OSDN Git Service

pg-rex/syncrep.git
16 years agoAlso build snprintf from pgport if needed.
Michael Meskes [Wed, 3 Oct 2007 13:40:52 +0000 (13:40 +0000)]
Also build snprintf from pgport if needed.

16 years agoRemove exports.list in clean target.
Michael Meskes [Wed, 3 Oct 2007 13:24:54 +0000 (13:24 +0000)]
Remove exports.list in clean target.

16 years agoMore detailed error msg (with stack trace) if a file copy fails.
Magnus Hagander [Wed, 3 Oct 2007 13:20:40 +0000 (13:20 +0000)]
More detailed error msg (with stack trace) if a file copy fails.

16 years agoAlso created export list for pgytpeslib and compatlib.
Michael Meskes [Wed, 3 Oct 2007 13:20:19 +0000 (13:20 +0000)]
Also created export list for pgytpeslib and compatlib.
Set pgtypes library version to 3.0.
Set compat library version to 3.0.

16 years agoUse snprintf from libpgport in ecpg compatlib. Required since
Magnus Hagander [Wed, 3 Oct 2007 12:34:29 +0000 (12:34 +0000)]
Use snprintf from libpgport in ecpg compatlib. Required since
we restricted exports from ecpglib.

16 years agoFix commandlines for final two ECPG regression tests
Magnus Hagander [Wed, 3 Oct 2007 12:31:18 +0000 (12:31 +0000)]
Fix commandlines for final two ECPG regression tests

16 years agoFix typo
Magnus Hagander [Wed, 3 Oct 2007 12:11:37 +0000 (12:11 +0000)]
Fix typo

16 years agoBuild DEF file for libecpg instead of generating a default one.
Magnus Hagander [Wed, 3 Oct 2007 12:11:00 +0000 (12:11 +0000)]
Build DEF file for libecpg instead of generating a default one.

16 years agoMake ECPG regression tests run with -c only for array_of_struct.pgc
Magnus Hagander [Wed, 3 Oct 2007 11:57:36 +0000 (11:57 +0000)]
Make ECPG regression tests run with -c only for array_of_struct.pgc
on MSVC.

Fix strange nonstandard version of __stdcall specifyer in thread
tests on win32.

16 years agoFix typo
Magnus Hagander [Wed, 3 Oct 2007 11:34:11 +0000 (11:34 +0000)]
Fix typo

16 years agoCleaned up ecpglib and renamed functions that do not need to be exported.
Michael Meskes [Wed, 3 Oct 2007 11:11:12 +0000 (11:11 +0000)]
Cleaned up ecpglib and renamed functions that do not need to be exported.
Created export list for ecpglib.

16 years agoHopefully fixed some stuff that causes Windows builds to fail.
Michael Meskes [Wed, 3 Oct 2007 08:55:23 +0000 (08:55 +0000)]
Hopefully fixed some stuff that causes Windows builds to fail.

16 years agoFirst cut at 8.3 release notes. SGML markup is nonexistent, and
Tom Lane [Wed, 3 Oct 2007 03:32:31 +0000 (03:32 +0000)]
First cut at 8.3 release notes.  SGML markup is nonexistent, and
the 'overview' and 'incompatibilities' summary lists remain to be
written.  But I think all the raw info is there (indeed maybe too
verbose).

16 years agoFix a compiler warning on Win32. Hannes Eder.
Neil Conway [Tue, 2 Oct 2007 22:01:02 +0000 (22:01 +0000)]
Fix a compiler warning on Win32. Hannes Eder.

16 years agoDefault to thread safety on, and support more CPU options. Also
Magnus Hagander [Tue, 2 Oct 2007 19:01:45 +0000 (19:01 +0000)]
Default to thread safety on, and support more CPU options. Also
make sure that a CPU option is actually chosen.

Hiroshi Saito

16 years agoMade new test also work without threading enabled.
Michael Meskes [Tue, 2 Oct 2007 14:01:57 +0000 (14:01 +0000)]
Made new test also work without threading enabled.

16 years agoNo idea where this file came from.
Michael Meskes [Tue, 2 Oct 2007 09:51:41 +0000 (09:51 +0000)]
No idea where this file came from.

