OSDN Git Service

pg-rex/syncrep.git
15 years agoDocument that Solaris can't use effective_io_concurrency because of an
Bruce Momjian [Thu, 2 Apr 2009 22:44:10 +0000 (22:44 +0000)]
Document that Solaris can't use effective_io_concurrency because of an
ineffective posix_fadvise().

15 years agoRefactor ExecProject and associated routines so that fast-path code is used
Tom Lane [Thu, 2 Apr 2009 22:39:30 +0000 (22:39 +0000)]
Refactor ExecProject and associated routines so that fast-path code is used
for simple Var targetlist entries all the time, even when there are other
entries that are not simple Vars.  Also, ensure that we prefetch attributes
(with slot_getsomeattrs) for all Vars in the targetlist, even those buried
within expressions.  In combination these changes seem to significantly
reduce the runtime for cases where tlists are mostly but not exclusively
Vars.  Per my proposal of yesterday.

15 years agoRevert DTrace patch from Robert Lor
Bruce Momjian [Thu, 2 Apr 2009 20:59:10 +0000 (20:59 +0000)]
Revert DTrace patch from Robert Lor

15 years agoMinor code beautification/consolidation.
Tom Lane [Thu, 2 Apr 2009 20:16:30 +0000 (20:16 +0000)]
Minor code beautification/consolidation.

15 years agoGive a better error message when trying to change
Bruce Momjian [Thu, 2 Apr 2009 19:57:19 +0000 (19:57 +0000)]
Give a better error message when trying to change
"effective_io_concurrency" on systems without posix_fadvise().

15 years agoHave PL/pgSQL FETCH set DIAGNOSTICS ROW_COUNT.
Bruce Momjian [Thu, 2 Apr 2009 19:20:45 +0000 (19:20 +0000)]
Have PL/pgSQL FETCH set DIAGNOSTICS ROW_COUNT.

Andrew Gierth

15 years agoAdd support for additional DTrace probes.
Bruce Momjian [Thu, 2 Apr 2009 19:14:34 +0000 (19:14 +0000)]
Add support for additional DTrace probes.

Robert Lor

15 years agoFix memory allocation for output of hstore type.
Teodor Sigaev [Thu, 2 Apr 2009 17:57:05 +0000 (17:57 +0000)]
Fix memory allocation for output of hstore type.
Per "maosen.zhang" <maosen.zhang@alibaba-inc.com> report.

15 years agoDo not show information_schema in \d* commands, unless 'S' or pattern is
Bruce Momjian [Thu, 2 Apr 2009 17:38:26 +0000 (17:38 +0000)]
Do not show information_schema in \d* commands, unless 'S' or pattern is
specified.

Martin Pihlak

15 years agoFix SetClientEncoding() to maintain a cache of previously selected encoding
Tom Lane [Thu, 2 Apr 2009 17:30:53 +0000 (17:30 +0000)]
Fix SetClientEncoding() to maintain a cache of previously selected encoding
conversion functions.  This allows transaction rollback to revert to a
previous client_encoding setting without doing fresh catalog lookups.
I believe that this explains and fixes the recent report of "failed to commit
client_encoding" failures.

This bug is present in 8.3.x, but it doesn't seem prudent to back-patch
the fix, at least not till it's had some time for field testing in HEAD.

In passing, remove SetDefaultClientEncoding(), which was used nowhere.

15 years agoChange psql \d* display so 'S' _or_ a pattern include system objects.
Bruce Momjian [Thu, 2 Apr 2009 15:15:32 +0000 (15:15 +0000)]
Change psql \d* display so 'S' _or_ a pattern include system objects.

15 years agoFix GUC's reports of assign_hook failure to always include the parameter value
Tom Lane [Thu, 2 Apr 2009 03:51:43 +0000 (03:51 +0000)]
Fix GUC's reports of assign_hook failure to always include the parameter value
we failed to assign, even in "can't happen" cases.  Motivated by wondering
what's going on in a recent trouble report where "failed to commit" did
happen.

