"Fri, 16 Jan 1998 10:41:09 -0500 (EST)","msg_from":"The Hermit Hacker ","msg_from_op":false,"msg_subject":"RE: [HACKERS] ODBC & LGPL license..."}],"string":"[\n {\n \"msg_contents\": \"> On 12-Jan-98 The Hermit Hacker wrote:\\n> > Does anyone here *understand* the LGPL? If we put the ODBC\\n> >drivers *under* src/interfaces, does that risk contaminating the rest of\\n> >the code *in any way*? Anyone here done a reasonably thorough study of\\n> >the LGPL and can comment on it?\\n> \\n> My understanding from Stallman's statements on the matter are: Distribution of\\n> GPL'd source with non-GPL'd source is fine, as long as it is simple to figure\\n> out which is which. By definition, GPL'd sources can be distributed freely.\\n> For binaries which fall under the GPL, again, mixing them with other stuff is\\n> OK, as long as GPL'd stuff is identified as such. Sources must be available,\\n> of course.\\n> \\n> LGPL is completely different. LGPL is what you use when you link your\\n> non-GPL'd sources against a library built with GPL'd sources. In that case,\\n> you are legal IFF you stuff can be re-linked against a different, non-GPL'd\\n> library without recompilation. Actually, there's a bit of confusion on my\\n> part about how much recompilation is permitted.\\n> \\n> Companies like DG/Sequent/Sun/etc wouldn't be able to include FSF software on\\n> the distributions if the above were not the case.\\n> \\n> ObCaveat: I'm not a lawyer. I don't look like a lawyer, I don't smell like a\\n> lawyer, and I don't lie like a lawyer.\\n> \\n> \\nMy understanding is pretty much the same. Originally there was only GPL. This\\nreally says that anything you link with GPL code must be distributed under\\nGPL - effectively your source becomes part of the original GPL'd product.\\n\\nClearly this is ridiculous when you are linking against, say, the GNU\\nC-library, so Stallman introduced LGPL which effectively says that any\\nmodifications or additions you make to the library fall under the LGPL,\\nbut anything which calls the LGPL'd library can have whatever copyright\\nyou want. Thus it is possible to produce commercial products which use\\nthe GNU C-library, etc., etc.\\n\\n\\nAndrew\\n\\n(Same caveat as above :-)\\n\\n\\n----------------------------------------------------------------------------\\nDr. Andrew C.R. Martin University College London\\nEMAIL: (Work) martin@biochem.ucl.ac.uk (Home) andrew@stagleys.demon.co.uk\\nURL: http://www.biochem.ucl.ac.uk/~martin\\nTel: (Work) +44(0)171 419 3890 (Home) +44(0)1372 275775\\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 15:06:12 GMT\",\n \"msg_from\": \"Andrew Martin \",\n \"msg_from_op\": true,\n \"msg_subject\": \"RE: [HACKERS] ODBC & LGPL license...\"\n },\n {\n \"msg_contents\": \"On Fri, 16 Jan 1998, Andrew Martin wrote:\\n\\n> > On 12-Jan-98 The Hermit Hacker wrote:\\n> > > Does anyone here *understand* the LGPL? If we put the ODBC\\n> > >drivers *under* src/interfaces, does that risk contaminating the rest of\\n> > >the code *in any way*? Anyone here done a reasonably thorough study of\\n> > >the LGPL and can comment on it?\\n> > \\n> > My understanding from Stallman's statements on the matter are: Distribution of\\n> > GPL'd source with non-GPL'd source is fine, as long as it is simple to figure\\n> > out which is which. By definition, GPL'd sources can be distributed freely.\\n> > For binaries which fall under the GPL, again, mixing them with other stuff is\\n> > OK, as long as GPL'd stuff is identified as such. Sources must be available,\\n> > of course.\\n> > \\n> > LGPL is completely different. LGPL is what you use when you link your\\n> > non-GPL'd sources against a library built with GPL'd sources. In that case,\\n> > you are legal IFF you stuff can be re-linked against a different, non-GPL'd\\n> > library without recompilation. Actually, there's a bit of confusion on my\\n> > part about how much recompilation is permitted.