]> granicus.if.org Git - postgresql/log
postgresql
17 years agoMove full text search operators, functions, and data type sections into
Bruce Momjian [Wed, 29 Aug 2007 20:37:14 +0000 (20:37 +0000)]
Move full text search operators, functions, and data type sections into
the main documentation, out of its own text search chapter.

17 years agoRelax permissions checks on dbsize functions, per discussion. Revert out all
Tom Lane [Wed, 29 Aug 2007 17:24:29 +0000 (17:24 +0000)]
Relax permissions checks on dbsize functions, per discussion.  Revert out all
checks for individual-table-size functions, since anyone in the database could
get approximate values from pg_class.relpages anyway.  Allow database-size to
users with CONNECT privilege for the target database (note that this is
granted by default).  Allow tablespace-size if the user has CREATE privilege
on the tablespace (which is *not* granted by default), or if the tablespace is
the default tablespace for the current database (since we treat that as
implicitly allowing use of the tablespace).

17 years agoFix aboriginal bug in _tarAddFile(): when complaining that the amount of data
Tom Lane [Wed, 29 Aug 2007 16:31:36 +0000 (16:31 +0000)]
Fix aboriginal bug in _tarAddFile(): when complaining that the amount of data
read from the temp file didn't match the file length reported by ftello(),
the wrong variable's value was printed, and so the message made no sense.
Clean up a couple other coding infelicities while at it.

17 years agoFixed bug in Informix define handling.
Michael Meskes [Wed, 29 Aug 2007 13:58:13 +0000 (13:58 +0000)]
Fixed bug in Informix define handling.

17 years agoProperly indent SGML in textsearch.sgml.
Bruce Momjian [Wed, 29 Aug 2007 02:37:04 +0000 (02:37 +0000)]
Properly indent SGML in textsearch.sgml.

17 years agoAdd a debug logging message when a resource manager rejects an attempted
Tom Lane [Tue, 28 Aug 2007 23:17:47 +0000 (23:17 +0000)]
Add a debug logging message when a resource manager rejects an attempted
restart point.  Per suggestion from Simon Riggs.

17 years agoReduce the permissions check needed to use pgrowlocks() to having
Tom Lane [Tue, 28 Aug 2007 22:59:30 +0000 (22:59 +0000)]
Reduce the permissions check needed to use pgrowlocks() to having
SELECT on the target table.  Per discussion.

17 years agoImprove behavior of log_lock_waits patch. Ensure that something gets logged
Tom Lane [Tue, 28 Aug 2007 03:23:44 +0000 (03:23 +0000)]
Improve behavior of log_lock_waits patch.  Ensure that something gets logged
even if the "deadlock detected" ERROR message is suppressed by an exception
catcher.  Be clearer about the event sequence when a soft deadlock is fixed:
the fixing process might or might not still have to wait, so log that
separately.  Fix race condition when someone releases us from the lock partway
through printing all this junk --- we'd not get confused about our state, but
the log message sequence could have been misleading, ie, a "still waiting"
message with no subsequent "acquired" message.  Greg Stark and Tom Lane.

17 years agoMention configurations early in text search documentation to table/index
Bruce Momjian [Tue, 28 Aug 2007 03:10:45 +0000 (03:10 +0000)]
Mention configurations early in text search documentation to table/index
section makes a little more sense.

17 years agoPlease ignore. Fresh checkout to see if this fixes it.
Bruce Momjian [Tue, 28 Aug 2007 03:09:01 +0000 (03:09 +0000)]
Please ignore.  Fresh checkout to see if this fixes it.

17 years agoCommit test, please ignore.
Bruce Momjian [Tue, 28 Aug 2007 03:06:07 +0000 (03:06 +0000)]
Commit test, please ignore.

17 years agoCVS commit test, ignore.
Bruce Momjian [Tue, 28 Aug 2007 03:05:31 +0000 (03:05 +0000)]
CVS commit test, ignore.

17 years agoCVS commit test for TODO. Please ignore.
Bruce Momjian [Tue, 28 Aug 2007 03:03:36 +0000 (03:03 +0000)]
CVS commit test for TODO.  Please ignore.

17 years agoSmall commit test for TODO.
Bruce Momjian [Tue, 28 Aug 2007 03:01:39 +0000 (03:01 +0000)]
Small commit test for TODO.

17 years agoUpdate TODO.html.
Bruce Momjian [Tue, 28 Aug 2007 03:01:08 +0000 (03:01 +0000)]
Update TODO.html.

17 years agoUpdate TODO with lost change, mostly completed items and whitespace
Bruce Momjian [Tue, 28 Aug 2007 03:00:36 +0000 (03:00 +0000)]
Update TODO with lost change, mostly completed items and whitespace
corrections.  Not sure why these were not in CVS.  Researching.