15 years agoplpgsql's exec_simple_cast_value() mistakenly supposed that it could bypass
Tom Lane [Thu, 2 Apr 2009 01:16:11 +0000 (01:16 +0000)]
plpgsql's exec_simple_cast_value() mistakenly supposed that it could bypass
casting effort whenever the input value was NULL.  However this prevents
application of not-null domain constraints in the cases that use this
function, as illustrated in bug #4741.  Since this function isn't meant
for use in performance-critical paths anyway, this certainly seems like
another case of "premature optimization is the root of all evil".

Back-patch as far as 8.2; older versions made no effort to enforce
domain constraints here anyway.

15 years agoImprove pg_dump's query for retrieving BLOB comments to be more efficient
Tom Lane [Wed, 1 Apr 2009 18:54:27 +0000 (18:54 +0000)]
Improve pg_dump's query for retrieving BLOB comments to be more efficient
when there are many blobs and not so many comments.  Tamas Vincze

15 years agoUpdate comment to reflect that LC_COLLATE and LC_CTYPE are now
Heikki Linnakangas [Wed, 1 Apr 2009 09:17:32 +0000 (09:17 +0000)]
Update comment to reflect that LC_COLLATE and LC_CTYPE are now
per-database settings.

15 years agoIndex some array functions, per Mario Splivalo.
Tom Lane [Wed, 1 Apr 2009 03:32:29 +0000 (03:32 +0000)]
Index some array functions, per Mario Splivalo.

15 years agoRemove last references to the crypt auth method, per Andreas Scherbaum.
Tom Lane [Wed, 1 Apr 2009 03:23:50 +0000 (03:23 +0000)]
Remove last references to the crypt auth method, per Andreas Scherbaum.

15 years agoFix contrib/pgstattuple and contrib/pageinspect to prevent attempts to read
Tom Lane [Tue, 31 Mar 2009 22:54:31 +0000 (22:54 +0000)]
Fix contrib/pgstattuple and contrib/pageinspect to prevent attempts to read
temporary tables of other sessions; that is unsafe because of the way our
buffer management works.  Per report from Stuart Bishop.
This is redundant with the bufmgr.c checks in HEAD, but not at all redundant
in the back branches.

15 years agoRemove some "Other" sections in the release notes by putting the items
Bruce Momjian [Tue, 31 Mar 2009 22:23:28 +0000 (22:23 +0000)]
Remove some "Other" sections in the release notes by putting the items
at the top of their sections.

15 years agoModify the relcache to record the temp status of both local and nonlocal
Tom Lane [Tue, 31 Mar 2009 22:12:48 +0000 (22:12 +0000)]
Modify the relcache to record the temp status of both local and nonlocal
temp relations; this is no more expensive than before, now that we have
pg_class.relistemp.  Insert tests into bufmgr.c to prevent attempting
to fetch pages from nonlocal temp relations.  This provides a low-level
defense against bugs-of-omission allowing temp pages to be loaded into shared
buffers, as in the contrib/pgstattuple problem reported by Stuart Bishop.
While at it, tweak a bunch of places to use new relcache tests (instead of
expensive probes into pg_namespace) to detect local or nonlocal temp tables.

15 years agoDon't crash initdb when we fail to get the current username.
Magnus Hagander [Tue, 31 Mar 2009 18:58:16 +0000 (18:58 +0000)]
Don't crash initdb when we fail to get the current username.
Give an error message and exit instead, like we do elsewhere...

Per report from Wez Furlong and Robert Treat.

15 years agoAdd a "relistemp" boolean column to pg_class, which is true for temporary
Tom Lane [Tue, 31 Mar 2009 17:59:56 +0000 (17:59 +0000)]
Add a "relistemp" boolean column to pg_class, which is true for temporary
relations (including a temp table's indexes and toast table/index), and
false for normal relations.  For ease of checking, this commit just adds
the column and fills it correctly --- revising the relation access machinery
to use it will come separately.

15 years agoFix a rare race condition when commit_siblings > 0 and a transaction commits
Heikki Linnakangas [Tue, 31 Mar 2009 05:18:33 +0000 (05:18 +0000)]
Fix a rare race condition when commit_siblings > 0 and a transaction commits
at the same instant as a new backend is spawned. Since CountActiveBackends()
doesn't hold ProcArrayLock, it needs to be prepared for the case that a
pointer at the end of the proc array is still NULL even though numProcs says
it should be valid, since it doesn't hold ProcArrayLock. Backpatch to 8.1.
8.0 and earlier had this right, but it was broken in the split of PGPROC and
sinval shared memory arrays.