16 years agoITAGAKI Takahiro <itagaki.takahiro@oss.ntt.co.jp> added thread-safe
Michael Meskes [Tue, 2 Oct 2007 09:50:00 +0000 (09:50 +0000)]
ITAGAKI Takahiro <itagaki.takahiro@oss.ntt.co.jp> added thread-safe
descriptor handling

16 years agoCope with ERR_set_mark() and ERR_pop_to_mark() not existing in older
Tom Lane [Tue, 2 Oct 2007 00:25:20 +0000 (00:25 +0000)]
Cope with ERR_set_mark() and ERR_pop_to_mark() not existing in older
OpenSSL libraries --- just don't call them if they're not there.  This
might possibly lead to misleading error messages, but we'll just have
to live with that.

16 years agoAdd catalogs.sgml documentation for text search catalogs.
Tom Lane [Mon, 1 Oct 2007 21:10:40 +0000 (21:10 +0000)]
Add catalogs.sgml documentation for text search catalogs.

16 years agoUse BIO functions to avoid passing FILE * pointers to OpenSSL functions.
Magnus Hagander [Mon, 1 Oct 2007 20:30:06 +0000 (20:30 +0000)]
Use BIO functions to avoid passing FILE * pointers to OpenSSL functions.
This fixes potential crashes on old versions of OpenSSL and the requirement on
"Applink" in new versions when building with MSVC and using different
runtimes.

Dave Page with fixes from me.

16 years agoAdd some examples.
D'Arcy J.M. Cain [Mon, 1 Oct 2007 19:06:48 +0000 (19:06 +0000)]
Add some examples.

16 years agoAdd note warning against use of pre-8.4 multithreaded Tcl.
Tom Lane [Mon, 1 Oct 2007 16:43:28 +0000 (16:43 +0000)]
Add note warning against use of pre-8.4 multithreaded Tcl.

16 years agoAvoid assuming that struct varattrib_pointer doesn't get padded by the
Tom Lane [Mon, 1 Oct 2007 16:25:56 +0000 (16:25 +0000)]
Avoid assuming that struct varattrib_pointer doesn't get padded by the
compiler --- at least on ARM, it does.  I suspect that the varvarlena patch
has been creating larger-than-intended toast pointers all along on ARM,
but it wasn't exposed until the latest tweak added some Asserts that
calculated the expected size in a different way.  We could probably have
fixed this by adding __attribute__((packed)) as is done for ItemPointerData,
but struct varattrib_pointer isn't really all that useful anyway, so it
seems cleanest to just get rid of it and have only struct varattrib_1b_e.
Per results from buildfarm member quagga.

16 years agoBuild with /GS on 64-bit to work with modern Platform SDK.
Magnus Hagander [Mon, 1 Oct 2007 14:32:02 +0000 (14:32 +0000)]
Build with /GS on 64-bit to work with modern Platform SDK.

Hiroshi Saito

16 years agoUpdate pgcvslog text to simply "<branch>"
Bruce Momjian [Mon, 1 Oct 2007 13:04:55 +0000 (13:04 +0000)]
Update pgcvslog text to simply "<branch>"

16 years agoEnable __FUNCTION__ on MSVC builds.
Magnus Hagander [Mon, 1 Oct 2007 10:54:29 +0000 (10:54 +0000)]
Enable __FUNCTION__ on MSVC builds.

Hannes Eder

16 years agoFix a minor typo.
D'Arcy J.M. Cain [Mon, 1 Oct 2007 06:52:42 +0000 (06:52 +0000)]
Fix a minor typo.

16 years agoFixup for new pgcvslog feature.
Bruce Momjian [Mon, 1 Oct 2007 03:01:38 +0000 (03:01 +0000)]
Fixup for new pgcvslog feature.

16 years agoHave pgcvslog mark back-branch commits with "<back-patch>".
Bruce Momjian [Mon, 1 Oct 2007 02:59:03 +0000 (02:59 +0000)]
Have pgcvslog mark back-branch commits with "<back-patch>".

16 years agoAdd an extra header byte to TOAST-pointer datums to represent their size
Tom Lane [Sun, 30 Sep 2007 19:54:58 +0000 (19:54 +0000)]
Add an extra header byte to TOAST-pointer datums to represent their size
explicitly.  This means a TOAST pointer takes 18 bytes instead of 17 --- still
smaller than in 8.2 --- which seems a good tradeoff to ensure we won't have
painted ourselves into a corner if we want to support multiple types of TOAST
pointer later on.  Per discussion with Greg Stark.

