]> granicus.if.org Git - postgresql/log
postgresql
16 years agoDraft release notes for upcoming back-branch updates.
Tom Lane [Wed, 4 Jun 2008 03:16:02 +0000 (03:16 +0000)]
Draft release notes for upcoming back-branch updates.

16 years agoFix initdb to reject a relative path for -X (--xlogdir) argument. This
Tom Lane [Mon, 2 Jun 2008 03:48:00 +0000 (03:48 +0000)]
Fix initdb to reject a relative path for -X (--xlogdir) argument.  This
doesn't work, and the real reason why not is it's unclear where the path
is relative to (initdb's CWD, or the data directory?).  We could make an
arbitrary decision, but it seems best to make the user be unambiguous.
Per gripe from Devrim.

16 years agoUpdate time zone data files to tzdata release 2008c (DST law changes in
Tom Lane [Sun, 1 Jun 2008 18:23:00 +0000 (18:23 +0000)]
Update time zone data files to tzdata release 2008c (DST law changes in
Morocco, Iraq, Choibalsan, Pakistan, Syria, Cuba, Argentina/San_Luis).

16 years agoRefactor SPI_cursor_open/SPI_cursor_open_with_args so that the latter sets
Tom Lane [Sun, 1 Jun 2008 17:32:48 +0000 (17:32 +0000)]
Refactor SPI_cursor_open/SPI_cursor_open_with_args so that the latter sets
the PARAM_FLAG_CONST flag on the parameters that are passed into the portal,
while the former's behavior is unchanged.  This should only affect the case
where the portal is executing an EXPLAIN; it will cause the generated plan to
look more like what would be generated if the portal were actually executing
the command being explained.  Per gripe from Pavel.

16 years agoMarginal improvements to the documentation for PGOPTIONS.
Tom Lane [Sun, 1 Jun 2008 16:23:08 +0000 (16:23 +0000)]
Marginal improvements to the documentation for PGOPTIONS.

16 years agoDone: -Remove psql newline patch when we find out why mingw outputs an extra newline
Andrew Dunstan [Sat, 31 May 2008 15:05:46 +0000 (15:05 +0000)]
Done:  -Remove psql newline patch when we find out why mingw outputs an extra newline

16 years agoCopy refint.so and autoinc.so into the src/test/regress directory during
Tom Lane [Fri, 30 May 2008 00:04:32 +0000 (00:04 +0000)]
Copy refint.so and autoinc.so into the src/test/regress directory during
"make all", and then reference them there during the actual tests.  This
makes the handling of these files more parallel to that of regress.so,
and in particular simplifies use of the regression tests outside the
original build tree.  The PGDG and Red Hat RPMs have been doing this via
patches for a very long time.  Inclusion of the change in core was requested
by Jørgen Austvik of Sun, and I can't see any reason not to.

I attempted to fix the MSVC scripts for this too, but they may need
further tweaking ...

16 years agoImprove the documentation comment for replace(). Robert Treat
Tom Lane [Thu, 29 May 2008 22:48:07 +0000 (22:48 +0000)]
Improve the documentation comment for replace().  Robert Treat

16 years agoTweak libpq to avoid crashing due to incorrect buffer size calculation when
Tom Lane [Thu, 29 May 2008 22:02:44 +0000 (22:02 +0000)]
Tweak libpq to avoid crashing due to incorrect buffer size calculation when
we are on a 64-bit machine (ie, size_t is wider than int) and someone passes
in a query string that approaches or exceeds INT_MAX bytes.  Also, just for
paranoia's sake, guard against similar overflows in sizing the input buffer.

The backend will not in the foreseeable future be prepared to send or receive
strings exceeding 1GB, so I didn't take the more invasive step of switching
all the buffer index variables from int to size_t; though someday we might
want to do that.

I have a suspicion that this is not the only such bug in libpq, but this
fix is enough to take care of the crash reported by Francisco Reyes.

16 years agoFix some bugs introduced by the 8.2-era conversion of cube functions to V1
Tom Lane [Thu, 29 May 2008 18:46:40 +0000 (18:46 +0000)]
Fix some bugs introduced by the 8.2-era conversion of cube functions to V1
calling convention.  cube_inter and cube_distance could attempt to pfree
their input arguments, and cube_dim returned a value from a struct it
might have just pfree'd (which would only really cause a problem in a
debug build, but it's still wrong).  Per bug #4208 and additional code
reading.

In HEAD and 8.3, I also made a batch of cosmetic changes to bring these
functions into line with the preferred coding style for V1 functions,
ie declare and fetch all the arguments at the top so readers can easily
see what they are.

