Bruce Momjian [Sun, 25 Jul 2010 03:28:39 +0000 (03:28 +0000)]
Prevent pg_upgrade from migrating databases that use reg* data types
where the oid is not preserved by pg_upgrade (everything but pg_type).
Update documentation.
Tom Lane [Sun, 18 Jul 2010 23:43:37 +0000 (23:43 +0000)]
Remove unnecessary "Not safe to send CSV data" complaint from elog.c's fallback
path when CSV logging is configured but not yet operational. It's sufficient
to send the message to stderr, as we were already doing, and the "Not safe"
gripe has already confused at least two core members ...
Backpatch to 9.0, but not further --- doesn't seem appropriate to change
this behavior in stable branches.
Tom Lane [Sun, 18 Jul 2010 19:37:57 +0000 (19:37 +0000)]
Allow ORDER BY/GROUP BY/etc items to match targetlist items regardless of
any implicit casting previously applied to the targetlist item. This is
reasonable because the implicit cast, by definition, wasn't written by the
user; so we are preserving the expected behavior that ORDER BY items match
textually equivalent tlist items. The case never arose before because there
couldn't be any implicit casting of a top-level SELECT item before we process
ORDER BY etc. But now it can arise in the context of aggregates containing
ORDER BY clauses, since the "targetlist" is the already-casted list of
arguments for the aggregate. The net effect is that the datatype used for
ORDER BY/DISTINCT purposes is the aggregate's declared input type, not that
of the original input column; which is a bit debatable but not horrendous,
and to do otherwise would require major rework that doesn't seem justified.
Per bug #5564 from Daniel Grace. Back-patch to 9.0 where aggregate ORDER BY
was implemented.
Tom Lane [Wed, 14 Jul 2010 21:21:16 +0000 (21:21 +0000)]
Fix several problems in pg_dump's handling of SQL/MED objects, notably failure
to dump a PUBLIC user mapping correctly, as per bug #5560 from Shigeru Hanada.
Use the pg_user_mappings view rather than trying to access pg_user_mapping
directly, so that the code doesn't fail when run by a non-superuser. And
clean up some minor carelessness such as unsafe usage of fmtId().
Tom Lane [Wed, 14 Jul 2010 17:09:54 +0000 (17:09 +0000)]
Allow full SSL certificate verification (wherein libpq checks its host name
parameter against server cert's CN field) to succeed in the case where
both host and hostaddr are specified. As with the existing precedents
for Kerberos, GSSAPI, SSPI, it is the calling application's responsibility
that host and hostaddr match up --- we just use the host name as given.
Per bug #5559 from Christopher Head.
In passing, make the error handling and messages for the no-host-name-given
failure more consistent among these four cases, and correct a lie in the
documentation: we don't attempt to reverse-lookup host from hostaddr
if host is missing.
Back-patch to 8.4 where SSL cert verification was introduced.
Oops, in the previous fix to prevent a cursor that's being used in a FOR
loop from being dropped, I missed subtransaction cleanup. Pinned portals
must be dropped at subtransaction cleanup just as they are at main
transaction cleanup.
Per bug #5556 by Robert Walker. Backpatch to 8.0, 7.4 didn't have
subtransactions.
Tom Lane [Fri, 9 Jul 2010 22:57:47 +0000 (22:57 +0000)]
Avoid an Assert failure in deconstruct_array() by making get_attstatsslot()
use the actual element type of the array it's disassembling, rather than
trusting the type OID passed in by its caller. This is needed because
sometimes the planner passes in a type OID that's only binary-compatible
with the target column's type, rather than being an exact match. Per an
example from Bernd Helmle.
Possibly we should refactor get_attstatsslot/free_attstatsslot to not expect
the caller to supply type ID data at all, but for now I'll just do the
minimum-change fix.
Back-patch to 7.4. Bernd's test case only crashes back to 8.0, but since
these subroutines are the same in 7.4, I suspect there may be variant
cases that would crash 7.4 as well.
Tom Lane [Fri, 9 Jul 2010 21:11:57 +0000 (21:11 +0000)]
Fix ruleutils' get_variable() to print something useful for Vars referencing
resjunk outputs of subquery tlists, instead of throwing an error. Per bug
#5548 from Daniel Grace.
