OSDN Git Service

pg-rex/syncrep.git
15 years agoChange macros to make gcc quiet when parsing.
Michael Meskes [Fri, 29 May 2009 13:54:52 +0000 (13:54 +0000)]
Change macros to make gcc quiet when parsing.

15 years agoDocument that forking while having open libpq connections is not
Bruce Momjian [Thu, 28 May 2009 20:02:10 +0000 (20:02 +0000)]
Document that forking while having open libpq connections is not
advised.

15 years agoWhen archiving is enabled, rotate the last WAL segment at shutdown so that
Heikki Linnakangas [Thu, 28 May 2009 11:02:16 +0000 (11:02 +0000)]
When archiving is enabled, rotate the last WAL segment at shutdown so that
all transactions are archived.

Original patch by Guillaume Smet.

15 years agoImprove release note explanation of the change in libpq's handling of
Tom Lane [Wed, 27 May 2009 22:12:53 +0000 (22:12 +0000)]
Improve release note explanation of the change in libpq's handling of
default usernames versus Kerberos tickets.  Per confusion about what
bug #4824 was really about.

15 years agoProperly return the usermap result when doing gssapi authentication. Without
Magnus Hagander [Wed, 27 May 2009 21:08:22 +0000 (21:08 +0000)]
Properly return the usermap result when doing gssapi authentication. Without
this, the username was in practice never matched against the kerberos principal
used to log in.

15 years agoFix to use the same format specifiers in both branches of a ngettext().
Peter Eisentraut [Wed, 27 May 2009 20:47:55 +0000 (20:47 +0000)]
Fix to use the same format specifiers in both branches of a ngettext().

Zdenek Kotala

15 years agoIgnore RECHECK in CREATE OPERATOR CLASS, just throwing a NOTICE, instead of
Tom Lane [Wed, 27 May 2009 20:42:29 +0000 (20:42 +0000)]
Ignore RECHECK in CREATE OPERATOR CLASS, just throwing a NOTICE, instead of
throwing an error as 8.4 had been doing.  The error interfered with porting
old database definitions (particularly for pg_migrator) without really buying
any safety.  Per bug #4817 and subsequent discussion.

15 years agoFix compiler warnings on Sun Studio of the sort
Peter Eisentraut [Wed, 27 May 2009 19:41:58 +0000 (19:41 +0000)]
Fix compiler warnings on Sun Studio of the sort

"tsquery_op.c", line 193: warning: syntax error:  empty declaration

Zdenek Kotala

15 years agoReverting patch just in case a compiler treats this enum as signed.
Michael Meskes [Wed, 27 May 2009 14:16:51 +0000 (14:16 +0000)]
Reverting patch just in case a compiler treats this enum as signed.

15 years agoRemove tabs from SGML file.
Bruce Momjian [Wed, 27 May 2009 12:28:55 +0000 (12:28 +0000)]
Remove tabs from SGML file.

15 years agoImprove documentation about function volatility: mention the snapshot
Tom Lane [Wed, 27 May 2009 01:18:06 +0000 (01:18 +0000)]
Improve documentation about function volatility: mention the snapshot
visibility effects in a couple of places where people are likely to look
for it.  Per discussion of recent question from Karl Nack.

15 years agoAllow the second argument of pg_get_expr() to be just zero when deparsing
Tom Lane [Tue, 26 May 2009 17:36:05 +0000 (17:36 +0000)]
Allow the second argument of pg_get_expr() to be just zero when deparsing
an expression that's not supposed to contain variables.  Per discussion
with Gevik Babakhani, this eliminates the need for an ugly kluge (namely,
specifying some unrelated relation name).  Remove one such kluge from
pg_dump.

15 years agoRemove the useless and rather inconsistent return values of EncodeDateOnly,
Tom Lane [Tue, 26 May 2009 02:17:50 +0000 (02:17 +0000)]
Remove the useless and rather inconsistent return values of EncodeDateOnly,
EncodeTimeOnly, EncodeDateTime, EncodeInterval.  These don't have any good
reason to fail, and their callers were mostly not checking anyway.

15 years agoRemove unused declarations of EncodeTimeOnly and DecodeTimeOnly.
Tom Lane [Tue, 26 May 2009 01:39:49 +0000 (01:39 +0000)]
Remove unused declarations of EncodeTimeOnly and DecodeTimeOnly.

15 years agoAdd range checks to time_recv() and timetz_recv(), to prevent binary input
Tom Lane [Tue, 26 May 2009 01:29:09 +0000 (01:29 +0000)]
Add range checks to time_recv() and timetz_recv(), to prevent binary input
of time values that would not be accepted via textual input.
Per gripe from Andrew McNamara.

This is potentially a back-patchable bug fix, but for the moment it doesn't
seem sufficiently high impact to justify doing that.

15 years agoUse more-portable coding for the check on handing out the last available
Tom Lane [Sun, 24 May 2009 22:22:44 +0000 (22:22 +0000)]
Use more-portable coding for the check on handing out the last available
relopt_kind value in add_reloption_kind().  Per Zdenek Kotala.