16 years agoAdd description to:
Bruce Momjian [Thu, 29 May 2008 02:04:15 +0000 (02:04 +0000)]
Add description to:

* Add deferred trigger queue file

<   This item involves dumping large queues into files.
>   This item involves dumping large queues into files, or doing some
>   kind of join to process all the triggers, or some bulk operation.

16 years agoAdd URL for:
Bruce Momjian [Thu, 29 May 2008 01:58:29 +0000 (01:58 +0000)]
Add URL for:

* Add deferred trigger queue file

>   http://archives.postgresql.org/pgsql-hackers/2008-05/msg00876.php

16 years agoSet hidden field for guc enum missed in previous commit.
Magnus Hagander [Wed, 28 May 2008 15:22:05 +0000 (15:22 +0000)]
Set hidden field for guc enum missed in previous commit.

16 years agoAdd a field to guc enums to allow hiding of values from display while
Magnus Hagander [Wed, 28 May 2008 09:04:06 +0000 (09:04 +0000)]
Add a field to guc enums to allow hiding of values from display while
still accepting them as input, used to allow alternate syntax for the
same setting.

Alex Hunsaker

16 years agoImprove GRANT documentation to point out that UPDATE and DELETE typically
Tom Lane [Wed, 28 May 2008 00:45:40 +0000 (00:45 +0000)]
Improve GRANT documentation to point out that UPDATE and DELETE typically
require SELECT privilege as well, since you normally need to read existing
column values within such commands.  This behavior is according to spec,
but we'd never documented it before.  Per gripe from Volkan Yazici.

16 years agoRequire bind_textdomain_codeset() not just gettext() to enable NLS support.
Tom Lane [Tue, 27 May 2008 22:18:04 +0000 (22:18 +0000)]
Require bind_textdomain_codeset() not just gettext() to enable NLS support.
GNU gettext before 0.10.36 does not have that function, and is generally too
incomplete to be usable.

16 years agoClarify description of typmod input function, per Jeff Davis.
Tom Lane [Tue, 27 May 2008 18:05:13 +0000 (18:05 +0000)]
Clarify description of typmod input function, per Jeff Davis.

16 years agoExplicitly bind gettext() to the UTF8 locale when in use.
Magnus Hagander [Tue, 27 May 2008 12:24:42 +0000 (12:24 +0000)]
Explicitly bind gettext() to the  UTF8 locale when in use.
This is required on Windows due to the special locale
handling for UTF8 that doesn't change the full environment.

Fixes crash with translated error messages per bugs 4180
and 4196.

Tom Lane

16 years agoAlter the xxx_pattern_ops opclasses to use the regular equality operator of
Tom Lane [Tue, 27 May 2008 00:13:09 +0000 (00:13 +0000)]
Alter the xxx_pattern_ops opclasses to use the regular equality operator of
the associated datatype as their equality member.  This means that these
opclasses can now support plain equality comparisons along with LIKE tests,
thus avoiding the need for an extra index in some applications.  This
optimization was not possible when the pattern opclasses were first introduced,
because we didn't insist that text equality meant bitwise equality; but we
do now, so there is no semantic difference between regular and pattern
equality operators.

I removed the name_pattern_ops opclass altogether, since it's really useless:
name's regular comparisons are just strcmp() and are unlikely to become
something different.  Instead teach indxpath.c that btree name_ops can be
used for LIKE whether or not the locale is C.  This might lead to a useful
speedup in LIKE queries on the system catalogs in non-C locales.

The ~=~ and ~<>~ operators are gone altogether.  (It would have been nice to
keep them for backward compatibility's sake, but since the pg_amop structure
doesn't allow multiple equality operators per opclass, there's no way.)

A not-immediately-obvious incompatibility is that the sort order within
bpchar_pattern_ops indexes changes --- it had been identical to plain
strcmp, but is now trailing-blank-insensitive.  This will impact
in-place upgrades, if those ever happen.

Per discussions a couple months ago.

16 years agoFix an old corner-case bug in set_config_option: push_old_value has to be
Tom Lane [Mon, 26 May 2008 18:54:29 +0000 (18:54 +0000)]
Fix an old corner-case bug in set_config_option: push_old_value has to be
called before, not after, calling the assign_hook if any.  This is because
push_old_value might fail (due to palloc out-of-memory), and in that case
there would be no stack entry to tell transaction abort to undo the GUC
assignment.  Of course the actual assignment to the GUC variable hasn't
happened yet --- but the assign_hook might have altered subsidiary state.
Without a stack entry we won't call it again to make it undo such actions.
So this is necessary to make the world safe for assign_hooks with side
effects.  Per a discussion a couple weeks ago with Magnus.

Back-patch to 8.0.  7.x did not have the problem because it did not have
allocatable stacks of GUC values.