16 years agoIn hopes of un-breaking the buildfarm, add missing file from
Tom Lane [Sun, 30 Sep 2007 17:50:56 +0000 (17:50 +0000)]
In hopes of un-breaking the buildfarm, add missing file from
ITAGAKI Takahiro's patch.

16 years agoAdjust recovery PS display as agreed with Simon: 'waiting for XXX'
Tom Lane [Sun, 30 Sep 2007 17:28:56 +0000 (17:28 +0000)]
Adjust recovery PS display as agreed with Simon: 'waiting for XXX'
while the restore_command does its thing, then 'recovering XXX' while
processing the segment file.  These operations are heavyweight enough
that an extra PS display set shouldn't bother anyone.

16 years agoProperly mark mergeable/hashable equality operators (found by opr_sanity
Tom Lane [Sun, 30 Sep 2007 17:13:19 +0000 (17:13 +0000)]
Properly mark mergeable/hashable equality operators (found by opr_sanity
testing).  Combine the formerly independent opclasses for the various
ISN types into opfamilies.  The latter causes some extra bleating from
opr_sanity, since the module doesn't provide complete sets of cross-type
operators, but it's still a good idea because it will give the planner
more information to work with.  The missing cross-type operators no longer
pose a risk of unexpected planner errors in 8.3, so there's no need to
insist on filling them in (and I gather it wouldn't be very sound
semantically to add them all).

16 years agoSupport functions for index opclasses should be immutable.
Tom Lane [Sun, 30 Sep 2007 15:52:06 +0000 (15:52 +0000)]
Support functions for index opclasses should be immutable.
Found by running opr_sanity on contrib modules.

16 years agoApplied another patch by ITAGAKI Takahiro <itagaki.takahiro@oss.ntt.co.jp>
Michael Meskes [Sun, 30 Sep 2007 11:38:48 +0000 (11:38 +0000)]
Applied another patch by ITAGAKI Takahiro <itagaki.takahiro@oss.ntt.co.jp>
to get memory allocation thread-safe. He also did some cleaning up.

16 years agoSupport functions for index opclasses should be immutable.
Tom Lane [Sun, 30 Sep 2007 00:03:47 +0000 (00:03 +0000)]
Support functions for index opclasses should be immutable.
Found by running opr_sanity on contrib modules.

16 years agoSupport functions for index opclasses should be immutable.
Tom Lane [Sat, 29 Sep 2007 23:48:22 +0000 (23:48 +0000)]
Support functions for index opclasses should be immutable.
Found by running opr_sanity on contrib modules.

16 years agoSupport functions for index opclasses should be immutable.
Tom Lane [Sat, 29 Sep 2007 23:43:18 +0000 (23:43 +0000)]
Support functions for index opclasses should be immutable.
Found by running opr_sanity on contrib modules.

16 years agoSupport functions for index opclasses should be immutable.
Tom Lane [Sat, 29 Sep 2007 23:38:17 +0000 (23:38 +0000)]
Support functions for index opclasses should be immutable.
Found by running opr_sanity on contrib modules.

16 years agoRemove bogus commutator marking --- the module doesn't actually supply
Tom Lane [Sat, 29 Sep 2007 23:32:42 +0000 (23:32 +0000)]
Remove bogus commutator marking --- the module doesn't actually supply
any commutator operator for =(chkpass,text), so this was creating a
shell operator that would fail on use.  Found by opr_sanity testing.

16 years agoSupport functions for index opclasses should be immutable.
Tom Lane [Sat, 29 Sep 2007 23:24:39 +0000 (23:24 +0000)]
Support functions for index opclasses should be immutable.
Found by running opr_sanity on contrib modules.

16 years agoMake recovery show the current input WAL segment name in the startup
Tom Lane [Sat, 29 Sep 2007 18:32:56 +0000 (18:32 +0000)]
Make recovery show the current input WAL segment name in the startup
process' PS display.  After a suggestion by Simon (not exactly his
patch though).

