]> granicus.if.org Git - postgresql/log
postgresql
17 years agoOn second thought, the tests for what to do with stderr output are a
Tom Lane [Thu, 19 Jul 2007 21:58:12 +0000 (21:58 +0000)]
On second thought, the tests for what to do with stderr output are a
lot more sensible if we check the chunk-output case first.  Not
back-patched since it's just a cosmetic improvement.

17 years agoMake replace(), split_part(), and string_to_array() behave somewhat sanely
Tom Lane [Thu, 19 Jul 2007 20:34:20 +0000 (20:34 +0000)]
Make replace(), split_part(), and string_to_array() behave somewhat sanely
when handed an invalidly-encoded pattern.  The previous coding could get
into an infinite loop if pg_mb2wchar_with_len() returned a zero-length
string after we'd tested for nonempty pattern; which is exactly what it
will do if the string consists only of an incomplete multibyte character.
This led to either an out-of-memory error or a backend crash depending
on platform.  Per report from Wiktor Wodecki.

17 years agoOnly use the pipe chunking protocol if we know the syslogger should
Andrew Dunstan [Thu, 19 Jul 2007 19:13:43 +0000 (19:13 +0000)]
Only use the pipe chunking protocol if we know the syslogger should
be catching stderr output, and we are not ourselves the
syslogger. Otherwise, go directly to stderr.
Bug noticed by Tom Lane.
Backpatch as far as 8.0.

17 years agoAdjust configure script to print the bison and flex versions in use.
Tom Lane [Thu, 19 Jul 2007 17:15:30 +0000 (17:15 +0000)]
Adjust configure script to print the bison and flex versions in use.
Minor rearrangements to make a few tests in a more logical order.

17 years agoFix an old thinko in SS_make_initplan_from_plan, which is used when optimizing
Tom Lane [Wed, 18 Jul 2007 21:40:57 +0000 (21:40 +0000)]
Fix an old thinko in SS_make_initplan_from_plan, which is used when optimizing
a MIN or MAX aggregate call into an indexscan: the initplan is being made at
the current query nesting level and so we shouldn't increment query_level.
Though usually harmless, this mistake could lead to bogus "plan should not
reference subplan's variable" failures on complex queries.  Per bug report
from David Sanchez i Gregori.

17 years agoCast NULL to a pointer type in the execl() call, to avoid a compiler warning on
Alvaro Herrera [Wed, 18 Jul 2007 21:19:17 +0000 (21:19 +0000)]
Cast NULL to a pointer type in the execl() call, to avoid a compiler warning on
some platforms and possibly a bug.  Per report from Stefan and subsequent
discussion.

17 years agoGSSAPI documentation
Magnus Hagander [Wed, 18 Jul 2007 12:00:47 +0000 (12:00 +0000)]
GSSAPI documentation

17 years agoDocument that 'deleted' is also tracked by autovacuum.
Bruce Momjian [Wed, 18 Jul 2007 03:39:01 +0000 (03:39 +0000)]
Document that 'deleted' is also tracked by autovacuum.

17 years agoProperly adjust age() seconds to match the sign of the larger units.
Bruce Momjian [Wed, 18 Jul 2007 03:13:13 +0000 (03:13 +0000)]
Properly adjust age() seconds to match the sign of the larger units.
Patch from Tom.

17 years agoDocument that age() adds days, then full months.
Bruce Momjian [Wed, 18 Jul 2007 03:12:42 +0000 (03:12 +0000)]
Document that age() adds days, then full months.

17 years agoAdd:
Bruce Momjian [Wed, 18 Jul 2007 00:16:21 +0000 (00:16 +0000)]
Add:

>
>  o Allow GLOBAL temporary tables to exist as empty by default in
>    all sessions
>
>    http://archives.postgresql.org/pgsql-hackers/2007-07/msg00006.php
>

17 years agoFix incorrect optimization of foreign-key checks. When an UPDATE on the
Tom Lane [Tue, 17 Jul 2007 17:45:28 +0000 (17:45 +0000)]
Fix incorrect optimization of foreign-key checks.  When an UPDATE on the
referencing table does not change the tuple's FK column(s), we don't bother
to check the PK table since the constraint was presumably already valid.
However, the check is still necessary if the tuple was inserted by our own
transaction, since in that case the INSERT trigger will conclude it need not
make the check (since its version of the tuple has been deleted).  We got this
right for simple cases, but not when the insert and update are in different
subtransactions of the current top-level transaction; in such cases the FK
check would never be made at all.  (Hence, problem dates back to 8.0 when
subtransactions were added --- it's actually the subtransaction version of a
bug fixed in 7.3.5.)  Fix, and add regression test cases.  Report and fix by
Affan Salman.