17 years agoBatch of TODO updates, mostly completed item marking and whitespace
Bruce Momjian [Tue, 28 Aug 2007 02:57:10 +0000 (02:57 +0000)]
Batch of TODO updates, mostly completed item marking and whitespace
fixes.  Not sure how these weren't comitted before.

17 years agoExclude tsearch2 contrib tests in regression tests,
Magnus Hagander [Mon, 27 Aug 2007 12:10:47 +0000 (12:10 +0000)]
Exclude tsearch2 contrib tests in regression tests,
pending decision on exactly what will happen with
contrib/tsearch2 now that it's in core.

17 years agoInstall stopword files
Magnus Hagander [Mon, 27 Aug 2007 10:51:15 +0000 (10:51 +0000)]
Install stopword files

17 years agoFix generation of snowball_create.sql on msvc builds.
Magnus Hagander [Mon, 27 Aug 2007 10:29:49 +0000 (10:29 +0000)]
Fix generation of snowball_create.sql on msvc builds.

17 years agoFix a couple of misbehaviors rooted in the fact that the default creation
Tom Lane [Mon, 27 Aug 2007 03:36:08 +0000 (03:36 +0000)]
Fix a couple of misbehaviors rooted in the fact that the default creation
namespace isn't necessarily first in the search path (there could be implicit
schemas ahead of it).  Examples are

test=# set search_path TO s1;

test=# create view pg_timezone_names as select * from pg_timezone_names();
ERROR:  "pg_timezone_names" is already a view

test=# create table pg_class (f1 int primary key);
ERROR:  permission denied: "pg_class" is a system catalog

You'd expect these commands to create the requested objects in s1, since
names beginning with pg_ aren't supposed to be reserved anymore.  What is
happening is that we create the requested base table and then execute
additional commands (here, CREATE RULE or CREATE INDEX), and that code is
passed the same RangeVar that was in the original command.  Since that
RangeVar has schemaname = NULL, the secondary commands think they should do a
path search, and that means they find system catalogs that are implicitly in
front of s1 in the search path.

This is perilously close to being a security hole: if the secondary command
failed to apply a permission check then it'd be possible for unprivileged
users to make schema modifications to system catalogs.  But as far as I can
find, there is no code path in which a check doesn't occur.  Which makes it
just a weird corner-case bug for people who are silly enough to want to
name their tables the same as a system catalog.

The relevant code has changed quite a bit since 8.2, which means this patch
wouldn't work as-is in the back branches.  Since it's a corner case no one
has reported from the field, I'm not going to bother trying to back-patch.

17 years agoRemove the 'not in' operator (!!=). This was a hangover from Berkeley
Tom Lane [Mon, 27 Aug 2007 01:39:25 +0000 (01:39 +0000)]
Remove the 'not in' operator (!!=).  This was a hangover from Berkeley
days that was obsolete the moment we had IN (SELECT ...) capability.
It's arguably a security hole since it applied no permissions check to
the table it searched, and since it was never documented anywhere,
removing it seems more appropriate than fixing it.

17 years agoRequire SELECT privilege on a table to do dblink_get_pkey(). This is
Tom Lane [Mon, 27 Aug 2007 01:24:50 +0000 (01:24 +0000)]
Require SELECT privilege on a table to do dblink_get_pkey().  This is
not all that exciting when the system catalogs are readable by all,
but some people try to lock them down, and would not like this sort of
end run ...

17 years agoRestrict pg_relation_size to relation owner, pg_database_size to DB owner,
Tom Lane [Mon, 27 Aug 2007 01:19:14 +0000 (01:19 +0000)]
Restrict pg_relation_size to relation owner, pg_database_size to DB owner,
and pg_tablespace_size to superusers.  Perhaps we could weaken the first
case to just require SELECT privilege, but that doesn't work for the
other cases, so use ownership as the common concept.

17 years agoMake currtid() functions require SELECT privileges on the target table.
Tom Lane [Mon, 27 Aug 2007 00:57:36 +0000 (00:57 +0000)]
Make currtid() functions require SELECT privileges on the target table.
While it's not clear that TID linkage info is of any great use to a
nefarious user, it's certainly unexpected that these functions wouldn't
insist on read privileges.

17 years agoRestrict pgrowlocks function to superusers. (This might be too strict,
Tom Lane [Mon, 27 Aug 2007 00:13:51 +0000 (00:13 +0000)]
Restrict pgrowlocks function to superusers.  (This might be too strict,
but no permissions check at all is certainly no good.)  Clean up usage
of some deprecated APIs.