15 years agoFix LIKE's special-case code for % followed by _. I'm not entirely sure that
Tom Lane [Sun, 24 May 2009 18:10:38 +0000 (18:10 +0000)]
Fix LIKE's special-case code for % followed by _.  I'm not entirely sure that
this case is worth a special code path, but a special code path that gets
the boundary condition wrong is definitely no good.  Per bug #4821 from
Andrew Gierth.

In passing, clean up some minor code formatting issues (excess parentheses
and blank lines in odd places).

Back-patch to 8.3, where the bug was introduced.

15 years agoResort tsvector's lexemes in tsvectorrecv instead of emmiting an error.
Teodor Sigaev [Thu, 21 May 2009 20:09:36 +0000 (20:09 +0000)]
Resort tsvector's lexemes in tsvectorrecv instead of emmiting an error.
Basically, it's needed to support binary dump from 8.3 because ordering rule
was changed.

Per discussion with Bruce.

15 years agoRemoved comparison of unsigned expression < 0.
Michael Meskes [Thu, 21 May 2009 12:54:27 +0000 (12:54 +0000)]
Removed comparison of unsigned expression < 0.

15 years agoConvert some messages to use ngettext().
Peter Eisentraut [Thu, 21 May 2009 01:08:43 +0000 (01:08 +0000)]
Convert some messages to use ngettext().

Author: Euler Taveira de Oliveira <euler@timbira.com>

15 years agoFix sgml-mode example. thanks Andrew Dunstan
Peter Eisentraut [Wed, 20 May 2009 23:51:24 +0000 (23:51 +0000)]
Fix sgml-mode example.  thanks Andrew Dunstan

15 years agoRemoved unsigned/signed mismatches.
Michael Meskes [Wed, 20 May 2009 16:49:23 +0000 (16:49 +0000)]
Removed unsigned/signed mismatches.

15 years agoMore variables gcc moans about
Michael Meskes [Wed, 20 May 2009 16:39:56 +0000 (16:39 +0000)]
More variables gcc moans about

15 years agoRemoved some unneeded variables and comparisons
Michael Meskes [Wed, 20 May 2009 16:13:18 +0000 (16:13 +0000)]
Removed some unneeded variables and comparisons

15 years agoSilence a gcc compiler warning about non-literal format string with no args
Heikki Linnakangas [Wed, 20 May 2009 08:48:10 +0000 (08:48 +0000)]
Silence a gcc compiler warning about non-literal format string with no args
when compiling with -Wformat-security. Fujii Masao.

15 years agoAdd documentation string for LDFLAGS_SL in configure --help output.
Peter Eisentraut [Tue, 19 May 2009 22:32:41 +0000 (22:32 +0000)]
Add documentation string for LDFLAGS_SL in configure --help output.

15 years agoUpdate relpages and reltuples estimates in stand-alone ANALYZE, even if
Heikki Linnakangas [Tue, 19 May 2009 08:30:00 +0000 (08:30 +0000)]
Update relpages and reltuples estimates in stand-alone ANALYZE, even if
there's no analyzable attributes or indexes. We also used to report 0 live
and dead tuples for such tables, which messed with autovacuum threshold
calculations.

This fixes bug #4812 reported by George Su. Backpatch back to 8.1.

15 years agoFix bug #4814 (wrong subscript in consistent-function call), and add some
Tom Lane [Tue, 19 May 2009 02:48:26 +0000 (02:48 +0000)]
Fix bug #4814 (wrong subscript in consistent-function call), and add some
minimal regression test coverage for matchPartialInPendingList().

15 years agoMake pwdfMatchesString() a little more careful about matching * fields.
Tom Lane [Mon, 18 May 2009 16:15:22 +0000 (16:15 +0000)]
Make pwdfMatchesString() a little more careful about matching * fields.

15 years agoAdd some instructions on how to customize emacs for working on the SGML
Peter Eisentraut [Mon, 18 May 2009 12:47:44 +0000 (12:47 +0000)]
Add some instructions on how to customize emacs for working on the SGML
sources.

15 years agoUpdate SQL conformance entries for window functions functionality
Peter Eisentraut [Mon, 18 May 2009 12:04:59 +0000 (12:04 +0000)]
Update SQL conformance entries for window functions functionality

15 years agoRefer to tables by id, not by "the following table", because tables are in
Peter Eisentraut [Mon, 18 May 2009 11:08:24 +0000 (11:08 +0000)]
Refer to tables by id, not by "the following table", because tables are in
theory floating elements.

15 years agoSome documentation cleanup for the addition of the KOI8U encoding. Change
Peter Eisentraut [Mon, 18 May 2009 08:59:29 +0000 (08:59 +0000)]
Some documentation cleanup for the addition of the KOI8U encoding.  Change
all (remaining) mentions of KOI8 to the new canonical form KOI8R.  Add
information about the available conversions for KOI8U.

