]> granicus.if.org Git - postgresql/log
postgresql
19 years agoFlush any remaining statistics counts out to the collector at process
Tom Lane [Thu, 31 Mar 2005 23:20:49 +0000 (23:20 +0000)]
Flush any remaining statistics counts out to the collector at process
exit.  Without this, operations triggered during backend exit (such as
temp table deletions) won't be counted ... which given heavy usage of
temp tables can lead to pg_autovacuum falling way behind on the need
to vacuum pg_class and pg_attribute.  Per reports from Steve Crawford
and others.

19 years agoFirst phase of OUT-parameters project. We can now define and use SQL
Tom Lane [Thu, 31 Mar 2005 22:46:33 +0000 (22:46 +0000)]
First phase of OUT-parameters project.  We can now define and use SQL
functions with OUT parameters.  The various PLs still need work, as does
pg_dump.  Rudimentary docs and regression tests included.

19 years ago1 fix various comparing functions
Teodor Sigaev [Thu, 31 Mar 2005 15:08:08 +0000 (15:08 +0000)]
1 fix various comparing functions
2 implement gtsvector_out for use with gevel module (debug GiST indexes, http://www.sai.msu.su/~megera/postgres/gist/gevel/)

19 years agoAdjust SGML ulink tags to the URL is always displayed in our printed docs.
Bruce Momjian [Thu, 31 Mar 2005 03:54:38 +0000 (03:54 +0000)]
Adjust SGML ulink tags to the URL is always displayed in our printed docs.

19 years agoMinor code cleanup: ExecHash() was returning a null TupleTableSlot, and an
Neil Conway [Thu, 31 Mar 2005 02:02:52 +0000 (02:02 +0000)]
Minor code cleanup: ExecHash() was returning a null TupleTableSlot, and an
old comment in the code claimed that this was necessary. Since it is not
actually necessary any more, it is clearer to remove the comment and
just return NULL instead -- the return value of ExecHash() is not used.

19 years agoFix build error in pg_buffercache on Win32 (native and cygwin). From
Neil Conway [Thu, 31 Mar 2005 00:45:26 +0000 (00:45 +0000)]
Fix build error in pg_buffercache on Win32 (native and cygwin). From
Mark Kirkwood.

19 years agoDocument that the "-P" option to pg_autovacuum is insecure on many
Neil Conway [Wed, 30 Mar 2005 05:05:01 +0000 (05:05 +0000)]
Document that the "-P" option to pg_autovacuum is insecure on many
platforms, and suggest using ~/.pgpass instead.

19 years agoMinor tweak to documentation.
Neil Conway [Wed, 30 Mar 2005 04:52:49 +0000 (04:52 +0000)]
Minor tweak to documentation.

19 years agoFix typo.
Neil Conway [Wed, 30 Mar 2005 02:08:39 +0000 (02:08 +0000)]
Fix typo.

19 years agoAdd proallargtypes and proargmodes columns to pg_proc, as per my earlier
Tom Lane [Tue, 29 Mar 2005 19:44:23 +0000 (19:44 +0000)]
Add proallargtypes and proargmodes columns to pg_proc, as per my earlier
proposal for OUT parameter support.  The columns don't actually *do*
anything yet, they are just left NULLs.  But I thought I'd commit this
part separately as a fairly pure example of the tasks needed when adding
a column to pg_proc or one of the other core system tables.

19 years agoFix grammar for IN/OUT/INOUT parameters. This commit doesn't actually
Tom Lane [Tue, 29 Mar 2005 17:58:51 +0000 (17:58 +0000)]
Fix grammar for IN/OUT/INOUT parameters.  This commit doesn't actually
implement any new feature, it just pushes the 'not implemented' error
message deeper into the backend.  I also tweaked the grammar to accept
Oracle-ish parameter syntax (parameter name first), as well as the
SQL99 standard syntax (parameter mode first), since it was easy and
people will doubtless try to use both anyway.

19 years agoOfficially decouple FUNC_MAX_ARGS from INDEX_MAX_KEYS, and set the
Tom Lane [Tue, 29 Mar 2005 03:01:32 +0000 (03:01 +0000)]
Officially decouple FUNC_MAX_ARGS from INDEX_MAX_KEYS, and set the
former to 100 by default.  Clean up some of the less necessary
dependencies on FUNC_MAX_ARGS; however, the biggie (FunctionCallInfoData)
remains.

19 years agoAdd SPI_getnspname(), including documentation.
Neil Conway [Tue, 29 Mar 2005 02:53:53 +0000 (02:53 +0000)]
Add SPI_getnspname(), including documentation.