16 years agoAdjust timestamp regression tests to prevent two low-probability failure
Tom Lane [Sun, 25 May 2008 21:51:00 +0000 (21:51 +0000)]
Adjust timestamp regression tests to prevent two low-probability failure
cases.  Recent buildfarm experience shows that it is sometimes possible
to execute several SQL commands in less time than the granularity of
Windows' not-very-high-resolution gettimeofday(), leading to a failure
because the tests expect the value of now() to change and it doesn't.
Also, it was recognized some time ago that the same area of the tests
could fail if local midnight passes between the insertion and the checking
of the values for 'yesterday', 'tomorrow', etc.  Clean all this up per
ideas from myself and Greg Stark.

There remains a window for failure if the transaction block is entered
exactly at local midnight (so that 'now' and 'today' have the same value),
but that seems low-probability enough to live with.

Since the point of this change is mostly to eliminate buildfarm noise,
back-patch to all versions we are still actively testing.

16 years agoLink in keywords file instead of copying it.
Michael Meskes [Wed, 21 May 2008 19:51:01 +0000 (19:51 +0000)]
Link in keywords file instead of copying it.
Use #define/#ifdef instead of sed to fix include files, this should work on Windows too.

16 years agoAdd include directory required after the latest changes to the ecpg build.
Magnus Hagander [Wed, 21 May 2008 18:15:29 +0000 (18:15 +0000)]
Add include directory required after the latest changes to the ecpg build.
Per red Windows buildfarm members.

16 years agoFix function headers not matching prototype in header file, per
Magnus Hagander [Wed, 21 May 2008 16:00:10 +0000 (16:00 +0000)]
Fix function headers not matching prototype in header file, per
compiler warnings on msvc.

16 years agoUse CRITICAL_SECTION instead of Mutexes for thread-locking in libpq on
Magnus Hagander [Wed, 21 May 2008 14:20:48 +0000 (14:20 +0000)]
Use CRITICAL_SECTION instead of Mutexes for thread-locking in libpq on
Windows, for better performance.

Per suggestion from Andrew Chernow, but not his patch since the underlying
code was changed to deal with return values.

16 years agoecpg_keywords.o also depends on preproc.h
Michael Meskes [Wed, 21 May 2008 00:26:26 +0000 (00:26 +0000)]
ecpg_keywords.o also depends on preproc.h

16 years agoSynced parser.
Michael Meskes [Tue, 20 May 2008 23:17:32 +0000 (23:17 +0000)]
Synced parser.
Made ecpg parser use backend provided keyword list.
Changed whenever test so exit value is 0.

16 years agoAdjust -Wl,--asneeded test to avoid using the switch if it breaks
Tom Lane [Tue, 20 May 2008 03:30:22 +0000 (03:30 +0000)]
Adjust -Wl,--asneeded test to avoid using the switch if it breaks
libreadline.  What we will do for compatibility :-(

16 years agoSimplify code in formatting.c now that to upper/lower/initcase do not
Bruce Momjian [Tue, 20 May 2008 01:41:02 +0000 (01:41 +0000)]
Simplify code in formatting.c now that to upper/lower/initcase do not
modify the passed string.

16 years agoDone:
Bruce Momjian [Tue, 20 May 2008 00:35:45 +0000 (00:35 +0000)]
Done:

<  o Use LC_TIME for localized weekday/month names, rather than
>  o -Use LC_TIME for localized weekday/month names, rather than

16 years agoRemove arbitrary 10MB limit on two-phase state file size. It's not that hard
Heikki Linnakangas [Mon, 19 May 2008 18:16:26 +0000 (18:16 +0000)]
Remove arbitrary 10MB limit on two-phase state file size. It's not that hard
to go beoynd 10MB, as demonstrated by Gavin Sharry's example of dropping a
schema with ~25000 objects. The really bogus thing about the limit was that
it was enforced when a state file file was read in, not when it was written,
so you would end up with a prepared transaction that you can't commit or
abort, and the only recourse was to shut down the server and remove the file
by hand.

Raise the limit to MaxAllocSize, and enforce it also when a state file is
written. We could've removed the limit altogether, but reading in a file
larger than MaxAllocSize would fail anyway because we read it into a
palloc'd buffer.

Backpatch down to 8.1, where 2PC and this issue was introduced.

16 years agoMake to_char()'s localized month/day names depend on LC_TIME, not LC_MESSAGES.
Tom Lane [Mon, 19 May 2008 18:08:16 +0000 (18:08 +0000)]
Make to_char()'s localized month/day names depend on LC_TIME, not LC_MESSAGES.

Euler Taveira de Oliveira

16 years agoCoercion sanity check in ri_HashCompareOp failed to allow for enums, as per
Tom Lane [Mon, 19 May 2008 04:14:24 +0000 (04:14 +0000)]
Coercion sanity check in ri_HashCompareOp failed to allow for enums, as per
example from Rod Taylor.  On reflection the correct test here is for any
polymorphic type, not specifically ANYARRAY as in the original coding.