15 years agoMinor copy-editing for description of partial-table vacuuming.
Tom Lane [Sat, 16 May 2009 22:51:24 +0000 (22:51 +0000)]
Minor copy-editing for description of partial-table vacuuming.

15 years agoMinor editorialization on storage.sgml's documentation of free space
Tom Lane [Sat, 16 May 2009 22:03:53 +0000 (22:03 +0000)]
Minor editorialization on storage.sgml's documentation of free space
maps.

15 years agoMake an editorial pass over the Client Authentication material.
Tom Lane [Sat, 16 May 2009 21:17:21 +0000 (21:17 +0000)]
Make an editorial pass over the Client Authentication material.

15 years agoImprove comments in pg_ident.conf.sample.
Tom Lane [Sat, 16 May 2009 20:43:46 +0000 (20:43 +0000)]
Improve comments in pg_ident.conf.sample.

15 years agoFix all the server-side SIGQUIT handlers (grumble ... why so many identical
Tom Lane [Fri, 15 May 2009 15:56:39 +0000 (15:56 +0000)]
Fix all the server-side SIGQUIT handlers (grumble ... why so many identical
copies?) to ensure they really don't run proc_exit/shmem_exit callbacks,
as was intended.  I broke this behavior recently by installing atexit
callbacks without thinking about the one case where we truly don't want
to run those callback functions.  Noted in an example from Dave Page.

15 years agocommit for BETA2
Marc G. Fournier [Fri, 15 May 2009 02:18:27 +0000 (02:18 +0000)]
commit for BETA2

15 years agoInclude recovery_end_command in recovery.conf.sample.
Tom Lane [Thu, 14 May 2009 22:22:01 +0000 (22:22 +0000)]
Include recovery_end_command in recovery.conf.sample.
Per suggestion of Jaime Casanova.

15 years agoUpdate release notes to today.
Tom Lane [Thu, 14 May 2009 22:17:28 +0000 (22:17 +0000)]
Update release notes to today.

15 years agoClean up overly hasty docs patch for pg_standby.
Tom Lane [Thu, 14 May 2009 21:59:22 +0000 (21:59 +0000)]
Clean up overly hasty docs patch for pg_standby.

15 years agoTranslation updates
Alvaro Herrera [Thu, 14 May 2009 21:41:53 +0000 (21:41 +0000)]
Translation updates

15 years agoImprove a couple of comments.
Tom Lane [Thu, 14 May 2009 21:28:35 +0000 (21:28 +0000)]
Improve a couple of comments.

15 years agoAdd a note to release notes about the smart failover mode in pg_standby.
Heikki Linnakangas [Thu, 14 May 2009 21:01:49 +0000 (21:01 +0000)]
Add a note to release notes about the smart failover mode in pg_standby.

15 years agoAdd recovery_end_command option to recovery.conf. recovery_end_command
Heikki Linnakangas [Thu, 14 May 2009 20:31:09 +0000 (20:31 +0000)]
Add recovery_end_command option to recovery.conf. recovery_end_command
is run at the end of archive recovery, providing a chance to do external
cleanup. Modify pg_standby so that it no longer removes the trigger file,
that is to be done using the recovery_end_command now.

Provide a "smart" failover mode in pg_standby, where we don't fail over
immediately, but only after recovering all unapplied WAL from the archive.
That gives you zero data loss assuming all WAL was archived before
failover, which is what most users of pg_standby actually want.

recovery_end_command by Simon Riggs, pg_standby changes by Fujii Masao and
myself.

15 years agoAdd checks to DefineQueryRewrite() to prohibit attaching rules to relations
Tom Lane [Wed, 13 May 2009 22:32:55 +0000 (22:32 +0000)]
Add checks to DefineQueryRewrite() to prohibit attaching rules to relations
that aren't RELKIND_RELATION or RELKIND_VIEW, and to disallow attaching rules
to system relations unless allowSystemTableMods is on.  This is to make the
behavior of CREATE RULE more like CREATE TRIGGER, which disallows the
comparable cases.  Per discussion of bug #4808.

15 years agoRemove a useless backslash from a pattern-match example. Michael Toews
Tom Lane [Wed, 13 May 2009 21:53:41 +0000 (21:53 +0000)]
Remove a useless backslash from a pattern-match example.  Michael Toews

15 years agoRewrite xml.c's memory management (yet again). Give up on the idea of
Tom Lane [Wed, 13 May 2009 20:27:17 +0000 (20:27 +0000)]
Rewrite xml.c's memory management (yet again).  Give up on the idea of
redirecting libxml's allocations into a Postgres context.  Instead, just let
it use malloc directly, and add PG_TRY blocks as needed to be sure we release
libxml data structures in error recovery code paths.  This is ugly but seems
much more likely to play nicely with third-party uses of libxml, as seen in
recent trouble reports about using Perl XML facilities in pl/perl and bug
#4774 about contrib/xml2.

I left the code for allocation redirection in place, but it's only
built/used if you #define USE_LIBXMLCONTEXT.  This is because I found it
useful to corral libxml's allocations in a palloc context when hunting
for libxml memory leaks, and we're surely going to have more of those
in the future with this type of approach.  But we don't want it turned on
in a normal build because it breaks exactly what we need to fix.

