]> granicus.if.org Git - postgresql/log
postgresql
17 years agoFix backend crash in parsing incorrect tsquery.
Teodor Sigaev [Mon, 12 Feb 2007 14:14:33 +0000 (14:14 +0000)]
Fix backend crash in parsing incorrect tsquery.

Per report from Jon Rosebaugh <jon@inklesspen.com>

17 years agoAdd support for optionally escaping periods when converting SQL identifiers
Peter Eisentraut [Sun, 11 Feb 2007 22:18:16 +0000 (22:18 +0000)]
Add support for optionally escaping periods when converting SQL identifiers
to XML names, which will be required for supporting XML export.

17 years agoFix another erroneous =-for-:= substitution.
Tom Lane [Sun, 11 Feb 2007 19:31:45 +0000 (19:31 +0000)]
Fix another erroneous =-for-:= substitution.

17 years agoFixed multibyte handling as reported by <harada.toshi@oss.ntt.co.jp>.
Michael Meskes [Sun, 11 Feb 2007 15:18:17 +0000 (15:18 +0000)]
Fixed multibyte handling as reported by <harada.toshi@oss.ntt.co.jp>.

17 years agoFix for early log messages during postmaster startup getting lost when
Magnus Hagander [Sun, 11 Feb 2007 11:59:26 +0000 (11:59 +0000)]
Fix for early log messages during postmaster startup getting lost when
running as a service on Win32.

Per report from Harald Armin Massa.

17 years agoImprove documentation for CREATE CONSTRAINT TRIGGER.
Tom Lane [Sat, 10 Feb 2007 20:43:59 +0000 (20:43 +0000)]
Improve documentation for CREATE CONSTRAINT TRIGGER.

17 years agoFix pg_standby to build on msvc.
Magnus Hagander [Sat, 10 Feb 2007 19:52:45 +0000 (19:52 +0000)]
Fix pg_standby to build on msvc.

17 years agoAdd proper mapping of boolean type data to XML Schema.
Peter Eisentraut [Sat, 10 Feb 2007 18:47:41 +0000 (18:47 +0000)]
Add proper mapping of boolean type data to XML Schema.

17 years agoStrNCpy -> strlcpy (not complete)
Peter Eisentraut [Sat, 10 Feb 2007 14:58:55 +0000 (14:58 +0000)]
StrNCpy -> strlcpy (not complete)

17 years agoPut back some not-so-unnecessary-as-all-that := usages. Per buildfarm.
Tom Lane [Sat, 10 Feb 2007 04:26:24 +0000 (04:26 +0000)]
Put back some not-so-unnecessary-as-all-that := usages.  Per buildfarm.

17 years agoHm, seems my hack on rowtypes regression test has made its output row
Tom Lane [Sat, 10 Feb 2007 04:18:32 +0000 (04:18 +0000)]
Hm, seems my hack on rowtypes regression test has made its output row
order platform-specific.  Add an ORDER BY clause to stop buildfarm
failures.

17 years agoUnbreak the SGML doc build: ":=" is needed to assign to variables if
Neil Conway [Fri, 9 Feb 2007 20:40:13 +0000 (20:40 +0000)]
Unbreak the SGML doc build: ":=" is needed to assign to variables if
the RHS of the assignment expands to a reference to the LHS.

17 years agoMinor tweak to make rowtypes regression test run faster. We don't
Tom Lane [Fri, 9 Feb 2007 20:17:59 +0000 (20:17 +0000)]
Minor tweak to make rowtypes regression test run faster.  We don't
currently have any better strategy for this query than re-running the
sub-select over and over; it seems unlikely that doing so 10000 times
is a more useful test than doing it a few dozen times.

17 years agoRemove useless CPPFLAGS.
Peter Eisentraut [Fri, 9 Feb 2007 17:24:33 +0000 (17:24 +0000)]
Remove useless CPPFLAGS.

17 years agoAdd $PostgreSQL$ marker to contrib makefiles.
Peter Eisentraut [Fri, 9 Feb 2007 17:04:00 +0000 (17:04 +0000)]
Add $PostgreSQL$ marker to contrib makefiles.

17 years agoCall pgstat_drop_database during DROP DATABASE, so that any stats file
Tom Lane [Fri, 9 Feb 2007 16:12:19 +0000 (16:12 +0000)]
Call pgstat_drop_database during DROP DATABASE, so that any stats file
entries for the victim database go away sooner rather than later.  We already
did the equivalent thing at the per-relation level, not sure why it's not
been done for whole databases.  With this change, pgstat_vacuum_tabstat
should usually not find anything to do; though we still need it as a backstop
in case DROPDB or TABPURGE messages get lost under load.