16 years agoMake another try at using -Wl,--as-needed to suppress linking of unnecessary
Tom Lane [Sun, 18 May 2008 20:13:12 +0000 (20:13 +0000)]
Make another try at using -Wl,--as-needed to suppress linking of unnecessary
shared libraries.  We've tried this before and had problems with libreadline
not linking properly on some platforms, but that seems to be a libreadline
bug that may have been fixed by now.  In any case, it's early enough in the
8.4 devel cycle that we can afford to have some transient breakage while
we work out any portability problems.

On Darwin, we try -Wl,-dead_strip_dylibs, which seems to be the equivalent
incantation there.

16 years agoRemove old kluge put in to allow Windows regression tests to succeed, and now
Andrew Dunstan [Sun, 18 May 2008 06:50:08 +0000 (06:50 +0000)]
Remove old kluge put in to allow Windows regression tests to succeed, and now
found to have been made necessary by our skipping tty detection on Windows. Now
that we are doing tty detection on Windows the kluge is unnecessary and wrong.

16 years agookay, looks like ecpg/compatlib needs -lintl too.
Tom Lane [Sun, 18 May 2008 03:24:17 +0000 (03:24 +0000)]
okay, looks like ecpg/compatlib needs -lintl too.
Per buildfarm results.

16 years agoRewrite the warning about non-transaction-safety of TRUNCATE ... RESTART
Tom Lane [Sat, 17 May 2008 23:36:27 +0000 (23:36 +0000)]
Rewrite the warning about non-transaction-safety of TRUNCATE ... RESTART
IDENTITY to be more explicit about the possible hazards.  Per gripe from Neil
and subsequent discussion.  Eventually we may be able to get rid of this
warning, but for now it had better be there.

16 years agoUse isatty() test for pager on Win32; not sure why it was disabled for
Bruce Momjian [Sat, 17 May 2008 23:34:44 +0000 (23:34 +0000)]
Use isatty() test for pager on Win32;  not sure why it was disabled for
that platform.

16 years agoIn psql, test for output to stdout when forcing pager for wide output.
Bruce Momjian [Sat, 17 May 2008 21:40:44 +0000 (21:40 +0000)]
In psql, test for output to stdout when forcing pager for wide output.

16 years agoAnother fix for compiles on an empty 'po' directory.
Bruce Momjian [Sat, 17 May 2008 21:27:12 +0000 (21:27 +0000)]
Another fix for compiles on an empty 'po' directory.

16 years agoDon't call rm with empty file list.
Peter Eisentraut [Sat, 17 May 2008 20:24:05 +0000 (20:24 +0000)]
Don't call rm with empty file list.

16 years agoecpglib needs to link with libintl if it's in use.
Tom Lane [Sat, 17 May 2008 20:23:40 +0000 (20:23 +0000)]
ecpglib needs to link with libintl if it's in use.
Per buildfarm results.

16 years agoImprove pg_regress so that it reports the fact if any test process
Tom Lane [Sat, 17 May 2008 20:02:01 +0000 (20:02 +0000)]
Improve pg_regress so that it reports the fact if any test process
exits with nonzero status.  The Windows part of this is untested ...

16 years agoFix utterly-bogus code for computing row heights. Per crashes on
Tom Lane [Sat, 17 May 2008 17:52:14 +0000 (17:52 +0000)]
Fix utterly-bogus code for computing row heights.  Per crashes on
spoonbill, though one wonders why it didn't misbehave everywhere.
In passing remove some unnecessary modulo calculations.

16 years agoFix a subtle bug exposed by recent wal_sync_method rearrangements.
Tom Lane [Sat, 17 May 2008 17:24:57 +0000 (17:24 +0000)]
Fix a subtle bug exposed by recent wal_sync_method rearrangements.
Formerly, the default value of wal_sync_method was determined inside xlog.c,
but now it is determined inside guc.c.  guc.c was reading xlogdefs.h
without having read <fcntl.h>, leading to wrong determination of
DEFAULT_SYNC_METHOD.  Obviously xlogdefs.h needs to include <fcntl.h>
for itself to ensure stable results.

16 years agoRemove DEFAULT_SYNC_FLAGBIT ... not used anymore.
Tom Lane [Sat, 17 May 2008 16:49:23 +0000 (16:49 +0000)]
Remove DEFAULT_SYNC_FLAGBIT ... not used anymore.

