]> granicus.if.org Git - postgresql/log
postgresql
24 years agoA few more macro cleanups
Bruce Momjian [Wed, 12 Apr 2000 20:33:37 +0000 (20:33 +0000)]
A few more macro cleanups

24 years agoUpdate man page to prefer old over current.
Bruce Momjian [Wed, 12 Apr 2000 20:07:13 +0000 (20:07 +0000)]
Update man page to prefer old over current.

24 years agoYe-old pgindent run. Same 4-space tabs.
Bruce Momjian [Wed, 12 Apr 2000 17:17:23 +0000 (17:17 +0000)]
Ye-old pgindent run.  Same 4-space tabs.

24 years agoReverse out macro fix for the time being.
Bruce Momjian [Wed, 12 Apr 2000 05:29:10 +0000 (05:29 +0000)]
Reverse out macro fix for the time being.

24 years agoPrevent drop database failure from showing.
Bruce Momjian [Wed, 12 Apr 2000 05:24:51 +0000 (05:24 +0000)]
Prevent drop database failure from showing.

24 years agoFix assert with missing semicolon
Bruce Momjian [Wed, 12 Apr 2000 04:58:09 +0000 (04:58 +0000)]
Fix assert with missing semicolon

24 years agoFix up markup to allow building.
Thomas G. Lockhart [Wed, 12 Apr 2000 04:40:03 +0000 (04:40 +0000)]
Fix up markup to allow building.
Adjust two-word substitution parameters to be one word only to reduce
 possible ambiguity.

24 years agoUpdate pgindent
Bruce Momjian [Wed, 12 Apr 2000 01:01:49 +0000 (01:01 +0000)]
Update pgindent

24 years agoUpdate pgindent
Bruce Momjian [Tue, 11 Apr 2000 22:15:08 +0000 (22:15 +0000)]
Update pgindent

24 years agoUpdate pgindent for 7.0 release
Bruce Momjian [Tue, 11 Apr 2000 19:09:04 +0000 (19:09 +0000)]
Update pgindent for 7.0 release

24 years agoThomas pointed out this bug in my non-blocking stuff.
Bruce Momjian [Tue, 11 Apr 2000 19:00:31 +0000 (19:00 +0000)]
Thomas pointed out this bug in my non-blocking stuff.

Alfred Perlstein

24 years agoAllow libpq++ compile to fail
Bruce Momjian [Tue, 11 Apr 2000 18:58:18 +0000 (18:58 +0000)]
Allow libpq++ compile to fail

24 years agodisable plperl
Bruce Momjian [Tue, 11 Apr 2000 17:42:28 +0000 (17:42 +0000)]
disable plperl

24 years agoClean up temp files from \e.
Bruce Momjian [Tue, 11 Apr 2000 17:35:50 +0000 (17:35 +0000)]
Clean up temp files from \e.

24 years agoUppercase pg_options param
Bruce Momjian [Tue, 11 Apr 2000 17:23:35 +0000 (17:23 +0000)]
Uppercase pg_options param

24 years agoInternal functions to support newest ODBC driver {fn ...} conventions.
Thomas G. Lockhart [Tue, 11 Apr 2000 15:53:13 +0000 (15:53 +0000)]
Internal functions to support newest ODBC driver {fn ...} conventions.
Includes compiled code to support pre-7.0 backends, but for 7.0 only
requires executing odbc.sql.

24 years ago Attached is are diffs for CREATE/ALTER table doc I've
Bruce Momjian [Tue, 11 Apr 2000 14:43:54 +0000 (14:43 +0000)]
Attached  is  are  diffs  for  CREATE/ALTER  table  doc  I've
    forgotten in my mailbox (sorry). Haven't tried to  apply  and
    since I don't have working sgml stuff cannot check.

Jan

24 years agoUpdate porting list.
Thomas G. Lockhart [Tue, 11 Apr 2000 05:39:15 +0000 (05:39 +0000)]
Update porting list.
Fix markup to get clean compile.

24 years agoUpdate for default ISO date style
Bruce Momjian [Tue, 11 Apr 2000 03:56:19 +0000 (03:56 +0000)]
Update for default ISO date style

24 years agoUpdate SET manual pages for pg_options.
Bruce Momjian [Tue, 11 Apr 2000 03:49:04 +0000 (03:49 +0000)]
Update SET manual pages for pg_options.