Per report and proposal by Marko Kreen.

15 years agoUpdate release note introductory description.
Bruce Momjian [Tue, 31 Mar 2009 02:34:04 +0000 (02:34 +0000)]
Update release note introductory description.

15 years agoAdd PQinitOpenSSL() function to support applications that use libcrypto
Tom Lane [Tue, 31 Mar 2009 01:41:27 +0000 (01:41 +0000)]
Add PQinitOpenSSL() function to support applications that use libcrypto
but not OpenSSL (or perhaps vice versa, if that's possible).

Andrew Chernow, with minor editorialization by me.

15 years agoMore new subsections in release notes.
Bruce Momjian [Tue, 31 Mar 2009 01:26:13 +0000 (01:26 +0000)]
More new subsections in release notes.

15 years agoMore release note changes, including a lower level of subsections.
Bruce Momjian [Mon, 30 Mar 2009 22:01:15 +0000 (22:01 +0000)]
More release note changes, including a lower level of subsections.

15 years agoMore release note adjustments, reordering.
Bruce Momjian [Mon, 30 Mar 2009 20:32:49 +0000 (20:32 +0000)]
More release note adjustments, reordering.

15 years agoMore release note wording improvements; section order adjustments.
Bruce Momjian [Mon, 30 Mar 2009 19:59:19 +0000 (19:59 +0000)]
More release note wording improvements;  section order adjustments.

15 years agoReorder release note sections.
Bruce Momjian [Mon, 30 Mar 2009 18:34:56 +0000 (18:34 +0000)]
Reorder release note sections.

15 years agoFix window function plan generation to cope with volatile sort expressions.
Tom Lane [Mon, 30 Mar 2009 17:30:44 +0000 (17:30 +0000)]
Fix window function plan generation to cope with volatile sort expressions.
(Not clear how useful these really are, but failing is no good...)
Per report from David Fetter and Robert Treat.

15 years agoUpdate URL to Python bug tracker. Backpatch to 8.3; doesn't seem worthy of
Alvaro Herrera [Mon, 30 Mar 2009 16:15:43 +0000 (16:15 +0000)]
Update URL to Python bug tracker.  Backpatch to 8.3; doesn't seem worthy of
further backpatch.

15 years agoFix an oversight in the support for storing/retrieving "minimal tuples" in
Tom Lane [Mon, 30 Mar 2009 04:08:43 +0000 (04:08 +0000)]
Fix an oversight in the support for storing/retrieving "minimal tuples" in
TupleTableSlots.  We have functions for retrieving a minimal tuple from a slot
after storing a regular tuple in it, or vice versa; but these were implemented
by converting the internal storage from one format to the other.  The problem
with that is it invalidates any pass-by-reference Datums that were already
fetched from the slot, since they'll be pointing into the just-freed version
of the tuple.  The known problem cases involve fetching both a whole-row
variable and a pass-by-reference value from a slot that is fed from a
tuplestore or tuplesort object.  The added regression tests illustrate some
simple cases, but there may be other failure scenarios traceable to the same
bug.  Note that the added tests probably only fail on unpatched code if it's
built with --enable-cassert; otherwise the bug leads to fetching from freed
memory, which will not have been overwritten without additional conditions.

Fix by allowing a slot to contain both formats simultaneously; which turns out
not to complicate the logic much at all, if anything it seems less contorted
than before.

Back-patch to 8.2, where minimal tuples were introduced.

15 years agoMore release note markup.
Bruce Momjian [Sun, 29 Mar 2009 19:13:46 +0000 (19:13 +0000)]
More release note markup.

15 years agoMore release note markup.
Bruce Momjian [Sun, 29 Mar 2009 03:58:37 +0000 (03:58 +0000)]
More release note markup.

15 years agoAdd SGML markup for commands/literal/application/etc in release notes;
Bruce Momjian [Sun, 29 Mar 2009 03:01:56 +0000 (03:01 +0000)]
Add SGML markup for commands/literal/application/etc in release notes;
still more work to do.