We might at some point find we ought to back-patch this further than 9.0,
but I think that such Vars can only occur as resjunk members of upper-level
tlists, in which case the problem can't arise because prior versions didn't
print resjunk tlist items in EXPLAIN VERBOSE.
Tom Lane [Thu, 8 Jul 2010 19:00:11 +0000 (19:00 +0000)]
Minor kibitzing on previous patch: no need to run check more than once.
(_PG_init should be called only once anyway, but as long as it's got an
internal guard against repeat calls, that should be in front of the
version check.)
Magnus Hagander [Thu, 8 Jul 2010 16:19:50 +0000 (16:19 +0000)]
Make the Windows tcp keepalive support depend on the existance of the
SIO_KEEPALIVE_VALS define instead of just WIN32, since MingW doesn't
support this API (yet?).
Tom Lane [Thu, 8 Jul 2010 15:15:05 +0000 (15:15 +0000)]
Fix variant float8 expected files to have exactly the expected spacing.
This wasn't important when we used diff's -w (--ignore-all-space) option
to compare regression result files, but it is now. Per buildfarm member
canary, which evidently has been offline since we did that in November,
but came to life again today.
Tom Lane [Thu, 8 Jul 2010 00:14:04 +0000 (00:14 +0000)]
Fix "cannot handle unplanned sub-select" error that can occur when a
sub-select contains a join alias reference that expands into an expression
containing another sub-select. Per yesterday's report from Merlin Moncure
and subsequent off-list investigation.
Back-patch to 7.4. Older versions didn't attempt to flatten sub-selects in
ways that would trigger this problem.
Tom Lane [Wed, 7 Jul 2010 15:13:21 +0000 (15:13 +0000)]
Adjust mbutils.c so it won't get broken by future pgindent runs.
To do that, replace L'\0' by (WCHAR) 0. Perhaps someday we should teach
pgindent about wide-character literals, but so long as this is the only
use-case in the entire Postgres sources, a workaround seems easier.
Robert Haas [Tue, 6 Jul 2010 22:55:26 +0000 (22:55 +0000)]
Make log_temp_files based on kB, and revert docs & comments to match.
Per extensive discussion on pgsql-hackers. We are deliberately not
back-patching this even though the behavior of 8.3 and 8.4 is
unquestionably broken, for fear of breaking existing users of this
parameter. This incompatibility should be release-noted.
Tom Lane [Tue, 6 Jul 2010 22:03:05 +0000 (22:03 +0000)]
Accept slightly grotty coding in Makefile.global in order to keep the -L
flag for src/port/ in front of any -L flags placed in LDFLAGS by configure.
This undoes an L-flag-ordering change that I had thought would be safe,
but seems to be making at least one buildfarm member fail --- the only
theory for orca's failure that I can think of is that it's got an old
copy of libpgport.a in /usr/lib. Also allow for LDFLAGS_SL to be set by
contrib makefiles before they invoke Makefile.global.
Tom Lane [Tue, 6 Jul 2010 03:55:33 +0000 (03:55 +0000)]
Still more third thoughts: when linking shared libraries, LDFLAGS probably
needs to appear before anything placed in SHLIB_LINK. This is because
SHLIB_LINK is typically a subset of LIBS, and LIBS has to appear after
LDFLAGS on platforms that are sensitive to the relative order of -L and -l
switches.
Tom Lane [Mon, 5 Jul 2010 23:30:50 +0000 (23:30 +0000)]
Make sure LDFLAGS come before LIBS when linking contrib programs.
Solaris, at least, seems to be sensitive to the relative order of -L
and -l switches, so this is needed. Per buildfarm results.
Tom Lane [Mon, 5 Jul 2010 23:15:56 +0000 (23:15 +0000)]
Fix a few single-file (MODULES, not MODULE_big) contrib makefiles that were
supposing that they should set SHLIB_LINK rather than LDFLAGS_SL. Since these
don't go through Makefile.shlib that was a no-op on most platforms. Also
regularize the few platform-specific Makefiles that did pay attention to
SHLIB_LINK: it seems that the real value of that is to pull in BE_DLLLIBS,
so do that instead. Per buildfarm failures on cygwin.
