OSDN Git Service

pg-rex/syncrep.git
14 years agoBuild perlchunks.h for plperl on MSVC
Andrew Dunstan [Sat, 9 Jan 2010 14:45:47 +0000 (14:45 +0000)]
Build perlchunks.h for plperl on MSVC

14 years agoFix makefile so it works for VPATH case.
Tom Lane [Sat, 9 Jan 2010 03:53:40 +0000 (03:53 +0000)]
Fix makefile so it works for VPATH case.

14 years agoTidy up and refactor plperl.c.
Andrew Dunstan [Sat, 9 Jan 2010 02:40:50 +0000 (02:40 +0000)]
Tidy up and refactor plperl.c.

- Changed MULTIPLICITY check from runtime to compiletime.
    No loads the large Config module.
- Changed plperl_init_interp() to return new interp
    and not alter the global interp_state
- Moved plperl_safe_init() call into check_interp().
- Removed plperl_safe_init_done state variable
    as interp_state now covers that role.
- Changed plperl_create_sub() to take a plperl_proc_desc argument.
- Simplified return value handling in plperl_create_sub.
- Changed perl.com link in the docs to perl.org and tweaked
    wording to clarify that require, not use, is what's blocked.
- Moved perl code in large multi-line C string literal macros
    out to plc_*.pl files.
- Added a test2macro.pl utility to convert the plc_*.pl files to
    macros in a perlchunks.h file which is #included
- Simplifed plperl_safe_init() slightly
- Optimized pg_verifymbstr calls to avoid unneeded strlen()s.

Patch from Tim Bunce, with minor editing from me.

14 years agoAlso update ChangerLog file.
Michael Meskes [Fri, 8 Jan 2010 09:06:05 +0000 (09:06 +0000)]
Also update ChangerLog file.

14 years agoFix oversight in EvalPlanQualFetch: after failing to lock a tuple because
Tom Lane [Fri, 8 Jan 2010 02:44:00 +0000 (02:44 +0000)]
Fix oversight in EvalPlanQualFetch: after failing to lock a tuple because
someone else has just updated it, we have to set priorXmax to that tuple's
xmax (ie, the XID of the other xact that updated it) before looping back to
examine the next tuple.  Obviously, the next tuple in the update chain should
have that XID as its xmin, not the same xmin as the preceding tuple that we
had been trying to lock.  The mismatch would cause the EvalPlanQual logic to
decide that the tuple chain ended in a deletion, when actually there was a
live tuple that should have been found.

I inserted this error when recently adding logic to EvalPlanQual to make it
lock tuples before returning them (as opposed to the old method in which the
lock would occur much later, causing a great deal of work to be wasted if we
only then discover someone else updated it).  Sigh.  Per today's report from
Takahiro Itagaki of inconsistent results during pgbench runs.

14 years agopgBufferUsage needs PGDLLIMPORT for pg_stat_statements on Windows.
Itagaki Takahiro [Fri, 8 Jan 2010 00:48:56 +0000 (00:48 +0000)]
pgBufferUsage needs PGDLLIMPORT for pg_stat_statements on Windows.

14 years agoAdd buffer access counters to pg_stat_statements.
Itagaki Takahiro [Fri, 8 Jan 2010 00:38:20 +0000 (00:38 +0000)]
Add buffer access counters to pg_stat_statements.

This uses the same infrastructure with EXPLAIN BUFFERS to support
{shared|local}_blks_{hit|read|written} andtemp_blks_{read|written}
columns in the pg_stat_statements view. The dumped file format
also updated.

Thanks to Robert Haas for the review.

14 years agoDocument why we copy reloptions into CacheMemoryContext after-the-fact.
Robert Haas [Thu, 7 Jan 2010 20:39:45 +0000 (20:39 +0000)]
Document why we copy reloptions into CacheMemoryContext after-the-fact.

14 years agoFix 3-parameter form of bit substring() to throw error for negative length,
Tom Lane [Thu, 7 Jan 2010 20:17:44 +0000 (20:17 +0000)]
Fix 3-parameter form of bit substring() to throw error for negative length,
as required by SQL standard.

14 years agoMake bit/varbit substring() treat any negative length as meaning "all the rest
Tom Lane [Thu, 7 Jan 2010 19:53:11 +0000 (19:53 +0000)]
Make bit/varbit substring() treat any negative length as meaning "all the rest
of the string".  The previous coding treated only -1 that way, and would
produce an invalid result value for other negative values.