15 years agoConsistent 8.4 release note naming for Itagaki Takahiro
Bruce Momjian [Sat, 28 Mar 2009 22:05:22 +0000 (22:05 +0000)]
Consistent 8.4 release note naming for Itagaki Takahiro

15 years agoClarify variable naming: pq_initssllib -> pq_init_ssl_lib
Bruce Momjian [Sat, 28 Mar 2009 18:48:55 +0000 (18:48 +0000)]
Clarify variable naming: pq_initssllib -> pq_init_ssl_lib

15 years agoUpdate release notes to say citext is multi-byte aware, per suggestion
Bruce Momjian [Sat, 28 Mar 2009 14:15:15 +0000 (14:15 +0000)]
Update release notes to say citext is multi-byte aware, per suggestion
from patch author:

       Add /contrib/citext as a case-insensitive, multibyte-capable
text data type (David Wheeler)

15 years agoBetter document that SET ROLE does not uset ALTER ROLE SET settings;
Bruce Momjian [Sat, 28 Mar 2009 03:26:02 +0000 (03:26 +0000)]
Better document that SET ROLE does not uset ALTER ROLE SET settings;
suggested wording from Josh Berkus.

15 years agoBetter document PQinitSSL(0) behavior in regards to libcrypto.
Bruce Momjian [Sat, 28 Mar 2009 01:36:11 +0000 (01:36 +0000)]
Better document PQinitSSL(0) behavior in regards to libcrypto.

15 years agoAdd documentation of the fact that dtrace probes evaluate their parameters
Tom Lane [Sat, 28 Mar 2009 00:10:23 +0000 (00:10 +0000)]
Add documentation of the fact that dtrace probes evaluate their parameters
even when not active.  Explain how to prevent that with an ENABLED() check.

15 years agoDocument in release notes that NOT IN is only for NOT EXIST clauses.
Bruce Momjian [Fri, 27 Mar 2009 22:39:42 +0000 (22:39 +0000)]
Document in release notes that NOT IN is only for NOT EXIST clauses.

Andrew Gierth

15 years agoOn Solaris, we should only force use of our own getopt(); it's okay to use
Tom Lane [Fri, 27 Mar 2009 19:58:11 +0000 (19:58 +0000)]
On Solaris, we should only force use of our own getopt(); it's okay to use
the system's getopt_long().  The previous coding was the result of a sloppy
discussion that failed to draw this distinction.  The result was that PG
programs don't handle options as users of that platform expect.  Per
gripe from Chuck McDevitt.

Although this is a pre-existing bug, I'm not backpatching since I think we
could do with a bit of beta testing before concluding this is really OK.

15 years agoFix markup, per Devrim
Magnus Hagander [Fri, 27 Mar 2009 19:17:11 +0000 (19:17 +0000)]
Fix markup, per Devrim

15 years agoAdd an errdetail explaining why we reject infinite dates and timestamps
Tom Lane [Fri, 27 Mar 2009 18:56:57 +0000 (18:56 +0000)]
Add an errdetail explaining why we reject infinite dates and timestamps
while converting to XML.  Bernd Helmle

15 years agoFix possible failures when a tuplestore switches from in-memory to on-disk
Tom Lane [Fri, 27 Mar 2009 18:30:21 +0000 (18:30 +0000)]
Fix possible failures when a tuplestore switches from in-memory to on-disk
mode while callers hold pointers to in-memory tuples.  I reported this for
the case of nodeWindowAgg's primary scan tuple, but inspection of the code
shows that all of the calls in nodeWindowAgg and nodeCtescan are at risk.
For the moment, fix it with a rather brute-force approach of copying
whenever one of the at-risk callers requests a tuple.  Later we might
think of some sort of reference-count approach to reduce tuple copying.

15 years agoTeach reindex_index() to clear pg_index.indcheckxmin when possible.
Tom Lane [Fri, 27 Mar 2009 15:57:11 +0000 (15:57 +0000)]
Teach reindex_index() to clear pg_index.indcheckxmin when possible.
Greg Stark, slightly modified by me.

15 years agoFix tab completion of ANALYZE VERBOSE <tab>. It was previously confused
Heikki Linnakangas [Fri, 27 Mar 2009 14:58:46 +0000 (14:58 +0000)]
Fix tab completion of ANALYZE VERBOSE <tab>. It was previously confused
with EXPLAIN ANALYZE VERBOSE.