16 years agofix ecpg regression tests broken by over-eager adding of $PostgreSQL$ markers
Andrew Dunstan [Sat, 17 May 2008 02:43:16 +0000 (02:43 +0000)]
fix ecpg regression tests broken by over-eager adding of $PostgreSQL$ markers

16 years agoAdd $PostgreSQL$ markers to a lot of files that were missing them.
Andrew Dunstan [Sat, 17 May 2008 01:28:26 +0000 (01:28 +0000)]
Add $PostgreSQL$ markers to a lot of files that were missing them.

This particular batch was just for *.c and *.h file.

The changes were made with the following 2 commands:

find . \( \( -name 'libstemmer' -o -name 'expected' -o -name 'ppport.h' \) -prune \) -o  \( -name '*.[ch]'  \) \( -exec grep -q '\$PostgreSQL' {} \; -o -print \) | while read file ; do head -n 1 < $file | grep -q '^/\*' && echo $file; done | xargs -l sed -i -e '1s/^\// /' -e '1i/*\n * $PostgreSQL:$ \n *'

find . \( \( -name 'libstemmer' -o -name 'expected' -o -name 'ppport.h' \) -prune \) -o  \( -name '*.[ch]'  \) \( -exec grep -q '\$PostgreSQL' {} \; -o -print \) | xargs -l sed -i -e '1i/*\n * $PostgreSQL:$ \n */'

16 years agoAllow ALTER SEQUENCE START WITH to change the recorded start_value of a
Tom Lane [Sat, 17 May 2008 01:20:39 +0000 (01:20 +0000)]
Allow ALTER SEQUENCE START WITH to change the recorded start_value of a
sequence.  This seems an obvious extension to the recent patch, and it
makes the code noticeably cleaner and more orthogonal.

16 years agoFix declarations of pthread functions, missed in recent commit.
Tom Lane [Fri, 16 May 2008 23:57:42 +0000 (23:57 +0000)]
Fix declarations of pthread functions, missed in recent commit.

16 years agoAdd a RESTART (without parameter) option to ALTER SEQUENCE, allowing a
Tom Lane [Fri, 16 May 2008 23:36:05 +0000 (23:36 +0000)]
Add a RESTART (without parameter) option to ALTER SEQUENCE, allowing a
sequence to be reset to its original starting value.  This requires adding the
original start value to the set of parameters (columns) of a sequence object,
which is a user-visible change with potential compatibility implications;
it also forces initdb.

Also add hopefully-SQL-compatible RESTART/CONTINUE IDENTITY options to
TRUNCATE TABLE.  RESTART IDENTITY executes ALTER SEQUENCE RESTART for all
sequences "owned by" any of the truncated relations.  CONTINUE IDENTITY is
a no-op option.

Zoltan Boszormenyi

16 years agoReduce unnecessary PANIC to ERROR, improve a couple of comments.
Tom Lane [Fri, 16 May 2008 19:15:05 +0000 (19:15 +0000)]
Reduce unnecessary PANIC to ERROR, improve a couple of comments.

16 years agoRecent patch doesn't compile without <limits.h>
Tom Lane [Fri, 16 May 2008 18:35:38 +0000 (18:35 +0000)]
Recent patch doesn't compile without <limits.h>

16 years agoSuppress a possibly-uninitialized-variable warning. (I'm only seeing it
Tom Lane [Fri, 16 May 2008 18:34:51 +0000 (18:34 +0000)]
Suppress a possibly-uninitialized-variable warning.  (I'm only seeing it
on Apple's gcc and not my other machines, but still it seems worth
getting rid of.)

16 years agoImplement error checking for pthreads calls in thread-safe mode. They really
Magnus Hagander [Fri, 16 May 2008 18:30:53 +0000 (18:30 +0000)]
Implement error checking for pthreads calls in thread-safe mode. They really
should always succeed, but in the likely event of a failure we would
previously fall through *without locking* - the new code will exit(1).

Printing the error message on stderr will not work for all applications, but
it's better than nothing at all - and our API doesn't provide a way to return
the error to the caller.

16 years agoSuppress compiler warning induced by recent header file refactoring.
Tom Lane [Fri, 16 May 2008 17:31:17 +0000 (17:31 +0000)]
Suppress compiler warning induced by recent header file refactoring.

16 years agoUpdate contrib for tsearch changes.
Tom Lane [Fri, 16 May 2008 17:26:07 +0000 (17:26 +0000)]
Update contrib for tsearch changes.

16 years agoUpdate psql startup banner to be shorter, suggest "help" for help.
Bruce Momjian [Fri, 16 May 2008 17:17:00 +0000 (17:17 +0000)]
Update psql startup banner to be shorter, suggest "help" for help.

Add a few "help" entries.

Move \g help entry into "General".

Update psql version mismatch warning text.