We ought to fix it so that 2-parameter bit substring() is a different C
function and the 3-parameter form throws error for negative length, but
that takes a pg_proc change which is impractical in the back branches;
and in any case somebody might be relying on -1 working this way.
So just do this as a back-patchable fix.

14 years agoFix (some of the) breakage introduced into query-cancel processing by HS.
Tom Lane [Thu, 7 Jan 2010 16:29:58 +0000 (16:29 +0000)]
Fix (some of the) breakage introduced into query-cancel processing by HS.

It is absolutely not okay to throw an ereport(ERROR) in any random place in
the code just because DoingCommandRead is set; interrupting, say, OpenSSL
in the midst of its activities is guaranteed to result in heartache.

Instead of that, undo the original optimizations that threw away
QueryCancelPending anytime we were starting or finishing a command read, and
instead discard the cancel request within ProcessInterrupts if we find that
there is no HS reason for forcing a cancel and we are DoingCommandRead.

In passing, may I once again condemn the practice of changing the code
and not fixing the adjacent comment that you just turned into a lie?

14 years agoUse -Z for vacuumdb --analyze-only, rather than -o.
Bruce Momjian [Thu, 7 Jan 2010 14:35:44 +0000 (14:35 +0000)]
Use -Z for vacuumdb --analyze-only, rather than -o.

14 years agoRename new vacuumdb option to --analyze-only from --only-analyze.
Bruce Momjian [Thu, 7 Jan 2010 12:38:55 +0000 (12:38 +0000)]
Rename new vacuumdb option to --analyze-only from --only-analyze.

14 years agoRemove all the special-case code for INT64_IS_BUSTED, per decision that
Tom Lane [Thu, 7 Jan 2010 04:53:35 +0000 (04:53 +0000)]
Remove all the special-case code for INT64_IS_BUSTED, per decision that
we're not going to support that anymore.

I did keep the 64-bit-CRC-with-32-bit-arithmetic code, since it has a
performance excuse to live.  It's a bit moot since that's all ifdef'd
out, of course.

14 years agoMore tablespace.c comment improvements.
Bruce Momjian [Thu, 7 Jan 2010 04:10:39 +0000 (04:10 +0000)]
More tablespace.c comment improvements.

14 years agoClarify tablespace.c::TablespaceCreateDbspace() comments.
Bruce Momjian [Thu, 7 Jan 2010 04:05:39 +0000 (04:05 +0000)]
Clarify tablespace.c::TablespaceCreateDbspace() comments.

14 years agoFurther fixes for per-tablespace options patch.
Robert Haas [Thu, 7 Jan 2010 03:53:08 +0000 (03:53 +0000)]
Further fixes for per-tablespace options patch.

Add missing varlena header to TableSpaceOpts structure.  And, per
Tom Lane, instead of calling tablespace_reloptions in CacheMemoryContext,
call it in the caller's memory context and copy the value over
afterwards, to reduce the chances of a session-lifetime memory leak.

14 years agoFail in a nicer way if we have --with-perl and no Perl.
Tom Lane [Thu, 7 Jan 2010 03:24:57 +0000 (03:24 +0000)]
Fail in a nicer way if we have --with-perl and no Perl.

14 years agoImprove a couple of comments relating to large object snapshot management.
Robert Haas [Thu, 7 Jan 2010 02:41:16 +0000 (02:41 +0000)]
Improve a couple of comments relating to large object snapshot management.

14 years agoMake configure check the version of Perl we're building with, and reject
Tom Lane [Thu, 7 Jan 2010 01:41:11 +0000 (01:41 +0000)]
Make configure check the version of Perl we're building with, and reject
versions < 5.8.  Also, if there's no Perl, emit a warning informing the
user that he won't be able to build from a CVS pull.  This is exactly the
same treatment we give Bison and Perl, and for the same reasons.

14 years agoAlter the configure script to fail immediately if the C compiler does not
Tom Lane [Thu, 7 Jan 2010 00:25:05 +0000 (00:25 +0000)]
Alter the configure script to fail immediately if the C compiler does not
provide a working 64-bit integer datatype.  As recently noted, we've been
broken on such platforms since early in the 8.4 development cycle.  Since
it took nearly two years for anyone to even notice, it seems that the
rationale for continuing to support such platforms has reached the point
of non-existence.  Rather than thrashing around to try to make it work
again, we'll just admit up front that this no longer works.

Back-patch to 8.4 since that branch is also broken.

We should go around to remove INT64_IS_BUSTED support, but just in HEAD,
so that seems like material for a separate commit.

14 years agoPG_MAJORVERSION:
Bruce Momjian [Wed, 6 Jan 2010 23:23:51 +0000 (23:23 +0000)]
PG_MAJORVERSION:

For simplicity, use PG_MAJORVERSION rather than PG_VERSION for creation
of the PG_VERSION file.