Greg Sabino Mullane, reformatted by myself. Backpatch to 8.1, where the
bug was introduced.

15 years agoClearify new SSL certificate verification in libpq
Magnus Hagander [Fri, 27 Mar 2009 12:01:57 +0000 (12:01 +0000)]
Clearify new SSL certificate verification in libpq

15 years agoFix release notes about pg_hba changes
Magnus Hagander [Fri, 27 Mar 2009 11:58:21 +0000 (11:58 +0000)]
Fix release notes about pg_hba changes

15 years agoUpdated release wording, per Greg Stark:
Bruce Momjian [Fri, 27 Mar 2009 02:25:04 +0000 (02:25 +0000)]
Updated release wording, per Greg Stark:

       Previously EXPLAIN VERBOSE had output an internal
representation of the

15 years agoSecond batch of release note fixes by Guillaume Smet
Bruce Momjian [Fri, 27 Mar 2009 01:44:30 +0000 (01:44 +0000)]
Second batch of release note fixes by Guillaume Smet

15 years agoMark Greg as the instigator of the statistics target increase:
Bruce Momjian [Fri, 27 Mar 2009 01:26:36 +0000 (01:26 +0000)]
Mark Greg as the instigator of the statistics target increase:

       Increase the default value of default_statistics_target from 10 to 100
       (Greg Sabino Mullane, Tom)

15 years agoFirst round of release note changes from Guillaume Smet.
Bruce Momjian [Fri, 27 Mar 2009 00:45:40 +0000 (00:45 +0000)]
First round of release note changes from Guillaume Smet.

15 years agoFix release note attribution:
Bruce Momjian [Fri, 27 Mar 2009 00:11:34 +0000 (00:11 +0000)]
Fix release note attribution:

       Support the IS0 8601 time interval syntax (Tom, Kevin
Grittner)

per Ron Mayer

15 years agoImproved release notes interval wording:
Bruce Momjian [Fri, 27 Mar 2009 00:08:14 +0000 (00:08 +0000)]
Improved release notes interval wording:

       Made interval seconds rounding more consistent across output
       formats (Ron Mayer)

Ron Mayer

15 years agoRemove duplicate interval (fractional seconds) items; Ron Mayer
Bruce Momjian [Fri, 27 Mar 2009 00:06:14 +0000 (00:06 +0000)]
Remove duplicate interval (fractional seconds) items;  Ron Mayer

15 years agoDocument that Datestyle no longer controls interval output, per
Bruce Momjian [Fri, 27 Mar 2009 00:01:47 +0000 (00:01 +0000)]
Document that Datestyle no longer controls interval output, per
suggestion from Ron Mayer

15 years agoMake pg_standby's maxretries option do what one would expect. Fujii Masao
Tom Lane [Thu, 26 Mar 2009 22:29:13 +0000 (22:29 +0000)]
Make pg_standby's maxretries option do what one would expect.  Fujii Masao

15 years agoGettext plural support
Peter Eisentraut [Thu, 26 Mar 2009 22:26:08 +0000 (22:26 +0000)]
Gettext plural support

In the backend, I changed only a handful of exemplary or important-looking
instances to make use of the plural support; there is probably more work
there.  For the rest of the source, this should cover all relevant cases.

15 years agoFix a couple of places that still claimed LC_COLLATE and LC_CTYPE can't
Tom Lane [Thu, 26 Mar 2009 20:55:49 +0000 (20:55 +0000)]
Fix a couple of places that still claimed LC_COLLATE and LC_CTYPE can't
be changed after initdb.

15 years agoAdjust phrasing of complaints about multi-byte COPY delimiter strings.
Tom Lane [Thu, 26 Mar 2009 19:24:54 +0000 (19:24 +0000)]
Adjust phrasing of complaints about multi-byte COPY delimiter strings.
Per pgsql-hackers discussion of 2009-02-17.

15 years agoIf we expect a hash join to be performed in multiple batches, suppress
Tom Lane [Thu, 26 Mar 2009 17:15:35 +0000 (17:15 +0000)]
If we expect a hash join to be performed in multiple batches, suppress
"physical tlist" optimization on the outer relation (ie, force a projection
step to occur in its scan).  This avoids storing useless column values when
the outer relation's tuples are written to temporary batch files.