Tom Lane [Mon, 5 Jul 2010 18:54:38 +0000 (18:54 +0000)]
Split the LDFLAGS make variable into two parts: LDFLAGS is now used for
linking both executables and shared libraries, and we add on LDFLAGS_EX when
linking executables or LDFLAGS_SL when linking shared libraries. This
provides a significantly cleaner way of dealing with link-time switches than
the former behavior. Also, make sure that the various platform-specific
%.so: %.o rules incorporate LDFLAGS and LDFLAGS_SL; most of them missed that
before. (I did not add these variables for the platforms that invoke $(LD)
directly, however. It's not clear if we can do that safely, since for the
most part we assume these variables use CC command-line syntax.)
Per gripe from Aaron Swenson and subsequent investigation.
The previous fix in CVS HEAD and 8.4 for handling the case where a cursor
being used in a PL/pgSQL FOR loop is closed was inadequate, as Tom Lane
pointed out. The bug affects FOR statement variants too, because you can
close an implicitly created cursor too by guessing the "<unnamed portal X>"
name created for it.
To fix that, "pin" the portal to prevent it from being dropped while it's
being used in a PL/pgSQL FOR loop. Backpatch all the way to 7.4 which is
the oldest supported version.
Tom Lane [Sat, 3 Jul 2010 22:15:45 +0000 (22:15 +0000)]
Don't set recoveryLastXTime when replaying a checkpoint --- that was a bogus
idea from the start since the variable is only meant to track commit/abort
events. This patch reverts the logic around the variable to what it was in
8.4, except that the value is now kept in shared memory rather than a static
variable, so that it can be reported correctly by CreateRestartPoint (which is
executed in the bgwriter).
Tom Lane [Sat, 3 Jul 2010 21:23:58 +0000 (21:23 +0000)]
Make vacuum_defer_cleanup_age be PGC_SIGHUP level, since it's not sensible
to have different values in different processes of the primary server.
Also put it into the "Streaming Replication" GUC category; it doesn't belong
in "Standby Servers" because you use it on the master not the standby.
In passing also correct guc.c's idea of wal_keep_segments' category.
Tom Lane [Sat, 3 Jul 2010 20:43:58 +0000 (20:43 +0000)]
Replace max_standby_delay with two parameters, max_standby_archive_delay and
max_standby_streaming_delay, and revise the implementation to avoid assuming
that timestamps found in WAL records can meaningfully be compared to clock
time on the standby server. Instead, the delay limits are compared to the
elapsed time since we last obtained a new WAL segment from archive or since
we were last "caught up" to WAL data arriving via streaming replication.
This avoids problems with clock skew between primary and standby, as well
as other corner cases that the original coding would misbehave in, such
as the primary server having significant idle time between transactions.
Per my complaint some time ago and considerable ensuing discussion.
Do some desultory editing on the hot standby documentation, too.
Robert Haas [Sat, 3 Jul 2010 13:53:13 +0000 (13:53 +0000)]
Allow REASSIGNED OWNED to handle opclasses and opfamilies.
Backpatch to 8.3, which is as far back as we have opfamilies.
The opclass portion could probably be backpatched to 8.2, when
REASSIGN OWNED was added, but for now I have not done that.
Tom Lane [Sat, 3 Jul 2010 04:03:06 +0000 (04:03 +0000)]
Fix assorted misstatements and poor wording in the descriptions of the I/O
formats for geometric types. Per bug #5536 from Jon Strait, and my own
testing.
Back-patch to all supported branches, since this doco has been wrong right
along -- we certainly haven't changed the I/O behavior of these types in
many years.
Robert Haas [Fri, 2 Jul 2010 17:03:30 +0000 (17:03 +0000)]
Move copydir.c from src/port to src/backend/storage/file
The previous commit to make copydir() interruptible prevented
postgres.exe from linking on MinGW and Cygwin, because on those
platforms libpgport_srv.a can't freely reference symbols defined
by the backend. Since that code is already backend-specific anyway,
just move the whole file into the backend rather than adding further
kludges to deal with the symbols needed by CHECK_FOR_INTERRUPTS().