16 years agoDisallow CLUSTER using an invalid index (that is, one left over from a failed
Tom Lane [Sat, 29 Sep 2007 18:05:20 +0000 (18:05 +0000)]
Disallow CLUSTER using an invalid index (that is, one left over from a failed
CREATE INDEX CONCURRENTLY).  Such an index might not have entries for every
heap row and thus clustering with it would result in silent data loss.
The scenario requires a pretty foolish DBA, but still ...

16 years agoImprove consistency of the error messages generated when you try to use
Tom Lane [Sat, 29 Sep 2007 17:18:58 +0000 (17:18 +0000)]
Improve consistency of the error messages generated when you try to use
ALTER TABLE on a composite type or ALTER TYPE on a table's rowtype.
We already rejected these cases, but the error messages were a bit
random and didn't always provide a HINT to use the other command type.

16 years agoIn the release checklist, mention packagers will see the minor upgrade
Bruce Momjian [Sat, 29 Sep 2007 12:19:16 +0000 (12:19 +0000)]
In the release checklist, mention packagers will see the minor upgrade
numbering for additional functions.

16 years agoAdd chklocale.c to pgport files for msvc builds, per
Magnus Hagander [Sat, 29 Sep 2007 07:15:33 +0000 (07:15 +0000)]
Add chklocale.c to pgport files for msvc builds, per
buildfarm failures.

16 years agoDefend against openssl libraries that fail on keys longer than 128 bits;
Tom Lane [Sat, 29 Sep 2007 02:18:15 +0000 (02:18 +0000)]
Defend against openssl libraries that fail on keys longer than 128 bits;
which is the case at least on some Solaris versions.  Marko Kreen

16 years agoMake archive recovery always start a new timeline, rather than only when a
Tom Lane [Sat, 29 Sep 2007 01:36:10 +0000 (01:36 +0000)]
Make archive recovery always start a new timeline, rather than only when a
recovery stop time was used.  This avoids a corner-case risk of trying to
overwrite an existing archived copy of the last WAL segment, and seems
simpler and cleaner all around than the original definition.  Per example
from Jon Colverson and subsequent analysis by Simon.

16 years agoMake use of additional chklocale.c entries to reject locales that we
Tom Lane [Sat, 29 Sep 2007 00:14:40 +0000 (00:14 +0000)]
Make use of additional chklocale.c entries to reject locales that we
cannot support.

16 years agoTeach chklocale.c about a few names for frontend-only encodings,
Tom Lane [Sat, 29 Sep 2007 00:01:43 +0000 (00:01 +0000)]
Teach chklocale.c about a few names for frontend-only encodings,
since this will allow initdb to reject attempts to initdb in a locale
that uses such an encoding.  We'll probably find out more such names
during beta ...

16 years agoOn OS X, assume that an empty-string result for nl_langinfo(CODESET)
Tom Lane [Fri, 28 Sep 2007 23:36:06 +0000 (23:36 +0000)]
On OS X, assume that an empty-string result for nl_langinfo(CODESET)
means UTF-8.  Per examination of /usr/share/locale in 10.4.10.

16 years agoUpdate TCL comment:
Bruce Momjian [Fri, 28 Sep 2007 22:33:20 +0000 (22:33 +0000)]
Update TCL comment:

 * We can only fix this with Tcl >= 8.4, when Tcl_SetNotifier()

16 years agoChange initdb and CREATE DATABASE to actively reject attempts to create
Tom Lane [Fri, 28 Sep 2007 22:25:49 +0000 (22:25 +0000)]
Change initdb and CREATE DATABASE to actively reject attempts to create
databases with encodings that are incompatible with the server's LC_CTYPE
locale, when we can determine that (which we can on most modern platforms,
I believe).  C/POSIX locale is compatible with all encodings, of course,
so there is still some usefulness to CREATE DATABASE's ENCODING option,
but this will insulate us against all sorts of recurring complaints
caused by mismatched settings.

I moved initdb's existing LC_CTYPE-to-encoding mapping knowledge into
a new src/port/ file so it could be shared by CREATE DATABASE.

16 years agoRemove:
Bruce Momjian [Fri, 28 Sep 2007 21:32:29 +0000 (21:32 +0000)]
Remove:

< * SQL*Net listener that makes PostgreSQL appear as an Oracle database
<   to clients

16 years agoTweak initdb's text search configuration selection code so it can
Tom Lane [Fri, 28 Sep 2007 15:25:44 +0000 (15:25 +0000)]
Tweak initdb's text search configuration selection code so it can
cope with LANG settings like 'es_ES@euro'.