17 years agoReplace useless uses of := by = in makefiles.
Peter Eisentraut [Fri, 9 Feb 2007 15:56:00 +0000 (15:56 +0000)]
Replace useless uses of := by = in makefiles.

17 years agoRemove blank lines in code.
Bruce Momjian [Fri, 9 Feb 2007 04:17:58 +0000 (04:17 +0000)]
Remove blank lines in code.

17 years agoUpdate FAQ for new 24-byte header, down from 28.
Bruce Momjian [Fri, 9 Feb 2007 03:43:22 +0000 (03:43 +0000)]
Update FAQ for new 24-byte header, down from 28.

17 years agoAdd blank line.
Bruce Momjian [Fri, 9 Feb 2007 03:39:59 +0000 (03:39 +0000)]
Add blank line.

17 years agoAdd URL for:
Bruce Momjian [Fri, 9 Feb 2007 03:39:39 +0000 (03:39 +0000)]
Add URL for:

<
>   http://archives.postgresql.org/pgsql-hackers/2007-02/msg00435.php

17 years agoDone!
Bruce Momjian [Fri, 9 Feb 2007 03:37:45 +0000 (03:37 +0000)]
Done!

< * Merge xmin/xmax/cmin/cmax back into three header fields
<
<   Before subtransactions, there used to be only three fields needed to
<   store these four values. This was possible because only the current
<   transaction looks at the cmin/cmax values. If the current transaction
<   created and expired the row the fields stored where xmin (same as
<   xmax), cmin, cmax, and if the transaction was expiring a row from a
<   another transaction, the fields stored were xmin (cmin was not
<   needed), xmax, and cmax. Such a system worked because a transaction
<   could only see rows from another completed transaction. However,
<   subtransactions can see rows from outer transactions, and once the
<   subtransaction completes, the outer transaction continues, requiring
<   the storage of all four fields. With subtransactions, an outer
<   transaction can create a row, a subtransaction expire it, and when the
<   subtransaction completes, the outer transaction still has to have
<   proper visibility of the row's cmin, for example, for cursors.
<
<   One possible solution is to create a phantom cid which represents a
<   cmin/cmax pair and is stored in local memory.  Another idea is to
<   store both cmin and cmax only in local memory.
<
> * -Merge xmin/xmax/cmin/cmax back into three header fields

17 years agoCombine cmin and cmax fields of HeapTupleHeaders into a single field, by
Tom Lane [Fri, 9 Feb 2007 03:35:35 +0000 (03:35 +0000)]
Combine cmin and cmax fields of HeapTupleHeaders into a single field, by
keeping private state in each backend that has inserted and deleted the same
tuple during its current top-level transaction.  This is sufficient since
there is no need to be able to determine the cmin/cmax from any other
transaction.  This gets us back down to 23-byte headers, removing a penalty
paid in 8.0 to support subtransactions.  Patch by Heikki Linnakangas, with
minor revisions by moi, following a design hashed out awhile back on the
pghackers list.

17 years agoRemove blank line from C code.
Bruce Momjian [Fri, 9 Feb 2007 03:15:48 +0000 (03:15 +0000)]
Remove blank line from C code.

17 years agoUpdate:
Bruce Momjian [Fri, 9 Feb 2007 01:29:48 +0000 (01:29 +0000)]
Update:
< * Consider placing all sequences in a single table
> * Consider placing all sequences in a single table, or create a system
>   view

17 years agoUpdate:
Bruce Momjian [Fri, 9 Feb 2007 00:34:31 +0000 (00:34 +0000)]
Update:

< * Consider placing all sequences in a single table, now that system
<   tables are full transactional
> * Consider placing all sequences in a single table

17 years agoAdd:
Bruce Momjian [Fri, 9 Feb 2007 00:32:15 +0000 (00:32 +0000)]
Add:

> * Consider placing all sequences in a single table, now that system
>   tables are full transactional

17 years agoAdd URL for:
Bruce Momjian [Thu, 8 Feb 2007 21:51:50 +0000 (21:51 +0000)]
Add URL for:

* Add support for SQL-standard GENERATED/IDENTITY columns
>   http://archives.postgresql.org/pgsql-hackers/2006-08/msg00038.php

17 years agoAdd missing #define for mingw, per Magnus.
Tom Lane [Thu, 8 Feb 2007 19:48:28 +0000 (19:48 +0000)]
Add missing #define for mingw, per Magnus.