This probably needs some further cleanup, but this commit just moves
the file as-is, which should hopefully be enough to turn the
buildfarm green again.
Robert Haas [Thu, 1 Jul 2010 20:12:40 +0000 (20:12 +0000)]
Allow copydir() to be interrupted.
This makes ALTER DATABASE .. SET TABLESPACE and CREATE DATABASE more
sensitive to interrupts. Backpatch to 8.4, where ALTER DATABASE .. SET
TABLESPACE was introduced. We could go back further, but in the absence
of complaints about the CREATE DATABASE case it doesn't seem worth it.
stringToNode() and deparse_expression_pretty() crash on invalid input,
but we have nevertheless exposed them to users via pg_get_expr(). It would
be too much maintenance effort to rigorously check the input, so put a hack
in place instead to restrict pg_get_expr() so that the argument must come
from one of the system catalog columns known to contain valid expressions.
Per report from Rushabh Lathia. Backpatch to 7.4 which is the oldest
supported version at the moment.
Tom Lane [Tue, 29 Jun 2010 21:20:19 +0000 (21:20 +0000)]
Add compatibility note warning that plpgsql is now stricter about the column
datatypes of composite results, per gripe from Marcel Asio. Some desultory
copy-editing of plpgsql-related sections of the release notes.
Tom Lane [Mon, 28 Jun 2010 02:07:02 +0000 (02:07 +0000)]
Improve pg_dump's checkSeek() function to verify the functioning of ftello
as well as fseeko, and to not assume that fseeko(fp, 0, SEEK_CUR) proves
anything. Also improve some related comments. Per my observation that
the SEEK_CUR test didn't actually work on some platforms, and subsequent
discussion with Robert Haas.
Back-patch to 8.4. In earlier releases it's not that important whether
we get the hasSeek test right, but with parallel restore it matters.
Tom Lane [Sun, 27 Jun 2010 19:07:24 +0000 (19:07 +0000)]
Fix pg_restore so parallel restore doesn't fail when the input file doesn't
contain data offsets (which it won't, if pg_dump thought its output wasn't
seekable). To do that, remove an unnecessarily aggressive error check, and
instead fail if we get to the end of the archive without finding the desired
data item. Also improve the error message to be more specific about the
cause of the problem. Per discussion of recent report from Igor Neyman.
Back-patch to 8.4 where parallel restore was introduced.
Robert Haas [Fri, 25 Jun 2010 17:08:09 +0000 (17:08 +0000)]
Rewrite docs for new libpq keepalive parameters.
The revised documentation makes it more clear that these are client-side
parameters, rather than server side parameters. It also puts the main
point of each parameter first, and consolidates the conditions under which
it might be ignored in a single list at the end.
Tom Lane [Fri, 25 Jun 2010 16:40:13 +0000 (16:40 +0000)]
Fix thinko in tok_is_keyword(): it was looking at the wrong union variant
of YYSTYPE, and hence returning the wrong answer for cases where a plpgsql
"unreserved keyword" really does conflict with a variable name. Obviously
I didn't test this enough :-(. Per bug #5524 from Peter Gagarinov.
Simon Riggs [Fri, 25 Jun 2010 13:11:25 +0000 (13:11 +0000)]
Fix log_temp_files docs and comments to say bytes not kilobytes.
stat(2) field st_size returns bytes not kilobytes.
Bug found during recent performance tuning for PostgreSQL user.
Robert Haas [Wed, 23 Jun 2010 21:54:13 +0000 (21:54 +0000)]
Add TCP keepalive support to libpq.
This adds four additional connection parameters to libpq: keepalives,
keepalives_idle, keepalives_count, and keepalives_interval.
keepalives default to on, per discussion, but can be turned off by
specifying keepalives=0. The remaining parameters, where supported,
can be used to adjust how often keepalives are sent and how many
can be lost before the connection is broken.
The immediate motivation for this patch is to make sure that
walreceiver will eventually notice if the master reboots without
closing the connection cleanly, but it should be helpful in other
cases as well.