19 years agoConvert oidvector and int2vector into variable-length arrays. This
Tom Lane [Tue, 29 Mar 2005 00:17:27 +0000 (00:17 +0000)]
Convert oidvector and int2vector into variable-length arrays.  This
change saves a great deal of space in pg_proc and its primary index,
and it eliminates the former requirement that INDEX_MAX_KEYS and
FUNC_MAX_ARGS have the same value.  INDEX_MAX_KEYS is still embedded
in the on-disk representation (because it affects index tuple header
size), but FUNC_MAX_ARGS is not.  I believe it would now be possible
to increase FUNC_MAX_ARGS at little cost, but haven't experimented yet.
There are still a lot of vestigial references to FUNC_MAX_ARGS, which
I will clean up in a separate pass.  However, getting rid of it
altogether would require changing the FunctionCallInfoData struct,
and I'm not sure I want to buy into that.

19 years agoRemove dead push/pop rollback code. Vadim once planned to implement
Tom Lane [Mon, 28 Mar 2005 01:50:34 +0000 (01:50 +0000)]
Remove dead push/pop rollback code.  Vadim once planned to implement
transaction rollback via UNDO but I think that's highly unlikely to
happen, so we may as well remove the stubs.  (Someday we ought to
rip out the stub xxx_undo routines, too.)  Per Alvaro.

19 years agoRethink the order of expression preprocessing: eval_const_expressions
Tom Lane [Mon, 28 Mar 2005 00:58:26 +0000 (00:58 +0000)]
Rethink the order of expression preprocessing: eval_const_expressions
really ought to run before canonicalize_qual, because it can now produce
forms that canonicalize_qual knows how to improve (eg, NOT clauses).
Also, because eval_const_expressions already knows about flattening
nested ANDs and ORs into N-argument form, the initial flatten_andors
pass in canonicalize_qual is now completely redundant and can be
removed.  This doesn't save a whole lot of code, but the time and
palloc traffic eliminated is a useful gain on large expression trees.

19 years agoFirst steps towards index scans with heap access decoupled from index
Tom Lane [Sun, 27 Mar 2005 23:53:05 +0000 (23:53 +0000)]
First steps towards index scans with heap access decoupled from index
access: define new index access method functions 'amgetmulti' that can
fetch multiple TIDs per call.  (The functions exist but are totally
untested as yet.)  Since I was modifying pg_am anyway, remove the
no-longer-needed 'rel' parameter from amcostestimate functions, and
also remove the vestigial amowner column that was creating useless
work for Alvaro's shared-object-dependencies project.
Initdb forced due to changes in pg_am.

19 years agoTeach const-expression simplification to simplify boolean equality cases,
Tom Lane [Sun, 27 Mar 2005 19:18:02 +0000 (19:18 +0000)]
Teach const-expression simplification to simplify boolean equality cases,
that is 'x = true' becomes 'x' and 'x = false' becomes 'NOT x'.  This isn't
all that amazingly useful in itself, but it ensures that we will recognize
the different forms as being logically equivalent when checking partial
index predicates.  Per example from Patrick Clery.

19 years agoEliminate duplicate hasnulls bit testing in index tuple access, and
Tom Lane [Sun, 27 Mar 2005 18:38:27 +0000 (18:38 +0000)]
Eliminate duplicate hasnulls bit testing in index tuple access, and
clean up itup.h a little bit.

19 years agoAdd a back-link from IndexOptInfo structs to their parent RelOptInfo
Tom Lane [Sun, 27 Mar 2005 06:29:49 +0000 (06:29 +0000)]
Add a back-link from IndexOptInfo structs to their parent RelOptInfo
structs.  There are many places in the planner where we were passing
both a rel and an index to subroutines, and now need only pass the
index struct.  Notationally simpler, and perhaps a tad faster.

19 years agoExpand the 'special index operator' machinery to handle special cases
Tom Lane [Sat, 26 Mar 2005 23:29:20 +0000 (23:29 +0000)]
Expand the 'special index operator' machinery to handle special cases
for boolean indexes.  Previously we would only use such an index with
WHERE clauses like 'indexkey = true' or 'indexkey = false'.  The new
code transforms the cases 'indexkey', 'NOT indexkey', 'indexkey IS TRUE',
and 'indexkey IS FALSE' into one of these.  While this is only marginally
useful in itself, I intend soon to change constant-expression simplification
so that 'foo = true' and 'foo = false' are reduced to just 'foo' and
'NOT foo' ... which would lose the ability to use boolean indexes for
such queries at all, if the indexscan machinery couldn't make the
reverse transformation.

19 years agoFix a pair of related issues with estimation of inequalities that involve
Tom Lane [Sat, 26 Mar 2005 20:55:39 +0000 (20:55 +0000)]
Fix a pair of related issues with estimation of inequalities that involve
binary-compatible relabeling of one or both operands.  examine_variable
should avoid stripping RelabelType from non-variable expressions, so that
they will continue to have the correct type; and convert_to_scalar should
just use that type and ignore the other input type.  This isn't perfect
but it beats failing entirely.  Per example from Michael Fuhr.