I have not re-indented most of the code sections that are now wrapped
by PG_TRY(); that's for ease of review.  pg_indent will fix it.

This is a pre-existing bug in 8.3, but I don't dare back-patch this change
until it's gotten a reasonable amount of field testing.

15 years agoUpdate release notes for changes through 2009-05-11. Also some minor
Tom Lane [Tue, 12 May 2009 23:43:50 +0000 (23:43 +0000)]
Update release notes for changes through 2009-05-11.  Also some minor
copy-editing and reordering of items.

15 years agoFix intratransaction memory leaks in xml_recv, xmlconcat, xmlroot, and
Tom Lane [Tue, 12 May 2009 20:17:40 +0000 (20:17 +0000)]
Fix intratransaction memory leaks in xml_recv, xmlconcat, xmlroot, and
xml_parse, all arising from the same sloppy usage of parse_xml_decl.
The original coding had that function returning its output string
parameters in the libxml context, which is long-lived, and all but one
of its callers neglected to free the strings afterwards.  The easiest
and most bulletproof fix is to return the strings in the local palloc
context instead, since that's short-lived.  This was only costing a
dozen or two bytes per function call, but that adds up fast if the
function is called repeatedly ...

Noted while poking at the more general problem of what to do with our
libxml memory allocation hooks.  Back-patch to 8.3, which has the
identical coding.

15 years agoFix LOCK TABLE to eliminate the race condition that could make it give weird
Tom Lane [Tue, 12 May 2009 16:43:32 +0000 (16:43 +0000)]
Fix LOCK TABLE to eliminate the race condition that could make it give weird
errors when tables are concurrently dropped.  To do this we must take lock
on each relation before we check its privileges.  The old code was trying
to do that the other way around, which is a bit pointless when there are lots
of other commands that lock relations before checking privileges.  I did keep
it checking each relation's privilege before locking the next relation, which
is a detail that ALTER TABLE isn't too picky about.

15 years agoModify find_inheritance_children() and find_all_inheritors() to add the
Tom Lane [Tue, 12 May 2009 03:11:02 +0000 (03:11 +0000)]
Modify find_inheritance_children() and find_all_inheritors() to add the
ability to lock relations as they scan pg_inherits, and to ignore any
relations that have disappeared by the time we get lock on them.  This
makes uses of these functions safe against concurrent DROP operations
on child tables: we will effectively ignore any just-dropped child,
rather than possibly throwing an error as in recent bug report from
Thomas Johansson (and similar past complaints).  The behavior should
not change otherwise, since the code was acquiring those same locks
anyway, just a little bit later.

An exception is LockTableCommand(), which is still behaving unsafely;
but that seems to require some more discussion before we change it.

15 years agoDo some minor code refactoring in preparation for changing the APIs of
Tom Lane [Tue, 12 May 2009 00:56:05 +0000 (00:56 +0000)]
Do some minor code refactoring in preparation for changing the APIs of
find_inheritance_children() and find_all_inheritors().  I got annoyed that
these are buried inside the planner but mostly used elsewhere.  So, create
a new file catalog/pg_inherits.c and put them there, along with a couple
of other functions that search pg_inherits.

The code that modifies pg_inherits is (still) in tablecmds.c --- it's
kind of entangled with unrelated code that modifies pg_depend and other
stuff, so pulling it out seemed like a bigger change than I wanted to make
right now.  But this file provides a natural home for it if anyone ever
gets around to that.

This commit just moves code around; it doesn't change anything, except
I succumbed to the temptation to make a couple of trivial optimizations
in typeInheritsFrom().

15 years agoPartially revert my patch of 2008-11-12 that installed a limit on the number
Tom Lane [Mon, 11 May 2009 17:56:08 +0000 (17:56 +0000)]
Partially revert my patch of 2008-11-12 that installed a limit on the number
of AND/OR clause branches that predtest.c would attempt to deal with.  As
noted in bug #4721, that change disabled proof attempts for sizes of problems
that people are actually expecting it to work for.  The original complaint
it was trying to solve was O(N^2) behavior for long IN-lists, so let's try
applying the limit to just ScalarArrayOpExprs rather than everything.
Another case of "foolish consistency" I fear.

Back-patch to 8.2, same as the previous patch was.

15 years agoMove crypt auth comment to proper section.
Magnus Hagander [Mon, 11 May 2009 09:11:41 +0000 (09:11 +0000)]
Move crypt auth comment to proper section.
Add some details about the name=value format of auth options.

15 years agoEdit the SSL and Kerberos parts of the release notes a bit, and add
Magnus Hagander [Mon, 11 May 2009 09:00:10 +0000 (09:00 +0000)]
Edit the SSL and Kerberos parts of the release notes a bit, and add
a note about the certificates chains patch just applied.