\\n> > \\n> > Companies like DG/Sequent/Sun/etc wouldn't be able to include FSF software on\\n> > the distributions if the above were not the case.\\n> > \\n> > ObCaveat: I'm not a lawyer. I don't look like a lawyer, I don't smell like a\\n> > lawyer, and I don't lie like a lawyer.\\n> > \\n> > \\n> My understanding is pretty much the same. Originally there was only GPL. This\\n> really says that anything you link with GPL code must be distributed under\\n> GPL - effectively your source becomes part of the original GPL'd product.\\n> \\n> Clearly this is ridiculous when you are linking against, say, the GNU\\n> C-library, so Stallman introduced LGPL which effectively says that any\\n> modifications or additions you make to the library fall under the LGPL,\\n> but anything which calls the LGPL'd library can have whatever copyright\\n> you want. Thus it is possible to produce commercial products which use\\n> the GNU C-library, etc., etc.\\n\\n\\tOkay, then going back to the original...the PostODBC drivers that\\nI'd like to include as part of the src/interfaces directory falls under\\nLGPL...if we did include it, then we wouldn't/shouldn't be contaminating\\nthe source tree in any way?\\n\\n\\n\\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 10:41:09 -0500 (EST)\",\n \"msg_from\": \"The Hermit Hacker \",\n \"msg_from_op\": false,\n \"msg_subject\": \"RE: [HACKERS] ODBC & LGPL license...\"\n }\n]"}}},{"rowIdx":146,"cells":{"threads":{"kind":"list like","value":[{"msg_contents":"> Got my Linux Journal today and the first featured article is entitled\n> \n> \"PostgreSQL - The Linux of Databases\"\n> \n> Now scrappy, before you get bent out of joint, they mean this in a nice\n> way ;-)\n> \n> The author is Rolf Herzog from Germany. It seems like a good article,\n> with a few factual errors but on the whole complimentary without\n> ignoring the weak points (yes Bruce, subselects are mentioned).\n> \nHaven't seen it yet - guess my copy will arrive any day. It does seem\na shame that an article like this came from someone who isn't active on\nany of these lists (AFAIK). Maybe he could have posted the article here\nso any factual errors could have been pointed out before it went\nto LJ.\n\nLJ have featured PG/SQL on a couple of previous occassions --- it really\nis a very good Unix magazine; an awful lot of it is far from Linux\nspecific.\n\nBest Wishes,\n\nAndrew\n\n----------------------------------------------------------------------------\nDr. Andrew C.R. Martin University College London\nEMAIL: (Work) martin@biochem.ucl.ac.uk (Home) andrew@stagleys.demon.co.uk\nURL: http://www.biochem.ucl.ac.uk/~martin\nTel: (Work) +44(0)171 419 3890 (Home) +44(0)1372 275775\n","msg_date":"Fri, 16 Jan 1998 15:25:35 GMT","msg_from":"Andrew Martin ","msg_from_op":true,"msg_subject":"Re: [HACKERS] Linux Journal article on PostgreSQL"},{"msg_contents":"On Fri, 16 Jan 1998, Andrew Martin wrote:\n\n> > Got my Linux Journal today and the first featured article is entitled\n> > \n> > \"PostgreSQL - The Linux of Databases\"\n> > \n> > Now scrappy, before you get bent out of joint, they mean this in a nice\n> > way ;-)\n> > \n> > The author is Rolf Herzog from Germany. It seems like a good article,\n> > with a few factual errors but on the whole complimentary without\n> > ignoring the weak points (yes Bruce, subselects are mentioned).\n> > \n> Haven't seen it yet - guess my copy will arrive any day. It does seem\n> a shame that an article like this came from someone who isn't active on\n> any of these lists (AFAIK). Maybe he could have posted the article here\n> so any factual errors could have been pointed out before it went\n> to LJ.