19 years agoUse a bitmapset instead of a list for duplicate-column checking in
Tom Lane [Sat, 26 Mar 2005 06:28:59 +0000 (06:28 +0000)]
Use a bitmapset instead of a list for duplicate-column checking in
checkInsertTargets().  Avoids O(N^2) behavior on wide target lists.

19 years agoRewrite rewriteTargetList() to avoid O(N^2) behavior on wide target lists.
Tom Lane [Sat, 26 Mar 2005 05:53:01 +0000 (05:53 +0000)]
Rewrite rewriteTargetList() to avoid O(N^2) behavior on wide target lists.

19 years agoAdd Windows-specific variant comparison file.
Tom Lane [Sat, 26 Mar 2005 03:38:01 +0000 (03:38 +0000)]
Add Windows-specific variant comparison file.

19 years agoAdd another ORDER BY to rules test to eliminate platform-specific
Tom Lane [Sat, 26 Mar 2005 02:14:43 +0000 (02:14 +0000)]
Add another ORDER BY to rules test to eliminate platform-specific
output ordering.

19 years agoPrevent to_char(interval) from dumping core on month-related formats
Tom Lane [Sat, 26 Mar 2005 00:41:31 +0000 (00:41 +0000)]
Prevent to_char(interval) from dumping core on month-related formats
when a zero-month interval is given.  Per discussion with Karel.
Also, some desultory const-labeling of constant tables.  More could be
done along that line.

19 years agoUse 'cp' and 'chmod' in place of 'install' to install header files.
Bruce Momjian [Fri, 25 Mar 2005 23:22:54 +0000 (23:22 +0000)]
Use 'cp' and 'chmod' in place of 'install' to install header files.
This reduces header file install from 8 seconds to 0.40 seconds.

19 years agoAdd
Bruce Momjian [Fri, 25 Mar 2005 23:13:22 +0000 (23:13 +0000)]
Add

>  o Do VACUUM FULL if table is nearly empty?

19 years agoRemove lazy_update_relstats; go back to having VACUUM just record the
Tom Lane [Fri, 25 Mar 2005 22:51:31 +0000 (22:51 +0000)]
Remove lazy_update_relstats; go back to having VACUUM just record the
actual number of unremoved tuples as pg_class.reltuples.  The idea of
trying to estimate a steady state condition still seems attractive, but
this particular implementation crashed and burned ...

19 years agoImprove EXPLAIN ANALYZE to show the time spent in each trigger when
Tom Lane [Fri, 25 Mar 2005 21:58:00 +0000 (21:58 +0000)]
Improve EXPLAIN ANALYZE to show the time spent in each trigger when
executing a statement that fires triggers.  Formerly this time was
included in "Total runtime" but not otherwise accounted for.
As a side benefit, we avoid re-opening relations when firing non-deferred
AFTER triggers, because the trigger code can re-use the main executor's
ResultRelInfo data structure.

19 years agoFix resource owner code to generate catcache and relcache leak warnings
Tom Lane [Fri, 25 Mar 2005 18:30:28 +0000 (18:30 +0000)]
Fix resource owner code to generate catcache and relcache leak warnings
when open references remain during normal cleanup of a resource owner.
This restores the system's ability to warn about leaks to what it was
before 8.0.  Not really a user-level bug, but helpful for development.

19 years agoHave libpgport link before libpq so that PG client applications are more
Bruce Momjian [Fri, 25 Mar 2005 18:17:14 +0000 (18:17 +0000)]
Have libpgport link before libpq so that PG client applications are more
immunte to changes in libpq's usage of pgport between major versions.

19 years agoFix two bugs in change_owner_recurse_to_sequences: it was grabbing an
Tom Lane [Fri, 25 Mar 2005 18:04:34 +0000 (18:04 +0000)]
Fix two bugs in change_owner_recurse_to_sequences: it was grabbing an
overly strong lock on pg_depend, and it wasn't closing the rel when done.
The latter bug was masked by the ResourceOwner code, which is something
that should be changed.

19 years agoUpdate and copy-edit description of privileges.
Tom Lane [Fri, 25 Mar 2005 16:38:58 +0000 (16:38 +0000)]
Update and copy-edit description of privileges.

19 years agoAdd missing min/max parameters to DefineCustomIntVariable() and
Tom Lane [Fri, 25 Mar 2005 16:17:28 +0000 (16:17 +0000)]
Add missing min/max parameters to DefineCustomIntVariable() and
DefineCustomRealVariable().  Thomas Hallgren

19 years agoFix to_date to behave reasonably when CC and YY fields are both used.
Tom Lane [Fri, 25 Mar 2005 16:08:40 +0000 (16:08 +0000)]
Fix to_date to behave reasonably when CC and YY fields are both used.
Karel Zak

19 years agoMake initialization of special trigger variables a little more readable.
Tom Lane [Fri, 25 Mar 2005 01:45:42 +0000 (01:45 +0000)]
Make initialization of special trigger variables a little more readable.
Correct one mis-setting of freeval (which could at worst leak a few bytes
until the trigger exits, so it's no big deal).