16 years agoexclude contrib/tsearch2 from regression tests.
Andrew Dunstan [Fri, 28 Sep 2007 00:14:58 +0000 (00:14 +0000)]
exclude contrib/tsearch2 from regression tests.

16 years agoTurn build and vcregress .bat files into pure one line wrappers for
Andrew Dunstan [Thu, 27 Sep 2007 21:13:11 +0000 (21:13 +0000)]
Turn build and vcregress .bat files into pure one line wrappers for
the perl scripts. Remove the now superfluous getregress.pl.

16 years agoFix typos in two comments. Spotted by Brendan Jurd
Tom Lane [Thu, 27 Sep 2007 21:01:59 +0000 (21:01 +0000)]
Fix typos in two comments.  Spotted by Brendan Jurd

16 years agoTweak pgbench.c to remove the hidden assumption that a WIN32 machine
Tom Lane [Thu, 27 Sep 2007 20:39:43 +0000 (20:39 +0000)]
Tweak pgbench.c to remove the hidden assumption that a WIN32 machine
couldn't possibly HAVE_GETOPT.  I believe this is the most appropriate
form of the patch submitted 2007-08-07 by Hiroshi Saito, though not
having a Windows build environment I won't know for sure till I see
the buildfarm results.

16 years agoDefine the FRONTEND symbol in postgres_fe.h, which allows us to eliminate
Tom Lane [Thu, 27 Sep 2007 19:53:44 +0000 (19:53 +0000)]
Define the FRONTEND symbol in postgres_fe.h, which allows us to eliminate
duplicative -DFRONTEND flags from many Makefiles.  We still need Makefile
control of the symbol in a few places that compile frontend-or-backend
src/port/ files, but it's a lot cleaner than before.

Hiroshi Saito

16 years agoAdd virtual transaction IDs to CSVLOG output, so that messages coming from
Tom Lane [Thu, 27 Sep 2007 18:15:36 +0000 (18:15 +0000)]
Add virtual transaction IDs to CSVLOG output, so that messages coming from
the same transaction can be identified even when no regular XID was assigned.
This seems essential after addition of the lazy-XID patch.  Also some
minor code cleanup in write_csvlog().

16 years agoFix Assert failure in ExpandColumnRefStar --- what I thought was a can't
Tom Lane [Thu, 27 Sep 2007 17:42:03 +0000 (17:42 +0000)]
Fix Assert failure in ExpandColumnRefStar --- what I thought was a can't
happen condition can happen given incorrect input.  The real problem is that
gram.y should try harder to distinguish * from "*" --- the latter is a legal
column name per spec, and someday we ought to treat it that way.  However
fixing that is too invasive for a back-patch, and it's too late for the 8.3
cycle too.  So just reduce the Assert to a plain elog for now.  Per report
from NikhilS.

16 years agoLowercase all mentions of "PostgreSQL" in URLs, per Magnus.
Bruce Momjian [Thu, 27 Sep 2007 06:14:47 +0000 (06:14 +0000)]
Lowercase all mentions of "PostgreSQL" in URLs, per Magnus.

16 years agoSome small tuptoaster improvements from Greg Stark. Avoid unnecessary
Tom Lane [Wed, 26 Sep 2007 23:29:10 +0000 (23:29 +0000)]
Some small tuptoaster improvements from Greg Stark.  Avoid unnecessary
decompression of an already-compressed external value when we have to copy
it; save a few cycles when a value is too short for compression; and
annotate various lines that are currently unreachable.

16 years agoMinor improvements in backup and recovery:
Tom Lane [Wed, 26 Sep 2007 22:36:30 +0000 (22:36 +0000)]
Minor improvements in backup and recovery:

- create a separate archive_mode GUC, on which archive_command is dependent

- %r option in recovery.conf sends last restartpoint to recovery command

- %r used in pg_standby, updated README

- minor other code cleanup in pg_standby

- doc on Warm Standby now mentions pg_standby and %r

- log_restartpoints recovery option emits LOG message at each restartpoint

- end of recovery now displays last transaction end time, as requested
  by Warren Little; also shown at each restartpoint

- restart archiver if needed to carry away WAL files at shutdown

Simon Riggs

16 years agoRename "PostgreSQL" to "Postgres" in 3 places.
Bruce Momjian [Wed, 26 Sep 2007 20:38:28 +0000 (20:38 +0000)]
Rename "PostgreSQL" to "Postgres" in 3 places.