17 years agoFix an ancient logic error in plpgsql's exec_stmt_block: it thought it could
Tom Lane [Thu, 8 Feb 2007 18:37:30 +0000 (18:37 +0000)]
Fix an ancient logic error in plpgsql's exec_stmt_block: it thought it could
get away with not (re)initializing a local variable if the variable is marked
"isconst" and not "isnull".  Unfortunately it makes this decision after having
already freed the old value, meaning that something like

   for i in 1..10 loop
     declare c constant text := 'hi there';

leads to subsequent accesses to freed memory, and hence probably crashes.
(In particular, this is why Asif Ali Rehman's bug leads to crash and not
just an unexpectedly-NULL value for SQLERRM: SQLERRM is marked CONSTANT
and so triggers this error.)

The whole thing seems wrong on its face anyway: CONSTANT means that you can't
change the variable inside the block, not that the initializer expression is
guaranteed not to change value across successive block entries.  Hence,
remove the "optimization" instead of trying to fix it.

17 years agoRearrange use of plpgsql_add_initdatums() so that only the parsing of a
Tom Lane [Thu, 8 Feb 2007 18:37:14 +0000 (18:37 +0000)]
Rearrange use of plpgsql_add_initdatums() so that only the parsing of a
DECLARE section needs to know about it.  Formerly, everyplace besides DECLARE
that created variables needed to do "plpgsql_add_initdatums(NULL)" to prevent
those variables from being sucked up as part of a subsequent DECLARE block.
This is obviously error-prone, and in fact the SQLSTATE/SQLERRM patch had
failed to do it for those two variables, leading to the bug recently exhibited
by Asif Ali Rehman: a DECLARE within an exception handler tried to reinitialize
SQLERRM.

Although the SQLSTATE/SQLERRM patch isn't in any pre-8.1 branches, and so
I can't point to a demonstrable failure there, it seems wise to back-patch
this into the older branches anyway, just to keep the logic similar to HEAD.

17 years agoFix bug when localized to_char() day or month names were incorectly
Bruce Momjian [Thu, 8 Feb 2007 18:19:33 +0000 (18:19 +0000)]
Fix bug when localized to_char() day or month names were incorectly
trnasformed to lower or upper string.

Pavel Stehule

17 years agoThis patch fixes shared_preload_libraries on Windows hosts. It forces
Bruce Momjian [Thu, 8 Feb 2007 15:46:04 +0000 (15:46 +0000)]
This patch fixes shared_preload_libraries on Windows hosts.  It forces
ach backend to re-load all shared_preload_libraries.

Korry Douglas

17 years agoAdd lock matrix to documentation.
Bruce Momjian [Thu, 8 Feb 2007 15:32:11 +0000 (15:32 +0000)]
Add lock matrix to documentation.

Teodor Sigaev

17 years agoWin32 regression test fixes:
Bruce Momjian [Thu, 8 Feb 2007 15:28:58 +0000 (15:28 +0000)]
Win32 regression test fixes:

For win32 in general, this makes it possible to run the regression tests
as an admin user by using the same restricted token method that's used
by pg_ctl and initdb.

For vc++, it adds building of pg_regress.exe, adds a resultmap, and
fixes how it runs the install.

Magnus Hagander

17 years agoAdd /contrib/pg_standby:
Bruce Momjian [Thu, 8 Feb 2007 15:16:19 +0000 (15:16 +0000)]
Add /contrib/pg_standby:

  pg_standby is a production-ready program that can be used to
  create a Warm Standby server. Other configuration is required
  as well, all of which is described in the main server manual.

Simon Riggs

17 years agoAdd /contrib/pg_standby:
Bruce Momjian [Thu, 8 Feb 2007 15:09:47 +0000 (15:09 +0000)]
Add /contrib/pg_standby:

  pg_standby is a production-ready program that can be used to
  create a Warm Standby server. Other configuration is required
  as well, all of which is described in the main server manual.

Simon Riggs

17 years agoFix reference-after-free in the new btree page split code, as reported by
Alvaro Herrera [Thu, 8 Feb 2007 13:52:55 +0000 (13:52 +0000)]
Fix reference-after-free in the new btree page split code, as reported by
the buildfarm via Stefan Kaltenbrunner.

Patch from Heikki Linnakangas.

17 years agoNormalize fgets() calls to use sizeof() for calculating the buffer size
Peter Eisentraut [Thu, 8 Feb 2007 11:10:27 +0000 (11:10 +0000)]
Normalize fgets() calls to use sizeof() for calculating the buffer size
where possible, and fix some sites that apparently thought that fgets()
will overwrite the buffer by one byte.

Also add some strlcpy() to eliminate some weird memory handling.

17 years agoReduce WAL activity for page splits:
Bruce Momjian [Thu, 8 Feb 2007 05:05:53 +0000 (05:05 +0000)]
Reduce WAL activity for page splits:

> Currently, an index split writes all the data on the split page to
> WAL. That's a lot of WAL traffic. The tuples that are copied to the
> right page need to be WAL logged, but the tuples that stay on the
> original page don't.