24 years agoTweak smgrblindwrt per advice from Vadim: add parameter indicating
Tom Lane [Mon, 10 Apr 2000 23:41:52 +0000 (23:41 +0000)]
Tweak smgrblindwrt per advice from Vadim: add parameter indicating
whether to do fsync or not, and if so (which should be seldom) just
do the fsync immediately.  This way we need not build data structures
in md.c/fd.c for blind writes.

24 years agoThe constant in backend/utils/misc/trace.c, line 32 limits the length of
Bruce Momjian [Mon, 10 Apr 2000 19:25:29 +0000 (19:25 +0000)]
The constant in backend/utils/misc/trace.c, line 32 limits the length of
logged queries to 1024, truncating longer queries.  That is about half of
the size I need (I have a union that is 2K long).  Can someone consider
bumping it to 4K or so?  Patch attached...

Regards,
Ed Loehr

24 years agoNo revoke on indexes any more
Bruce Momjian [Mon, 10 Apr 2000 01:35:44 +0000 (01:35 +0000)]
No revoke on indexes any more

24 years agoFix an Assert bug(was my fault) pointed out by Tom Lane.
Hiroshi Inoue [Mon, 10 Apr 2000 00:45:42 +0000 (00:45 +0000)]
Fix an Assert bug(was my fault) pointed out by Tom Lane.

24 years agoUPdate faq/
Bruce Momjian [Sun, 9 Apr 2000 19:17:27 +0000 (19:17 +0000)]
UPdate faq/

24 years agoNo more need to create groups manually.
Bruce Momjian [Sun, 9 Apr 2000 12:14:25 +0000 (12:14 +0000)]
No more need to create groups manually.

24 years agoNo grant on indexes.
Bruce Momjian [Sun, 9 Apr 2000 12:08:33 +0000 (12:08 +0000)]
No grant on indexes.

24 years agoBuffer manager modifications to keep a local buffer-dirtied bit as well
Tom Lane [Sun, 9 Apr 2000 04:43:20 +0000 (04:43 +0000)]
Buffer manager modifications to keep a local buffer-dirtied bit as well
as a shared dirtybit for each shared buffer.  The shared dirtybit still
controls writing the buffer, but the local bit controls whether we need
to fsync the buffer's file.  This arrangement fixes a bug that allowed
some required fsyncs to be missed, and should improve performance as well.
For more info see my post of same date on pghackers.

24 years agoFurther tweaking of indexscan cost estimates.
Tom Lane [Sun, 9 Apr 2000 04:31:37 +0000 (04:31 +0000)]
Further tweaking of indexscan cost estimates.

24 years agoAdd explanation about 'Unrecognized variable client_encoding'
Tom Lane [Sat, 8 Apr 2000 23:32:34 +0000 (23:32 +0000)]
Add explanation about 'Unrecognized variable client_encoding'
message at startup.

24 years agoDocument FSYNC in pg_options sgml file.
Bruce Momjian [Sat, 8 Apr 2000 23:12:01 +0000 (23:12 +0000)]
Document FSYNC in pg_options sgml file.

24 years agoAdd new pg_options.sample file.
Bruce Momjian [Sat, 8 Apr 2000 19:38:00 +0000 (19:38 +0000)]
Add new pg_options.sample file.

24 years agoTweak TypeCategory to treat new BIT types as of STRING category, rather
Tom Lane [Sat, 8 Apr 2000 19:29:40 +0000 (19:29 +0000)]
Tweak TypeCategory to treat new BIT types as of STRING category, rather
than not knowing what they are at all.  Perhaps they should have their own
type category?  Hard to say.  In the meantime, doing it this way allows
SELECT 'unknown' || 'unknown' to continue being resolved as textcat,
instead of spitting out an ambiguous-operator error.

24 years agoI think we want machine pattern i.86 not i386 --- looks like config.guess
Tom Lane [Sat, 8 Apr 2000 19:20:10 +0000 (19:20 +0000)]
I think we want machine pattern i.86 not i386 --- looks like config.guess
could output several different high digits on most PC Unixen.

24 years agoAdd colo to display at end of initdb.
Bruce Momjian [Sat, 8 Apr 2000 18:35:30 +0000 (18:35 +0000)]
Add colo to display at end of initdb.