16 years agoRename "PostgreSQL" to "Postgres" in 7 places in the FAQ.
Bruce Momjian [Wed, 26 Sep 2007 20:36:50 +0000 (20:36 +0000)]
Rename "PostgreSQL" to "Postgres" in 7 places in the FAQ.

16 years agoAdjust the new memory limit in the lazy vacuum code to use MaxHeapTuplesPerPage
Alvaro Herrera [Wed, 26 Sep 2007 20:16:28 +0000 (20:16 +0000)]
Adjust the new memory limit in the lazy vacuum code to use MaxHeapTuplesPerPage
tuples per page instead of fixed 200, to better cope with systems that use a
different block size.

16 years agoCreate a function variable "join_search_hook" to let plugins override the
Tom Lane [Wed, 26 Sep 2007 18:51:51 +0000 (18:51 +0000)]
Create a function variable "join_search_hook" to let plugins override the
join search order portion of the planner; this is specifically intended to
simplify developing a replacement for GEQO planning.  Patch by Julius
Stroffek, editorialized on by me.  I renamed make_one_rel_by_joins to
standard_join_search and make_rels_by_joins to join_search_one_level to better
reflect their place within this scheme.

16 years agoProperly pass -c parameter to ecpg when building regression tests.
Magnus Hagander [Wed, 26 Sep 2007 11:19:09 +0000 (11:19 +0000)]
Properly pass -c parameter to ecpg when building regression tests.
Per Michael.

16 years agoApplied patch by ITAGAKI Takahiro <itagaki.takahiro@oss.ntt.co.jp> to get prepare...
Michael Meskes [Wed, 26 Sep 2007 10:57:01 +0000 (10:57 +0000)]
Applied patch by ITAGAKI Takahiro <itagaki.takahiro@oss.ntt.co.jp> to get prepare thread-safe.

16 years agoFix crash of to_tsvector() function on huge input: compareWORD()
Teodor Sigaev [Wed, 26 Sep 2007 10:09:57 +0000 (10:09 +0000)]
Fix crash of to_tsvector() function on huge input: compareWORD()
function didn't return correct result for word position greate than
limit.

Per report from Stuart Bishop <stuart@stuartbishop.net>

16 years agoAdd:
Bruce Momjian [Wed, 26 Sep 2007 08:47:43 +0000 (08:47 +0000)]
Add:

> * Implement Boyer-Moore searching in strpos()
>
>   http://archives.postgresql.org/pgsql-patches/2007-08/msg00012.php

16 years agoDocument that libpq's PQntuples() might overflow.
Bruce Momjian [Wed, 26 Sep 2007 08:45:50 +0000 (08:45 +0000)]
Document that libpq's PQntuples() might overflow.

16 years agoIn the integer-datetimes case, date2timestamp and date2timestamptz need
Tom Lane [Wed, 26 Sep 2007 01:10:42 +0000 (01:10 +0000)]
In the integer-datetimes case, date2timestamp and date2timestamptz need
to check for overflow because the legal range of type date is actually
wider than timestamp's.  Problem found by Neil Conway.

16 years agoUse SYSV semaphores rather than POSIX on Darwin >= 6.0 (i.e., OS X 10.2
Tom Lane [Wed, 26 Sep 2007 00:32:46 +0000 (00:32 +0000)]
Use SYSV semaphores rather than POSIX on Darwin >= 6.0 (i.e., OS X 10.2
and up), per Chris Marcellino.  This avoids consuming O(N^2) file
descriptors to support N backends.  Tests suggest it's about a wash for
small installations, but large ones would have a problem.

16 years agoChange on-disk representation of NUMERIC datatype so that the sign_dscale
Tom Lane [Tue, 25 Sep 2007 22:21:55 +0000 (22:21 +0000)]
Change on-disk representation of NUMERIC datatype so that the sign_dscale
word comes before the weight instead of after.  This will allow future
binary-compatible extension of the representation to support compact formats,
as discussed on pgsql-hackers around 2007/06/18.  The reason to do it now is
that we've already pretty well broken any chance of simple in-place upgrade
from 8.2 to 8.3, but it's possible that 8.3 to 8.4 (or whenever we get around
to squeezing NUMERIC) could otherwise be data-compatible.

