OSDN Git Service

pg-rex/syncrep.git
19 years agoHave psql escape bytes in strings for variables follow the backend
Bruce Momjian [Mon, 30 May 2005 14:50:35 +0000 (14:50 +0000)]
Have psql escape bytes in strings for variables follow the backend
conventions of only allowing octal, like \045.  Remove support for
\decimal, \0octal, and \0xhex which matches the strtol() function but
didn't make sense with backslashes.

These now return the same character:

test=> \set x '\54'
test=> \echo :x
,
test=> \set x '\054'
test=> \echo :x
,

THIS IS A BACKWARD COMPATIBILITY CHANGE.

19 years agoAdd missing <P>.
Bruce Momjian [Mon, 30 May 2005 13:11:06 +0000 (13:11 +0000)]
Add missing <P>.

19 years agoWhen enqueueing after-row triggers for updates of a table with a foreign
Neil Conway [Mon, 30 May 2005 07:20:59 +0000 (07:20 +0000)]
When enqueueing after-row triggers for updates of a table with a foreign
key, compare the new and old row versions. If the foreign key column has
not changed, we needn't enqueue the trigger, since the update cannot
violate the foreign key. This optimization was previously applied in the
RI trigger function, but it is more efficient to avoid firing the trigger
altogether. Per recent discussion on pgsql-hackers.

Also add a regression test for some unintuitive foreign key behavior, and
refactor some code that deals with the OIDs of the various RI trigger
functions.

19 years agoCreate separate ON INSERT and ON UPDATE triggers on tables with foreign
Neil Conway [Mon, 30 May 2005 06:52:38 +0000 (06:52 +0000)]
Create separate ON INSERT and ON UPDATE triggers on tables with foreign
keys, rather than a single trigger for both events. This should not change
functionality, but it is more consistent: previously, there were trigger
functions for both "check_insert" and "check_update", but the former was
used for both events.

Bump catalog version number (not strictly necessary, but best to be
cautious).

19 years agoFix longstanding oversight in ruleutils.c: it doesn't regurgitate
Tom Lane [Mon, 30 May 2005 01:57:27 +0000 (01:57 +0000)]
Fix longstanding oversight in ruleutils.c: it doesn't regurgitate
a FOR UPDATE clause, if one is present.

19 years agoChange the UNKNOWN type to have an internal representation matching
Tom Lane [Mon, 30 May 2005 01:20:50 +0000 (01:20 +0000)]
Change the UNKNOWN type to have an internal representation matching
cstring, rather than text, so as to eliminate useless conversions
inside the parser.  Per recent discussion.

19 years agoSkip eval_const_expressions when the query is such that the expression
Tom Lane [Mon, 30 May 2005 01:04:44 +0000 (01:04 +0000)]
Skip eval_const_expressions when the query is such that the expression
would be evaluated only once anyway (ie, it's just a SELECT with no
FROM or an INSERT ... VALUES).  The planner can't do it any faster than
the executor, so no point in an extra copying of the expression tree.

19 years agoAvoid unnecessary fetch from pg_shadow in the normal case in
Tom Lane [Sun, 29 May 2005 23:38:05 +0000 (23:38 +0000)]
Avoid unnecessary fetch from pg_shadow in the normal case in
pg_class_aclmask().  We only need to do this when we have to check
pg_shadow.usecatupd, and that's not relevant unless the target table
is a system catalog.  So we can usually avoid one syscache lookup.

19 years agoImprove LockAcquire API per my recent proposal. All error conditions
Tom Lane [Sun, 29 May 2005 22:45:02 +0000 (22:45 +0000)]
Improve LockAcquire API per my recent proposal.  All error conditions
are now reported via elog, eliminating the need to test the result code
at most call sites.  Make it possible for the caller to distinguish a
freshly acquired lock from one already held in the current transaction.
Use that capability to avoid redundant AcceptInvalidationMessages() calls
in LockRelation().

19 years agoMake superuser.c maintain a simple one-entry cache holding the superuser
Tom Lane [Sun, 29 May 2005 20:38:06 +0000 (20:38 +0000)]
Make superuser.c maintain a simple one-entry cache holding the superuser
status of the most recently queried userid.  Since the common pattern is
many successive queries about the same user (ie, the current user) this
can save a lot of syscache probes.