19 years agoPrint URL in printed output..
Bruce Momjian [Fri, 25 Mar 2005 01:01:14 +0000 (01:01 +0000)]
Print URL in printed output..

19 years agoAdd URL of how to report bugs:
Bruce Momjian [Fri, 25 Mar 2005 00:51:04 +0000 (00:51 +0000)]
Add URL of how to report bugs:

http://www.chiark.greenend.org.uk/~sgtatham/bugs.html

19 years agoKerberos fixes from Magnus Hagander --- in theory Kerberos 5 auth
Tom Lane [Fri, 25 Mar 2005 00:34:31 +0000 (00:34 +0000)]
Kerberos fixes from Magnus Hagander --- in theory Kerberos 5 auth
should work on Windows now.  Also, rename set_noblock to pg_set_noblock;
since it is included in libpq, the former name polluted application
namespace.

19 years agoRevert ill-conceived change of libpq linkage --- breaks ecpg.
Tom Lane [Thu, 24 Mar 2005 23:53:48 +0000 (23:53 +0000)]
Revert ill-conceived change of libpq linkage --- breaks ecpg.

19 years agoarray_map can't use the fn_extra field of the provided fcinfo struct as
Tom Lane [Thu, 24 Mar 2005 21:50:38 +0000 (21:50 +0000)]
array_map can't use the fn_extra field of the provided fcinfo struct as
its private storage, because that belongs to the function that it is
supposed to call.  Per report from Ezequiel Tolnay.

19 years agoForce PG client applications to link to non-shared libpgport before
Bruce Momjian [Thu, 24 Mar 2005 19:33:32 +0000 (19:33 +0000)]
Force PG client applications to link to non-shared libpgport before
linking to libpq.  This insulates applications from changes in libpq's
usage of libpgport functions.

Backpatched to 8.0.X.

19 years agoTweak planner to use a minimum size estimate of 10 pages for a
Tom Lane [Thu, 24 Mar 2005 19:14:49 +0000 (19:14 +0000)]
Tweak planner to use a minimum size estimate of 10 pages for a
never-yet-vacuumed relation.  This restores the pre-8.0 behavior of
avoiding seqscans during initial data loading, while still allowing
reasonable optimization after a table has been vacuumed.  Several
regression test cases revert to 7.4-like behavior, which is probably
a good sign.  Per gripes from Keith Browne and others.

19 years agoSet socket timer to 58 instead of 60 minutes for hour-old cleaners:
Bruce Momjian [Thu, 24 Mar 2005 18:16:17 +0000 (18:16 +0000)]
Set socket timer to 58 instead of 60 minutes for hour-old cleaners:

         * Touch the socket and lock file at least every hour, to
         * ensure that they are not removed by overzealous /tmp-cleaning
         * tasks.  Set to 58 minutes so a cleaner never sees the
         * file as an hour old.

19 years agoFix python regression testing script to bail out early if language
Tom Lane [Thu, 24 Mar 2005 17:23:34 +0000 (17:23 +0000)]
Fix python regression testing script to bail out early if language
creation fails ... no point in running the tests.

19 years agoAdjust plpython to convert \r\n and \r to \n in Python scripts,
Tom Lane [Thu, 24 Mar 2005 17:22:34 +0000 (17:22 +0000)]
Adjust plpython to convert \r\n and \r to \n in Python scripts,
per recent discussion concluding that this is the Right Thing.  Add
regression test check for this behavior.  Michael Fuhr

19 years agoTouch postmaster log file every hour, rather than every 10 minutes, to
Bruce Momjian [Thu, 24 Mar 2005 05:19:05 +0000 (05:19 +0000)]
Touch postmaster log file every hour, rather than every 10 minutes, to
prevent complaints from laptop users who don't like their hard drives
starting up every 10 minutes.

19 years agoAdd:
Bruce Momjian [Thu, 24 Mar 2005 05:15:59 +0000 (05:15 +0000)]
Add:
<
> * Add GUC to issue notice about queries that use unjoined tables

19 years agoChange Win32 O_SYNC method to O_DSYNC because that is what the method
Bruce Momjian [Thu, 24 Mar 2005 04:36:20 +0000 (04:36 +0000)]
Change Win32 O_SYNC method to O_DSYNC because that is what the method
currently does.  This is now the default Win32 wal sync method because
we perfer o_datasync to fsync.

Also, change Win32 fsync to a new wal sync method called
fsync_writethrough because that is the behavior of _commit, which is
what is used for fsync on Win32.

Backpatch to 8.0.X.

19 years agoUpdate INTERVAL items.
Bruce Momjian [Thu, 24 Mar 2005 03:50:24 +0000 (03:50 +0000)]
Update INTERVAL items.