17 years agoRestrict pgstattuple functions to superusers. (This might be too strict,
Tom Lane [Sun, 26 Aug 2007 23:59:50 +0000 (23:59 +0000)]
Restrict pgstattuple functions to superusers.  (This might be too strict,
but no permissions check at all is certainly no good.)  Clean up usage
of some deprecated APIs.

17 years agoClean up usage of some deprecated APIs, particularly schema-unsafe
Tom Lane [Sun, 26 Aug 2007 23:36:32 +0000 (23:36 +0000)]
Clean up usage of some deprecated APIs, particularly schema-unsafe
usage of RelationNameGetTupleDesc().

17 years agoCode review for btreefuncs additions: restrict to superusers to avoid
Tom Lane [Sun, 26 Aug 2007 23:22:49 +0000 (23:22 +0000)]
Code review for btreefuncs additions: restrict to superusers to avoid
exposing user data to others, and clean up usage of deprecated APIs.

17 years agoMake ARRAY(SELECT ...) return an empty array, rather than a NULL, when the
Tom Lane [Sun, 26 Aug 2007 21:44:25 +0000 (21:44 +0000)]
Make ARRAY(SELECT ...) return an empty array, rather than a NULL, when the
sub-select returns zero rows.  Per complaint from Jens Schicke.  Since this
is more in the nature of a definition change than a bug, not back-patched.

17 years agoAdjust with-system-tzdata patch to not attempt to install a symlink,
Tom Lane [Sat, 25 Aug 2007 20:29:25 +0000 (20:29 +0000)]
Adjust with-system-tzdata patch to not attempt to install a symlink,
but just hardwire the specified timezone database path into the executable.
Per discussion, this avoids some packaging disadvantages of using a
symlink.

17 years agoFix brain fade in DefineIndex(): it was continuing to access the table's
Tom Lane [Sat, 25 Aug 2007 19:08:19 +0000 (19:08 +0000)]
Fix brain fade in DefineIndex(): it was continuing to access the table's
relcache entry after having heap_close'd it.  This could lead to misbehavior
if a relcache flush wiped out the cache entry meanwhile.  In 8.2 there is a
very real risk of CREATE INDEX CONCURRENTLY using the wrong relid for locking
and waiting purposes.  I think the bug is only cosmetic in 8.0 and 8.1,
because their transgression is limited to using RelationGetRelationName(rel)
in an ereport message immediately after heap_close, and there's no way (except
with special debugging options) for a cache flush to occur in that interval.
Not quite sure that it's cosmetic in 7.4, but seems best to patch anyway.

Found by trying to run the regression tests with CLOBBER_CACHE_ALWAYS enabled.
Maybe we should try to do that on a regular basis --- it's awfully slow,
but perhaps some fast buildfarm machine could do it once in awhile.

17 years agoSimplify implementation of ts_debug() function --- use a join instead
Tom Lane [Sat, 25 Aug 2007 17:47:44 +0000 (17:47 +0000)]
Simplify implementation of ts_debug() function --- use a join instead
of redundant sub-selects.  initdb not forced, since this is just a
cosmetic change, but the new code won't show up till you do one.

17 years agoThe upper limit for -c option of pgbench is now obtained from
Tatsuo Ishii [Sat, 25 Aug 2007 09:21:14 +0000 (09:21 +0000)]
The upper limit for -c option of pgbench is now obtained from
(FD_SETSIZE - 10) rather than a hardwired number.

17 years agoSome more tsearch docs work --- sync names with CVS-tip reality, some
Tom Lane [Sat, 25 Aug 2007 06:26:57 +0000 (06:26 +0000)]
Some more tsearch docs work --- sync names with CVS-tip reality, some
minor rewording, some markup fixups.  Lots left to do here ...

17 years agoFix synonym-dict breakage introduced in last patch :-(.
Tom Lane [Sat, 25 Aug 2007 02:29:45 +0000 (02:29 +0000)]
Fix synonym-dict breakage introduced in last patch :-(.
Minor other cleanups.

17 years agoRename built-in Snowball stemmer dictionaries to be english_stem,
Tom Lane [Sat, 25 Aug 2007 01:06:25 +0000 (01:06 +0000)]
Rename built-in Snowball stemmer dictionaries to be english_stem,
russian_stem, etc.  Per discussion.

17 years agoCleanup for some problems in tsearch patch:
Tom Lane [Sat, 25 Aug 2007 00:03:59 +0000 (00:03 +0000)]
Cleanup for some problems in tsearch patch:
- ispell initialization crashed on empty dictionary file
- ispell initialization crashed on affix file with prefixes but no suffixes
- stop words file was run through pg_verify_mbstr, with database
  encoding, but it's supposed to be UTF-8; similar bug for synonym files
- bunch of comments added, typos fixed, and other cleanup

Introduced consistent encoding checking/conversion of data read from tsearch
configuration files, by doing this in a single t_readline() subroutine
(replacing direct usages of fgets).  Cleaned up API for readstopwords too.