19 years agoMarginal hack to save a little bit of time in bpcharin() when typmod is -1,
Tom Lane [Sun, 29 May 2005 20:15:59 +0000 (20:15 +0000)]
Marginal hack to save a little bit of time in bpcharin() when typmod is -1,
which is a common case.

19 years agoAvoid unnecessary call of rangeTableEntry_used() for the result relation
Tom Lane [Sun, 29 May 2005 18:34:57 +0000 (18:34 +0000)]
Avoid unnecessary call of rangeTableEntry_used() for the result relation
of a query.

19 years agoRemove typeidIsValid() checks in can_coerce_type(). These checks
Tom Lane [Sun, 29 May 2005 18:24:14 +0000 (18:24 +0000)]
Remove typeidIsValid() checks in can_coerce_type().  These checks
were pretty expensive and I believe the case they were put in to
defend against can no longer arise, now that we have dependency checks
to prevent deletion of a type entry that is still referenced.  Certainly
the example given in the CVS log entry can't happen anymore.
Since this was the only use of typeidIsValid(), remove the routine too.

19 years agoexpandRTE and get_rte_attribute_type mistakenly always imputed typmod -1
Tom Lane [Sun, 29 May 2005 17:10:23 +0000 (17:10 +0000)]
expandRTE and get_rte_attribute_type mistakenly always imputed typmod -1
to columns of an RTE that was a function returning RECORD with a column
definition list.  Apparently no one has tried to use non-default typmod
with a function returning RECORD before.

19 years agoModify hash_search() API to prevent future occurrences of the error
Tom Lane [Sun, 29 May 2005 04:23:07 +0000 (04:23 +0000)]
Modify hash_search() API to prevent future occurrences of the error
spotted by Qingqing Zhou.  The HASH_ENTER action now automatically
fails with elog(ERROR) on out-of-memory --- which incidentally lets
us eliminate duplicate error checks in quite a bunch of places.  If
you really need the old return-NULL-on-out-of-memory behavior, you
can ask for HASH_ENTER_NULL.  But there is now an Assert in that path
checking that you aren't hoping to get that behavior in a palloc-based
hash table.
Along the way, remove the old HASH_FIND_SAVE/HASH_REMOVE_SAVED actions,
which were not being used anywhere anymore, and were surely too ugly
and unsafe to want to see revived again.

19 years agoAdd reference to libpq environment variables for client applications.
Bruce Momjian [Sun, 29 May 2005 03:32:18 +0000 (03:32 +0000)]
Add reference to libpq environment variables for client applications.

19 years agoBgwriter should PANIC if it runs out of memory for pending-fsyncs
Tom Lane [Sat, 28 May 2005 17:21:32 +0000 (17:21 +0000)]
Bgwriter should PANIC if it runs out of memory for pending-fsyncs
hash table.  This is a pretty unlikely scenario, since the table
should be tiny, but we can't guarantee continued correct operation
if it does occur.  Spotted by Qingqing Zhou.

19 years agoget_expr_result_type has to be prepared to pull type information
Tom Lane [Sat, 28 May 2005 05:10:47 +0000 (05:10 +0000)]
get_expr_result_type has to be prepared to pull type information
from a RECORD Const node, because that's what it may be faced with
after constant-folding of a function returning RECORD.  Per example
from Michael Fuhr.

19 years agoRemove:
Bruce Momjian [Sat, 28 May 2005 04:12:13 +0000 (04:12 +0000)]
Remove:

<
< * Add XML output to pg_dump and COPY
<
<   We already allow XML to be stored in the database, and XPath queries
<   can be used on that data using /contrib/xml2. It also supports XSLT
<   transformations.

19 years agoArrange to cache fmgr lookup information for an index's access method
Tom Lane [Fri, 27 May 2005 23:31:21 +0000 (23:31 +0000)]
Arrange to cache fmgr lookup information for an index's access method
routines in the index's relcache entry, instead of doing a fresh fmgr_info
on every index access.  We were already doing this for the index's opclass
support functions; not sure why we didn't think to do it for the AM
functions too.  This supersedes the former method of caching (only)
amgettuple in indexscan scan descriptors; it's an improvement because the
function lookup can be amortized across multiple statements instead of
being repeated for each statement.  Even though lookup for builtin
functions is pretty cheap, this seems to drop a percent or two off some
simple benchmarks.