Heikki Linnakangas

17 years agoCheck if the role exists before doing more complex ident and Kerberos
Bruce Momjian [Thu, 8 Feb 2007 04:52:18 +0000 (04:52 +0000)]
Check if the role exists before doing more complex ident and Kerberos
authentication checks in the backend.

Gavin Sherry

17 years agoUpdate URL for "Generalized Partial Indexes" paper to point to a cached
Bruce Momjian [Thu, 8 Feb 2007 04:31:37 +0000 (04:31 +0000)]
Update URL for "Generalized Partial Indexes" paper to point to a cached
version.

Backpatch to 8.2.X.

17 years agoDocument that wal_sync_method open_* methods use O_DIRECT, if available.
Bruce Momjian [Thu, 8 Feb 2007 03:56:42 +0000 (03:56 +0000)]
Document that wal_sync_method open_* methods use O_DIRECT, if available.

Backpatch to 8.2.X.

17 years agoFix bug in our code when using to_timestamp() or to_date() without "TM".
Bruce Momjian [Thu, 8 Feb 2007 03:22:28 +0000 (03:22 +0000)]
Fix bug in our code when using to_timestamp() or to_date() without "TM".

Assume "TM" when input fields are variable-length, like month or day
names.  This matches Oracle behavior.

17 years agoAdd a function pg_stat_clear_snapshot() that discards any statistics snapshot
Tom Lane [Wed, 7 Feb 2007 23:11:30 +0000 (23:11 +0000)]
Add a function pg_stat_clear_snapshot() that discards any statistics snapshot
already collected in the current transaction; this allows plpgsql functions to
watch for stats updates even though they are confined to a single transaction.
Use this instead of the previous kluge involving pg_stat_file() to wait for
the stats collector to update in the stats regression test.  Internally,
decouple storage of stats snapshots from transaction boundaries; they'll
now stick around until someone calls pgstat_clear_snapshot --- which xact.c
still does at transaction end, to maintain the previous behavior.  This makes
the logic a lot cleaner, at the price of a couple dozen cycles per transaction
exit.

17 years agoModify the stats regression test to delay until the stats file actually
Tom Lane [Wed, 7 Feb 2007 18:34:56 +0000 (18:34 +0000)]
Modify the stats regression test to delay until the stats file actually
changes (with an upper limit of 30 seconds), and record the delay time in
the postmaster log.  This should give us some info about what's happening
with the intermittent stats failures in buildfarm.  After an idea of
Andrew Dunstan's.

17 years agoRemove the xlog-centric "database system is ready" message and replace it with
Tom Lane [Wed, 7 Feb 2007 16:44:48 +0000 (16:44 +0000)]
Remove the xlog-centric "database system is ready" message and replace it with
"database system is ready to accept connections", which is issued by the
postmaster when it really is ready to accept connections.  Per proposal from
Markus Schiltknecht and subsequent discussion.

17 years agoThe VC++ build needs to compile the new strlcat.c file. Patch from Magnus
Alvaro Herrera [Wed, 7 Feb 2007 13:37:42 +0000 (13:37 +0000)]
The VC++ build needs to compile the new strlcat.c file.  Patch from Magnus
Hagander.

17 years agoStamp bug template for 8.3.
Bruce Momjian [Wed, 7 Feb 2007 05:14:08 +0000 (05:14 +0000)]
Stamp bug template for 8.3.

17 years agoStamp releases notes for 8.2.3, 8.1.8, 8.0.12.
Bruce Momjian [Wed, 7 Feb 2007 04:22:44 +0000 (04:22 +0000)]
Stamp releases notes for 8.2.3, 8.1.8, 8.0.12.

17 years agoUpdate for 8.2.3.
Bruce Momjian [Wed, 7 Feb 2007 03:16:22 +0000 (03:16 +0000)]
Update for 8.2.3.

17 years agoReplace some strncpy() by strlcpy().
Peter Eisentraut [Wed, 7 Feb 2007 00:52:35 +0000 (00:52 +0000)]
Replace some strncpy() by strlcpy().

17 years agoUse memcpy() instead of strncpy() for copying into varlena structures.
Peter Eisentraut [Wed, 7 Feb 2007 00:32:15 +0000 (00:32 +0000)]
Use memcpy() instead of strncpy() for copying into varlena structures.

17 years agoAdd strlcat() from OpenBSD, to be used for replacing strncat and other
Peter Eisentraut [Wed, 7 Feb 2007 00:28:55 +0000 (00:28 +0000)]
Add strlcat() from OpenBSD, to be used for replacing strncat and other
strange coding practices.