17 years agoRemove http://www.benchmarkresources.com, no longer resolves to a
Bruce Momjian [Tue, 17 Jul 2007 05:03:55 +0000 (05:03 +0000)]
Remove http://www.benchmarkresources.com, no longer resolves to a
meaningful site.

17 years agoImplement CREATE TABLE LIKE ... INCLUDING INDEXES. Patch from NikhilS,
Neil Conway [Tue, 17 Jul 2007 05:02:03 +0000 (05:02 +0000)]
Implement CREATE TABLE LIKE ... INCLUDING INDEXES. Patch from NikhilS,
based in part on an earlier patch from Trevor Hardcastle, and reviewed
by myself.

17 years agoAdd CVS Wiki URL to docs.
Bruce Momjian [Tue, 17 Jul 2007 01:52:34 +0000 (01:52 +0000)]
Add CVS Wiki URL to docs.

17 years agoFix outfuncs.c to dump A_Const nodes representing NULLs correctly. This has
Tom Lane [Tue, 17 Jul 2007 01:21:43 +0000 (01:21 +0000)]
Fix outfuncs.c to dump A_Const nodes representing NULLs correctly.  This has
been broken since forever, but was not noticed because people seldom look
at raw parse trees.  AFAIK, no impact on users except that debug_print_parse
might fail; but patch it all the way back anyway.  Per report from Jeff Ross.

17 years agoAdd:
Bruce Momjian [Tue, 17 Jul 2007 00:07:54 +0000 (00:07 +0000)]
Add:

> * Allow multiple indexes to be created concurrently, ideally via a
>   single heap scan, and have a restore of a pg_dump somehow use it
>
>   http://archives.postgresql.org/pgsql-general/2007-05/msg01274.php

Small blank line additions.

17 years agoUpdate docs that GNU tar versions >=1.16 exit with 1 on files changed, 2
Bruce Momjian [Mon, 16 Jul 2007 22:20:51 +0000 (22:20 +0000)]
Update docs that GNU tar versions >=1.16 exit with 1 on files changed, 2
on other errors.

17 years agoFix pg_buffercache to release buffer partition locks in reverse order,
Tom Lane [Mon, 16 Jul 2007 21:20:36 +0000 (21:20 +0000)]
Fix pg_buffercache to release buffer partition locks in reverse order,
and add a note about why.  This is not tremendously important right now,
probably, but it will get more urgent if NUM_BUFFER_PARTITIONS is increased
as much as proposed.

17 years agoAdd comments spelling out why it's a good idea to release multiple
Tom Lane [Mon, 16 Jul 2007 21:09:50 +0000 (21:09 +0000)]
Add comments spelling out why it's a good idea to release multiple
partition locks in reverse order.

17 years agoWith the native compiler on Unixware, disable optimization if
Neil Conway [Mon, 16 Jul 2007 17:38:48 +0000 (17:38 +0000)]
With the native compiler on Unixware, disable optimization if
--enable-debug is used, to avoid complaints about debugging and
optimization being mutually exclusive. Patch from Stefan Kaltenbrunner.

17 years agoAllow plpgsql function parameter names to be qualified with the function's
Tom Lane [Mon, 16 Jul 2007 17:01:11 +0000 (17:01 +0000)]
Allow plpgsql function parameter names to be qualified with the function's
name.  With this patch, it is always possible for the user to qualify a
plpgsql variable name if needed to avoid ambiguity.  While there is much more
work to be done in this area, this simple change removes one unnecessary
incompatibility with Oracle.  Per discussion.

17 years agoHmm, so evidently _check_lock and _clear_lock take an argument of type
Tom Lane [Mon, 16 Jul 2007 14:02:22 +0000 (14:02 +0000)]
Hmm, so evidently _check_lock and _clear_lock take an argument of type
int not unsigned int.  Third try to get grebe building without warnings...

17 years agoQuote pathnames so pg_standby works with paths that have
Magnus Hagander [Mon, 16 Jul 2007 08:40:52 +0000 (08:40 +0000)]
Quote pathnames so pg_standby works with paths that have
spaces in them.

ISHIDA Akio

17 years agoSo our reward for including <sys/atomic_op.h> seems to be a bunch of
Tom Lane [Mon, 16 Jul 2007 04:57:57 +0000 (04:57 +0000)]
So our reward for including <sys/atomic_op.h> seems to be a bunch of
nattering about casting away volatile.  Losers.