19 years agoUpdate Farsi FAQ.
Bruce Momjian [Fri, 27 May 2005 22:09:42 +0000 (22:09 +0000)]
Update Farsi FAQ.

Mahmoud Taghizadeh

19 years agoAdd:
Bruce Momjian [Fri, 27 May 2005 22:07:26 +0000 (22:07 +0000)]
Add:

> * Consider sorting hash buckets so entries can be found using a binary
>   search, rather than a linear scan
> * In hash indexes, consider storing the hash value with or instead
>   of the key itself

19 years agoAdd:
Bruce Momjian [Fri, 27 May 2005 22:01:18 +0000 (22:01 +0000)]
Add:

> * Add the features of packages
>  o  Make private objects accessable only to objects in the same schema
>  o  Allow current_schema.objname to access current schema objects
>  o  Add session variables
>  o  Allow nested schemas

19 years agoDisplay only 9 subsecond digits instead of 10 for time values, for
Bruce Momjian [Fri, 27 May 2005 21:31:23 +0000 (21:31 +0000)]
Display only 9 subsecond digits instead of 10 for time values, for
consistency and to prevent rounding for days < 30.  Also round off all
trailing zeros, rather than leaving an even number of digits.

19 years agoRe-order 'else' clause for clarity.
Bruce Momjian [Fri, 27 May 2005 18:33:30 +0000 (18:33 +0000)]
Re-order 'else' clause for clarity.

19 years agoClean up bogus checking of date and numeric fields in DBF files,
Tom Lane [Fri, 27 May 2005 15:44:04 +0000 (15:44 +0000)]
Clean up bogus checking of date and numeric fields in DBF files,
per report from Boris van Schooten.

19 years agoBack out part of patch that should be applied later.
Bruce Momjian [Fri, 27 May 2005 15:16:45 +0000 (15:16 +0000)]
Back out part of patch that should be applied later.

19 years agoFix compile of entab to use stdarg.h. Clean up includes.
Bruce Momjian [Fri, 27 May 2005 15:15:31 +0000 (15:15 +0000)]
Fix compile of entab to use stdarg.h.  Clean up includes.

Marko Kreen

19 years agoRemove second argument from textToQualifiedNameList(), as it is no longer
Neil Conway [Fri, 27 May 2005 00:57:49 +0000 (00:57 +0000)]
Remove second argument from textToQualifiedNameList(), as it is no longer
used. From Jaime Casanova.

19 years agoDocument that utility commands such as DROP FUNCTION allow argument names
Tom Lane [Thu, 26 May 2005 20:05:03 +0000 (20:05 +0000)]
Document that utility commands such as DROP FUNCTION allow argument names
and modes to be included when naming an existing function.

19 years agoUse Abs() macro to replace ? :.
Bruce Momjian [Thu, 26 May 2005 16:44:05 +0000 (16:44 +0000)]
Use Abs() macro to replace ? :.

19 years agoBack out:
Bruce Momjian [Thu, 26 May 2005 15:26:00 +0000 (15:26 +0000)]
Back out:

Display only 9 not 10 digits of precision for timestamp values when
using non-integer timestamps.  This prevents the display of rounding
errors for common values like days < 32.

19 years agoBack out SQLSTATE and SQLERRM support.
Bruce Momjian [Thu, 26 May 2005 04:08:32 +0000 (04:08 +0000)]
Back out SQLSTATE and SQLERRM support.

19 years agoDisplay only 9 not 10 digits of precision for timestamp values when
Bruce Momjian [Thu, 26 May 2005 03:48:25 +0000 (03:48 +0000)]
Display only 9 not 10 digits of precision for timestamp values when
using non-integer timestamps.  This prevents the display of rounding
errors for common values like days < 32.

19 years agoMinor cleanup for recent SQLSTATE / SQLERRM patch: spell "successful"
Neil Conway [Thu, 26 May 2005 03:18:53 +0000 (03:18 +0000)]
Minor cleanup for recent SQLSTATE / SQLERRM patch: spell "successful"
correctly, style fixes.