Modified version of a patch by Michael Henderson and Ramon Lawrence.

15 years agoCorrection: ansi-join ->anti-join.
Bruce Momjian [Thu, 26 Mar 2009 12:38:03 +0000 (12:38 +0000)]
Correction:  ansi-join ->anti-join.

15 years agoFix markup so that 'make HISTORY' works. A couple very minor
Tom Lane [Thu, 26 Mar 2009 03:46:14 +0000 (03:46 +0000)]
Fix markup so that 'make HISTORY' works.  A couple very minor
editorial improvements.

15 years agoReorder 8.4 release note sections.
Bruce Momjian [Thu, 26 Mar 2009 02:48:53 +0000 (02:48 +0000)]
Reorder 8.4 release note sections.

15 years agoRe-add release notes for release 8.3.7.
Bruce Momjian [Thu, 26 Mar 2009 02:40:37 +0000 (02:40 +0000)]
Re-add release notes for release 8.3.7.

15 years agoAdjust AS OF release notes markup.
Bruce Momjian [Thu, 26 Mar 2009 01:48:27 +0000 (01:48 +0000)]
Adjust AS OF release notes markup.

15 years agoMention release note items current as of 2009-03-16.
Bruce Momjian [Thu, 26 Mar 2009 01:31:15 +0000 (01:31 +0000)]
Mention release note items current as of 2009-03-16.

15 years agoFirst version of 8.4 release notes; markup/cleanup/reorganization still
Bruce Momjian [Thu, 26 Mar 2009 01:19:07 +0000 (01:19 +0000)]
First version of 8.4 release notes;  markup/cleanup/reorganization still
required.

15 years agoAdd btree_gin contrib module.
Tom Lane [Wed, 25 Mar 2009 23:20:01 +0000 (23:20 +0000)]
Add btree_gin contrib module.

Teodor Sigaev and Oleg Bartunov

15 years agoAdjust the APIs for GIN opclass support functions to allow the extractQuery()
Tom Lane [Wed, 25 Mar 2009 22:19:02 +0000 (22:19 +0000)]
Adjust the APIs for GIN opclass support functions to allow the extractQuery()
method to pass extra data to the consistent() and comparePartial() methods.
This is the core infrastructure needed to support the soon-to-appear
contrib/btree_gin module.  The APIs are still upward compatible with the
definitions used in 8.3 and before, although *not* with the previous 8.4devel
function definitions.

catversion bump for changes in pg_proc entries (although these are just
cosmetic, since GIN doesn't actually look at the function signature before
calling it...)

Teodor Sigaev and Oleg Bartunov

15 years agoFix old thinko in pgp.h: the idea is to declare some named enum types,
Tom Lane [Wed, 25 Mar 2009 15:03:13 +0000 (15:03 +0000)]
Fix old thinko in pgp.h: the idea is to declare some named enum types,
not global variables of anonymous enum types.  This didn't actually hurt
much because most linkers will just merge the duplicated definitions ...
but some will complain.  Per bug #4731 from Ceriel Jacobs.

Backpatch to 8.1 --- the declarations don't exist before that.

15 years agoUse double quotes instead of single quotes.
Peter Eisentraut [Wed, 25 Mar 2009 14:12:02 +0000 (14:12 +0000)]
Use double quotes instead of single quotes.

15 years agoComplete list of valid fork names, and use double quotes.
Peter Eisentraut [Wed, 25 Mar 2009 14:11:48 +0000 (14:11 +0000)]
Complete list of valid fork names, and use double quotes.

15 years agoUpdate line count for \? output (obtained from wc -l)
Peter Eisentraut [Wed, 25 Mar 2009 13:15:55 +0000 (13:15 +0000)]
Update line count for \? output (obtained from wc -l)

15 years agoIn the \? output, align the columns consistently.
Peter Eisentraut [Wed, 25 Mar 2009 13:14:17 +0000 (13:14 +0000)]
In the \? output, align the columns consistently.