17 years agoOn AIX, include <sys/atomic_op.h> so that the functions we use for
Tom Lane [Mon, 16 Jul 2007 02:03:14 +0000 (02:03 +0000)]
On AIX, include <sys/atomic_op.h> so that the functions we use for
TAS support are properly declared.

17 years agoGet rid of overly cute, unportable, probably not very efficient substitute
Tom Lane [Sun, 15 Jul 2007 23:57:13 +0000 (23:57 +0000)]
Get rid of overly cute, unportable, probably not very efficient substitute
for 'bool'.  Per buildfarm warnings.

17 years agoFix CHECK_RELATION_BLOCK_RANGE macro, which was not merely producing
Tom Lane [Sun, 15 Jul 2007 23:46:20 +0000 (23:46 +0000)]
Fix CHECK_RELATION_BLOCK_RANGE macro, which was not merely producing
a warning but was outright wrong.

17 years agoFix a passel of signed vs unsigned char warnings.
Tom Lane [Sun, 15 Jul 2007 23:30:19 +0000 (23:30 +0000)]
Fix a passel of signed vs unsigned char warnings.

17 years agoSilence a rather odd compiler warning. In passing, make this file's
Tom Lane [Sun, 15 Jul 2007 23:09:26 +0000 (23:09 +0000)]
Silence a rather odd compiler warning.  In passing, make this file's
error messages look at least a little bit like the message style
guidelines say.

17 years agoAvoid possibly-unportable initializer, per buildfarm warning.
Tom Lane [Sun, 15 Jul 2007 22:57:48 +0000 (22:57 +0000)]
Avoid possibly-unportable initializer, per buildfarm warning.

17 years agoChange a couple of exit(0) to return 0 to suppress complaints from
Tom Lane [Sun, 15 Jul 2007 22:54:21 +0000 (22:54 +0000)]
Change a couple of exit(0) to return 0 to suppress complaints from
not-too-bright compilers.  Per buildfarm results.

17 years agoSilence Solaris compiler warnings, per buildfarm.
Tom Lane [Sun, 15 Jul 2007 22:49:36 +0000 (22:49 +0000)]
Silence Solaris compiler warnings, per buildfarm.

17 years agoSilence Solaris compiler warning, per buildfarm.
Tom Lane [Sun, 15 Jul 2007 22:43:40 +0000 (22:43 +0000)]
Silence Solaris compiler warning, per buildfarm.

17 years agoFix compile warning on Solaris, per buildfarm. (Why have we got
Tom Lane [Sun, 15 Jul 2007 22:40:28 +0000 (22:40 +0000)]
Fix compile warning on Solaris, per buildfarm.  (Why have we got
three slightly different copies of this file?)

17 years agoFix possible portability problem, per buildfarm warnings.
Tom Lane [Sun, 15 Jul 2007 22:34:26 +0000 (22:34 +0000)]
Fix possible portability problem, per buildfarm warnings.

17 years agoFix unportable use of isspace(), per buildfarm results.
Tom Lane [Sun, 15 Jul 2007 22:32:53 +0000 (22:32 +0000)]
Fix unportable use of isspace(), per buildfarm results.

17 years agoBecause plpgsql's scanner uses %option case-insensitive, flex's results could
Tom Lane [Sun, 15 Jul 2007 22:18:24 +0000 (22:18 +0000)]
Because plpgsql's scanner uses %option case-insensitive, flex's results could
theoretically vary depending on what the compile-time locale setting is.
Hence, force it to see LC_CTYPE=C to ensure consistent build results.
(It's likely that this makes no difference in practice, since our
specification for "identifier" surely includes both ends of any possible
uppercase/lowercase pair anyway.  But it should silence warnings about
ambiguous character classes that are reported by some buildfarm members.)

17 years agoReject zero or negative BY step in plpgsql integer FOR-loops, and behave
Tom Lane [Sun, 15 Jul 2007 02:15:04 +0000 (02:15 +0000)]
Reject zero or negative BY step in plpgsql integer FOR-loops, and behave
sanely if the loop value overflows int32 on the way to the end value.
Avoid useless computation of "SELECT 1" when BY is omitted.  Avoid some
type-punning between Datum and int4 that dates from the original coding.

17 years agoNote incompatibility with Oracle's version of FOR ... REVERSE, per
Tom Lane [Sun, 15 Jul 2007 00:45:16 +0000 (00:45 +0000)]
Note incompatibility with Oracle's version of FOR ... REVERSE, per
Andrew Dunstan.  Minor other improvements in documentation of integer
FOR loops.