14 years agoFix spccache.c to not suppose that a cache entry will live across database
Tom Lane [Wed, 6 Jan 2010 23:00:02 +0000 (23:00 +0000)]
Fix spccache.c to not suppose that a cache entry will live across database
access, per testing with CLOBBER_CACHE_ALWAYS.  Minor other editorialization.

14 years agoAccess hash entry before freeing it, not after.
Tom Lane [Wed, 6 Jan 2010 22:27:09 +0000 (22:27 +0000)]
Access hash entry before freeing it, not after.

14 years agoMake error messages for bad --set-version argument more useful.
Tom Lane [Wed, 6 Jan 2010 22:02:45 +0000 (22:02 +0000)]
Make error messages for bad --set-version argument more useful.
Per Stefan.

14 years agoMake the makefile pass $MAJORVERSION to genbki.pl, not $VERSION which is
Tom Lane [Wed, 6 Jan 2010 19:56:29 +0000 (19:56 +0000)]
Make the makefile pass $MAJORVERSION to genbki.pl, not $VERSION which is
overridden in the snapshot build script.  $MAJORVERSION is what it really
wanted anyway, so we can tighten up the parsing of --set-version's argument.

14 years agoFix missing <manvolnum> markers in dblink reference entries.
Tom Lane [Wed, 6 Jan 2010 19:07:05 +0000 (19:07 +0000)]
Fix missing <manvolnum> markers in dblink reference entries.
Silences complaints from "make man", and allows these man pages
to be included in man3/.

14 years agoAdd missing 'static' keywords.
Tom Lane [Wed, 6 Jan 2010 18:07:19 +0000 (18:07 +0000)]
Add missing 'static' keywords.

14 years agoCorrected CVS entry:
Bruce Momjian [Wed, 6 Jan 2010 16:04:05 +0000 (16:04 +0000)]
Corrected CVS entry:

--only-analyze mode is for _vacuumdb_, not pg_dump.

14 years agoReplaced int64_t with int64 as Andrew suggested.
Michael Meskes [Wed, 6 Jan 2010 15:15:55 +0000 (15:15 +0000)]
Replaced int64_t with int64 as Andrew suggested.

14 years agoRemove __FUNCTION__ keyword that is not recognized by som compilers.
Michael Meskes [Wed, 6 Jan 2010 15:10:21 +0000 (15:10 +0000)]
Remove __FUNCTION__ keyword that is not recognized by som compilers.

14 years agoRemoved test case using nan as float value because printf's output for nan is
Michael Meskes [Wed, 6 Jan 2010 11:59:52 +0000 (11:59 +0000)]
Removed test case using nan as float value because printf's output for nan is
OS specific with some distinguishing between signaling and quiet nans. It's not
really importnat for us here anyway.

14 years agoRemoved more inttypes.h stuff.
Michael Meskes [Wed, 6 Jan 2010 11:49:43 +0000 (11:49 +0000)]
Removed more inttypes.h stuff.

14 years agoSilence compiler warning about uninitialized variables. This initialization
Itagaki Takahiro [Wed, 6 Jan 2010 11:25:39 +0000 (11:25 +0000)]
Silence compiler warning about uninitialized variables. This initialization
is not necessary needed, but some compilers complain about it.

14 years agoApplied Zoltan's patch to remove hardware dependant offset logging and
Michael Meskes [Wed, 6 Jan 2010 09:44:05 +0000 (09:44 +0000)]
Applied Zoltan's patch to remove hardware dependant offset logging and
superfluous include files.

14 years agoSupport rewritten-based full vacuum as VACUUM FULL. Traditional
Itagaki Takahiro [Wed, 6 Jan 2010 05:31:14 +0000 (05:31 +0000)]
Support rewritten-based full vacuum as VACUUM FULL. Traditional
VACUUM FULL was renamed to VACUUM FULL INPLACE. Also added a new
option -i, --inplace for vacuumdb to perform FULL INPLACE vacuuming.

Since the new VACUUM FULL uses CLUSTER infrastructure, we cannot
use it for system tables. VACUUM FULL for system tables always
fall back into VACUUM FULL INPLACE silently.

Itagaki Takahiro, reviewed by Jeff Davis and Simon Riggs.

14 years agobinary upgrade:
Bruce Momjian [Wed, 6 Jan 2010 05:18:18 +0000 (05:18 +0000)]
binary upgrade:

Preserve relfilenodes for views and composite types --- even though we
don't store data in, them, they do consume relfilenodes.