Joshua D. Drake

16 years agoDone in TODO:
Bruce Momjian [Fri, 16 May 2008 17:03:40 +0000 (17:03 +0000)]
Done in TODO:

>  o -Have \l+ show database size, if permissions allow

16 years agoAdd detection of psql pager to trigger on wide output. Also add pager
Bruce Momjian [Fri, 16 May 2008 16:59:05 +0000 (16:59 +0000)]
Add detection of psql pager to trigger on wide output.  Also add pager
detection for wrapped lines or lines with newlines that need pager to
display.

16 years agoExtend GIN to support partial-match searches, and extend tsquery to support
Tom Lane [Fri, 16 May 2008 16:31:02 +0000 (16:31 +0000)]
Extend GIN to support partial-match searches, and extend tsquery to support
prefix matching using this facility.

Teodor Sigaev and Oleg Bartunov

16 years agoAdd localization support to ecpg.
Peter Eisentraut [Fri, 16 May 2008 15:20:04 +0000 (15:20 +0000)]
Add localization support to ecpg.

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

16 years agoDone:
Bruce Momjian [Fri, 16 May 2008 02:23:44 +0000 (02:23 +0000)]
Done:

PL/pgSQL

>  o -Add CASE capability to language (already in SQL)

16 years agoPersuade GIN to react to control-C in a reasonable amount of time
Tom Lane [Fri, 16 May 2008 01:27:06 +0000 (01:27 +0000)]
Persuade GIN to react to control-C in a reasonable amount of time
while building a GIN index.

16 years agoSupport SQL/PSM-compatible CASE statement in plpgsql.
Tom Lane [Thu, 15 May 2008 22:39:49 +0000 (22:39 +0000)]
Support SQL/PSM-compatible CASE statement in plpgsql.

Pavel Stehule

16 years agoAdd code to eval_const_expressions() to support const-simplification of
Tom Lane [Thu, 15 May 2008 17:37:49 +0000 (17:37 +0000)]
Add code to eval_const_expressions() to support const-simplification of
CoerceViaIO nodes.  This improves the ability of the planner to deal with
cases where the node input is a constant.  Per bug #4170.

16 years agoAdd support for tracking call counts and elapsed runtime for user-defined
Tom Lane [Thu, 15 May 2008 00:17:41 +0000 (00:17 +0000)]
Add support for tracking call counts and elapsed runtime for user-defined
functions.

Note that because this patch changes FmgrInfo, any external C functions
you might be testing with 8.4 will need to be recompiled.

Patch by Martin Pihlak, some editorialization by me (principally, removing
tracking of getrusage() numbers)

16 years agoMove the "instr_time" typedef and associated macros into a new header
Tom Lane [Wed, 14 May 2008 19:10:29 +0000 (19:10 +0000)]
Move the "instr_time" typedef and associated macros into a new header
file portability/instr_time.h, and add a couple more macros to eliminate
some abstraction leakage we formerly had.  Also update psql to use this
header instead of its own copy of nearly the same code.

This commit in itself is just code cleanup and shouldn't change anything.
It lays some groundwork for the upcoming function-stats patch, though.

16 years agoReorgnize psql \? help output, to reduce the size of the "General"
Bruce Momjian [Wed, 14 May 2008 15:30:22 +0000 (15:30 +0000)]
Reorgnize psql \? help output, to reduce the size of the "General"
heading at the top;   broken into more sections now.

16 years agoFix a few warnings that have crept into CVS HEAD.
Bruce Momjian [Wed, 14 May 2008 15:16:27 +0000 (15:16 +0000)]
Fix a few warnings that have crept into CVS HEAD.

16 years agoRemove the special variable for open_sync_bit used in O_SYNC and O_DSYNC
Magnus Hagander [Wed, 14 May 2008 14:02:57 +0000 (14:02 +0000)]
Remove the special variable for open_sync_bit used in O_SYNC and O_DSYNC
modes, replacing it with a call to a function that derives it from the
sync_method variable, now that it has distinct values for these two cases.

This means that assign_xlog_sync_method() no longer changes any settings,
thus fixing the bug introduced in the change to use a guc enum for
wal_sync_method.

16 years agoMake the win32 implementation of getrusage() return EINVAL if being
Magnus Hagander [Wed, 14 May 2008 07:28:13 +0000 (07:28 +0000)]
Make the win32 implementation of getrusage() return EINVAL if being
asked for anything other than RUSAGE_SELF, since it's not supported.

This is never called anywhere in the code today, but might be in
the future.

Not backpatching, since it's not called anywhere today.

16 years agoRemove odd duplicate wording in psql pager section.
Bruce Momjian [Wed, 14 May 2008 04:07:01 +0000 (04:07 +0000)]
Remove odd duplicate wording in psql pager section.