19 years agoAdd missing error checking in readdir() loops.
Tom Lane [Thu, 24 Mar 2005 02:11:20 +0000 (02:11 +0000)]
Add missing error checking in readdir() loops.

19 years agoChange ANSI to ISO standard:
Bruce Momjian [Thu, 24 Mar 2005 02:11:06 +0000 (02:11 +0000)]
Change ANSI to ISO standard:

< * Add ANSI INTERVAL handling
> * Add ISo INTERVAL handling
<  o Interpret syntax that isn't uniquely ANSI or PG, like '1:30' or
<    '1' as ANSI syntax, e.g. interpret '1:30' MINUTE TO SECOND as
>  o Interpret syntax that isn't uniquely ISO or PG, like '1:30' or
>    '1' as ISO syntax, e.g. interpret '1:30' MINUTE TO SECOND as
649c649
< * Add pre-parsing phase that converts non-ANSI syntax to supported
> * Add pre-parsing phase that converts non-ISO syntax to supported

19 years agoRevert changes to CREATE TRIGGER and ALTER TABLE ADD FOREIGN KEY locking,
Neil Conway [Thu, 24 Mar 2005 00:03:26 +0000 (00:03 +0000)]
Revert changes to CREATE TRIGGER and ALTER TABLE ADD FOREIGN KEY locking,
per request from Tom.

19 years agoUpdate wording:
Bruce Momjian [Wed, 23 Mar 2005 22:57:02 +0000 (22:57 +0000)]
Update wording:

<  o Process mixed ANSI/PG syntax, and round value to requested
<    precision or generate an error
<  o Interpret INTERVAL '1 year' MONTH as CAST (INTERVAL '1 year' AS
<    INTERVAL MONTH), and this should return '12 months'
194a191,194
>  o Interpret INTERVAL '1 year' MONTH as CAST (INTERVAL '1 year' AS
>    INTERVAL MONTH), and this should return '12 months'
>  o Round or truncate values to the requested precision, e.g.
>    INTERVAL '11 months' AS YEAR should return one or zero

19 years agoUpdate item on ambiguous INTERVAL syntax:
Bruce Momjian [Wed, 23 Mar 2005 21:49:52 +0000 (21:49 +0000)]
Update item on ambiguous INTERVAL syntax:

<  o Add support for day-time syntax, INTERVAL '1 2:03:04'
>  o Add support for day-time syntax, INTERVAL '1 2:03:04'
192c192,194
<  o Interpret INTERVAL '1:30' MINUTE TO SECOND as '1 minute 30 seconds'
>  o Interpret syntax that isn't uniquely ANSI or PG, like '1:30' or
>    '1' as ANSI syntax, e.g. interpret '1:30' MINUTE TO SECOND as
>    '1 minute 30 seconds'

19 years agoUpdate ANSI INTERVAL section:
Bruce Momjian [Wed, 23 Mar 2005 21:10:02 +0000 (21:10 +0000)]
Update ANSI INTERVAL section:

< * Add support for ANSI time INTERVAL syntax, INTERVAL '1 2:03:04' DAY TO SECOND
< * Add support for ANSI date INTERVAL syntax, INTERVAL '20-6' YEAR TO MONTH
< * Process mixed ANSI/PG INTERVAL syntax, and round value to requested precision
<
<   Interpret INTERVAL '1 year' MONTH as CAST (INTERVAL '1 year' AS INTERVAL
<   MONTH), and this should return '12 months'
<
< * Interpret INTERVAL '1:30' MINUTE TO SECOND as '1 minute 30 seconds'
> * Add ANSI INTERVAL handling
>  o Add support for day-time syntax, INTERVAL '1 2:03:04'
>    DAY TO SECOND
>  o Add support for year-month syntax, INTERVAL '50-6' YEAR TO MONTH
>  o Process mixed ANSI/PG syntax, and round value to requested
>    precision or generate an error
>  o Interpret INTERVAL '1 year' MONTH as CAST (INTERVAL '1 year' AS
>    INTERVAL MONTH), and this should return '12 months'
>  o Interpret INTERVAL '1:30' MINUTE TO SECOND as '1 minute 30 seconds'
>  o Support precision, CREATE TABLE foo (a INTERVAL MONTH(3))

19 years agoAdd note about risks involved in replaying CREATE TABLESPACE commands
Tom Lane [Wed, 23 Mar 2005 19:38:53 +0000 (19:38 +0000)]
Add note about risks involved in replaying CREATE TABLESPACE commands
from WAL.  A couple other grammatical improvements too.