17 years agoFix an error in the original coding of holdable cursors: PersistHoldablePortal
Tom Lane [Tue, 6 Feb 2007 22:49:24 +0000 (22:49 +0000)]
Fix an error in the original coding of holdable cursors: PersistHoldablePortal
thought that it didn't have to reposition the underlying tuplestore if the
portal is atEnd.  But this is not so, because tuplestores have separate read
and write cursors ... and the read cursor hasn't moved from the start.
This mistake explains bug #2970 from William Zhang.

Note: the coding here is pretty inefficient, but given that no one has noticed
this bug until now, I'd say hardly anyone uses the case where the cursor has
been advanced before being persisted.  So maybe it's not worth worrying about.

17 years agoUpdate for 8.2.2 as most recent release.
Bruce Momjian [Tue, 6 Feb 2007 18:36:26 +0000 (18:36 +0000)]
Update for 8.2.2 as most recent release.

17 years agoUpdate workding for daylight savings time.
Bruce Momjian [Tue, 6 Feb 2007 18:33:20 +0000 (18:33 +0000)]
Update workding for daylight savings time.

17 years agoUpdate timezone FAQ item:
Bruce Momjian [Tue, 6 Feb 2007 18:31:26 +0000 (18:31 +0000)]
Update timezone FAQ item:

    <P>USA saving time changes are included in PostgreSQL release 8.0.[4+],
    and all later major releases, e.g. 8.1.  Canada and Western Australia
    changes are included in 8.0.[10+], 8.1.[6+], and all later major
    releases.  PostgreSQL releases prior to 8.0 use the operating system's
    timezone database for daylight saving information.</P>

17 years agoRemove typmod checking from the recent security-related patches. It turns
Tom Lane [Tue, 6 Feb 2007 17:35:20 +0000 (17:35 +0000)]
Remove typmod checking from the recent security-related patches.  It turns
out that ExecEvalVar and friends don't necessarily have access to a tuple
descriptor with correct typmod: it definitely can contain -1, and possibly
might contain other values that are different from the Var's value.
Arguably this should be cleaned up someday, but it's not a simple change,
and in any case typmod discrepancies don't pose a security hazard.
Per reports from numerous people :-(

I'm not entirely sure whether the failure can occur in 8.0 --- the simple
test cases reported so far don't trigger it there.  But back-patch the
change all the way anyway.

17 years agoSplit apart entries, one done now:
Bruce Momjian [Tue, 6 Feb 2007 16:44:14 +0000 (16:44 +0000)]
Split apart entries, one done now:

* -Move NAMEDATALEN from postgres_ext.h to pg_config_manual.h
* Consider making NAMEDATALEN more configurable in future releases

17 years agoFix typo in comment.
Tom Lane [Tue, 6 Feb 2007 16:20:23 +0000 (16:20 +0000)]
Fix typo in comment.

17 years agoRemove some dead code, per Heikki.
Tom Lane [Tue, 6 Feb 2007 14:55:11 +0000 (14:55 +0000)]
Remove some dead code, per Heikki.

17 years agoMove NAMEDATALEN definition from postgres_ext.h to pg_config_manual.h. It
Peter Eisentraut [Tue, 6 Feb 2007 09:16:08 +0000 (09:16 +0000)]
Move NAMEDATALEN definition from postgres_ext.h to pg_config_manual.h. It
used to be part of libpq's exported interface many releases ago, but now
it's no longer necessary to make it accessible to clients.

17 years agoFix a performance regression in 8.2: optimization of MIN/MAX into indexscans
Tom Lane [Tue, 6 Feb 2007 06:50:26 +0000 (06:50 +0000)]
Fix a performance regression in 8.2: optimization of MIN/MAX into indexscans
had stopped working for tables buried inside views or sub-selects.  This is
because I had gotten rid of the simplify_jointree() preprocessing step, and
optimize_minmax_aggregates() wasn't smart enough to deal with a non-canonical
FromExpr.  Per gripe from Bill Howe.

17 years agoDocs updates for cross-type hashing.
Tom Lane [Tue, 6 Feb 2007 04:38:31 +0000 (04:38 +0000)]
Docs updates for cross-type hashing.

17 years agoCome to think of it, we should check that commutator pairs have the same
Tom Lane [Tue, 6 Feb 2007 03:52:59 +0000 (03:52 +0000)]
Come to think of it, we should check that commutator pairs have the same
merges/hashes property settings.