Bump catalog version.

14 years agoRemove erroneous comma added to pg_dumpall tablespace query when running
Bruce Momjian [Wed, 6 Jan 2010 03:34:41 +0000 (03:34 +0000)]
Remove erroneous comma added to pg_dumpall tablespace query when running
with PG version 8.2-8.4.

14 years agoUpdate catalog version for recent relfilenode patch, so pg_migrator can
Bruce Momjian [Wed, 6 Jan 2010 03:07:24 +0000 (03:07 +0000)]
Update catalog version for recent relfilenode patch, so pg_migrator can
identify the new API.

14 years agoPreserve relfilenodes:
Bruce Momjian [Wed, 6 Jan 2010 03:04:03 +0000 (03:04 +0000)]
Preserve relfilenodes:

Add support to pg_dump --binary-upgrade to preserve all relfilenodes,
for use by pg_migrator.

14 years agopg_dump --only-analyze
Bruce Momjian [Wed, 6 Jan 2010 02:59:46 +0000 (02:59 +0000)]
pg_dump --only-analyze

Implement pg_dump --only-analyze for use by pg_migrator to only analyze
all databases.

14 years agoRemove tabs in SGML.
Bruce Momjian [Wed, 6 Jan 2010 02:41:37 +0000 (02:41 +0000)]
Remove tabs in SGML.
Move OIDCHARS to proper include file.

14 years agoUse OIDCHARS:
Bruce Momjian [Wed, 6 Jan 2010 01:48:09 +0000 (01:48 +0000)]
Use OIDCHARS:

Use OIDCHARS for oid character length, rather than '10', in tablespace
code.

14 years agoAdjust headers to allow pgbench to be built with PGXS.
Itagaki Takahiro [Wed, 6 Jan 2010 01:30:03 +0000 (01:30 +0000)]
Adjust headers to allow pgbench to be built with PGXS.

14 years agoAdd verification of variable names in pgbench.
Itagaki Takahiro [Wed, 6 Jan 2010 01:12:14 +0000 (01:12 +0000)]
Add verification of variable names in pgbench.

Variables must consist of only alphabets, numerals and underscores.
We had allowed to set variables with invalid names, but could not
refer them in queries.

Thanks to Robert Haas for the review.

14 years agoAdd support for doing FULL JOIN ON FALSE. While this is really a rather
Tom Lane [Tue, 5 Jan 2010 23:25:36 +0000 (23:25 +0000)]
Add support for doing FULL JOIN ON FALSE.  While this is really a rather
peculiar variant of UNION ALL, and so wouldn't likely get written directly
as-is, it's possible for it to arise as a result of simplification of
less-obviously-silly queries.  In particular, now that we can do flattening
of subqueries that have constant outputs and are underneath an outer join,
it's possible for the case to result from simplification of queries of the
type exhibited in bug #5263.  Back-patch to 8.4 to avoid a functionality
regression for this type of query.

14 years agoSupport ALTER TABLESPACE name SET/RESET ( tablespace_options ).
Robert Haas [Tue, 5 Jan 2010 21:54:00 +0000 (21:54 +0000)]
Support ALTER TABLESPACE name SET/RESET ( tablespace_options ).

This patch only supports seq_page_cost and random_page_cost as parameters,
but it provides the infrastructure to scalably support many more.
In particular, we may want to add support for effective_io_concurrency,
but I'm leaving that as future work for now.

Thanks to Tom Lane for design help and Alvaro Herrera for the review.

14 years agoFix genbki.pl and Gen_fmgrtab.pl to use PID-specific temp file names,
Tom Lane [Tue, 5 Jan 2010 20:23:32 +0000 (20:23 +0000)]
Fix genbki.pl and Gen_fmgrtab.pl to use PID-specific temp file names,
so that it's safe if a parallel make chooses to run two concurrent copies.
Also, work around a memory leak in some versions of Perl.

14 years agoAnd another one
Michael Meskes [Tue, 5 Jan 2010 17:10:23 +0000 (17:10 +0000)]
And another one

14 years agoAnd another one of the same problem.
Michael Meskes [Tue, 5 Jan 2010 17:07:51 +0000 (17:07 +0000)]
And another one of the same problem.

14 years agoAnd then forgot to commit the fixed files. ARGH!
Michael Meskes [Tue, 5 Jan 2010 17:05:49 +0000 (17:05 +0000)]
And then forgot to commit the fixed files. ARGH!