17 years agoEditorial overhaul of plpgsql documentation. Provide detailed documentation
Tom Lane [Sat, 14 Jul 2007 23:02:25 +0000 (23:02 +0000)]
Editorial overhaul of plpgsql documentation.  Provide detailed documentation
of variable substitution and plan caching behavior in dedicated sections.
(A lot of this material existed already, but was scattered in various places
in the chapter.)  Reorganize material a little bit, mostly to try to avoid
diving into deep details in the first introductory sections.  Document some
fine points that had escaped treatment before, notably the ability to qualify
plpgsql variable names with block labels.  Some minor wordsmithing here and
there.

17 years agoSupport for finding gssapi functions in the library "gss", as required
Magnus Hagander [Sat, 14 Jul 2007 11:13:28 +0000 (11:13 +0000)]
Support for finding gssapi functions in the library "gss", as required
by Solaris 10 and possibly others.

Stefan Kaltenbrunner

17 years agoVolatile-qualify a dozen variables in plpython.c to eliminate warnings
Tom Lane [Fri, 13 Jul 2007 04:57:59 +0000 (04:57 +0000)]
Volatile-qualify a dozen variables in plpython.c to eliminate warnings
from old versions of gcc.  It's not clear to me that this is really
necessary for correctness, but less warnings are always good.
Per buildfarm results and local testing.

17 years agoFix map_sql_typecoll_to_xmlschema_types() to not fail on dropped
Tom Lane [Fri, 13 Jul 2007 03:43:23 +0000 (03:43 +0000)]
Fix map_sql_typecoll_to_xmlschema_types() to not fail on dropped
columns, per my gripe earlier today.  Make it look a bit less like
someone's first effort at backend coding.

17 years agoAdd casts to suppress warnings about m68k-specific kluge in fmgr.c.
Tom Lane [Fri, 13 Jul 2007 02:25:48 +0000 (02:25 +0000)]
Add casts to suppress warnings about m68k-specific kluge in fmgr.c.

17 years agoFix a portability bug (ye olde not casting a <ctype.h> argument to
Tom Lane [Thu, 12 Jul 2007 23:51:10 +0000 (23:51 +0000)]
Fix a portability bug (ye olde not casting a <ctype.h> argument to
unsigned char).  Fortunately we still have buildfarm machines that
will flag this.  Seems to be new in CVS HEAD, so no backpatch.

17 years agoGet dirmod.c on the same page as port.h about whether we use pgsymlink
Tom Lane [Thu, 12 Jul 2007 23:28:49 +0000 (23:28 +0000)]
Get dirmod.c on the same page as port.h about whether we use pgsymlink
on Cygwin (answer: we don't).  Also try to unwind the #ifdef spaghetti
a little bit.  Untested but hopefully I didn't break anything.

17 years agoFixup the indentation of a comment that was mangled by pgindent, and
Neil Conway [Thu, 12 Jul 2007 23:25:26 +0000 (23:25 +0000)]
Fixup the indentation of a comment that was mangled by pgindent, and
add dashes to the start/end of the comment block to try to prevent
this happening in the future.

17 years agoSimplify overly-cute array coding to avoid an apparent gcc bug, which
Tom Lane [Thu, 12 Jul 2007 23:10:57 +0000 (23:10 +0000)]
Simplify overly-cute array coding to avoid an apparent gcc bug, which
may or may not be harmless.  Report from Stefan, patch from Heikki.

17 years agoSuppress Sun Studio warnings, per Stefan.
Tom Lane [Thu, 12 Jul 2007 21:27:09 +0000 (21:27 +0000)]
Suppress Sun Studio warnings, per Stefan.

17 years agoSuppress an integer-overflow warning.
Tom Lane [Thu, 12 Jul 2007 21:17:09 +0000 (21:17 +0000)]
Suppress an integer-overflow warning.

17 years agoSome of our port-specific dynloader implementations are careful to
Tom Lane [Thu, 12 Jul 2007 21:13:27 +0000 (21:13 +0000)]
Some of our port-specific dynloader implementations are careful to
define pg_dlsym() as returning a PGFunction pointer, not just any
pointer-to-function.  But many are not.  Suppress compiler warnings
on platforms that aren't careful by inserting explicit casts at the
two call sites that didn't have a cast already.  Per Stefan.

17 years agoCompute max and min int8 values using unsigned arithmetic, in hopes of
Tom Lane [Thu, 12 Jul 2007 21:04:45 +0000 (21:04 +0000)]
Compute max and min int8 values using unsigned arithmetic, in hopes of
suppressing Sun Studio compiler warnings.  Per Stefan.