15 years agoSupport SSL certificate chains in the server certificate file.
Magnus Hagander [Mon, 11 May 2009 08:06:21 +0000 (08:06 +0000)]
Support SSL certificate chains in the server certificate file.

Andrew Gierth

15 years agoMake a marginal performance improvement in predicate_implied_by and
Tom Lane [Sun, 10 May 2009 22:45:28 +0000 (22:45 +0000)]
Make a marginal performance improvement in predicate_implied_by and
predicate_refuted_by: if either top-level input is a single-element list,
reduce it to its lone member before proceeding.  This avoids
a useless level of AND-recursion within the recursive proof routines.
It's worth doing because, for example, if the clause is a 100-element
list and the predicate is a 1-element list then we'd otherwise strip
the predicate's list structure 100 times as we iterate through the clause.
It's only needed at top level because there won't be any trivial ANDs below
that --- this situation is an artifact of the decision to represent even
single-item conditions as Lists in the "implicit AND" format, and that format
is only used at the top level of any predicate or restriction condition.

15 years agoAdjust pg_dumpall so that it emits ENCODING, LC_COLLATE, and LC_CTYPE options
Tom Lane [Sun, 10 May 2009 02:51:44 +0000 (02:51 +0000)]
Adjust pg_dumpall so that it emits ENCODING, LC_COLLATE, and LC_CTYPE options
in its CREATE DATABASE commands only for databases that have settings
different from the installation defaults.  This is a low-tech method of
avoiding unnecessary platform dependencies in dump files.  Eventually we ought
to have a platform-independent way of specifying LC_COLLATE and LC_CTYPE, but
that's not going to happen for 8.4, and this patch at least avoids the issue
for people who aren't setting up per-database locales.  ENCODING doesn't have
the platform dependency problem, but it seems consistent to make it act the
same as the locale settings.

15 years agoFix cost_nestloop and cost_hashjoin to model the behavior of semi and anti
Tom Lane [Sat, 9 May 2009 22:51:41 +0000 (22:51 +0000)]
Fix cost_nestloop and cost_hashjoin to model the behavior of semi and anti
joins a bit better, ie, understand the differing cost functions for matched
and unmatched outer tuples.  There is more that could be done in cost_hashjoin
but this already helps a great deal.  Per discussions with Robert Haas.

15 years agoAdd alternative expected output files for cs_CZ locale for btree_gist and
Heikki Linnakangas [Fri, 8 May 2009 14:48:06 +0000 (14:48 +0000)]
Add alternative expected output files for cs_CZ locale for btree_gist and
tsearch2 tests. This should make 'comet_moth' buildfarm member pass
contrib check. Zdenek Kotala.

15 years agoAdd missing third argument to open().
Bruce Momjian [Fri, 8 May 2009 14:06:27 +0000 (14:06 +0000)]
Add missing third argument to open().

15 years ago'PGDLLIMPORT' ShmemVariableCache, needed for pg_migrator.so function
Bruce Momjian [Fri, 8 May 2009 03:21:35 +0000 (03:21 +0000)]
'PGDLLIMPORT' ShmemVariableCache, needed for pg_migrator.so function
linkage on Win32.

Tested by Hiroshi Saito

15 years agoAdd an option to AlterTableCreateToastTable() to allow its caller to force
Tom Lane [Thu, 7 May 2009 22:58:28 +0000 (22:58 +0000)]
Add an option to AlterTableCreateToastTable() to allow its caller to force
a toast table to be built, even if the sum-of-column-widths calculation
indicates one isn't needed.  This is needed by pg_migrator because if the
old table has a toast table, we have to migrate over the toast table since
it might contain some live data, even though subsequent column drops could
mean that no recently-added rows could require toasting.

15 years agoChange pgbench to use the table names pgbench_accounts, pgbench_branches,
Tom Lane [Thu, 7 May 2009 22:01:18 +0000 (22:01 +0000)]
Change pgbench to use the table names pgbench_accounts, pgbench_branches,
pgbench_history, and pgbench_tellers, rather than just accounts, branches,
history, and tellers.  This is to prevent accidental conflicts with real
application tables, as has been reported to happen at least once.  Also
remove the automatic "SET search_path = public" that it did at startup,
as this seems to restrict testing flexibility without actually buying much.
Per proposal by Joshua Drake and ensuing discussion.

Joshua Drake and Tom Lane

15 years agoOoops ... make_outerjoininfo wasn't actually enforcing the join order
Tom Lane [Thu, 7 May 2009 20:13:09 +0000 (20:13 +0000)]
Ooops ... make_outerjoininfo wasn't actually enforcing the join order
restrictions specified for semijoins in optimizer/README, to wit that
you can't reassociate outer joins into or out of the RHS of a semijoin.
Per report from Heikki.