19 years agoUpdate numbers in example:
Bruce Momjian [Wed, 23 Mar 2005 17:05:22 +0000 (17:05 +0000)]
Update numbers in example:

< * Add support for ANSI date INTERVAL syntax, INTERVAL '9-3' YEAR TO MONTH
> * Add support for ANSI date INTERVAL syntax, INTERVAL '20-6' YEAR TO MONTH

19 years agoUpdate:
Bruce Momjian [Wed, 23 Mar 2005 16:38:55 +0000 (16:38 +0000)]
Update:

< * Add support for ANSI date INTERVAL syntax, INTERVAL '1-2' YEAR TO MONTH
> * Add support for ANSI date INTERVAL syntax, INTERVAL '9-3' YEAR TO MONTH

19 years agoAdjust CREATE TRIGGER and ALTER TABLE ... ADD FOREIGN KEY to acquire
Neil Conway [Wed, 23 Mar 2005 07:44:57 +0000 (07:44 +0000)]
Adjust CREATE TRIGGER and ALTER TABLE ... ADD FOREIGN KEY to acquire
ExclusiveLock rather than AccessExclusiveLock. This will allow concurrent
SELECT queries to proceed on the table. Per discussion with Andrew at
SuperNews.

19 years agoAdd:
Bruce Momjian [Wed, 23 Mar 2005 06:11:57 +0000 (06:11 +0000)]
Add:

>  o Add MOVE to PL/pgSQL

19 years agoAssign:
Bruce Momjian [Wed, 23 Mar 2005 05:50:58 +0000 (05:50 +0000)]
Assign:
>  o Update pg_dump and psql to use the new COPY libpq API (Christopher)

19 years agoAdd:
Bruce Momjian [Wed, 23 Mar 2005 05:49:22 +0000 (05:49 +0000)]
Add:

>
> * Interpret INTERVAL '1:30' MINUTE TO SECOND as '1 minute 30 seconds'

19 years agoAdd INTERVAL items:
Bruce Momjian [Wed, 23 Mar 2005 05:42:40 +0000 (05:42 +0000)]
Add INTERVAL items:

> * Add support for ANSI time INTERVAL syntax, INTERVAL '1 2:03:04' DAY TO SECOND
> * Add support for ANSI date INTERVAL syntax, INTERVAL '1-2' YEAR TO MONTH
> * Process mixed ANSI/PG INTERVAL syntax, and round value to requested precision
184a188,189
>   Interpret INTERVAL '1 year' MONTH as CAST (INTERVAL '1 year' AS INTERVAL
>   MONTH), and this should return '12 months'

19 years agoRemove duplicate:
Bruce Momjian [Wed, 23 Mar 2005 02:50:56 +0000 (02:50 +0000)]
Remove duplicate:

< * Allow a warm standby system to also allow read-only queries

19 years agoAdd:
Bruce Momjian [Wed, 23 Mar 2005 02:49:42 +0000 (02:49 +0000)]
Add:

> * Allow WAL replay of CREATE TABLESPACE to work when the directory
>   structure on the recovery computer is different from the original

19 years agoAdd:
Bruce Momjian [Wed, 23 Mar 2005 02:41:29 +0000 (02:41 +0000)]
Add:

>  o Update pg_dump and psql to use the new COPY libpq API

19 years agoRemove:
Bruce Momjian [Wed, 23 Mar 2005 02:21:01 +0000 (02:21 +0000)]
Remove:
<  o Update to use the newer COPY syntax

19 years agoAdd for pg_dump:
Bruce Momjian [Wed, 23 Mar 2005 02:14:11 +0000 (02:14 +0000)]
Add for pg_dump:

>  o Update to use the newer COPY syntax

19 years agoAdd:
Bruce Momjian [Wed, 23 Mar 2005 01:25:07 +0000 (01:25 +0000)]
Add:

>
> * Support table partitioning that allows a single table to be stored
>   in subtables that are partitioned based on the primary key or a WHERE
>   clause

19 years agoRemove:
Bruce Momjian [Wed, 23 Mar 2005 00:36:14 +0000 (00:36 +0000)]
Remove:

<
<  o Prevent conflicting SET options from being set
<
<    This requires a checking function to be called after the server
<    configuration file is read.

19 years agoWAL must log CREATE and DROP DATABASE operations *without* using any
Tom Lane [Wed, 23 Mar 2005 00:03:37 +0000 (00:03 +0000)]
WAL must log CREATE and DROP DATABASE operations *without* using any
explicit paths, so that the log can be replayed in a data directory
with a different absolute path than the original had.  To avoid forcing
initdb in the 8.0 branch, continue to accept the old WAL log record
types; they will never again be generated however, and the code can be
dropped after the next forced initdb.  Per report from Oleg Bartunov.
We still need to think about what it really means to WAL-log CREATE
TABLESPACE commands: we more or less have to put the absolute path
into those, but how to replay in a different context??

19 years agoUse InitFunctionCallInfoData() macro instead of MemSet in performance
Tom Lane [Tue, 22 Mar 2005 20:13:09 +0000 (20:13 +0000)]
Use InitFunctionCallInfoData() macro instead of MemSet in performance
critical places in execQual.  By Atsushi Ogawa; some minor cleanup by moi.