16 years agoDept. of second thoughts: fix loop in BgBufferSync so that the exit when
Tom Lane [Tue, 25 Sep 2007 22:11:48 +0000 (22:11 +0000)]
Dept. of second thoughts: fix loop in BgBufferSync so that the exit when
bgwriter_lru_maxpages is exceeded leaves the loop variables in the
expected state.  In the original coding, we'd fail to advance
next_to_clean, causing that buffer to be probably-uselessly rechecked next
time, and also have an off-by-one idea of the number of buffers scanned.

16 years agoJust-in-time background writing strategy. This code avoids re-scanning
Tom Lane [Tue, 25 Sep 2007 20:03:38 +0000 (20:03 +0000)]
Just-in-time background writing strategy.  This code avoids re-scanning
buffers that cannot possibly need to be cleaned, and estimates how many
buffers it should try to clean based on moving averages of recent allocation
requests and density of reusable buffers.  The patch also adds a couple
more columns to pg_stat_bgwriter to help measure the effectiveness of the
bgwriter.

Greg Smith, building on his own work and ideas from several other people,
in particular a much older patch from Itagaki Takahiro.

16 years agoSmall string tweaks
Peter Eisentraut [Tue, 25 Sep 2007 16:29:34 +0000 (16:29 +0000)]
Small string tweaks

16 years agoUse proper search for contrib makefiles in vcregress.pl
Andrew Dunstan [Mon, 24 Sep 2007 21:42:34 +0000 (21:42 +0000)]
Use proper search for contrib makefiles in vcregress.pl

16 years agofix typo in vcregress.pl
Andrew Dunstan [Mon, 24 Sep 2007 21:14:54 +0000 (21:14 +0000)]
fix typo in vcregress.pl

16 years agoAdd comments re text <-> bytea internal equivalence in convert routines.
Andrew Dunstan [Mon, 24 Sep 2007 16:38:24 +0000 (16:38 +0000)]
Add comments re text <-> bytea internal equivalence in convert routines.

16 years agoUse correct PG_GETARG macro in pg_convert
Andrew Dunstan [Mon, 24 Sep 2007 14:59:37 +0000 (14:59 +0000)]
Use correct PG_GETARG macro in pg_convert

16 years agoAvoid having autovacuum read pgstats data too many times in quick succession.
Alvaro Herrera [Mon, 24 Sep 2007 04:12:01 +0000 (04:12 +0000)]
Avoid having autovacuum read pgstats data too many times in quick succession.
This is problematic for the autovac launcher when there are many databases,
so we keep data for a full second before reading it again.

16 years agoReduce the size of memory allocations by lazy vacuum when processing a small
Alvaro Herrera [Mon, 24 Sep 2007 03:52:55 +0000 (03:52 +0000)]
Reduce the size of memory allocations by lazy vacuum when processing a small
table, by allocating just enough for a hardcoded number of dead tuples per
page.  The current estimate is 200 dead tuples per page.

Per reports from Jeff Amiel, Erik Jones and Marko Kreen, and subsequent
discussion.
CVS: ----------------------------------------------------------------------
CVS: Enter Log.  Lines beginning with `CVS:' are removed automatically
CVS:
CVS: Committing in .
CVS:
CVS: Modified Files:
CVS:  commands/vacuumlazy.c
CVS: ----------------------------------------------------------------------

16 years agoSimplify and rename some GUC variables, per various recent discussions:
Tom Lane [Mon, 24 Sep 2007 03:12:23 +0000 (03:12 +0000)]
Simplify and rename some GUC variables, per various recent discussions:

* stats_start_collector goes away; we always start the collector process,
unless prevented by a problem with setting up the stats UDP socket.

* stats_reset_on_server_start goes away; it seems useless in view of the
availability of pg_stat_reset().

* stats_block_level and stats_row_level are merged into a single variable
"track_counts", which controls all reports sent to the collector process.

* stats_command_string is renamed to track_activities.

* log_autovacuum is renamed to log_autovacuum_min_duration to better reflect
its meaning.