16 years agoImprove plpgsql's RAISE command. It is now possible to attach DETAIL and
Tom Lane [Tue, 13 May 2008 22:10:30 +0000 (22:10 +0000)]
Improve plpgsql's RAISE command.  It is now possible to attach DETAIL and
HINT fields to a user-thrown error message, and to specify the SQLSTATE
error code to use.  The syntax has also been tweaked so that the
Oracle-compatible case "RAISE exception_name" works (though you won't get a
very nice error message if you just write that much).  Lastly, support
the Oracle-compatible syntax "RAISE" with no parameters to re-throw
the current error from within an EXCEPTION block.

In passing, allow the syntax SQLSTATE 'nnnnn' within EXCEPTION lists,
so that there is a way to trap errors with custom SQLSTATE codes.

Pavel Stehule and Tom Lane

16 years agoDon't try to close negative file descriptors, since this can cause
Magnus Hagander [Tue, 13 May 2008 20:53:52 +0000 (20:53 +0000)]
Don't try to close negative file descriptors, since this can cause
crashes on certain platforms. In particular, the MSVC runtime is known
to do this.

Fixes bug #4162, reported and diagnosed by Javier Pimas

16 years agoThis is the patch replace offnum++ by OffsetNumberNext, to be
Bruce Momjian [Tue, 13 May 2008 15:44:08 +0000 (15:44 +0000)]
This is the patch replace offnum++ by OffsetNumberNext, to be
consistent.  OffsetNumberNext() has some casting that makes it useful.

Fujii Masao

16 years agoChange \du to use the new printTable API, improving the situation of very wide
Alvaro Herrera [Tue, 13 May 2008 00:23:17 +0000 (00:23 +0000)]
Change \du to use the new printTable API, improving the situation of very wide
tables.

Brendan Jurd, with some help from me.

16 years agoFix a bug in the previous patch, which caused the title pointer to be used
Alvaro Herrera [Tue, 13 May 2008 00:14:11 +0000 (00:14 +0000)]
Fix a bug in the previous patch, which caused the title pointer to be used
before it was actually set.

16 years agoImprove psql's internal print.c code by introducing an actual print API.
Alvaro Herrera [Mon, 12 May 2008 22:59:58 +0000 (22:59 +0000)]
Improve psql's internal print.c code by introducing an actual print API.
Provides for better code readability, but mainly this is infrastructure changes
to allow further changes such as arbitrary footers on printed tables.  Also,
the translation status of each element in the table is more easily customized.

Brendan Jurd, with some editorialization by me.

16 years agoTodo done:
Bruce Momjian [Mon, 12 May 2008 20:50:52 +0000 (20:50 +0000)]
Todo done:

> * -Improve dead row detection during multi-statement transactions usage

16 years agoImprove snapshot manager by keeping explicit track of snapshots.
Alvaro Herrera [Mon, 12 May 2008 20:02:02 +0000 (20:02 +0000)]
Improve snapshot manager by keeping explicit track of snapshots.

There are two ways to track a snapshot: there's the "registered" list, which
is used for arbitrary long-lived snapshots; and there's the "active stack",
which is used for the snapshot that is considered "active" at any time.
This also allows users of snapshots to stop worrying about snapshot memory
allocation and freeing, and about using PG_TRY blocks around ActiveSnapshot
assignment.  This is all done automatically now.

As a consequence, this allows us to reset MyProc->xmin when there are no
more snapshots registered in the current backend, reducing the impact that
long-running transactions have on VACUUM.

16 years agoFix breakage by the wal_sync_method patch in installations that use
Magnus Hagander [Mon, 12 May 2008 19:45:23 +0000 (19:45 +0000)]
Fix breakage by the wal_sync_method patch in installations that use
O_DSYNC (specifically this broke all the Windows buildfarm members)

16 years agoCheck for non-existant connection in prepare statement handling.
Michael Meskes [Mon, 12 May 2008 16:29:04 +0000 (16:29 +0000)]
Check for non-existant connection in prepare statement handling.
Do not close files that weren't opened.

16 years agoDocument that "ROLLBACK TO savepoint" does not un-close cursors.
Alvaro Herrera [Mon, 12 May 2008 16:23:26 +0000 (16:23 +0000)]
Document that "ROLLBACK TO savepoint" does not un-close cursors.

16 years agoPut back bufmgr.h in bufpage.h -- it is needed by some macros.
Alvaro Herrera [Mon, 12 May 2008 16:06:10 +0000 (16:06 +0000)]
Put back bufmgr.h in bufpage.h -- it is needed by some macros.

Remove #include bufmgr.h from (most?) source files which already include
bufpage.h.

16 years agoReport which WAL sync method we are trying to change *to* when it fails,
Magnus Hagander [Mon, 12 May 2008 14:27:47 +0000 (14:27 +0000)]
Report which WAL sync method we are trying to change *to* when it fails,
not which one we had before (that worked, and thus is completley irrelevant)

16 years agoConvert wal_sync_method to guc enum.
Magnus Hagander [Mon, 12 May 2008 08:35:05 +0000 (08:35 +0000)]
Convert wal_sync_method to guc enum.