19 years agoCreate a routine PageIndexMultiDelete() that replaces a loop around
Tom Lane [Tue, 22 Mar 2005 06:17:03 +0000 (06:17 +0000)]
Create a routine PageIndexMultiDelete() that replaces a loop around
PageIndexTupleDelete() with a single pass of compactification ---
logic mostly lifted from PageRepairFragmentation.  I noticed while
profiling that a VACUUM that's cleaning up a whole lot of deleted
tuples would spend as much as a third of its CPU time in
PageIndexTupleDelete; not too surprising considering the loop method
was roughly O(N^2) in the number of tuples involved.

19 years agoFix quote_ident to use quote_identifier rather than its own, not quite
Tom Lane [Mon, 21 Mar 2005 16:29:20 +0000 (16:29 +0000)]
Fix quote_ident to use quote_identifier rather than its own, not quite
up-to-speed logic; in particular this will cause it to quote names that
match keywords.  Remove unnecessary multibyte cruft from quote_literal
(all backend-internal encodings are 8-bit-safe).

19 years agopgcrypto update:
Neil Conway [Mon, 21 Mar 2005 05:24:52 +0000 (05:24 +0000)]
pgcrypto update:

* test error handling
* add tests for des, 3des, cast5
* add some tests to blowfish, rijndael
* Makefile: ability to specify different tests for different crypto
  libraries, so we can skip des, 3des and cast5 for builtin.

Marko Kreen

19 years agopgcrypto update:
Neil Conway [Mon, 21 Mar 2005 05:22:14 +0000 (05:22 +0000)]
pgcrypto update:

Reserve px_get_random_bytes() for strong randomness,
add new function px_get_pseudo_random_bytes() for
weak randomness and use it in gen_salt().

On openssl case, use RAND_pseudo_bytes() for
px_get_pseudo_random_bytes().

Final result is that is user has not configured random
souce but kept the 'silly' one, gen_salt() keeps
working, but pgp_encrypt() will throw error.

Marko Kreen

19 years agopgcrypto update:
Neil Conway [Mon, 21 Mar 2005 05:21:04 +0000 (05:21 +0000)]
pgcrypto update:

* openssl.c: Add 3des and AES support
* README.pgcrypto: list only supported ciphers for openssl

OpenSSL has pre-processor symbol OPENSSL_NO_AES, which
isn't that helpful for detecting if it _does_ exist.
Thus the hack with AES_ENCRYPT.

Marko Kreen

19 years agopgcrypto update:
Neil Conway [Mon, 21 Mar 2005 05:19:55 +0000 (05:19 +0000)]
pgcrypto update:

* Use error codes instead of -1
* px_strerror for new error codes
* calling convention change for px_gen_salt - return error code
* use px_strerror in pgcrypto.c

Marko Kreen

19 years ago* construct "struct {} list [] = {}" confuses pgindent - split those.
Neil Conway [Mon, 21 Mar 2005 05:18:46 +0000 (05:18 +0000)]
* construct "struct {} list [] = {}" confuses pgindent - split those.
  It was a bad style to begin with, and now several loops can be clearer.
* pgcrypto.c: Fix function comments
* crypt-gensalt.c, crypt-blowfish.c: stop messing with errno
* openssl.c: use px_free instead pfree
* px.h: make redefining px_alloc/px_realloc/px_free easier

Marko Kreen

19 years agoRemove support for libmhash/libmcrypt.
Neil Conway [Mon, 21 Mar 2005 05:17:16 +0000 (05:17 +0000)]
Remove support for libmhash/libmcrypt.

libmcrypt seems to dead, maintainer address bounces,
and cast-128 fails on 2 of the 3 test vectors from RFC2144.

So I see no reason to keep around stuff I don't trust
anymore.

Support for several crypto libraries is probably only
confusing to users, although it was good for initial
developing - it helped to find hidden assumptions and
forced me to create regression tests for all functionality.

Marko Kreen

19 years agoConvert index-related tuple handling routines from char 'n'/' ' to bool
Tom Lane [Mon, 21 Mar 2005 01:24:04 +0000 (01:24 +0000)]
Convert index-related tuple handling routines from char 'n'/' ' to bool
convention for isnull flags.  Also, remove the useless InsertIndexResult
return struct from index AM aminsert calls --- there is no reason for
the caller to know where in the index the tuple was inserted, and we
were wasting a palloc cycle per insert to deliver this uninteresting
value (plus nontrivial complexity in some AMs).
I forced initdb because of the change in the signature of the aminsert
routines, even though nothing really looks at those pg_proc entries...