19 years agoAdjust datetime parsing to be more robust. We now pass the length of the
Neil Conway [Thu, 26 May 2005 02:04:14 +0000 (02:04 +0000)]
Adjust datetime parsing to be more robust. We now pass the length of the
working buffer into ParseDateTime() and reject too-long input there,
rather than checking the length of the input string before calling
ParseDateTime(). The old method was bogus because ParseDateTime() can use
a variable amount of working space, depending on the content of the
input string (e.g. how many fields need to be NUL terminated). This fixes
a minor stack overrun -- I don't _think_ it's exploitable, although I
won't claim to be an expert.

Along the way, fix a bug reported by Mark Dilger: the working buffer
allocated by interval_in() was too short, which resulted in rejecting
some perfectly valid interval input values. I added a regression test for
this fix.

19 years agoTweak the backend scanner (and psqlscan.l, which must track the backend
Tom Lane [Thu, 26 May 2005 01:24:29 +0000 (01:24 +0000)]
Tweak the backend scanner (and psqlscan.l, which must track the backend
scanner anyway) to avoid having any backup states.  According to the
flex manual, this should speed things up, and indeed the backend scanner
is about a third faster according to some quick profiling checks.
I haven't tried to measure the speed change in psql, but it probably
is similar.

19 years agoAdd PL/pgSQL SQLSTATE and SQLERRM support which sets these values on
Bruce Momjian [Thu, 26 May 2005 00:16:31 +0000 (00:16 +0000)]
Add PL/pgSQL SQLSTATE and SQLERRM support which sets these values on
error.

Pavel Stehule

19 years agoAt the head of wchareq, length of (multibyte) character is compared by
Bruce Momjian [Wed, 25 May 2005 22:59:33 +0000 (22:59 +0000)]
At the head of wchareq, length of (multibyte) character is compared by
using pg_mblen. Therefore, pg_mblen is executed many times, and it
becomes a bottleneck.

This patch makes a short cut, and reduces execution frequency of
pg_mblen by comparing the first byte first.

a_ogawa

19 years agoQuick patch to adress a recent concern on the mailing list
Bruce Momjian [Wed, 25 May 2005 22:12:05 +0000 (22:12 +0000)]
Quick patch to adress a recent concern on the mailing list
about adding an errant "TO" when we already have a TO. Since
TO cannot be a valid column name (we must quote it), we can
simply ignore the tab-completion if the previous word
was a "TO".

Greg Sabino Mullane

19 years agoAdd parentheses to macros when args are used in computations. Without
Bruce Momjian [Wed, 25 May 2005 21:40:43 +0000 (21:40 +0000)]
Add parentheses to macros when args are used in computations.  Without
them, the executation behavior could be unexpected.

19 years agoEditing for the new "Encryption Options" section of the manual.
Neil Conway [Wed, 25 May 2005 02:56:15 +0000 (02:56 +0000)]
Editing for the new "Encryption Options" section of the manual.

19 years agoPrevious fix for "x FULL JOIN y ON true" failed to handle the case
Tom Lane [Tue, 24 May 2005 18:02:31 +0000 (18:02 +0000)]
Previous fix for "x FULL JOIN y ON true" failed to handle the case
where there was also a WHERE-clause restriction that applied to the
join.  The check on restrictlist == NIL is really unnecessary anyway,
because select_mergejoin_clauses already checked for and complained
about any unmergejoinable join clauses.  So just take it out.

19 years agoFix PL makefiles to support running regression tests in VPATH builds.
Tom Lane [Tue, 24 May 2005 17:07:41 +0000 (17:07 +0000)]
Fix PL makefiles to support running regression tests in VPATH builds.

19 years agoAdd -I$(srcdir) to CPPFLAGS to make psqlscan.c compile in vpath builds.
Tom Lane [Tue, 24 May 2005 16:45:23 +0000 (16:45 +0000)]
Add -I$(srcdir) to CPPFLAGS to make psqlscan.c compile in vpath builds.
Not sure why this hasn't been reported before; perhaps it is not needed
with newer gcc versions, but it definitely fails here.

19 years agoInserting 5 characters into char(10) does not produce 5 padding spaces
Tatsuo Ishii [Tue, 24 May 2005 15:45:34 +0000 (15:45 +0000)]
Inserting 5 characters into char(10) does not produce 5 padding spaces
if they are two-byte multibyte characters. Same thing can be happen
if octet_length(multibyte_chars) == n where n is char(n).
Long standing bug since 7.3 days. Per report and fix from Yoshiyuki Asaba.