Heikki Linnakangas

17 years agoFix combo_decrypt() to throw an error for zero-length input when using a
Tom Lane [Thu, 23 Aug 2007 16:15:51 +0000 (16:15 +0000)]
Fix combo_decrypt() to throw an error for zero-length input when using a
padded encryption scheme.  Formerly it would try to access res[(unsigned) -1],
which resulted in core dumps on 64-bit machines, and was certainly trouble
waiting to happen on 32-bit machines (though in at least the known case
it was harmless because that byte would be overwritten after return).
Per report from Ken Colson; fix by Marko Kreen.

17 years agoReduce memory requirements for writing CSVlogs, so it will work with about
Andrew Dunstan [Thu, 23 Aug 2007 01:24:43 +0000 (01:24 +0000)]
Reduce memory requirements for writing CSVlogs, so it will work with about
the same amount of memory in ErrorContext as standard logs.

17 years agounbreak ECPG tests for VPATH builds.
Andrew Dunstan [Thu, 23 Aug 2007 00:38:17 +0000 (00:38 +0000)]
unbreak ECPG tests for VPATH builds.

17 years agoFix typo in FAQ_DEV.
Bruce Momjian [Thu, 23 Aug 2007 00:10:07 +0000 (00:10 +0000)]
Fix typo in FAQ_DEV.

17 years agoAdd book to FAQ_DEV:
Bruce Momjian [Thu, 23 Aug 2007 00:09:34 +0000 (00:09 +0000)]
Add book to FAQ_DEV:

Transactional Information Systems by Gerhard Weikum, Kaufmann

17 years agoFix possible core dump from pgbench -d option. Julius Stroffek
Tom Lane [Wed, 22 Aug 2007 23:03:27 +0000 (23:03 +0000)]
Fix possible core dump from pgbench -d option.  Julius Stroffek

17 years agoSuppress testing the options of CREATE TEXT SEARCH DICTIONARY during
Tom Lane [Wed, 22 Aug 2007 22:30:20 +0000 (22:30 +0000)]
Suppress testing the options of CREATE TEXT SEARCH DICTIONARY during
initdb.  We should create all the standard dictionaries even though
some of them may not work in template1's encoding.  Per Teodor.

17 years agoMore cleaning up and removed some duplicates.
Michael Meskes [Wed, 22 Aug 2007 08:20:58 +0000 (08:20 +0000)]
More cleaning up and removed some duplicates.

17 years agoPropagate Magnus' fix for Windows snowball build into Makefile.cygwin.
Tom Lane [Wed, 22 Aug 2007 06:34:18 +0000 (06:34 +0000)]
Propagate Magnus' fix for Windows snowball build into Makefile.cygwin.

17 years agoFix cut-n-paste-o.
Tom Lane [Wed, 22 Aug 2007 06:26:32 +0000 (06:26 +0000)]
Fix cut-n-paste-o.

17 years agoFix VPATH-build problem in new tsearch makefile, per Chad Wagner.
Tom Lane [Wed, 22 Aug 2007 06:11:56 +0000 (06:11 +0000)]
Fix VPATH-build problem in new tsearch makefile, per Chad Wagner.

17 years agoRemove option to change parser of an existing text search configuration.
Tom Lane [Wed, 22 Aug 2007 05:13:50 +0000 (05:13 +0000)]
Remove option to change parser of an existing text search configuration.
This prevents needing to do complex and poorly-defined updates of the
mapping table if the new parser has different token types than the old.
Per discussion.

17 years agoText search doc updates --- first cut at
Tom Lane [Wed, 22 Aug 2007 04:45:20 +0000 (04:45 +0000)]
Text search doc updates --- first cut at
syncing the existing docs with the final syntax decisions.

17 years agoWhoops, missed updating dsynonym_init for new dictionary parameter method.
Tom Lane [Wed, 22 Aug 2007 04:13:15 +0000 (04:13 +0000)]
Whoops, missed updating dsynonym_init for new dictionary parameter method.

17 years agoMake \dFt and \dFt+ do something different --- omitting the function
Tom Lane [Wed, 22 Aug 2007 02:25:34 +0000 (02:25 +0000)]
Make \dFt and \dFt+ do something different --- omitting the function
names in the former case seems more consistent with the behavior of
other \dF commands.

17 years agoSimplify the syntax of CREATE/ALTER TEXT SEARCH DICTIONARY by treating the
Tom Lane [Wed, 22 Aug 2007 01:39:46 +0000 (01:39 +0000)]
Simplify the syntax of CREATE/ALTER TEXT SEARCH DICTIONARY by treating the
init options of the template as top-level options in the syntax.  This also
makes ALTER a bit easier to use, since options can be replaced individually.
I also made these statements verify that the tmplinit method will accept
the new settings before they get stored; in the original coding you didn't
find out about mistakes until the dictionary got invoked.