19 years agoChange the return value of HeapTupleSatisfiesUpdate() to be an enum,
Neil Conway [Sun, 20 Mar 2005 23:40:34 +0000 (23:40 +0000)]
Change the return value of HeapTupleSatisfiesUpdate() to be an enum,
rather than an integer, and fix the associated fallout. From Alvaro
Herrera.

19 years agoOn Windows, use QueryPerformanceCounter instead of gettimeofday for
Tom Lane [Sun, 20 Mar 2005 22:27:52 +0000 (22:27 +0000)]
On Windows, use QueryPerformanceCounter instead of gettimeofday for
EXPLAIN ANALYZE instrumentation.  Magnus Hagander

19 years agoRemove unnecessary calls of FlushRelationBuffers: there is no need
Tom Lane [Sun, 20 Mar 2005 22:00:54 +0000 (22:00 +0000)]
Remove unnecessary calls of FlushRelationBuffers: there is no need
to write out data that we are about to tell the filesystem to drop.
smgr_internal_unlink already had a DropRelFileNodeBuffers call to
get rid of dead buffers without a write after it's no longer possible
to roll back the deleting transaction.  Adding a similar call in
smgrtruncate simplifies callers and makes the overall division of
labor clearer.  This patch removes the former behavior that VACUUM
would write all dirty buffers of a relation unconditionally.

19 years agoAdd mention of why malloc() has to be used in snprintf.c.
Bruce Momjian [Sun, 20 Mar 2005 13:54:53 +0000 (13:54 +0000)]
Add mention of why malloc() has to be used in snprintf.c.

19 years agoDepartment of second thoughts. Remove FRONTEND from snprintf.c because
Bruce Momjian [Sun, 20 Mar 2005 03:53:39 +0000 (03:53 +0000)]
Department of second thoughts.  Remove FRONTEND from snprintf.c because
snprintf is called before the memory system is started.  We have to just
malloc/free.  There are no elogs in the code so we should be fine.

19 years agoFix typo in Makefile.
Bruce Momjian [Sun, 20 Mar 2005 02:44:45 +0000 (02:44 +0000)]
Fix typo in Makefile.

19 years agoAnother change for FRONTEND snprintf.c.
Bruce Momjian [Sun, 20 Mar 2005 02:40:28 +0000 (02:40 +0000)]
Another change for FRONTEND snprintf.c.

19 years agoMark snprintf.c as a file that uses FRONTEND and needs to a version in
Bruce Momjian [Sun, 20 Mar 2005 02:39:33 +0000 (02:39 +0000)]
Mark snprintf.c as a file that uses FRONTEND and needs to a version in
the server-side port library.  Somehow I missed that change when I added
memory allocation to snprintf.c.

19 years agoDone:
Bruce Momjian [Sun, 20 Mar 2005 01:57:11 +0000 (01:57 +0000)]
Done:
> * -Allow the size of the buffer cache used by temporary objects to be

19 years agoAdd temp_buffers GUC variable to allow users to determine the size
Tom Lane [Sat, 19 Mar 2005 23:27:11 +0000 (23:27 +0000)]
Add temp_buffers GUC variable to allow users to determine the size
of the local buffer arena for temporary table access.

19 years agoUpgrade localbuf.c to use a hash table instead of linear search to
Tom Lane [Sat, 19 Mar 2005 17:39:43 +0000 (17:39 +0000)]
Upgrade localbuf.c to use a hash table instead of linear search to
find already-allocated local buffers.  This is the last obstacle
in the way of setting NLocBuffer to something reasonably large.

19 years agoPut 'dump complete' message in the right place, so it comes out where
Tom Lane [Fri, 18 Mar 2005 17:32:55 +0000 (17:32 +0000)]
Put 'dump complete' message in the right place, so it comes out where
it's supposed to when --file option is used.

19 years agoNeed to reset local buffer pin counts, not only shared buffer pins,
Tom Lane [Fri, 18 Mar 2005 16:16:09 +0000 (16:16 +0000)]
Need to reset local buffer pin counts, not only shared buffer pins,
before we attempt any file deletions in ShutdownPostgres.  Per Tatsuo.

19 years agoAdded patch by Christof Petig <christof@petig-baender.de> to work around gcc bug...
Michael Meskes [Fri, 18 Mar 2005 10:00:43 +0000 (10:00 +0000)]
Added patch by Christof Petig <christof@petig-baender.de> to work around gcc bug on powerpc and amd64.

19 years agoAvoid infinite loop in InvalidateBuffer if we ourselves are holding
Tom Lane [Fri, 18 Mar 2005 05:25:23 +0000 (05:25 +0000)]
Avoid infinite loop in InvalidateBuffer if we ourselves are holding
a pin on the victim buffer.

19 years agoNeed to release buffer pins before attempting to drop files during
Tom Lane [Fri, 18 Mar 2005 05:24:13 +0000 (05:24 +0000)]
Need to release buffer pins before attempting to drop files during
backend exit.  Per report from Bruce.