15 years agoPolish SQL/MED terminology
Peter Eisentraut [Wed, 25 Mar 2009 13:11:43 +0000 (13:11 +0000)]
Polish SQL/MED terminology

15 years agoNo period if it's not a sentence
Peter Eisentraut [Wed, 25 Mar 2009 13:07:26 +0000 (13:07 +0000)]
No period if it's not a sentence

15 years agoInstall a search tree depth limit in GIN bulk-insert operations, to prevent
Tom Lane [Tue, 24 Mar 2009 22:06:03 +0000 (22:06 +0000)]
Install a search tree depth limit in GIN bulk-insert operations, to prevent
them from degrading badly when the input is sorted or nearly so.  In this
scenario the tree is unbalanced to the point of becoming a mere linked list,
so insertions become O(N^2).  The easiest and most safely back-patchable
solution is to stop growing the tree sooner, ie limit the growth of N.  We
might later consider a rebalancing tree algorithm, but it's not clear that
the benefit would be worth the cost and complexity.  Per report from Sergey
Burladyan and an earlier complaint from Heikki.

Back-patch to 8.2; older versions didn't have GIN indexes.

15 years agoFix stupid parenthesization mistake. Per bug #4728 from Bruce Toll.
Tom Lane [Tue, 24 Mar 2009 21:12:56 +0000 (21:12 +0000)]
Fix stupid parenthesization mistake.  Per bug #4728 from Bruce Toll.

15 years agoImplement "fastupdate" support for GIN indexes, in which we try to accumulate
Tom Lane [Tue, 24 Mar 2009 20:17:18 +0000 (20:17 +0000)]
Implement "fastupdate" support for GIN indexes, in which we try to accumulate
multiple index entries in a holding area before adding them to the main index
structure.  This helps because bulk insert is (usually) significantly faster
than retail insert for GIN.

This patch also removes GIN support for amgettuple-style index scans.  The
API defined for amgettuple is difficult to support with fastupdate, and
the previously committed partial-match feature didn't really work with
it either.  We might eventually figure a way to put back amgettuple
support, but it won't happen for 8.4.

catversion bumped because of change in GIN's pg_am entry, and because
the format of GIN indexes changed on-disk (there's a metapage now,
and possibly a pending list).

Teodor Sigaev

15 years agoFix my typo
Peter Eisentraut [Tue, 24 Mar 2009 12:57:31 +0000 (12:57 +0000)]
Fix my typo

15 years agoRemove munging of xml and xpath params to xpath(). The XML must now be a well formed...
Andrew Dunstan [Mon, 23 Mar 2009 21:00:39 +0000 (21:00 +0000)]
Remove munging of xml and xpath params to xpath(). The XML must now be a well formed XML document.

15 years agoConst-ify the parse table passed to fillRelOptions. The previous coding
Tom Lane [Mon, 23 Mar 2009 16:36:27 +0000 (16:36 +0000)]
Const-ify the parse table passed to fillRelOptions.  The previous coding
meant it had to be built on-the-fly at each entry to default_reloptions.

15 years agoUpdate constraint_exclusion docs because parameter is now 3-valued, not
Bruce Momjian [Mon, 23 Mar 2009 15:14:14 +0000 (15:14 +0000)]
Update constraint_exclusion docs because parameter is now 3-valued, not
just enable/disable.

15 years agoAdd expected file for locale-enabled installations.
Peter Eisentraut [Mon, 23 Mar 2009 08:07:32 +0000 (08:07 +0000)]
Add expected file for locale-enabled installations.

15 years agoMore fixes for 8.4 DTrace probes. Remove useless BUFFER_HIT/BUFFER_MISS
Tom Lane [Mon, 23 Mar 2009 01:52:38 +0000 (01:52 +0000)]
More fixes for 8.4 DTrace probes.  Remove useless BUFFER_HIT/BUFFER_MISS
probes --- the BUFFER_READ_DONE probe provides the same information and more
besides.  Expand the LOCK_WAIT_START/DONE probe arguments so that there's
actually some chance of telling what is being waited for.  Update and
clean up the documentation.

15 years agoClarify libpq 'sslverify' documentation wording.
Bruce Momjian [Mon, 23 Mar 2009 01:45:29 +0000 (01:45 +0000)]
Clarify libpq 'sslverify' documentation wording.