Under the hood, init methods now get options as a List of DefElem instead
of a raw text string --- that lets tsearch use existing options-pushing code
instead of duplicating functionality.

17 years agoSimplify CREATE TEXT SEARCH CONFIGURATION by eliminating the separate
Tom Lane [Tue, 21 Aug 2007 21:24:00 +0000 (21:24 +0000)]
Simplify CREATE TEXT SEARCH CONFIGURATION by eliminating the separate
'with map' parameter; as things now stand there's really not much point
in specifying a config-to-copy if you don't copy its map.  Also, use
COPY instead of TEMPLATE as the key word for a config-to-copy, so as
to avoid confusion with text search templates.  Per discussion; the
just-committed reference page for the command already describes it
this way.

17 years agoFirst rough cut at text search documentation: bare bones reference
Tom Lane [Tue, 21 Aug 2007 21:08:47 +0000 (21:08 +0000)]
First rough cut at text search documentation: bare bones reference
pages for the new SQL commands.  I also committed Bruce's text search
introductory chapter, as-is except for fixing some markup errors,
so that there would be a place for the reference pages to link to.

17 years agoExclude snowball/libstemmer/ files from the set processed by pgindent.
Tom Lane [Tue, 21 Aug 2007 16:08:23 +0000 (16:08 +0000)]
Exclude snowball/libstemmer/ files from the set processed by pgindent.
There's not much point in prettifying machine-generated code, and it
seems best to keep these files exactly like upstream anyway.  Also add
some notes about why various files are excluded.

17 years agoUppercase keywords in regression tsearch test scripts.
Bruce Momjian [Tue, 21 Aug 2007 15:41:13 +0000 (15:41 +0000)]
Uppercase keywords in regression tsearch test scripts.

17 years agoAvoid using TEXT as a Bison symbol, since this provokes warnings on
Tom Lane [Tue, 21 Aug 2007 15:13:42 +0000 (15:13 +0000)]
Avoid using TEXT as a Bison symbol, since this provokes warnings on
Windows builds.  In passing, fix an obsolete comment, per gripe from
Greg Stark.

17 years agoRemove tabs from SGML files.
Bruce Momjian [Tue, 21 Aug 2007 15:13:16 +0000 (15:13 +0000)]
Remove tabs from SGML files.

17 years agoBuild snowball DLL for tsearch-in-core.
Magnus Hagander [Tue, 21 Aug 2007 15:10:41 +0000 (15:10 +0000)]
Build snowball DLL for tsearch-in-core.
(Still needs to build the .sql output files, but this handles the C part
of the build)

17 years agoDon't define BUILDING_DLL for snowball lib. Should fix build problems
Magnus Hagander [Tue, 21 Aug 2007 13:32:33 +0000 (13:32 +0000)]
Don't define BUILDING_DLL for snowball lib. Should fix build problems
on mingw and probably cygwin.

17 years agoRemove extraneous semicolon --- buildfarm member bear, for one,
Tom Lane [Tue, 21 Aug 2007 06:34:42 +0000 (06:34 +0000)]
Remove extraneous semicolon --- buildfarm member bear, for one,
objects to it.

17 years agoFix cash_mul_int4 and cash_div_int4 for overenthusiastic substitution
Tom Lane [Tue, 21 Aug 2007 03:56:07 +0000 (03:56 +0000)]
Fix cash_mul_int4 and cash_div_int4 for overenthusiastic substitution
of int64 for int32.  Per reports from Merlin Moncure and Andrew Chernow.

17 years agoFix money type's send/receive functions to conform to recent widening
Tom Lane [Tue, 21 Aug 2007 03:14:36 +0000 (03:14 +0000)]
Fix money type's send/receive functions to conform to recent widening
of the datatype to int64.  Per Andrew Chernow.

17 years agoFix potential access-off-the-end-of-memory in varbit_out(): it fetched the
Tom Lane [Tue, 21 Aug 2007 02:40:06 +0000 (02:40 +0000)]
Fix potential access-off-the-end-of-memory in varbit_out(): it fetched the
byte after the last full byte of the bit array, regardless of whether that
byte was part of the valid data or not.  Found by buildfarm testing.
Thanks to Stefan Kaltenbrunner for nailing down the cause.

17 years agoSuppress uninitialized-variable warning.
Tom Lane [Tue, 21 Aug 2007 01:47:19 +0000 (01:47 +0000)]
Suppress uninitialized-variable warning.

17 years agoFix a small 64-bit problem in tsearch patch.
Tom Lane [Tue, 21 Aug 2007 01:45:33 +0000 (01:45 +0000)]
Fix a small 64-bit problem in tsearch patch.