14 years agoAh, should read the file completely. I got so used to git that I almost forgot about...
Michael Meskes [Tue, 5 Jan 2010 17:04:18 +0000 (17:04 +0000)]
Ah, should read the file completely. I got so used to git that I almost forgot about the cvs stuff.

14 years agoThrid try. It seems my two checkouts need some cleaning up.
Michael Meskes [Tue, 5 Jan 2010 16:58:37 +0000 (16:58 +0000)]
Thrid try. It seems my two checkouts need some cleaning up.

14 years agoStill not correct.
Michael Meskes [Tue, 5 Jan 2010 16:52:15 +0000 (16:52 +0000)]
Still not correct.

14 years agoAdjusted CVS headers.
Michael Meskes [Tue, 5 Jan 2010 16:49:46 +0000 (16:49 +0000)]
Adjusted CVS headers.

14 years agoApplied patch by Boszormenyi Zoltan <zb@cybertec.at> to add sqlda support to
Michael Meskes [Tue, 5 Jan 2010 16:38:23 +0000 (16:38 +0000)]
Applied patch by Boszormenyi Zoltan <zb@cybertec.at> to add sqlda support to
ecpg in both native and compatiblity mode.

14 years agoMove the default configuration for the MSVC build system to config_default.pl,
Magnus Hagander [Tue, 5 Jan 2010 13:31:58 +0000 (13:31 +0000)]
Move the default configuration for the MSVC build system to config_default.pl,
and allow using config.pl to override the defaults. config.pl is removed from
the repository, so changes there will no longer show up when doing diff, and
will not prevent switching branches and such things.

config.pl would normally be used to override single values, but if an
old-style config.pl is read, it will override the entire default configuration,
making it backwards compatible.

14 years agoDisable linker warning 4197 on Win64, that is caused by the fact that we export
Magnus Hagander [Tue, 5 Jan 2010 11:12:50 +0000 (11:12 +0000)]
Disable linker warning 4197 on Win64, that is caused by the fact that we export
symbols both using __declspec(dllexport) (via the PGDLLIMPORT macro) and using
full-dll-export. This works without warning on Win32, but not on Win64.

In passing, fix the fact that the framework could never deal with more than
one disbled linker warning - because MSVC wants commas between linker warnings,
and semicolons between compiler warnings...

14 years agoUse _mm_pause() for win64 spin_delay(), per note from Tsutomu Yamada.
Magnus Hagander [Tue, 5 Jan 2010 11:06:28 +0000 (11:06 +0000)]
Use _mm_pause() for win64 spin_delay(), per note from Tsutomu Yamada.

14 years agoFurther code review for genbki.pl. Improve comments, fix some
Tom Lane [Tue, 5 Jan 2010 06:41:44 +0000 (06:41 +0000)]
Further code review for genbki.pl.  Improve comments, fix some
rather random code choices, don't slavishly duplicate the original
pg_attribute.h's failure to put an OID into Schema_pg_index entries.

14 years agoFix a few places where we needed -I. in CPPFLAGS to work properly in
Tom Lane [Tue, 5 Jan 2010 03:56:52 +0000 (03:56 +0000)]
Fix a few places where we needed -I. in CPPFLAGS to work properly in
VPATH builds.  We had this already in several places, but not all.

14 years agoRemove too-smart-for-its-own-good optimization of not overwriting the output
Tom Lane [Tue, 5 Jan 2010 02:34:03 +0000 (02:34 +0000)]
Remove too-smart-for-its-own-good optimization of not overwriting the output
files when they haven't changed.  This confuses make because the build fails
to update the file timestamps, and so it keeps on doing the action over again.

14 years agoVerify input in pg_read_file().
Itagaki Takahiro [Tue, 5 Jan 2010 01:29:36 +0000 (01:29 +0000)]
Verify input in pg_read_file().

14 years agoFix parallel-make timing problem.
Tom Lane [Tue, 5 Jan 2010 01:20:35 +0000 (01:20 +0000)]
Fix parallel-make timing problem.

14 years agoGet rid of the need for manual maintenance of the initial contents of
Tom Lane [Tue, 5 Jan 2010 01:06:57 +0000 (01:06 +0000)]
Get rid of the need for manual maintenance of the initial contents of
pg_attribute, by having genbki.pl derive the information from the various
catalog header files.  This greatly simplifies modification of the
"bootstrapped" catalogs.

This patch finally kills genbki.sh and Gen_fmgrtab.sh; we now rely entirely on
Perl scripts for those build steps.  To avoid creating a Perl build dependency
where there was not one before, the output files generated by these scripts
are now treated as distprep targets, ie, they will be built and shipped in
tarballs.  But you will need a reasonably modern Perl (probably at least
5.6) if you want to build from a CVS pull.