17 years agoFix some warnings (probably actual bugs) generated by new GSSAPI code
Tom Lane [Thu, 12 Jul 2007 20:36:11 +0000 (20:36 +0000)]
Fix some warnings (probably actual bugs) generated by new GSSAPI code
when built on a 64-bit machine.  Per buildfarm results extracted by Stefan.

17 years agoFix mistaken Assert in adjust_appendrel_attr_needed, per Greg Stark.
Tom Lane [Thu, 12 Jul 2007 18:27:01 +0000 (18:27 +0000)]
Fix mistaken Assert in adjust_appendrel_attr_needed, per Greg Stark.

17 years agoSilence compile warning on win32, per Stefan and Buildfarm.
Magnus Hagander [Thu, 12 Jul 2007 18:15:52 +0000 (18:15 +0000)]
Silence compile warning on win32, per Stefan and Buildfarm.

17 years agoEnable GSSAPI to build using MSVC. Always build GSSAPI when Kerberos is
Magnus Hagander [Thu, 12 Jul 2007 14:43:21 +0000 (14:43 +0000)]
Enable GSSAPI to build using MSVC. Always build GSSAPI when Kerberos is
enabled, because the only Kerberos library supported always contains it.

17 years agoSupport GSSAPI builds where the header is <gssapi.h> and not <gssapi/gssapi.h>,
Magnus Hagander [Thu, 12 Jul 2007 14:36:52 +0000 (14:36 +0000)]
Support GSSAPI builds where the header is <gssapi.h> and not <gssapi/gssapi.h>,
such as OpenBSD (possibly all Heimdal).

Stefan Kaltenbrunner

17 years agoFix freenig of names in Kerberos when using MIT - need to use the
Magnus Hagander [Thu, 12 Jul 2007 14:10:39 +0000 (14:10 +0000)]
Fix freenig of names in Kerberos when using MIT - need to use the
free function provided in the Kerberos library.
This fixes a very hard to track down heap corruption on windows
when using debug runtimes.

17 years agoSuppress a warning that some versions of gcc emit about %x in strftime.
Tom Lane [Wed, 11 Jul 2007 23:15:38 +0000 (23:15 +0000)]
Suppress a warning that some versions of gcc emit about %x in strftime.
Per suggestion from Alvaro.

17 years agoA bunch of GSSAPI fixes per comments from Tom:
Magnus Hagander [Wed, 11 Jul 2007 08:27:33 +0000 (08:27 +0000)]
A bunch of GSSAPI fixes per comments from Tom:

* use elog not ereport for debug
* fix debug levels for some output
* properly check for memory allocation errors in a couple of missed places

17 years agoAdd note that building from CVS requires bison and flex, whereas
Tom Lane [Tue, 10 Jul 2007 23:03:18 +0000 (23:03 +0000)]
Add note that building from CVS requires bison and flex, whereas
building from a distribution tarball does not.

17 years agoFix misspelling.
Tom Lane [Tue, 10 Jul 2007 16:41:01 +0000 (16:41 +0000)]
Fix misspelling.

17 years agoAdd support for GSSAPI authentication.
Magnus Hagander [Tue, 10 Jul 2007 13:14:22 +0000 (13:14 +0000)]
Add support for GSSAPI authentication.

Documentation still being written, will be committed later.

Henry B. Hotz and Magnus Hagander

17 years agoAdjust processSQLNamePattern() so that $ within the pattern is always matched
Tom Lane [Tue, 10 Jul 2007 00:21:31 +0000 (00:21 +0000)]
Adjust processSQLNamePattern() so that $ within the pattern is always matched
literally, whether quoted or not.  Since we allow $ as a character within
identifiers, this behavior is useful, whereas the previous behavior of
treating it as the regexp ending anchor was nearly useless given that the
pattern is automatically anchored anyway.  This affects the arguments of
psql's \d commands as well as pg_dump's -n and -t switches.  Per discussion.

17 years agoFix stddev_pop(numeric) and var_pop(numeric), which were incorrectly producing
Tom Lane [Mon, 9 Jul 2007 16:13:57 +0000 (16:13 +0000)]
Fix stddev_pop(numeric) and var_pop(numeric), which were incorrectly producing
the same outputs as stddev_samp() and var_samp() respectively.

17 years agoFix single-user mode so that interrupts (particularly SIGTERM and
Tom Lane [Mon, 9 Jul 2007 01:15:14 +0000 (01:15 +0000)]
Fix single-user mode so that interrupts (particularly SIGTERM and
SIGQUIT) will be recognized and processed while waiting for input,
rather than only after something has been typed.  Also make SIGQUIT
do the same thing as SIGTERM in single-user mode, ie, do a normal
shutdown and exit.  Since it's relatively easy to provoke SIGQUIT
from the keyboard, people may try that instead of control-D, and we'd
rather this leads to orderly shutdown.  Per report from Leon Mergen
and subsequent discussion.