17 years agocontrib/tsearch2 now fails its regression test because of conflicts with
Tom Lane [Tue, 21 Aug 2007 01:12:28 +0000 (01:12 +0000)]
contrib/tsearch2 now fails its regression test because of conflicts with
the core additions.  For the moment I diked it out of contrib/Makefile.
We should look at turning it into a backward-compatibility package.

17 years agoTsearch2 functionality migrates to core. The bulk of this work is by
Tom Lane [Tue, 21 Aug 2007 01:11:32 +0000 (01:11 +0000)]
Tsearch2 functionality migrates to core.  The bulk of this work is by
Oleg Bartunov and Teodor Sigaev, but I did a lot of editorializing,
so anything that's broken is probably my fault.

Documentation is nonexistent as yet, but let's land the patch so we can
get some portability testing done.

17 years agoAdd configure option --with-system-tzdata to use operating system time zone
Peter Eisentraut [Mon, 20 Aug 2007 08:53:12 +0000 (08:53 +0000)]
Add configure option --with-system-tzdata to use operating system time zone
database.

17 years agominor typo fixes in CSVlog docs, per Michael Glaesemann
Andrew Dunstan [Sun, 19 Aug 2007 03:23:30 +0000 (03:23 +0000)]
minor typo fixes in CSVlog docs, per Michael Glaesemann

17 years agoDone:
Bruce Momjian [Sun, 19 Aug 2007 02:36:08 +0000 (02:36 +0000)]
Done:

< * Allow server log information to be output as INSERT statements
> * -Allow server log information to be output as CSV format

17 years agoProvide for logfiles in machine readable CSV format. In consequence, rename
Andrew Dunstan [Sun, 19 Aug 2007 01:41:25 +0000 (01:41 +0000)]
Provide for logfiles in machine readable CSV format. In consequence, rename
redirect_stderr to logging_collector.
Original patch from Arul Shaji, subsequently modified by Greg Smith, and then
heavily modified by me.

17 years agoAdd:
Bruce Momjian [Thu, 16 Aug 2007 19:02:15 +0000 (19:02 +0000)]
Add:

>
> * Reduce XID consumption of read-only queries
>
>    http://archives.postgresql.org/pgsql-hackers/2007-08/msg00516.php
>

17 years agoArrange to cache a ResultRelInfo in the executor's EState for relations that
Tom Lane [Wed, 15 Aug 2007 21:39:50 +0000 (21:39 +0000)]
Arrange to cache a ResultRelInfo in the executor's EState for relations that
are not one of the query's defined result relations, but nonetheless have
triggers fired against them while the query is active.  This was formerly
impossible but can now occur because of my recent patch to fix the firing
order for RI triggers.  Caching a ResultRelInfo avoids duplicating work by
repeatedly opening and closing the same relation, and also allows EXPLAIN
ANALYZE to "see" and report on these extra triggers.  Use the same mechanism
to cache open relations when firing deferred triggers at transaction shutdown;
this replaces the former one-element-cache strategy used in that case, and
should improve performance a bit when there are deferred triggers on a number
of relations.

17 years agoRepair problems occurring when multiple RI updates have to be done to the same
Tom Lane [Wed, 15 Aug 2007 19:15:47 +0000 (19:15 +0000)]
Repair problems occurring when multiple RI updates have to be done to the same
row within one query: we were firing check triggers before all the updates
were done, leading to bogus failures.  Fix by making the triggers queued by
an RI update go at the end of the outer query's trigger event list, thereby
effectively making the processing "breadth-first".  This was indeed how it
worked pre-8.0, so the bug does not occur in the 7.x branches.
Per report from Pavel Stehule.

17 years agoAdd third idea about pulling data from indexes.
Bruce Momjian [Wed, 15 Aug 2007 15:53:30 +0000 (15:53 +0000)]
Add third idea about pulling data from indexes.

>   A third idea would be for a heap scan to check if all rows are visible
>   and if so set a per-table flag which can be checked by index scans.
>   Any change to the table would have to clear the flag.  To detect
>   changes during the heap scan a counter could be set at the start and
>   checked at the end --- if it is the same, the table has not been
>   modified --- any table change would increment the counter.

17 years agoFix whitespace in TODO.
Bruce Momjian [Wed, 15 Aug 2007 15:36:31 +0000 (15:36 +0000)]
Fix whitespace in TODO.

17 years agoFix oversight in async-commit patch: there were some places in heapam.c
Tom Lane [Tue, 14 Aug 2007 17:35:18 +0000 (17:35 +0000)]
Fix oversight in async-commit patch: there were some places in heapam.c
that still thought they could set HEAP_XMAX_COMMITTED immediately after
seeing the other transaction commit.  Make them use the same logic as
tqual.c does to determine if the hint bit can be set yet.