The changes to the MSVC build process are untested, and may well break ---
we'll soon find out from the buildfarm.

John Naylor, based on ideas from Robert Haas and others

14 years agoRemove sometimes inaccurate error hint about source of wrongly encoded data.
Andrew Dunstan [Mon, 4 Jan 2010 20:38:31 +0000 (20:38 +0000)]
Remove sometimes inaccurate error hint about source of wrongly encoded data.

14 years agoCheck values passed back from PLPerl to the database, via function return,
Andrew Dunstan [Mon, 4 Jan 2010 20:29:59 +0000 (20:29 +0000)]
Check values passed back from PLPerl to the database, via function return,
trigger tuple modification or SPI call, to ensure they are valid in the
server encoding. Along the way, replace uses of SvPV(foo, PL_na)
with SvPV_nolen(foo) as recommended in the perl docs. Bug report from
Hannu Krosing.

14 years agoAdd a Win64-specific spin_delay() function.
Magnus Hagander [Mon, 4 Jan 2010 17:10:24 +0000 (17:10 +0000)]
Add a Win64-specific spin_delay() function.
We can't use the same as before, since MSVC on Win64 doesn't
support inline assembly.

14 years agoImprove PGXS makefile system to allow the module's makefile to specify
Tom Lane [Mon, 4 Jan 2010 16:34:11 +0000 (16:34 +0000)]
Improve PGXS makefile system to allow the module's makefile to specify
where to install DATA and DOCS files.  This is mainly intended to allow
versioned installation, eg, install into contrib/fooM.N/ rather than
directly into contrib/.

Mark Cave-Ayland

14 years agoWrite an end-of-backup WAL record at pg_stop_backup(), and wait for it at
Heikki Linnakangas [Mon, 4 Jan 2010 12:50:50 +0000 (12:50 +0000)]
Write an end-of-backup WAL record at pg_stop_backup(), and wait for it at
recovery instead of reading the backup history file. This is more robust,
as it stops you from prematurely starting up an inconsisten cluster if the
backup history file is lost for some reason, or if the base backup was
never finished with pg_stop_backup().

This also paves the way for a simpler streaming replication patch, which
doesn't need to care about backup history files anymore.

The backup history file is still created and archived as before, but it's
not used by the system anymore. It's just for informational purposes now.

Bump PG_CONTROL_VERSION as the location of the backup startpoint is now
written to a new field in pg_control, and catversion because initdb is
required

Original patch by Fujii Masao per Simon's idea, with further fixes by me.

14 years agoWhen estimating the selectivity of an inequality "column > constant" or
Tom Lane [Mon, 4 Jan 2010 02:44:40 +0000 (02:44 +0000)]
When estimating the selectivity of an inequality "column > constant" or
"column < constant", and the comparison value is in the first or last
histogram bin or outside the histogram entirely, try to fetch the actual
column min or max value using an index scan (if there is an index on the
column).  If successful, replace the lower or upper histogram bound with
that value before carrying on with the estimate.  This limits the
estimation error caused by moving min/max values when the comparison
value is close to the min or max.  Per a complaint from Josh Berkus.

It is tempting to consider using this mechanism for mergejoinscansel as well,
but that would inject index fetches into main-line join estimation not just
endpoint cases.  I'm refraining from that until we can get a better handle
on the costs of doing this type of lookup.

14 years agoSilence compiler warning about printf format for HANDLE.
Itagaki Takahiro [Mon, 4 Jan 2010 01:06:21 +0000 (01:06 +0000)]
Silence compiler warning about printf format for HANDLE.

14 years agoDept of second thoughts: my first cut at supporting "x IS NOT NULL" btree
Tom Lane [Sun, 3 Jan 2010 05:39:08 +0000 (05:39 +0000)]
Dept of second thoughts: my first cut at supporting "x IS NOT NULL" btree
indexscans would do the wrong thing if index_rescan() was called with a
NULL instead of a new set of scankeys and the index was DESC order,
because sk_strategy would not get flipped a second time.  I think
that those provisions for a NULL argument are dead code now as far as the
core backend goes, but possibly somebody somewhere is still using it.
In any case, this refactoring seems clearer, and it's definitely shorter.

14 years agoMake ssize_t 64-bit on Win64, for compatibility with for example plpython.
Magnus Hagander [Sat, 2 Jan 2010 22:47:37 +0000 (22:47 +0000)]
Make ssize_t 64-bit on Win64, for compatibility with for example plpython.