17 years agoMinor copy-editing.
Tom Lane [Mon, 9 Jul 2007 01:08:09 +0000 (01:08 +0000)]
Minor copy-editing.

17 years agoRemove the pgstat_drop_relation() call from smgr_internal_unlink(), because
Tom Lane [Sun, 8 Jul 2007 22:23:16 +0000 (22:23 +0000)]
Remove the pgstat_drop_relation() call from smgr_internal_unlink(), because
we don't know at that point which relation OID to tell pgstat to forget.
The code was passing the relfilenode, which is incorrect, and could possibly
cause some other relation's stats to be zeroed out.  While we could try to
clean this up, it seems much simpler and more reliable to let the next
invocation of pgstat_vacuum_tabstat() fix things; which indeed is how it
worked before I introduced the buggy code into 8.1.3 and later :-(.
Problem noticed by Itagaki Takahiro, fix is per subsequent discussion.

17 years agoGet rid of client-code dependencies on the exact text of the no-password
Tom Lane [Sun, 8 Jul 2007 19:07:38 +0000 (19:07 +0000)]
Get rid of client-code dependencies on the exact text of the no-password
error message, by using PQconnectionUsedPassword() instead.  Someday
we might be able to localize that error message, but not until this
coding technique has disappeared everywhere.

17 years agoCloser code review for PQconnectionUsedPassword() patch: in particular,
Tom Lane [Sun, 8 Jul 2007 18:28:56 +0000 (18:28 +0000)]
Closer code review for PQconnectionUsedPassword() patch: in particular,
not OK to include postgres_fe.h into libpq-fe.h, hence declare it as
returning int not bool.

17 years agoFix broken markup.
Tom Lane [Sun, 8 Jul 2007 17:47:38 +0000 (17:47 +0000)]
Fix broken markup.

17 years agoRestrict non-superusers to password authenticated connections
Joe Conway [Sun, 8 Jul 2007 17:12:38 +0000 (17:12 +0000)]
Restrict non-superusers to password authenticated connections
to prevent possible escalation of privilege. Provide new SECURITY
DEFINER functions with old behavior, but initially REVOKE ALL
from public for these functions. Per list discussion and design
proposed by Tom Lane. A different approach will be used for
back-branches, committed separately.

17 years agoArrange for the authentication request type to be preserved in
Joe Conway [Sun, 8 Jul 2007 17:11:51 +0000 (17:11 +0000)]
Arrange for the authentication request type to be preserved in
PGconn. Invent a new libpq connection-status function,
PQconnectionUsedPassword() that returns true if the server
demanded a password during authentication, false otherwise.
This may be useful to clients in general, but is immediately
useful to help plug a privilege escalation path in dblink.
Per list discussion and design proposed by Tom Lane.

17 years agoRemove TODO.detail/win32intr, no longer needed.
Bruce Momjian [Sun, 8 Jul 2007 00:42:07 +0000 (00:42 +0000)]
Remove TODO.detail/win32intr, no longer needed.

17 years agoRemove, per Magnus:
Bruce Momjian [Sun, 8 Jul 2007 00:41:11 +0000 (00:41 +0000)]
Remove, per Magnus:

<  o Check WSACancelBlockingCall() for interrupts [win32intr]

17 years agoFix a couple of planner bugs introduced by the new ability to discard
Tom Lane [Sat, 7 Jul 2007 20:46:45 +0000 (20:46 +0000)]
Fix a couple of planner bugs introduced by the new ability to discard
ORDER BY <constant> as redundant.  One is that this means query_planner()
has to canonicalize pathkeys even when the query jointree is empty;
the canonicalization was always a no-op in such cases before, but no more.
Also, we have to guard against thinking that a set-returning function is
"constant" for this purpose.  Add a couple of regression tests for these
evidently under-tested cases.  Per report from Greg Stark and subsequent
experimentation.

17 years agoMove parse.h into src/backend on msvc, which is where it is on Unix.
Magnus Hagander [Sat, 7 Jul 2007 07:43:21 +0000 (07:43 +0000)]
Move parse.h into src/backend on msvc, which is where it is on Unix.
Fixes builds from tarballs where the file is pre-generated.

Yoshiyuki Asaba

17 years agoChanged new \usleep command into \sleep with an optional time unit
Jan Wieck [Fri, 6 Jul 2007 20:17:02 +0000 (20:17 +0000)]
Changed new \usleep command into \sleep with an optional time unit
argument to specify us, ms or s. As per suggestion by Peter E.