The log_autovacuum change is not a compatibility issue since it didn't exist
before 8.3 anyway.  The other changes need to be release-noted.

16 years agoRemove "convert 'blah' using conversion_name" facility, because if it
Andrew Dunstan [Mon, 24 Sep 2007 01:29:30 +0000 (01:29 +0000)]
Remove "convert 'blah' using conversion_name" facility, because if it
produces text it is an encoding hole and if not it's incompatible
with the spec, whatever the spec means (which we're not sure about anyway).

16 years agoReplace linear searches with binary searches in pg_dump's code to
Tom Lane [Sun, 23 Sep 2007 23:39:36 +0000 (23:39 +0000)]
Replace linear searches with binary searches in pg_dump's code to
lookup objects by OID.  Per gripe from nikitathespider.

16 years agoAdd perl replacements for build.bat and vcregress.bat. In due course
Andrew Dunstan [Sun, 23 Sep 2007 21:52:56 +0000 (21:52 +0000)]
Add perl replacements for build.bat and vcregress.bat. In due course
the .bat files will be altered to become tiny wrappers for these scripts,
and one or two other .bat files will disappear.

16 years agoFix bugs in XML binary I/O functions. Heikki and Tom
Tom Lane [Sun, 23 Sep 2007 21:36:42 +0000 (21:36 +0000)]
Fix bugs in XML binary I/O functions.  Heikki and Tom

16 years agoRestrict overly broad searches, and fix typo, in recent fix. Per Hannes Eder.
Andrew Dunstan [Sun, 23 Sep 2007 20:32:40 +0000 (20:32 +0000)]
Restrict overly broad searches, and fix typo, in recent fix. Per Hannes Eder.

16 years agoMake autovacuum report the start time of its current activity in
Tom Lane [Sun, 23 Sep 2007 20:07:33 +0000 (20:07 +0000)]
Make autovacuum report the start time of its current activity in
pg_stat_activity.  Per gripe from Jim Nasby.

16 years agoTransactionIdIsInProgress can skip scanning the ProcArray if the target XID is
Tom Lane [Sun, 23 Sep 2007 18:50:38 +0000 (18:50 +0000)]
TransactionIdIsInProgress can skip scanning the ProcArray if the target XID is
later than latestCompletedXid, per Florian Pflug.  Also some minor
improvements in the XIDCACHE_DEBUG code --- make sure each call of
TransactionIdIsInProgress is counted one way or another.

16 years agoTypo fix from Brendan Jurd.
Bruce Momjian [Sun, 23 Sep 2007 18:06:47 +0000 (18:06 +0000)]
Typo fix from Brendan Jurd.

16 years agoTemporarily modify tsearch regression tests to suppress notice that comes
Tom Lane [Sun, 23 Sep 2007 15:58:58 +0000 (15:58 +0000)]
Temporarily modify tsearch regression tests to suppress notice that comes
out at erratic times, because it is creating a totally unacceptable level
of noise in our buildfarm results.  This patch can be reverted when and if
the code is fixed to not issue notices during cache reload events.

16 years agoFix cost estimates for EXISTS subqueries that are evaluated as initPlans
Tom Lane [Sat, 22 Sep 2007 21:36:40 +0000 (21:36 +0000)]
Fix cost estimates for EXISTS subqueries that are evaluated as initPlans
(because they are uncorrelated with the immediate parent query).  We were
charging the full run cost to the parent node, disregarding the fact that
only one row need be fetched for EXISTS.  While this would only be a
cosmetic issue in most cases, it might possibly affect planning outcomes
if the parent query were itself a subquery to some upper query.
Per recent discussion with Steve Crawford.

16 years agoReplace calls to external dir program with perlish globs and File::Find
Andrew Dunstan [Sat, 22 Sep 2007 20:38:10 +0000 (20:38 +0000)]
Replace calls to external dir program with perlish globs and File::Find
calls. Fixes complaint fron Hannes Eder, whose environment found a different
dir program.

16 years agoDocument the translations from Postgres message severity levels to
Tom Lane [Sat, 22 Sep 2007 19:10:44 +0000 (19:10 +0000)]
Document the translations from Postgres message severity levels to
syslog and eventlog severity levels, per suggestion from Josh Drake.
Also, some wordsmithing for the csvlog documentation.