\n\n\tI'm borrowing a copy from one of the profs here at the University,\nbut from Thomas' \"review\", sounds like he was pretty accurate :)\n\n\n","msg_date":"Fri, 16 Jan 1998 10:44:24 -0500 (EST)","msg_from":"The Hermit Hacker ","msg_from_op":false,"msg_subject":"Re: [HACKERS] Linux Journal article on PostgreSQL"},{"msg_contents":"> > Got my Linux Journal today and the first featured article is entitled\n> > \"PostgreSQL - The Linux of Databases\"\n> Haven't seen it yet - guess my copy will arrive any day. It does seem\n> a shame that an article like this came from someone who isn't active on\n> any of these lists (AFAIK). Maybe he could have posted the article here\n> so any factual errors could have been pointed out before it went\n> to LJ.\n\nYes, that was my initial reaction too. However, to look at it another way, it\nis great that someone, who is not as wrapped up in Postgres as those of us\nactive on the lists seem to be, can look at it, use it, and say nice things\nabout it. I have _never_ seen an article written about something with which I\nam familiar which seemed to be 100% correct, except (perhaps :) in refereed\njournals.\n\n - Tom\n\n","msg_date":"Fri, 16 Jan 1998 16:23:10 +0000","msg_from":"\"Thomas G. Lockhart\" ","msg_from_op":false,"msg_subject":"Re: [HACKERS] Linux Journal article on PostgreSQL"}],"string":"[\n {\n \"msg_contents\": \"> Got my Linux Journal today and the first featured article is entitled\\n> \\n> \\\"PostgreSQL - The Linux of Databases\\\"\\n> \\n> Now scrappy, before you get bent out of joint, they mean this in a nice\\n> way ;-)\\n> \\n> The author is Rolf Herzog from Germany. It seems like a good article,\\n> with a few factual errors but on the whole complimentary without\\n> ignoring the weak points (yes Bruce, subselects are mentioned).\\n> \\nHaven't seen it yet - guess my copy will arrive any day. It does seem\\na shame that an article like this came from someone who isn't active on\\nany of these lists (AFAIK). Maybe he could have posted the article here\\nso any factual errors could have been pointed out before it went\\nto LJ.\\n\\nLJ have featured PG/SQL on a couple of previous occassions --- it really\\nis a very good Unix magazine; an awful lot of it is far from Linux\\nspecific.\\n\\nBest Wishes,\\n\\nAndrew\\n\\n----------------------------------------------------------------------------\\nDr. Andrew C.R. Martin University College London\\nEMAIL: (Work) martin@biochem.ucl.ac.uk (Home) andrew@stagleys.demon.co.uk\\nURL: http://www.biochem.ucl.ac.uk/~martin\\nTel: (Work) +44(0)171 419 3890 (Home) +44(0)1372 275775\\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 15:25:35 GMT\",\n \"msg_from\": \"Andrew Martin \",\n \"msg_from_op\": true,\n \"msg_subject\": \"Re: [HACKERS] Linux Journal article on PostgreSQL\"\n },\n {\n \"msg_contents\": \"On Fri, 16 Jan 1998, Andrew Martin wrote:\\n\\n> > Got my Linux Journal today and the first featured article is entitled\\n> > \\n> > \\\"PostgreSQL - The Linux of Databases\\\"\\n> > \\n> > Now scrappy, before you get bent out of joint, they mean this in a nice\\n> > way ;-)\\n> > \\n> > The author is Rolf Herzog from Germany. It seems like a good article,\\n> > with a few factual errors but on the whole complimentary without\\n> > ignoring the weak points (yes Bruce, subselects are mentioned).\\n> > \\n> Haven't seen it yet - guess my copy will arrive any day. It does seem\\n> a shame that an article like this came from someone who isn't active on\\n> any of these lists (AFAIK). Maybe he could have posted the article here\\n> so any factual errors could have been pointed out before it went\\n> to LJ.\\n\\n\\tI'm borrowing a copy from one of the profs here at the University,\\nbut from Thomas' \\\"review\\\", sounds like he was pretty accurate :)\\n\\n\\n\",\n \"msg_date\": \"Fri, 16 Jan 1998 10:44:24 -0500 (EST)\",\n \"msg_from\": \"The Hermit Hacker