Jan

17 years agoAdded a new scripting meta command
Jan Wieck [Fri, 6 Jul 2007 13:36:55 +0000 (13:36 +0000)]
Added a new scripting meta command

    \usleep [milliseconds|:variable]

which can be used in -f scripts to insert a thinking time between
other commands.

Jan

17 years agoFix up hash functions for datetime datatypes so that they don't take
Tom Lane [Fri, 6 Jul 2007 04:16:00 +0000 (04:16 +0000)]
Fix up hash functions for datetime datatypes so that they don't take
unwarranted liberties with int8 vs float8 values for these types.
Specifically, be sure to apply either hashint8 or hashfloat8 depending
on HAVE_INT64_TIMESTAMP.  Per my gripe of even date.

17 years agoImprove description of SPI_scroll_cursor_move's return values, per a
Tom Lane [Thu, 5 Jul 2007 19:15:12 +0000 (19:15 +0000)]
Improve description of SPI_scroll_cursor_move's return values, per a
gripe by Pavel Stehule awhile ago.

17 years agoFix incorrect comment about the timing of AbsorbFsyncRequests() during
Tom Lane [Tue, 3 Jul 2007 14:51:24 +0000 (14:51 +0000)]
Fix incorrect comment about the timing of AbsorbFsyncRequests() during
checkpoint.  The comment claimed that we could do this anytime after
setting the checkpoint REDO point, but actually BufferSync is relying
on the assumption that buffers dumped by other backends will be fsync'd
too.  So we really could not do it any sooner than we are doing it.

17 years agoAdd ALTER VIEW ... RENAME TO, and a RENAME TO clause to ALTER SEQUENCE.
Neil Conway [Tue, 3 Jul 2007 01:30:37 +0000 (01:30 +0000)]
Add ALTER VIEW ... RENAME TO, and a RENAME TO clause to ALTER SEQUENCE.
Sequences and views could previously be renamed using ALTER TABLE, but
this was a repeated source of confusion for users. Update the docs,
and psql tab completion. Patch from David Fetter; various minor fixes
by myself.

17 years ago- Fix the -w (wait) option to work in Windows service mode, per bug #3382.
Magnus Hagander [Mon, 2 Jul 2007 21:58:31 +0000 (21:58 +0000)]
- Fix the -w (wait) option to work in Windows service mode, per bug #3382.
- Prevent the -w option being passed to the postmaster.
- Read the postmaster options file when starting as a Windows service.

Dave Page

17 years agoFix failure to restart Postgres when Linux kernel returns EIDRM for shmctl().
Tom Lane [Mon, 2 Jul 2007 20:11:55 +0000 (20:11 +0000)]
Fix failure to restart Postgres when Linux kernel returns EIDRM for shmctl().

This is a Linux kernel bug that apparently exists in every extant kernel
version: sometimes shmctl() will fail with EIDRM when EINVAL is correct.
We were assuming that EIDRM indicates a possible conflict with pre-existing
backends, and refusing to start the postmaster when this happens.  Fortunately,
there does not seem to be any case where Linux can legitimately return EIDRM
(it doesn't track shmem segments in a way that would allow that), so we can
get away with just assuming that EIDRM means EINVAL on this platform.

Per reports from Michael Fuhr and Jon Lapham --- it's a bit surprising
we have not seen more reports, actually.

17 years agoAdd notes about configuring Visual Studio Express for use with the
Magnus Hagander [Sun, 1 Jul 2007 18:40:37 +0000 (18:40 +0000)]
Add notes about configuring Visual Studio Express for use with the
Platform SDK.

17 years agoReduce the maximum sleep interval in the autovac launcher to 1 second,
Tom Lane [Sun, 1 Jul 2007 18:30:54 +0000 (18:30 +0000)]
Reduce the maximum sleep interval in the autovac launcher to 1 second,
so that it responds to SIGQUIT reasonably promptly even on machines where
SA_RESTART signals restart a sleep from scratch.  (This whole area could
stand some rethinking, but for now make it work like the other processes
do.)  Also some marginal stylistic cleanups.

17 years agoTreat the autovac launcher more like a regular backend, in that we wait
Tom Lane [Sun, 1 Jul 2007 18:28:41 +0000 (18:28 +0000)]
Treat the autovac launcher more like a regular backend, in that we wait
for it to die before telling the bgwriter to initiate shutdown checkpoint.
Since it's connected to shared memory, this seems more prudent than the
alternative of letting it quit asynchronously.  Resolves my complaint
of yesterday about repeated shutdown checkpoints in CVS HEAD.