16 years agoRestructure some header files a bit, in particular heapam.h, by removing some
Alvaro Herrera [Mon, 12 May 2008 00:00:54 +0000 (00:00 +0000)]
Restructure some header files a bit, in particular heapam.h, by removing some
unnecessary #include lines in it.  Also, move some tuple routine prototypes and
macros to htup.h, which allows removal of heapam.h inclusion from some .c
files.

For this to work, a new header file access/sysattr.h needed to be created,
initially containing attribute numbers of system columns, for pg_dump usage.

While at it, make contrib ltree, intarray and hstore header files more
consistent with our header style.

16 years agoRemaining pieces of fix for contrib makefiles
Andrew Dunstan [Sat, 10 May 2008 15:30:11 +0000 (15:30 +0000)]
Remaining pieces of fix for contrib makefiles

16 years agoRestore psql's former behavior that padding spaces to the right of the last
Tom Lane [Sat, 10 May 2008 03:31:58 +0000 (03:31 +0000)]
Restore psql's former behavior that padding spaces to the right of the last
output column are not emitted.  (That change already caused more noise in
the regression test output files than I would like.)  Provide some needed
editorial help for comments, clean up code formatting.

16 years agoAdd URL for:
Bruce Momjian [Sat, 10 May 2008 01:50:51 +0000 (01:50 +0000)]
Add URL for:

        o Allow an existing index to be marked as a table's primary key
>
>    http://archives.postgresql.org/pgsql-hackers/2008-04/msg00500.php

16 years agoAdd to TODO:
Bruce Momjian [Sat, 10 May 2008 01:50:14 +0000 (01:50 +0000)]
Add to TODO:

>  o Allow an existing index to be marked as a table's primary key

16 years agoAdd URL for:
Bruce Momjian [Sat, 10 May 2008 01:40:09 +0000 (01:40 +0000)]
Add URL for:

* Add column to pg_stat_activity that shows the progress of long-running
  commands like CREATE INDEX and VACUUM
>
>   http://archives.postgresql.org/pgsql-patches/2008-04/msg00203.php
>

16 years agoUpdate alternative output regression files for new float8 test of power().
Bruce Momjian [Sat, 10 May 2008 01:04:59 +0000 (01:04 +0000)]
Update alternative output regression files for new float8 test of power().

16 years agoAdd TODO item:
Bruce Momjian [Sat, 10 May 2008 00:33:33 +0000 (00:33 +0000)]
Add TODO item:

>  o Clear table counters on TRUNCATE
>
>    http://archives.postgresql.org/pgsql-hackers/2008-04/msg00169.php

16 years agoTODOs done:
Bruce Momjian [Sat, 10 May 2008 00:25:08 +0000 (00:25 +0000)]
TODOs done:

>  o -Prevent parent tables from altering or dropping constraints
>  o -Prevent child tables from altering or dropping constraints

16 years agoChange the rules for inherited CHECK constraints to be essentially the same
Tom Lane [Fri, 9 May 2008 23:32:05 +0000 (23:32 +0000)]
Change the rules for inherited CHECK constraints to be essentially the same
as those for inherited columns; that is, it's no longer allowed for a child
table to not have a check constraint matching one that exists on a parent.
This satisfies the principle of least surprise (rows selected from the parent
will always appear to meet its check constraints) and eliminates some
longstanding bogosity in pg_dump, which formerly had to guess about whether
check constraints were really inherited or not.

The implementation involves adding conislocal and coninhcount columns to
pg_constraint (paralleling attislocal and attinhcount in pg_attribute)
and refactoring various ALTER TABLE actions to be more like those for
columns.

Alex Hunsaker, Nikhil Sontakke, Tom Lane

16 years agoAdjust power() error messages to be more descriptive.
Bruce Momjian [Fri, 9 May 2008 21:31:23 +0000 (21:31 +0000)]
Adjust power() error messages to be more descriptive.

16 years agoImprove logic for finding object files on OBJS lines in contrib Makefiles. If this...
Andrew Dunstan [Fri, 9 May 2008 16:01:05 +0000 (16:01 +0000)]
Improve logic for finding object files on OBJS lines in contrib Makefiles. If this unbreaks buildfarm mastodon, apply everywhere.

16 years agoGive a warning if -s switch is used with a non-custom pgbench test.
Tom Lane [Fri, 9 May 2008 15:53:07 +0000 (15:53 +0000)]
Give a warning if -s switch is used with a non-custom pgbench test.
Also, clean up the code that assigned the scale into :scale variables.
Greg Smith and Tom Lane