19 years agoAdd regression tests for previously-untested PL/Perl features. From
Neil Conway [Tue, 24 May 2005 08:05:36 +0000 (08:05 +0000)]
Add regression tests for previously-untested PL/Perl features. From
Andrew Dunstan.

19 years agoFix weird indentation of function return type.
Neil Conway [Tue, 24 May 2005 07:16:27 +0000 (07:16 +0000)]
Fix weird indentation of function return type.

19 years agoLog queries for client-side prepare/execute. Simon Riggs
Bruce Momjian [Tue, 24 May 2005 04:18:04 +0000 (04:18 +0000)]
Log queries for client-side prepare/execute.  Simon Riggs
Log prepare query during execute.  Bruce Momjian

19 years agoPut parentheses around use of macro arguments in FMODULO and TMODULO.
Bruce Momjian [Tue, 24 May 2005 04:03:01 +0000 (04:03 +0000)]
Put parentheses around use of macro arguments in FMODULO and TMODULO.

19 years agoRemove more extraneous parentheses in date/time functions.
Bruce Momjian [Tue, 24 May 2005 02:09:45 +0000 (02:09 +0000)]
Remove more extraneous parentheses in date/time functions.

19 years agoCorrect a thinko in pgbench that might result in incorrectly ignoring an
Neil Conway [Tue, 24 May 2005 00:26:40 +0000 (00:26 +0000)]
Correct a thinko in pgbench that might result in incorrectly ignoring an
error condition when executing some DDL. Per report from ITAGAKI Takahiro.

19 years agoMore macro cleanups for date/time.
Bruce Momjian [Mon, 23 May 2005 21:54:02 +0000 (21:54 +0000)]
More macro cleanups for date/time.

19 years agoAdd datetime macros for constants, for clarity:
Bruce Momjian [Mon, 23 May 2005 18:56:55 +0000 (18:56 +0000)]
Add datetime macros for constants, for clarity:

#define SECS_PER_DAY  86400
#define USECS_PER_DAY INT64CONST(86400000000)
#define USECS_PER_HOUR    INT64CONST(3600000000)
#define USECS_PER_MINUTE INT64CONST(60000000)
#define USECS_PER_SEC INT64CONST(1000000)

19 years agoRemove unnecessary parentheses in datetime/timestamp code.
Bruce Momjian [Mon, 23 May 2005 17:13:14 +0000 (17:13 +0000)]
Remove unnecessary parentheses in datetime/timestamp code.

19 years agoAvoid redundant relation lock grabs during planning, and make sure
Tom Lane [Mon, 23 May 2005 03:01:14 +0000 (03:01 +0000)]
Avoid redundant relation lock grabs during planning, and make sure
that we acquire a lock on relations added to the query due to inheritance.
Formerly, no such lock was held throughout planning, which meant that
a schema change could occur to invalidate the plan before it's even
been completed.

19 years agoFix typo in PL/Perl Safe.pm initialization that prevented the proper
Neil Conway [Mon, 23 May 2005 01:57:51 +0000 (01:57 +0000)]
Fix typo in PL/Perl Safe.pm initialization that prevented the proper
sharing of %_SHARED. From Andrew Dunstan.

19 years agoRemove some verbiage describing how min() and max() are slow when applied
Neil Conway [Mon, 23 May 2005 01:50:01 +0000 (01:50 +0000)]
Remove some verbiage describing how min() and max() are slow when applied
to the entire table: as of current sources, they are no longer slow
provided there is an index on the column.

19 years agoConsistently do not include a terminating period in single-sentence
Neil Conway [Mon, 23 May 2005 01:29:54 +0000 (01:29 +0000)]
Consistently do not include a terminating period in single-sentence
function descriptions in func.sgml; other minor SGML tweaks.

19 years agoTeach the planner to remove SubqueryScan nodes from the plan if they
Tom Lane [Sun, 22 May 2005 22:30:20 +0000 (22:30 +0000)]
Teach the planner to remove SubqueryScan nodes from the plan if they
aren't doing anything useful (ie, neither selection nor projection).
Also, extend to SubqueryScan the hacks already in place to avoid
unnecessary ExecProject calls when the result would just be the same
tuple the subquery already delivered.  This saves some overhead in
UNION and other set operations, as well as avoiding overhead for
unflatten-able subqueries.  Per example from Sokolov Yura.