15 years agoRequest XLOG switch before writing checkpoint in pg_start_backup(). Otherwise
Heikki Linnakangas [Thu, 7 May 2009 11:25:25 +0000 (11:25 +0000)]
Request XLOG switch before writing checkpoint in pg_start_backup(). Otherwise
you can end up with an unrecoverable backup if you start a new base backup
right after finishing archive recovery. In that scenario, the redo pointer of
the checkpoint that pg_start_backup() writes points to the XLOG segment where
the timeline-changing end-of-archive-recovery checkpoint is. The beginning
of that segment contains pages with the old timeline ID, and we don't accept
that in recovery unless we find a history file covering the old timeline ID.
If you omit pg_xlog from the base backup and clear the archive directory
before starting the backup, there will be no such history file available.

The bug is present in all versions since PITR was introduced in 8.0, but I'm
back-patching only back to 8.2. Earlier versions didn't have XLOG switch
records, making this fix unfeasible. Given the lack of reports until now,
it doesn't seem worthwhile to spend more effort to fix 8.0 and 8.1.

Per report and suggestion by Mikael Krantz

15 years agoTweak distribute_qual_to_rels so that when we decide a pseudoconstant qual
Tom Lane [Wed, 6 May 2009 20:31:18 +0000 (20:31 +0000)]
Tweak distribute_qual_to_rels so that when we decide a pseudoconstant qual
can be pushed to the top of the join tree, we update both the relids and
qualscope variables to keep them in sync.  This prevents a possible later
failure of an Assert clause, and affects nothing else since qualscope isn't
used later except for that Assert.  At the moment the Assert shouldn't be
reachable when we've pushed the qual up; but this is cheap insurance, and
it's more sensible anyway in terms of the overall logic of the routine.
Per analysis of a bug report from Stefan Huehner.

I'm not back-patching this since it's just future-proofing; but if anyone
gets tempted to change check_outerjoin_delay again in the back branches,
this might be needed.

15 years agoModify CREATE DATABASE to enforce that the source database's encoding setting
Tom Lane [Wed, 6 May 2009 16:15:21 +0000 (16:15 +0000)]
Modify CREATE DATABASE to enforce that the source database's encoding setting
must be used for the new database, except when copying from template0.
This is the same rule that we now enforce for locale settings, and it has
the same motivation: databases other than template0 might contain data that
would be invalid according to a different setting.  This represents another
step in a continuing process of locking down ways in which encoding violations
could occur inside the backend.  Per discussion of a few days ago.

In passing, fix pre-existing breakage of mbregress.sh, and fix up a couple
of ereport() calls in dbcommands.c that failed to specify sqlstate codes.

15 years agoFix ecpg tests for change that disallowed Unicode literals unless
Tom Lane [Wed, 6 May 2009 03:22:15 +0000 (03:22 +0000)]
Fix ecpg tests for change that disallowed Unicode literals unless
standard_conforming_strings is on.

15 years agoTweak a comment to agree a bit better with the new dispensation that
Tom Lane [Tue, 5 May 2009 23:39:55 +0000 (23:39 +0000)]
Tweak a comment to agree a bit better with the new dispensation that
locales are database-wide, not server-wide.

15 years agoMinor improvement: avoid assuming that GetLastError value cannot be
Tom Lane [Tue, 5 May 2009 21:51:46 +0000 (21:51 +0000)]
Minor improvement: avoid assuming that GetLastError value cannot be
affected by CloseHandle() or Sleep().

15 years agoMake new complaint about unsafe Unicode literals include an error location.
Tom Lane [Tue, 5 May 2009 21:09:23 +0000 (21:09 +0000)]
Make new complaint about unsafe Unicode literals include an error location.
Every other ereport in scan.l has one, this should too.

15 years agoInstall an atexit(2) callback that ensures that proc_exit's cleanup processing
Tom Lane [Tue, 5 May 2009 20:06:07 +0000 (20:06 +0000)]
Install an atexit(2) callback that ensures that proc_exit's cleanup processing
will still be performed if something in a backend process calls exit()
directly, instead of going through proc_exit() as we prefer.  This is a second
response to the issue that we might load third-party code that doesn't know it
should not call exit().  Such a call will now cause a reasonably graceful
backend shutdown, if possible.  (Of course, if the reason for the exit() call
is out-of-memory or some such, we might not be able to recover, but at least
we will try.)

15 years agoInstall a "dead man switch" to allow the postmaster to detect cases where
Tom Lane [Tue, 5 May 2009 19:59:00 +0000 (19:59 +0000)]
Install a "dead man switch" to allow the postmaster to detect cases where
a backend has done exit(0) or exit(1) without having disengaged itself
from shared memory.  We are at risk for this whenever third-party code is
loaded into a backend, since such code might not know it's supposed to go
through proc_exit() instead.  Also, it is reported that under Windows
there are ways to externally kill a process that cause the status code
returned to the postmaster to be indistinguishable from a voluntary exit
(thank you, Microsoft).  If this does happen then the system is probably
hosed --- for instance, the dead session might still be holding locks.
So the best recovery method is to treat this like a backend crash.

The dead man switch is armed for a particular child process when it
acquires a regular PGPROC, and disarmed when the PGPROC is released;
these should be the first and last touches of shared memory resources
in a backend, or close enough anyway.  This choice means there is no
coverage for auxiliary processes, but I doubt we need that, since they
shouldn't be executing any user-provided code anyway.