24 years agoModify pgbench.c not to depend on configure, since it's not shipped with one.
Tom Lane [Sat, 8 Apr 2000 18:32:24 +0000 (18:32 +0000)]
Modify pgbench.c not to depend on configure, since it's not shipped with one.

24 years ago*** empty log message ***
Michael Meskes [Sat, 8 Apr 2000 12:20:27 +0000 (12:20 +0000)]
*** empty log message ***

24 years agoLink already-existing COMMENT and REINDEX command docs into documentation.
Tom Lane [Sat, 8 Apr 2000 05:11:03 +0000 (05:11 +0000)]
Link already-existing COMMENT and REINDEX command docs into documentation.

24 years agoFix relcache refcount leakage when inv_drop is applied
Tom Lane [Sat, 8 Apr 2000 04:37:07 +0000 (04:37 +0000)]
Fix relcache refcount leakage when inv_drop is applied
to a non-LO relation.

24 years agoMention cube root.
Tom Lane [Sat, 8 Apr 2000 03:42:29 +0000 (03:42 +0000)]
Mention cube root.

24 years agoupdate
Bruce Momjian [Sat, 8 Apr 2000 02:44:55 +0000 (02:44 +0000)]
update

24 years agoFix cross-references, update examples, copy-edit.
Tom Lane [Sat, 8 Apr 2000 02:39:02 +0000 (02:39 +0000)]
Fix cross-references, update examples, copy-edit.

24 years agoAdd reference page for pg_ctl.
Thomas G. Lockhart [Sat, 8 Apr 2000 02:16:26 +0000 (02:16 +0000)]
Add reference page for pg_ctl.

24 years agoUpdate some porting info.
Thomas G. Lockhart [Sat, 8 Apr 2000 02:16:08 +0000 (02:16 +0000)]
Update some porting info.
Add mention of transcendental functions.

24 years agoAdd zpbit and varbit data types from Adrian Joubert
Thomas G. Lockhart [Sat, 8 Apr 2000 02:13:11 +0000 (02:13 +0000)]
Add zpbit and varbit data types from Adrian Joubert
 <a.joubert@albourne.com>.

24 years agoUpdate for to_char change.
Tom Lane [Sat, 8 Apr 2000 02:02:26 +0000 (02:02 +0000)]
Update for to_char change.

24 years agoTweak parallel test script so that command line options for postmaster
Tom Lane [Sat, 8 Apr 2000 01:54:47 +0000 (01:54 +0000)]
Tweak parallel test script so that command line options for postmaster
can be set via environment variable PMOPTIONS.  Default is -o -F.

24 years agoAdd copyObject logic for TruncateStmt and a few other utility-statement
Tom Lane [Sat, 8 Apr 2000 00:21:15 +0000 (00:21 +0000)]
Add copyObject logic for TruncateStmt and a few other utility-statement
parse node types.  This allows these statements to be placed in a plpgsql
function.  Also, see to it that statement types not handled by the copy
logic will draw an appropriate elog(ERROR), instead of leaving a null
pointer that will cause coredump later on.  More utility statements could
be added if anyone felt like turning the crank.

24 years agoMore man updates to fix bad tags
Bruce Momjian [Fri, 7 Apr 2000 19:20:48 +0000 (19:20 +0000)]
More man updates to fix bad tags

24 years agoUpdate create_rule manual page.
Bruce Momjian [Fri, 7 Apr 2000 19:17:51 +0000 (19:17 +0000)]
Update create_rule manual page.

24 years agoresultmap additions for solaris x86 ...
Marc G. Fournier [Fri, 7 Apr 2000 19:00:44 +0000 (19:00 +0000)]
resultmap additions for solaris x86 ...

24 years agoUse initdb --noclean so that test database is available for examination
Tom Lane [Fri, 7 Apr 2000 17:51:13 +0000 (17:51 +0000)]
Use initdb --noclean so that test database is available for examination
if initdb fails.

24 years agoAgain update of rule manual page.
Bruce Momjian [Fri, 7 Apr 2000 17:37:24 +0000 (17:37 +0000)]
Again update of rule manual page.

24 years agoUpdate rule examples
Bruce Momjian [Fri, 7 Apr 2000 17:35:08 +0000 (17:35 +0000)]
Update rule examples

24 years agoRemove mention of INSTANCE rule system, because it is gone.
Bruce Momjian [Fri, 7 Apr 2000 17:23:11 +0000 (17:23 +0000)]
Remove mention of INSTANCE rule system, because it is gone.