19 years agoINT4 is probably enough:
Bruce Momjian [Sat, 21 May 2005 21:31:26 +0000 (21:31 +0000)]
INT4 is probably enough:

< * Allow INET + INT4/INT8 to increment the host part of the address, or
> * Allow INET + INT4 to increment the host part of the address, or

19 years agoCleanup of GiST extensions in contrib/: now that we always invoke GiST
Neil Conway [Sat, 21 May 2005 12:08:06 +0000 (12:08 +0000)]
Cleanup of GiST extensions in contrib/: now that we always invoke GiST
methods in a short-lived memory context, there is no need for GiST methods
to do their own manual (and error-prone) memory management.

19 years agoRemove excess parens, use Abs instead of : ?.
Bruce Momjian [Sat, 21 May 2005 03:38:05 +0000 (03:38 +0000)]
Remove excess parens, use Abs instead of : ?.

19 years agoMention overflow:
Bruce Momjian [Sat, 21 May 2005 03:12:44 +0000 (03:12 +0000)]
Mention overflow:

<   throw an error
>   throw an error on overflow

19 years agoAdd:
Bruce Momjian [Sat, 21 May 2005 03:11:27 +0000 (03:11 +0000)]
Add:

> * Allow INET + INT4/INT8 to increment the host part of the address, or
>   throw an error

19 years agoRemove 2-phase description, because it isn't accurate anymore:
Bruce Momjian [Fri, 20 May 2005 19:18:15 +0000 (19:18 +0000)]
Remove 2-phase description, because it isn't accurate anymore:
<
<   This will involve adding a way to respond to commit failure by either
<   taking the server into offline/readonly mode or notifying the
<   administrator

19 years agoUpdate Japanese FAQ.
Bruce Momjian [Fri, 20 May 2005 15:53:06 +0000 (15:53 +0000)]
Update Japanese FAQ.

Jun Kuwamura

19 years agoAdd support for wal_fsync_writethrough for Darwin, and restructure the
Bruce Momjian [Fri, 20 May 2005 14:53:26 +0000 (14:53 +0000)]
Add support for wal_fsync_writethrough for Darwin, and restructure the
code to better handle writethrough.

Chris Campbell

19 years agoFix typo in ECPG docs, per Andreas Seltenreich.
Neil Conway [Fri, 20 May 2005 12:36:48 +0000 (12:36 +0000)]
Fix typo in ECPG docs, per Andreas Seltenreich.

19 years agoAdd some links to the CREATE FUNCTION reference page when describing
Neil Conway [Fri, 20 May 2005 01:52:25 +0000 (01:52 +0000)]
Add some links to the CREATE FUNCTION reference page when describing
function definition for particular PLs. Original patch from David
Fetter, editorializing by Neil Conway.

19 years agoMake the CREATE RULE syntax description in rules.sgml more consistent
Neil Conway [Fri, 20 May 2005 01:37:08 +0000 (01:37 +0000)]
Make the CREATE RULE syntax description in rules.sgml more consistent
with the syntax description in the CREATE RULE reference page. From
Kris Jurka.

19 years agoImplement md5(bytea), update regression tests and documentation. Patch
Neil Conway [Fri, 20 May 2005 01:29:56 +0000 (01:29 +0000)]
Implement md5(bytea), update regression tests and documentation. Patch
from Abhijit Menon-Sen, minor editorialization from Neil Conway. Also,
improve md5(text) to allocate a constant-sized buffer on the stack
rather than via palloc.

Catalog version bumped.

19 years agoMake a comment pgindent-proof, per suggestion from Alvaro.
Tom Lane [Thu, 19 May 2005 23:58:51 +0000 (23:58 +0000)]
Make a comment pgindent-proof, per suggestion from Alvaro.

19 years agoUpdate comment that I missed the first time around.
Tom Lane [Thu, 19 May 2005 23:57:11 +0000 (23:57 +0000)]
Update comment that I missed the first time around.