17 years agoDone for ecpg:
Bruce Momjian [Tue, 14 Aug 2007 17:02:42 +0000 (17:02 +0000)]
Done for ecpg:

<  o Use backend PREPARE/EXECUTE facility for ecpg where possible

17 years agoIt seems we simply expected the same output for failure and success.
Michael Meskes [Tue, 14 Aug 2007 11:21:38 +0000 (11:21 +0000)]
It seems we simply expected the same output for failure and success.

17 years agoIt seems I always forget .in files.
Michael Meskes [Tue, 14 Aug 2007 10:56:20 +0000 (10:56 +0000)]
It seems I always forget .in files.

17 years agoHopefully fixed missing CHAR_BIT symbol
Michael Meskes [Tue, 14 Aug 2007 10:54:57 +0000 (10:54 +0000)]
Hopefully fixed missing CHAR_BIT symbol

17 years agoFixed compiler warning for enum handling
Michael Meskes [Tue, 14 Aug 2007 10:32:47 +0000 (10:32 +0000)]
Fixed compiler warning for enum handling

17 years agoAs usual I forgot some files. :-)
Michael Meskes [Tue, 14 Aug 2007 10:19:14 +0000 (10:19 +0000)]
As usual I forgot some files. :-)

17 years ago- Finished major rewrite to use new protocol version
Michael Meskes [Tue, 14 Aug 2007 10:01:54 +0000 (10:01 +0000)]
- Finished major rewrite to use new protocol version
- Really prepare statements
- Added more regression tests
- Added auto-prepare mode
- Use '$n' for positional variables, '?' is still possible via ecpg option
- Cleaned up the sources a little bit

17 years agoTEMPORARILY make synchronous_commit default to OFF, so that we can get more
Tom Lane [Mon, 13 Aug 2007 19:27:12 +0000 (19:27 +0000)]
TEMPORARILY make synchronous_commit default to OFF, so that we can get more
thorough testing of async-commit mode from the buildfarm.  This patch MUST
get reverted before 8.3 release!

17 years agoFix two bugs induced in VACUUM FULL by async-commit patch.
Tom Lane [Mon, 13 Aug 2007 19:08:26 +0000 (19:08 +0000)]
Fix two bugs induced in VACUUM FULL by async-commit patch.

First, we cannot assume that XLogAsyncCommitFlush guarantees hint bits will be
settable, because clog.c's inexact LSN bookkeeping results in windows where a
previously flushed transaction is considered unhintable because it shares an
LSN slot with a later unflushed transaction.  But repair_frag requires
XMIN_COMMITTED to be correct so that it can distinguish tuples moved by the
current vacuum.  Since not being able to set the bit is an uncommon corner
case, the most practical way of dealing with it seems to be to abandon
shrinking (ie, don't invoke repair_frag) when we find a non-dead tuple whose
XMIN_COMMITTED bit couldn't be set.

Second, it is possible for the same reason that a RECENTLY_DEAD tuple does not
get its XMAX_COMMITTED bit set during scan_heap.  But by the time repair_frag
examines the tuple it might be possible to set the bit.  We therefore must
take buffer content lock when calling HeapTupleSatisfiesVacuum a second time,
else we can get an Assert failure in SetBufferCommitInfoNeedsSave.  This
latter bug is latent in existing releases, but I think it cannot actually
occur without async commit, since the first HeapTupleSatisfiesVacuum call
should always have set the bit.  So I'm not going to back-patch it.

In passing, reduce the existing "cannot shrink relation" messages from NOTICE
to LOG level.  The new message must be no higher than LOG if we don't want
unpredictable regression test failures, and consistency seems like a good
idea.  Also arrange that only one such message is reported per VACUUM FULL;
in typical scenarios you could get spammed with many such messages, which
seems a bit useless.

17 years agoDocument that the regexp split functions ignore zero-length matches in
Tom Lane [Mon, 13 Aug 2007 01:18:47 +0000 (01:18 +0000)]
Document that the regexp split functions ignore zero-length matches in
certain corner cases.  Per discussion, the code does what we want, but
it really needs to be documented that these functions act differently
from regexp_matches.

17 years agoRemove an "optimization" I installed in 2001, to make repalloc() attempt to
Tom Lane [Sun, 12 Aug 2007 20:39:14 +0000 (20:39 +0000)]
Remove an "optimization" I installed in 2001, to make repalloc() attempt to
enlarge the memory chunk in-place when it was feasible to do so.  This turns
out to not work well at all for scenarios involving repeated cycles of
palloc/repalloc/pfree: the eventually freed chunks go into the wrong freelist
for the next initial palloc request, and so we consume memory indefinitely.
While that could be defended against, the number of cases where the
optimization can still be applied drops significantly, and adjusting the
initial sizes of StringInfo buffers makes it drop to almost nothing.
Seems better to just remove the extra complexity.
Per recent discussion and testing.