14 years agoAdd missing schema-qualification in tab completion query.
Tom Lane [Sat, 2 Jan 2010 21:28:46 +0000 (21:28 +0000)]
Add missing schema-qualification in tab completion query.

14 years agoFix similar_escape() to convert parentheses to non-capturing style.
Tom Lane [Sat, 2 Jan 2010 20:59:16 +0000 (20:59 +0000)]
Fix similar_escape() to convert parentheses to non-capturing style.
This is needed to avoid unwanted interference with SUBSTRING behavior,
as per bug #5257 from Roman Kononov.  Also, add some basic intelligence
about character classes (bracket expressions) since we now have several
behaviors that aren't appropriate inside a character class.

As with the previous patch in this area, I'm reluctant to back-patch
since it might affect applications that are relying on the prior
behavior.

14 years agocheck_exclusion_constraint didn't actually work correctly for index
Tom Lane [Sat, 2 Jan 2010 17:53:57 +0000 (17:53 +0000)]
check_exclusion_constraint didn't actually work correctly for index
expressions: FormIndexDatum requires the estate's scantuple to already point
at the tuple the values are supposedly being extracted from.  Adjust test
case so that this type of confusion will be exposed.
Per report from hubert depesz lubaczewski.

14 years agoUpdate copyright for the year 2010.
Bruce Momjian [Sat, 2 Jan 2010 16:58:17 +0000 (16:58 +0000)]
Update copyright for the year 2010.

14 years agoSilence compiler warning about size of size_t being larger than the result
Magnus Hagander [Sat, 2 Jan 2010 15:18:42 +0000 (15:18 +0000)]
Silence compiler warning about size of size_t being larger than the result
variable it's stored in. We know this can never happen. Per discussion.

14 years agoSet proper sizes for size_t and void* on 64-bit Windows builds.
Magnus Hagander [Sat, 2 Jan 2010 13:56:37 +0000 (13:56 +0000)]
Set proper sizes for size_t and void* on 64-bit Windows builds.

Tsutomu Yamada

14 years agoSupport 64-bit shared memory when building on 64-bit Windows.
Magnus Hagander [Sat, 2 Jan 2010 12:18:45 +0000 (12:18 +0000)]
Support 64-bit shared memory when building on 64-bit Windows.

Tsutomu Yamada

14 years agoFix one more cast for _open_osfhandle().
Magnus Hagander [Sat, 2 Jan 2010 12:01:29 +0000 (12:01 +0000)]
Fix one more cast for _open_osfhandle().

Tsutomu Yamada

14 years agoFix cast for _open_osfhandle().
Magnus Hagander [Sat, 2 Jan 2010 12:00:08 +0000 (12:00 +0000)]
Fix cast for _open_osfhandle().

Tsutomu Yamada

14 years agoAdd an "argisrow" field to NullTest nodes, following a plan made way back in
Tom Lane [Fri, 1 Jan 2010 23:03:10 +0000 (23:03 +0000)]
Add an "argisrow" field to NullTest nodes, following a plan made way back in
8.2beta but never carried out.  This avoids repetitive tests of whether the
argument is of scalar or composite type.  Also, be a bit more paranoid about
composite arguments in some places where we previously weren't checking.

14 years agoSupport "x IS NOT NULL" clauses as indexscan conditions. This turns out
Tom Lane [Fri, 1 Jan 2010 21:53:49 +0000 (21:53 +0000)]
Support "x IS NOT NULL" clauses as indexscan conditions.  This turns out
to be just a minor extension of the previous patch that made "x IS NULL"
indexable, because we can treat the IS NOT NULL condition as if it were
"x < NULL" or "x > NULL" (depending on the index's NULLS FIRST/LAST option),
just like IS NULL is treated like "x = NULL".  Aside from any possible
usefulness in its own right, this is an important improvement for
index-optimized MAX/MIN aggregates: it is now reliably possible to get
a column's min or max value cheaply, even when there are a lot of nulls
cluttering the interesting end of the index.

14 years agoSilence compiler warning on 64-bit windows build
Magnus Hagander [Fri, 1 Jan 2010 19:57:07 +0000 (19:57 +0000)]
Silence compiler warning on 64-bit windows build

14 years agoDetect a 64-bit build environment on Windows, and generate the appropriate
Magnus Hagander [Fri, 1 Jan 2010 17:34:25 +0000 (17:34 +0000)]
Detect a 64-bit build environment on Windows, and generate the appropriate
project files.

Based on the work of Tsutomu Yamada, but much refactored.

14 years agoDelete solution cache file on clean, if it exists.
Magnus Hagander [Fri, 1 Jan 2010 16:58:53 +0000 (16:58 +0000)]
Delete solution cache file on clean, if it exists.