This patch also improves the management of the EXEC_BACKEND
ShmemBackendArray array a bit, by reducing search costs.

Although this problem is of long standing, the lack of field complaints
seems to mean it's not critical enough to risk back-patching; at least
not till we get some more testing of this mechanism.

15 years agoInsert CHECK_FOR_INTERRUPTS() calls into btree and hash index scans at the
Tom Lane [Tue, 5 May 2009 19:36:32 +0000 (19:36 +0000)]
Insert CHECK_FOR_INTERRUPTS() calls into btree and hash index scans at the
points where we step right or left to the next page.  This should ensure
reasonable response time to a query cancel request during an unsuccessful
index scan, as seen in recent gripe from Marc Cousin.  It's a bit trickier
than it might seem at first glance, because CHECK_FOR_INTERRUPTS() is a no-op
if executed while holding a buffer lock.  So we have to do it just at the
point where we've dropped one page lock and not yet acquired the next.

Remove CHECK_FOR_INTERRUPTS calls at the top level of btgetbitmap and
hashgetbitmap, since they're pointless given the added checks.

I think that GIST is okay already --- at least, there's a CHECK_FOR_INTERRUPTS
at a plausible-looking place in gistnext().  I don't claim to know GIN well
enough to try to poke it for this, if indeed it has a problem at all.

This is a pre-existing issue, but in view of the lack of prior complaints
I'm not going to risk back-patching.

15 years agoUpdate comment for _bt_relandgetbuf.
Tom Lane [Tue, 5 May 2009 19:02:22 +0000 (19:02 +0000)]
Update comment for _bt_relandgetbuf.

15 years agoDisable the use of Unicode escapes in string constants (U&'') when
Peter Eisentraut [Tue, 5 May 2009 18:32:17 +0000 (18:32 +0000)]
Disable the use of Unicode escapes in string constants (U&'') when
standard_conforming_strings is not on, for security reasons.

15 years agoAvoid integer overflow in the loop that extracts histogram entries from
Tom Lane [Tue, 5 May 2009 18:02:11 +0000 (18:02 +0000)]
Avoid integer overflow in the loop that extracts histogram entries from
ANALYZE's total sample.  The original coding is at risk of overflow for
statistics targets exceeding about 2675; this was not a problem before
8.4 but it is now.  Per bug #4793 from Dennis Noordsij.

15 years agoMake the win32 shared memory code try 10 times instead of one if
Magnus Hagander [Tue, 5 May 2009 09:48:51 +0000 (09:48 +0000)]
Make the win32 shared memory code try 10 times instead of one if
it fails because the shared memory segment already exists. This
means it can take up to 10 seconds before it reports the error
if it *does* exist, but hopefully it will make the system capable
of restarting even when the server is under high load.

15 years agoCode review for \df rewrite: fix assorted bugs, make type and
Tom Lane [Tue, 5 May 2009 02:29:06 +0000 (02:29 +0000)]
Code review for \df rewrite: fix assorted bugs, make type and
volatility columns localizable.

15 years agoFix the query used for \d against 8.2 and 8.3 servers.
Heikki Linnakangas [Mon, 4 May 2009 17:31:35 +0000 (17:31 +0000)]
Fix the query used for \d against 8.2 and 8.3 servers.

15 years agoUpdate no longer current comments in header.
Magnus Hagander [Mon, 4 May 2009 09:30:06 +0000 (09:30 +0000)]
Update no longer current comments in header.

15 years agoCall SetLastError(0) before calling the file mapping functions
Magnus Hagander [Mon, 4 May 2009 08:36:40 +0000 (08:36 +0000)]
Call SetLastError(0) before calling the file mapping functions
to make sure that the error code is reset, as a precaution in
case the API doesn't properly reset it on success. This could
be necessary, since we check the error value even if the function
doesn't fail for specific success cases.

15 years agoReplace a couple of references to files that no longer exist in the source
Peter Eisentraut [Mon, 4 May 2009 08:08:47 +0000 (08:08 +0000)]
Replace a couple of references to files that no longer exist in the source
tree with references to the appropriate URLs.

Robert Haas

15 years agoFix missed usage of DLNewElem()
Tom Lane [Mon, 4 May 2009 02:46:36 +0000 (02:46 +0000)]
Fix missed usage of DLNewElem()

15 years agoAvoid a memory allocation in the backend startup code, to avoid having to check
Alvaro Herrera [Mon, 4 May 2009 02:24:17 +0000 (02:24 +0000)]
Avoid a memory allocation in the backend startup code, to avoid having to check
whether it failed.  Modelled after catcache.c's usage of DlList, per suggestion
from Tom.

15 years agoFix pg_resetxlog to remove archive status files along with WAL segment files.
Tom Lane [Sun, 3 May 2009 23:13:37 +0000 (23:13 +0000)]
Fix pg_resetxlog to remove archive status files along with WAL segment files.