15 years agoAdd isExtend to the parameters of the buffer_read_start and buffer_read_done
Tom Lane [Sun, 22 Mar 2009 22:39:05 +0000 (22:39 +0000)]
Add isExtend to the parameters of the buffer_read_start and buffer_read_done
DTrace probes, so that ordinary reads can be distinguished from relation
extension operations.  Move buffer_read_start probe to before the
smgrnblocks() call that's needed in the isExtend case, since really that step
should be charged as part of the time needed for the extension operation.
(This makes it slightly harder to match the read_start with the associated
read_done, since now you can't match them on blockNumber, but it should still
be possible since isExtend operations on the same relation can never be
interleaved.)  Per recent discussion.

In passing, add the page identity (forkNum/blockNum) to the parameters of the
buffer_flush_start/buffer_flush_done probes, which were unaccountably lacking
the info.

15 years agoClarify 'cert' authentication documention.
Bruce Momjian [Sun, 22 Mar 2009 21:55:07 +0000 (21:55 +0000)]
Clarify 'cert' authentication documention.

15 years agoClean up pg_SSPI_error() coding a little bit: make the messages more
Tom Lane [Sun, 22 Mar 2009 18:06:35 +0000 (18:06 +0000)]
Clean up pg_SSPI_error() coding a little bit: make the messages more
consistent, translate where intended, const-ify declarations.
Resolves a gripe from Alvaro as well as some stuff I didn't like.

15 years agoRemove the -d and -D options of pg_dump and pg_dumpall. The functionality
Tom Lane [Sun, 22 Mar 2009 16:44:26 +0000 (16:44 +0000)]
Remove the -d and -D options of pg_dump and pg_dumpall.  The functionality
is still available, but you must now write the long equivalent --inserts
or --column-inserts.  This change is made to eliminate confusion with the
use of -d to specify a database name in most other Postgres client programs.
Original patch by Greg Mullane, modified per subsequent discussion.

15 years agoRemove the datetime keywords ABSTIME and RELTIME, which we'd been treating as
Tom Lane [Sun, 22 Mar 2009 01:12:32 +0000 (01:12 +0000)]
Remove the datetime keywords ABSTIME and RELTIME, which we'd been treating as
noise words for the last twelve years, for compatibility with Berkeley-era
output formatting of the special INVALID values for those datatypes.
Considering that the datatypes themselves have been deprecated for awhile,
this is taking backwards compatibility a little far.  Per gripe from Josh
Berkus.

15 years agoOptimize multi-batch hash joins when the outer relation has a nonuniform
Tom Lane [Sat, 21 Mar 2009 00:04:40 +0000 (00:04 +0000)]
Optimize multi-batch hash joins when the outer relation has a nonuniform
distribution, by creating a special fast path for the (first few) most common
values of the outer relation.  Tuples having hashvalues matching the MCVs
are effectively forced to be in the first batch, so that we never write
them out to the batch temp files.

Bryce Cutt and Ramon Lawrence, with some editorialization by me.

15 years agoRename pg_restore -m to -j, and add documentation about what good numbers
Peter Eisentraut [Fri, 20 Mar 2009 09:21:09 +0000 (09:21 +0000)]
Rename pg_restore -m to -j, and add documentation about what good numbers
are.  per discussion on hackers

15 years agoDon't set the signal handler for SIGQUIT on Windows. Buildfarm shows that
Heikki Linnakangas [Wed, 18 Mar 2009 20:30:35 +0000 (20:30 +0000)]
Don't set the signal handler for SIGQUIT on Windows. Buildfarm shows that
reinstalling the default signal handler doesn't work as it is on Windows.
Presumably core dumps on SIGQUIT are not a problem on Windows, so rather
than figure out what header files or other changes are required to make it
work, just don't bother.

15 years agoBe more clear about when to use gist__int_ops vs. gist__intbig_ops.
Tom Lane [Wed, 18 Mar 2009 20:18:18 +0000 (20:18 +0000)]
Be more clear about when to use gist__int_ops vs. gist__intbig_ops.
Per suggestion from Ron Mayer.

15 years agoDon't intercept SIGQUIT as a signal to trigger failover; that's what
Heikki Linnakangas [Wed, 18 Mar 2009 19:27:28 +0000 (19:27 +0000)]
Don't intercept SIGQUIT as a signal to trigger failover; that's what
postmaster uses for immediate shutdown. Trap SIGUSR1 as the preferred
signal for that.

Per report by Fujii Masao and subsequent discussion on -hackers.