17 years agoNot only did we agree that this 'hint' doesn't belong here, but the
Tom Lane [Tue, 6 Feb 2007 03:03:11 +0000 (03:03 +0000)]
Not only did we agree that this 'hint' doesn't belong here, but the
markup's broken.  So just remove it...

17 years agoAdd support for cross-type hashing in hashed subplans (hashed IN/NOT IN cases
Tom Lane [Tue, 6 Feb 2007 02:59:15 +0000 (02:59 +0000)]
Add support for cross-type hashing in hashed subplans (hashed IN/NOT IN cases
that aren't turned into true joins).  Since this is the last missing bit of
infrastructure, go ahead and fill out the hash integer_ops and float_ops
opfamilies with cross-type operators.  The operator family project is now
DONE ... er, except for documentation ...

17 years agoTrim down environment variable instructions for Win32, backpatch to 8.2.X.
Bruce Momjian [Mon, 5 Feb 2007 22:18:11 +0000 (22:18 +0000)]
Trim down environment variable instructions for Win32, backpatch to 8.2.X.

17 years agoUpdated TODO item:
Bruce Momjian [Mon, 5 Feb 2007 17:17:13 +0000 (17:17 +0000)]
Updated TODO item:

>  o Add a \set variable to control whether \s displays line numbers

>    Another option is to add \# which lists line numbers, and
>    allows command execution.
>    http://archives.postgresql.org/pgsql-hackers/2006-12/msg00255.php

17 years agoModify:
Bruce Momjian [Mon, 5 Feb 2007 16:49:13 +0000 (16:49 +0000)]
Modify:

<  o Add \# to list command history like \s, but with line numbers
>  o Add \# to list and execute command history

17 years agoPass modern COPY syntax to backend, since copy (query) does not accept old syntax...
Andrew Dunstan [Mon, 5 Feb 2007 15:22:18 +0000 (15:22 +0000)]
Pass modern COPY syntax to backend, since copy (query) does not accept old syntax. Per complaint from Michael Fuhr.

17 years agoRename MaxTupleSize to MaxHeapTupleSize to clarify that it's not meant to
Tom Lane [Mon, 5 Feb 2007 04:22:18 +0000 (04:22 +0000)]
Rename MaxTupleSize to MaxHeapTupleSize to clarify that it's not meant to
describe the maximum size of index tuples (which is typically AM-dependent
anyway); and consequently remove the bogus deduction for "special space"
that was built into it.

Adjust TOAST_TUPLE_THRESHOLD and TOAST_MAX_CHUNK_SIZE to avoid wasting two
bytes per toast chunk, and to ensure that the calculation correctly tracks any
future changes in page header size.  The computation had been inaccurate in a
way that didn't cause any harm except space wastage, but future changes could
have broken it more drastically.

Fix the calculation of BTMaxItemSize, which was formerly computed as 1 byte
more than it could safely be.  This didn't cause any harm in practice because
it's only compared against maxalign'd lengths, but future changes in the size
of page headers or btree special space could have exposed the problem.

initdb forced because of change in TOAST_MAX_CHUNK_SIZE, which alters the
storage of toast tables.

17 years agoDon't MAXALIGN in the checks to decide whether a tuple is over TOAST's
Tom Lane [Sun, 4 Feb 2007 20:00:37 +0000 (20:00 +0000)]
Don't MAXALIGN in the checks to decide whether a tuple is over TOAST's
threshold for tuple length.  On 4-byte-MAXALIGN machines, the toast code
creates tuples that have t_len exactly TOAST_TUPLE_THRESHOLD ... but this
number is not itself maxaligned, so if heap_insert maxaligns t_len before
comparing to TOAST_TUPLE_THRESHOLD, it'll uselessly recurse back to
tuptoaster.c, wasting cycles.  (It turns out that this does not happen on
8-byte-MAXALIGN machines, because for them the outer MAXALIGN in the
TOAST_MAX_CHUNK_SIZE macro reduces TOAST_MAX_CHUNK_SIZE so that toast tuples
will be less than TOAST_TUPLE_THRESHOLD in size.  That MAXALIGN is really
incorrect, but we can't remove it now, see below.)  There isn't any particular
value in maxaligning before comparing to the thresholds, so just don't do
that, which saves a small number of cycles in itself.

These numbers should be rejiggered to minimize wasted space on toast-relation
pages, but we can't do that in the back branches because changing
TOAST_MAX_CHUNK_SIZE would force an initdb (by changing the contents of toast
tables).  We can move the toast decision thresholds a bit, though, which is
what this patch effectively does.

Thanks to Pavan Deolasee for discovering the unintended recursion.

Back-patch into 8.2, but not further, pending more testing.  (HEAD is about
to get a further patch modifying the thresholds, so it won't help much
for testing this form of the patch.)