Fujii Masao

15 years agoFix mis-description of XML Schema functions, per discussion.
Tom Lane [Sun, 3 May 2009 21:10:27 +0000 (21:10 +0000)]
Fix mis-description of XML Schema functions, per discussion.

15 years agoFix unintelligible description created by removing only part of
Tom Lane [Sun, 3 May 2009 20:45:43 +0000 (20:45 +0000)]
Fix unintelligible description created by removing only part of
a parenthetical remark.

15 years agoFix assign_pgstat_temp_directory() to ensure the directory path is
Tom Lane [Sun, 3 May 2009 20:09:54 +0000 (20:09 +0000)]
Fix assign_pgstat_temp_directory() to ensure the directory path is
canonicalized.  Avoid the need to elog(FATAL) on out-of-memory.

15 years agoFix already-obsolete hint message ... sslverify parameter is no more.
Tom Lane [Sun, 3 May 2009 17:16:58 +0000 (17:16 +0000)]
Fix already-obsolete hint message ... sslverify parameter is no more.

15 years agoUpdate UTF-8 <--> EUC_KR, JOHAB, UHC mappings.
Tatsuo Ishii [Sun, 3 May 2009 01:17:41 +0000 (01:17 +0000)]
Update UTF-8 <--> EUC_KR, JOHAB, UHC mappings.
Patch contributed by Chuck McDevitt

15 years agoInstall some simple defenses in postmaster startup to help ensure a useful
Tom Lane [Sat, 2 May 2009 22:02:37 +0000 (22:02 +0000)]
Install some simple defenses in postmaster startup to help ensure a useful
error message if the installation directory layout is messed up (or at least,
something more useful than the behavior exhibited in bug #4787).  During
postmaster startup, check that get_pkglib_path resolves as a readable
directory; and if ParseTzFile() fails to open the expected timezone
abbreviation file, check the possibility that the directory is missing rather
than just the specified file.  In case of either failure, issue a hint
suggesting that the installation is broken.  These two checks cover the lib/
and share/ trees of a full installation, which should take care of most
scenarios where a sysadmin decides to get cute.

15 years agoWe don't need major_release_split any more.
Tom Lane [Sat, 2 May 2009 20:28:17 +0000 (20:28 +0000)]
We don't need major_release_split any more.

15 years agoSplit the release notes into a separate file for each (active) major branch,
Tom Lane [Sat, 2 May 2009 20:17:19 +0000 (20:17 +0000)]
Split the release notes into a separate file for each (active) major branch,
as per my recent proposal.  release.sgml itself is now just a stub that should
change rarely; ideally, only once per major release to add a new include line.
Most editing work will occur in the release-N.N.sgml files.  To update a back
branch for a minor release, just copy the appropriate release-N.N.sgml
file(s) into the back branch.

This commit doesn't change the end-product documentation at all, only the
source layout.  However, it makes it easy to start omitting ancient information
from newer branches' documentation, should we ever decide to do that.

15 years agoFix plpgsql's EXIT so that an EXIT without a label only matches a loop,
Tom Lane [Sat, 2 May 2009 17:27:57 +0000 (17:27 +0000)]
Fix plpgsql's EXIT so that an EXIT without a label only matches a loop,
never a BEGIN block.  This is required for Oracle compatibility and is
also plainly stated to be the behavior by our original documentation
(up until 8.1, in which the docs were adjusted to match the code's behavior;
but actually the old docs said the correct thing and the code was wrong).

Not back-patched because this introduces an incompatibility that could
break working applications.  Requires release note.

15 years agoFix a couple of cases where the plpgsql grammar looked for T_WORD and
Tom Lane [Fri, 1 May 2009 23:57:34 +0000 (23:57 +0000)]
Fix a couple of cases where the plpgsql grammar looked for T_WORD and
failed to consider the possibility that it would get T_SCALAR, T_RECORD,
or T_ROW instead because the word happens to match a plpgsql variable name.
In particular, give "duplicate declaration" rather than generic "syntax error"
if the same identifier is declared twice in the same block, as per my recent
complaint.  Also behave more sanely when decl_aliasitem or proc_condition or
opt_lblname is coincidentally not T_WORD.  Refactor the related productions a
bit to reduce duplication.

This is a longstanding bug, but it doesn't seem critical enough to
back-patch.

15 years agoWhen checking for datetime field overflow, we should allow a fractional-second
Tom Lane [Fri, 1 May 2009 19:29:07 +0000 (19:29 +0000)]
When checking for datetime field overflow, we should allow a fractional-second
part that rounds up to exactly 1.0 second.  The previous coding rejected input
like "00:12:57.9999999999999999999999999999", with the exact number of nines
needed to cause failure varying depending on float-timestamp option and
possibly on platform.  Obviously this should round up to the next integral
second, if we don't have enough precision to distinguish the value from that.
Per bug #4789 from Robert Kruus.

In passing, fix a missed check for fractional seconds in one copy of the
"is it greater than 24:00:00" code.

Broken all the way back, so patch all the way back.