14 years agoUpdate SQL features supported list
Peter Eisentraut [Fri, 1 Jan 2010 16:54:48 +0000 (16:54 +0000)]
Update SQL features supported list

14 years agoMake the win32 putenv() override update *all* present versions of the
Magnus Hagander [Fri, 1 Jan 2010 14:57:16 +0000 (14:57 +0000)]
Make the win32 putenv() override update *all* present versions of the
MSVCRxx runtime, not just the current + Visual Studio 6 (MSVCRT). Clearly
there can be an almost unlimited number of runtimes loaded at the same
time.

Per report from Hiroshi Inoue

14 years agoSuppress compiler warning, per buildfarm member narwhal
Tom Lane [Thu, 31 Dec 2009 22:11:10 +0000 (22:11 +0000)]
Suppress compiler warning, per buildfarm member narwhal

14 years agoSuppress compiler warning (pid_t isn't int everywhere)
Tom Lane [Thu, 31 Dec 2009 22:07:36 +0000 (22:07 +0000)]
Suppress compiler warning (pid_t isn't int everywhere)

14 years agoAdd missing 'static' tag.
Tom Lane [Thu, 31 Dec 2009 21:47:12 +0000 (21:47 +0000)]
Add missing 'static' tag.

14 years agoRedefine Datum as uintptr_t, instead of unsigned long.
Tom Lane [Thu, 31 Dec 2009 19:41:37 +0000 (19:41 +0000)]
Redefine Datum as uintptr_t, instead of unsigned long.

This is more in keeping with modern practice, and is a first step towards
porting to Win64 (which has sizeof(pointer) > sizeof(long)).

Tsutomu Yamada, Magnus Hagander, Tom Lane

14 years agoUpdate SQL features list for aggregate ORDER BY support
Peter Eisentraut [Thu, 31 Dec 2009 14:51:16 +0000 (14:51 +0000)]
Update SQL features list for aggregate ORDER BY support

14 years agoAdd information_schema.triggered_update_columns
Peter Eisentraut [Thu, 31 Dec 2009 14:41:23 +0000 (14:41 +0000)]
Add information_schema.triggered_update_columns

This reflects the recently added support for triggers on columns.

14 years agoenable examples to compile on Windows. patch from Hiroshi Saito.
Andrew Dunstan [Thu, 31 Dec 2009 00:16:47 +0000 (00:16 +0000)]
enable examples to compile on Windows. patch from Hiroshi Saito.

14 years agoFill in information schema column for trigger WHEN condition
Peter Eisentraut [Wed, 30 Dec 2009 22:48:10 +0000 (22:48 +0000)]
Fill in information schema column for trigger WHEN condition

14 years agoDept of second thoughts: recursive case in ANALYZE shouldn't emit a
Tom Lane [Wed, 30 Dec 2009 21:21:33 +0000 (21:21 +0000)]
Dept of second thoughts: recursive case in ANALYZE shouldn't emit a
pgstats message.  This might need to be done differently later, but
with the current logic that's what should happen.

14 years agoRevise pgstat's tracking of tuple changes to improve the reliability of
Tom Lane [Wed, 30 Dec 2009 20:32:14 +0000 (20:32 +0000)]
Revise pgstat's tracking of tuple changes to improve the reliability of
decisions about when to auto-analyze.

The previous code depended on n_live_tuples + n_dead_tuples - last_anl_tuples,
where all three of these numbers could be bad estimates from ANALYZE itself.
Even worse, in the presence of a steady flow of HOT updates and matching
HOT-tuple reclamations, auto-analyze might never trigger at all, even if all
three numbers are exactly right, because n_dead_tuples could hold steady.

To fix, replace last_anl_tuples with an accurately tracked count of the total
number of committed tuple inserts + updates + deletes since the last ANALYZE
on the table.  This can still be compared to the same threshold as before, but
it's much more trustworthy than the old computation.  Tracking this requires
one more intra-transaction counter per modified table within backends, but no
additional memory space in the stats collector.  There probably isn't any
measurable speed difference; if anything it might be a bit faster than before,
since I was able to eliminate some per-tuple arithmetic operations in favor of
adding sums once per (sub)transaction.

Also, simplify the logic around pgstat vacuum and analyze reporting messages
by not trying to fold VACUUM ANALYZE into a single pgstat message.

The original thought behind this patch was to allow scheduling of analyzes
on parent tables by artificially inflating their changes_since_analyze count.
I've left that for a separate patch since this change seems to stand on its
own merit.