19 years agoFactor out lock cleanup code that is needed in several places in lock.c.
Tom Lane [Thu, 19 May 2005 23:30:18 +0000 (23:30 +0000)]
Factor out lock cleanup code that is needed in several places in lock.c.
Also, remove the rather useless return value of LockReleaseAll.  Change
response to detection of corruption in the shared lock tables to PANIC,
since that is the only way of cleaning up fully.
Originally an idea of Heikki Linnakangas, variously hacked on by
Alvaro Herrera and Tom Lane.

19 years agoSplit the shared-memory array of PGPROC pointers out of the sinval
Tom Lane [Thu, 19 May 2005 21:35:48 +0000 (21:35 +0000)]
Split the shared-memory array of PGPROC pointers out of the sinval
communication structure, and make it its own module with its own lock.
This should reduce contention at least a little, and it definitely makes
the code seem cleaner.  Per my recent proposal.

19 years agoUpon closer inspection, Greg's psql tab completion patch leaks memory.
Neil Conway [Wed, 18 May 2005 05:01:10 +0000 (05:01 +0000)]
Upon closer inspection, Greg's psql tab completion patch leaks memory.
Fix the leak, and add a comment to note that the return value of
previous_word must be free'd.

19 years agopsql tab completion improvements, from Greg Sabino Mullane:
Neil Conway [Wed, 18 May 2005 04:47:40 +0000 (04:47 +0000)]
psql tab completion improvements, from Greg Sabino Mullane:

* Made DELETE into "DELETE FROM"
* Moved ANALZYE to the end of the list to ease EXPLAIN / VACUUM
  conflicts
* Removed the ANALYZE xx semicolon completion: we don't do that anywhere
  else
* Add DECLARE support
* Add parens for DROP AGGREGATE
* Add "CASCADE | RESTRICT" for DROP xx
* Make EXPLAIN <tab> a lot smarter
* GROUP "BY" and ORDER "BY"
* "ISOLATION" becomes "ISOLATION LEVEL"
* Fix error in which REVOKE xx ON yy was receiving "TO", now gets "FROM"
* Add GRANT/REVOKE xx ON yy TO/FROM choices: usernames, GROUP, PUBLIC
* PREPARE xx <tab> AS "SELECT | INSERT | UPDATE | DELETE"
* Add = at end of UPDATE xx SET yy
* Beef up VACUUM stuff

19 years agoExtend the pg_locks system view so that it can fully display all lock
Tom Lane [Tue, 17 May 2005 21:46:11 +0000 (21:46 +0000)]
Extend the pg_locks system view so that it can fully display all lock
types, as per recent discussion.

19 years agoAdd a --dbname option to the pg_regress script, and use pl_regression
Tom Lane [Tue, 17 May 2005 18:26:23 +0000 (18:26 +0000)]
Add a --dbname option to the pg_regress script, and use pl_regression
for testing PLs and contrib_regression for testing contrib, instead of
overwriting the core system's regression database as formerly done.
Andrew Dunstan

19 years agoGuard against duplicate IDs in input file in SortTocFromFile().
Tom Lane [Tue, 17 May 2005 17:30:29 +0000 (17:30 +0000)]
Guard against duplicate IDs in input file in SortTocFromFile().
Per report from Brian Hackett.

19 years agoUpdate:
Bruce Momjian [Tue, 17 May 2005 04:18:20 +0000 (04:18 +0000)]
Update:

< * All ability to monitor the use of temporary sort files
> * Add ability to monitor the use of temporary sort files

19 years agoConvert Chinese FAQ to valid XHTML, finally.
Bruce Momjian [Tue, 17 May 2005 04:17:47 +0000 (04:17 +0000)]
Convert Chinese FAQ to valid XHTML, finally.

19 years agoFix Chinese markup some more.
Bruce Momjian [Tue, 17 May 2005 03:40:35 +0000 (03:40 +0000)]
Fix Chinese markup some more.

19 years agoCleanup GiST header files. Since GiST extensions are often written as
Neil Conway [Tue, 17 May 2005 03:34:18 +0000 (03:34 +0000)]
Cleanup GiST header files. Since GiST extensions are often written as
external projects, we should be careful about what parts of the GiST
API are considered implementation details, and which are part of the
public API. Therefore, I've moved internal-only declarations into
gist_private.h -- future backward-incompatible changes to gist.h should
be made with care, to avoid needlessly breaking external GiST extensions.

Also did some related header cleanup: remove some unnecessary #includes
from gist.h, and remove some unused definitions: isAttByVal(), _gistdump(),
and GISTNStrategies.