17 years agoIncrease the initial size of StringInfo buffers to 1024 bytes (from 256);
Tom Lane [Sun, 12 Aug 2007 20:18:06 +0000 (20:18 +0000)]
Increase the initial size of StringInfo buffers to 1024 bytes (from 256);
likewise increase the initial size of the scanner's literal buffer to 1024
(from 128).  Instrumentation of the regression tests suggests that this
saves a useful amount of repalloc() traffic --- the number of calls occurring
during one set of tests drops from about 6900 to about 3900.  The old sizes
were chosen in the late 90's with an eye to machines much smaller than
are common today.

17 years agoAvoid memory leakage across successive calls of regexp_matches() or
Tom Lane [Sat, 11 Aug 2007 19:16:41 +0000 (19:16 +0000)]
Avoid memory leakage across successive calls of regexp_matches() or
regexp_split_to_table() within a single query.  This is only a partial
solution, as it turns out that with enough matches per string these
functions can also tickle a repalloc() misbehavior.  But fixing that
is a topic for a separate patch.

17 years agoCode review for regexp_matches/regexp_split patch. Refactor to avoid assuming
Tom Lane [Sat, 11 Aug 2007 03:56:24 +0000 (03:56 +0000)]
Code review for regexp_matches/regexp_split patch.  Refactor to avoid assuming
that cached compiled patterns will still be there when the function is next
called.  Clean up looping logic, thereby fixing bug identified by Pavel
Stehule.  Share setup code between the two functions, add some comments, and
avoid risky mixing of int and size_t variables.  Clean up the documentation a
tad, and accept all the flag characters mentioned in table 9-19 rather than
just a subset.

17 years agoFix uninitialized-memory bug in plpython proargnames patch. Per bug #3523
Tom Lane [Fri, 10 Aug 2007 03:16:04 +0000 (03:16 +0000)]
Fix uninitialized-memory bug in plpython proargnames patch.  Per bug #3523

17 years agoFix unintended change of output format for createlang/droplang -l. Missed
Tom Lane [Fri, 10 Aug 2007 00:39:31 +0000 (00:39 +0000)]
Fix unintended change of output format for createlang/droplang -l.  Missed
these uses of printQuery() in FETCH_COUNT patch a year ago :-(.  Per report
from Tomoaki Sato.

17 years agoFix search for SGML stylesheets to include the place where Gentoo keeps them.
Tom Lane [Thu, 9 Aug 2007 02:33:58 +0000 (02:33 +0000)]
Fix search for SGML stylesheets to include the place where Gentoo keeps them.
Brendan Jurd

17 years agoRevise postmaster startup/shutdown logic to eliminate the problem that a
Tom Lane [Thu, 9 Aug 2007 01:18:43 +0000 (01:18 +0000)]
Revise postmaster startup/shutdown logic to eliminate the problem that a
constant flow of new connection requests could prevent the postmaster from
completing a shutdown or crash restart.  This is done by labeling child
processes that are "dead ends", that is, we know that they were launched only
to tell a client that it can't connect.  These processes are managed
separately so that they don't confuse us into thinking that we can't advance
to the next stage of a shutdown or restart sequence, until the very end
where we must wait for them to drain out so we can delete the shmem segment.
Per discussion of a misbehavior reported by Keaton Adams.

Since this code was baroque already, and my first attempt at fixing the
problem made it entirely impenetrable, I took the opportunity to rewrite it
in a state-machine style.  That eliminates some duplicated code sections and
hopefully makes everything a bit clearer.

17 years agoFix a gradual memory leak in ExecReScanAgg(). Because the aggregation
Neil Conway [Wed, 8 Aug 2007 18:07:05 +0000 (18:07 +0000)]
Fix a gradual memory leak in ExecReScanAgg(). Because the aggregation
hash table is allocated in a child context of the agg node's memory
context, MemoryContextReset() will reset but *not* delete the child
context. Since ExecReScanAgg() proceeds to build a new hash table
from scratch (in a new sub-context), this results in leaking the
header for the previous memory context. Therefore, use
MemoryContextResetAndDeleteChildren() instead.

Credit: My colleague Sailesh Krishnamurthy at Truviso for isolating
the cause of the leak.

17 years agoFix thinko in multi-autovac-workers code: validity checks made by
Tom Lane [Wed, 8 Aug 2007 16:00:46 +0000 (16:00 +0000)]
Fix thinko in multi-autovac-workers code: validity checks made by
GUC assign hooks are supposed to be made whether doit is true or not.