17 years agoAvoid memory leakage when a series of subtransactions invoke AFTER triggers
Tom Lane [Sun, 1 Jul 2007 17:45:42 +0000 (17:45 +0000)]
Avoid memory leakage when a series of subtransactions invoke AFTER triggers
that are fired at end-of-statement (as is the normal case for foreign keys,
for example).  In this situation the per-subxact deferred trigger context
is always empty when subtransaction exit is reached; so we could free it,
but were not doing so, leading to an intratransaction leak of 8K or more
per subtransaction.  Per off-list example from Viatcheslav Kalinin
subsequent to bug #3418 (his original bug report omitted a foreign key
constraint needed to cause this leak).

Back-patch to 8.2; prior versions were not using per-subxact contexts
for deferred triggers, so did not have this leak.

17 years agoFix comments not updated in recent patch.
Tom Lane [Sun, 1 Jul 2007 02:22:23 +0000 (02:22 +0000)]
Fix comments not updated in recent patch.

17 years agoAdd 'volatile' to suppress 'variable might be clobbered by longjmp'
Tom Lane [Sun, 1 Jul 2007 02:20:59 +0000 (02:20 +0000)]
Add 'volatile' to suppress 'variable might be clobbered by longjmp'
warning emitted by some versions of gcc.

17 years agoImprove logging of checkpoints. Patch by Greg Smith, worked over
Tom Lane [Sat, 30 Jun 2007 19:12:02 +0000 (19:12 +0000)]
Improve logging of checkpoints.  Patch by Greg Smith, worked over
by Heikki and a little bit by me.

17 years agoAvoid crash in interrupted autovacuum worker, caused by leaving the current
Alvaro Herrera [Sat, 30 Jun 2007 04:08:05 +0000 (04:08 +0000)]
Avoid crash in interrupted autovacuum worker, caused by leaving the current
memory context pointing at a context not long lived enough.

Also, create a fake PortalContext where to store the vac_context, if only
to avoid having it be a top-level memory context.

17 years agoArrange for SIGINT in autovacuum workers to cancel the current table and
Alvaro Herrera [Fri, 29 Jun 2007 17:07:39 +0000 (17:07 +0000)]
Arrange for SIGINT in autovacuum workers to cancel the current table and
continue with the schedule.  Change current uses of SIGINT to abort a worker
into SIGTERM, which keeps the old behaviour of terminating the process.

Patch from ITAGAKI Takahiro, with some editorializing of my own.

17 years agoFix computation of PG_VERSION_NUM by configure: remove unnecessary and
Tom Lane [Fri, 29 Jun 2007 16:18:43 +0000 (16:18 +0000)]
Fix computation of PG_VERSION_NUM by configure: remove unnecessary and
unportable backslashes in awk script (per Patrick Welche), and add
brackets to prevent autoconf from mangling sed's regexp (the sed call
here never did what was expected).

17 years agoAdd a note that pg_start_backup will take awhile because of new
Tom Lane [Fri, 29 Jun 2007 15:46:21 +0000 (15:46 +0000)]
Add a note that pg_start_backup will take awhile because of new
distributed checkpoint behavior.  Explain how to work around this
by issuing a manual CHECKPOINT command.  Per discussion with Heikki.

17 years agoFix a passel of ancient bugs in to_char(), including two distinct buffer
Tom Lane [Fri, 29 Jun 2007 01:51:35 +0000 (01:51 +0000)]
Fix a passel of ancient bugs in to_char(), including two distinct buffer
overruns (neither of which seem likely to be exploitable as security holes,
fortunately, since the provoker can't control the data written).  One of
these is due to choosing to stomp on the output of a called function, which
is bad news in any case; make it treat the called functions' results as
read-only.  Avoid some unnecessary palloc/pfree traffic too; it's not
really helpful to free small temporary objects, and again this is presuming
more than it ought to about the nature of the results of called functions.
Per report from Patrick Welche and additional code-reading by Imad.

17 years agoDone:
Bruce Momjian [Thu, 28 Jun 2007 21:55:00 +0000 (21:55 +0000)]
Done:

> * -Reduce checkpoint performance degredation by forcing data to disk

17 years agoFix incorrect tests for undef Perl values in some places in plperl.c.
Tom Lane [Thu, 28 Jun 2007 17:49:59 +0000 (17:49 +0000)]
Fix incorrect tests for undef Perl values in some places in plperl.c.
The correct test for defined-ness is SvOK(sv), not anything involving
SvTYPE.  Per bug #3415 from Matt Taylor.
Back-patch as far as 8.0; no apparent problem in 7.x.