17 years agoAdd:
Bruce Momjian [Sun, 4 Feb 2007 04:06:08 +0000 (04:06 +0000)]
Add:

>
> * Allow custom variable classes that can restrict who can set the values
>
>   http://archives.postgresql.org/pgsql-hackers/2006-11/msg00911.php
>

17 years agoUpdate wording.
Bruce Momjian [Sun, 4 Feb 2007 04:00:28 +0000 (04:00 +0000)]
Update wording.

17 years agoAdd documentation for Windows on how to set an environment variable.
Bruce Momjian [Sun, 4 Feb 2007 03:55:51 +0000 (03:55 +0000)]
Add documentation for Windows on how to set an environment variable.
Backpatch to 8.2.X.

17 years agoChange vacuum lazy "compacting" warning message to:
Bruce Momjian [Sun, 4 Feb 2007 03:10:55 +0000 (03:10 +0000)]
Change vacuum lazy "compacting" warning message to:

  errhint("Consider using VACUUM FULL on this relation or increasing the configuration parameter \"max_fsm_pages\".")));

17 years agoUpdate URL for:
Bruce Momjian [Sun, 4 Feb 2007 02:32:03 +0000 (02:32 +0000)]
Update URL for:

* Allow sequential scans to take advantage of other concurrent
  sequential scans, also called "Synchronised Scanning"

17 years agoAdd:
Bruce Momjian [Sun, 4 Feb 2007 01:50:59 +0000 (01:50 +0000)]
Add:

>
>  o Add \# to list command history like \s, but with line numbers
>
>    http://archives.postgresql.org/pgsql-hackers/2006-12/msg00255.php
>

17 years agoAdd URLs for:
Bruce Momjian [Sat, 3 Feb 2007 23:52:19 +0000 (23:52 +0000)]
Add URLs for:

* Allow sequential scans to take advantage of other concurrent
  sequential scans, also called "Synchronised Scanning"
>   http://archives.postgresql.org/pgsql-patches/2006-12/msg00076.php
>   http://archives.postgresql.org/pgsql-hackers/2006-12/msg00408.php

17 years agoDocument that a client-only install using:
Bruce Momjian [Sat, 3 Feb 2007 23:01:06 +0000 (23:01 +0000)]
Document that a client-only install using:

gmake -C src/bin install

does install a few server-only binaries.

17 years agoAdd:
Bruce Momjian [Sat, 3 Feb 2007 22:32:49 +0000 (22:32 +0000)]
Add:

>    o Allow recovery.conf to allow the same syntax as
>      postgresql.conf, including quoting
>
>      http://archives.postgresql.org/pgsql-hackers/2006-12/msg00497.php

17 years agoUpdate SQL conformance information about XML features.
Peter Eisentraut [Sat, 3 Feb 2007 17:59:36 +0000 (17:59 +0000)]
Update SQL conformance information about XML features.

17 years agoImplement XMLSERIALIZE for real. Analogously, make the xml to text cast
Peter Eisentraut [Sat, 3 Feb 2007 14:06:56 +0000 (14:06 +0000)]
Implement XMLSERIALIZE for real.  Analogously, make the xml to text cast
observe the xmloption.

Reorganize the representation of the XML option in the parse tree and the
API to make it easier to manage and understand.

Add regression tests for parsing back XML expressions.

17 years agoFix configure detection code when --with-ldap and --enable-thread-safety
Bruce Momjian [Sat, 3 Feb 2007 02:43:38 +0000 (02:43 +0000)]
Fix configure detection code when --with-ldap and --enable-thread-safety
are both used.

Albe Laurenz

17 years agoAdd patches thread for:
Bruce Momjian [Sat, 3 Feb 2007 01:59:27 +0000 (01:59 +0000)]
Add patches thread for:

* Reduce checkpoint performance degredation by forcing data to disk
  more evenly

>   http://archives.postgresql.org/pgsql-patches/2006-12/msg00104.php

17 years agoAdd URL for:
Bruce Momjian [Fri, 2 Feb 2007 23:05:36 +0000 (23:05 +0000)]
Add URL for:

* Allow sequential scans to take advantage of other concurrent
  sequential scans, also called "Synchronised Scanning"
>
>   http://archives.postgresql.org/pgsql-hackers/2006-12/msg00784.php

17 years agoAdd:
Bruce Momjian [Fri, 2 Feb 2007 22:55:08 +0000 (22:55 +0000)]
Add:

> * Reduce checkpoint performance degredation by forcing data to disk
>   more evenly
>
>   http://archives.postgresql.org/pgsql-hackers/2006-12/msg00337.php
>   http://archives.postgresql.org/pgsql-hackers/2007-01/msg00079.php