19 years agoGiST improvements:
Neil Conway [Tue, 17 May 2005 00:59:30 +0000 (00:59 +0000)]
GiST improvements:

- make sure we always invoke user-supplied GiST methods in a short-lived
  memory context. This means the backend isn't exposed to any memory leaks
  that be in those methods (in fact, it is probably a net loss for most
  GiST methods to bother manually freeing memory now). This also means
  we can do away with a lot of ugly manual memory management in the
  GiST code itself.

- keep the current page of a GiST index scan pinned, rather than doing a
  ReadBuffer() for each tuple produced by the scan. Since ReadBuffer() is
  expensive, this is a perf. win

- implement dead tuple killing for GiST indexes (which is easy to do, now
  that we keep a pin on the current scan page). Now all the builtin indexes
  implement dead tuple killing.

- cleanup a lot of ugly code in GiST

19 years agoFix markup:
Bruce Momjian [Tue, 17 May 2005 00:51:58 +0000 (00:51 +0000)]
Fix markup:
<  * Add internationalized message strings
>  o Add internationalized message strings

19 years agoUpdate italics.
Bruce Momjian [Tue, 17 May 2005 00:51:10 +0000 (00:51 +0000)]
Update italics.

19 years agoModify tidbitmap.c to avoid creating a hash table until there is more
Tom Lane [Tue, 17 May 2005 00:43:47 +0000 (00:43 +0000)]
Modify tidbitmap.c to avoid creating a hash table until there is more
than one heap page represented in the bitmap.  This is a bit ugly but
it cuts overhead fairly effectively in simple join cases.  Per example
from Sergey Koposov.

19 years agoUpdate italics on web site.
Bruce Momjian [Tue, 17 May 2005 00:24:22 +0000 (00:24 +0000)]
Update italics on web site.

19 years agoUpdate italics on web site.
Bruce Momjian [Mon, 16 May 2005 23:35:42 +0000 (23:35 +0000)]
Update italics on web site.

19 years agoUpdate Chinese FAQ to validate properly.
Bruce Momjian [Mon, 16 May 2005 22:27:54 +0000 (22:27 +0000)]
Update Chinese FAQ to validate properly.

19 years agoFix typo in comment.
Neil Conway [Mon, 16 May 2005 06:45:57 +0000 (06:45 +0000)]
Fix typo in comment.

19 years agoFix a few minor mistakes in header comments. From Qingqing Zhou.
Neil Conway [Mon, 16 May 2005 05:52:13 +0000 (05:52 +0000)]
Fix a few minor mistakes in header comments. From Qingqing Zhou.

19 years agoUpdate Chinese FAQ to xhtml.
Bruce Momjian [Mon, 16 May 2005 02:50:20 +0000 (02:50 +0000)]
Update Chinese FAQ to xhtml.

19 years agoAdjust out-of-date comment.
Tom Lane [Mon, 16 May 2005 00:19:04 +0000 (00:19 +0000)]
Adjust out-of-date comment.

19 years agoFix latent bug in ExecSeqRestrPos: it leaves the plan node's result slot
Tom Lane [Sun, 15 May 2005 21:19:55 +0000 (21:19 +0000)]
Fix latent bug in ExecSeqRestrPos: it leaves the plan node's result slot
in an inconsistent state.  (This is only latent because in reality
ExecSeqRestrPos is dead code at the moment ... but someday maybe it won't
be.)  Add some comments about what the API for plan node mark/restore
actually is, because it's not immediately obvious.

19 years agoVarious style cleanups for GiST; no changes to functionality.
Neil Conway [Sun, 15 May 2005 04:08:29 +0000 (04:08 +0000)]
Various style cleanups for GiST; no changes to functionality.

19 years agoRename Rendezvous to Bonjour to match OS/X renaming.
Bruce Momjian [Sun, 15 May 2005 00:26:19 +0000 (00:26 +0000)]
Rename Rendezvous to Bonjour to match OS/X renaming.

19 years agoFurther marginal speed hacking: in MemoryContextReset, don't call
Tom Lane [Sat, 14 May 2005 23:16:29 +0000 (23:16 +0000)]
Further marginal speed hacking: in MemoryContextReset, don't call
MemoryContextResetChildren unless necessary.