17 years agoAdd URL for:
Bruce Momjian [Fri, 2 Feb 2007 21:34:39 +0000 (21:34 +0000)]
Add URL for:

                o Fix RENAME to work on variables other than OLD/NEW
>    http://archives.postgresql.org/pgsql-hackers/2007-01/msg01587.php

17 years agoThis patch changes the installscript for vcbuild to actually parse the
Neil Conway [Fri, 2 Feb 2007 16:25:34 +0000 (16:25 +0000)]
This patch changes the installscript for vcbuild to actually parse the
generated solution files for what to install, instead of blindly copying
everything as it previously did. With the previous quick-n-dirty
version, it would copy old DLLs if you reconfigured in a way that didn't
include subprojects like a PL for example.

Magnus Hagander.

17 years agoReword suggestion that libpq.dll be installed in WINNT\SYSTEM32 under
Neil Conway [Fri, 2 Feb 2007 16:10:08 +0000 (16:10 +0000)]
Reword suggestion that libpq.dll be installed in WINNT\SYSTEM32 under
Windows. Per Magnus Hagander, this is not recommended.

17 years agoApplied Magnus Hagander's patch to take away some compiler warnings.
Michael Meskes [Fri, 2 Feb 2007 09:31:10 +0000 (09:31 +0000)]
Applied Magnus Hagander's patch to take away some compiler warnings.

17 years agoCleaned up va_list handling. Hopefully this now works on all archs.
Michael Meskes [Fri, 2 Feb 2007 08:58:23 +0000 (08:58 +0000)]
Cleaned up va_list handling. Hopefully this now works on all archs.

17 years agoAdd:
Bruce Momjian [Fri, 2 Feb 2007 05:42:56 +0000 (05:42 +0000)]
Add:

>  o Allow column display reordering by recording a display,
>    storage, and permanent id for every column?
>
>    http://archives.postgresql.org/pgsql-hackers/2006-12/msg00782.php
>

17 years agoUpdate release notes for security-related releases in all active branches.
Tom Lane [Fri, 2 Feb 2007 00:10:23 +0000 (00:10 +0000)]
Update release notes for security-related releases in all active branches.

Security: CVE-2007-0555, CVE-2007-0556

17 years agoRepair failure to check that a table is still compatible with a previously
Tom Lane [Fri, 2 Feb 2007 00:07:03 +0000 (00:07 +0000)]
Repair failure to check that a table is still compatible with a previously
made query plan.  Use of ALTER COLUMN TYPE creates a hazard for cached
query plans: they could contain Vars that claim a column has a different
type than it now has.  Fix this by checking during plan startup that Vars
at relation scan level match the current relation tuple descriptor.  Since
at that point we already have at least AccessShareLock, we can be sure the
column type will not change underneath us later in the query.  However,
since a backend's locks do not conflict against itself, there is still a
hole for an attacker to exploit: he could try to execute ALTER COLUMN TYPE
while a query is in progress in the current backend.  Seal that hole by
rejecting ALTER TABLE whenever the target relation is already open in
the current backend.

This is a significant security hole: not only can one trivially crash the
backend, but with appropriate misuse of pass-by-reference datatypes it is
possible to read out arbitrary locations in the server process's memory,
which could allow retrieving database content the user should not be able
to see.  Our thanks to Jeff Trout for the initial report.

Security: CVE-2007-0556

17 years agoRepair insufficiently careful type checking for SQL-language functions:
Tom Lane [Fri, 2 Feb 2007 00:02:55 +0000 (00:02 +0000)]
Repair insufficiently careful type checking for SQL-language functions:
we should check that the function code returns the claimed result datatype
every time we parse the function for execution.  Formerly, for simple
scalar result types we assumed the creation-time check was sufficient, but
this fails if the function selects from a table that's been redefined since
then, and even more obviously fails if check_function_bodies had been OFF.

This is a significant security hole: not only can one trivially crash the
backend, but with appropriate misuse of pass-by-reference datatypes it is
possible to read out arbitrary locations in the server process's memory,
which could allow retrieving database content the user should not be able
to see.  Our thanks to Jeff Trout for the initial report.

Security: CVE-2007-0555

17 years agoFix broken markup.
Tom Lane [Thu, 1 Feb 2007 22:06:14 +0000 (22:06 +0000)]
Fix broken markup.

17 years agoAdd "May/Can/Might" section to error message style guidlines, and
Bruce Momjian [Thu, 1 Feb 2007 21:28:34 +0000 (21:28 +0000)]
Add "May/Can/Might" section to error message style guidlines, and
"can't" -> "cannot" section.