24 years agoAdd transcendental math functions (sine, cosine, etc)
Thomas G. Lockhart [Fri, 7 Apr 2000 13:40:45 +0000 (13:40 +0000)]
Add transcendental math functions (sine, cosine, etc)
Add a random number generator and seed setter (random(), SET SEED)
Fix up the interval*float8 math to carry partial months
 into the time field.
Add float8*interval so we have symmetry in the available math.
Fix the parser and define.c to accept SQL92 types as field arguments.
Fix the parser to accept SQL92 types for CREATE TYPE, etc. This is
 necessary to allow...
Bit/varbit support in contrib/bit cleaned up to compile and load
 cleanly. Still needs some work before final release.
Implement the "SOME" keyword as a synonym for "ANY" per SQL92.
Implement ascii(text), ichar(int4), repeat(text,int4) to help
 support the ODBC driver.
Enable the TRUNCATE() function mapping in the ODBC driver.

24 years agoDocument new SET SEED command.
Thomas G. Lockhart [Fri, 7 Apr 2000 13:31:18 +0000 (13:31 +0000)]
Document new SET SEED command.

24 years agoFix path to initdb in installation instructions.
Thomas G. Lockhart [Fri, 7 Apr 2000 13:30:58 +0000 (13:30 +0000)]
Fix path to initdb in installation instructions.
General cleanup for 7.0.

24 years agoClean up directory to compile and run on my Linux box at least.
Thomas G. Lockhart [Fri, 7 Apr 2000 13:28:37 +0000 (13:28 +0000)]
Clean up directory to compile and run on my Linux box at least.
Note that there is some trouble with inconsistant input/output formats.

24 years agoCommebts & one check in EndEvalPlanQual().
Vadim B. Mikheev [Fri, 7 Apr 2000 07:24:47 +0000 (07:24 +0000)]
Commebts & one check in EndEvalPlanQual().

24 years agoFix (I hope) resource leakage in EvalPlanQual: open subplans must be
Tom Lane [Fri, 7 Apr 2000 00:59:17 +0000 (00:59 +0000)]
Fix (I hope) resource leakage in EvalPlanQual: open subplans must be
properly shut down in EndPlan, else we fail to free buffers and so forth
that they hold.

24 years agoPartial fix for EvalPlanQual bugs reported by Magnus Hagander, 3-Apr.
Tom Lane [Fri, 7 Apr 2000 00:30:41 +0000 (00:30 +0000)]
Partial fix for EvalPlanQual bugs reported by Magnus Hagander, 3-Apr.
Ensure that outer tuple link needed for inner indexscan qual evaluation
gets set in the EvalPlanQual case.  This stops coredump, but we still
have resource leaks due to failure to clean up EvalPlanQual properly...

24 years agoUpdate initdb display of postmaster startup
Bruce Momjian [Thu, 6 Apr 2000 18:27:01 +0000 (18:27 +0000)]
Update initdb display of postmaster startup

24 years agoAllow vacuum of temporary tables
Bruce Momjian [Thu, 6 Apr 2000 18:12:07 +0000 (18:12 +0000)]
Allow vacuum of temporary tables

24 years agoAdd POLLUTE to interfaces/Makefile
Bruce Momjian [Thu, 6 Apr 2000 03:34:55 +0000 (03:34 +0000)]
Add POLLUTE to interfaces/Makefile

24 years agoRepair assert failure in tuple-chain-moving logic (introduced by yours
Tom Lane [Thu, 6 Apr 2000 00:29:51 +0000 (00:29 +0000)]
Repair assert failure in tuple-chain-moving logic (introduced by yours
truly, I'm afraid).

24 years agoUpdate vacuum to mention analyze
Bruce Momjian [Wed, 5 Apr 2000 21:51:11 +0000 (21:51 +0000)]
Update vacuum to mention analyze

24 years ago*** empty log message ***
Michael Meskes [Wed, 5 Apr 2000 15:51:28 +0000 (15:51 +0000)]
*** empty log message ***

24 years agoPlease forget all I said about gcc and AIX in my previous mail.
Bruce Momjian [Wed, 5 Apr 2000 14:47:21 +0000 (14:47 +0000)]
Please forget all I said about gcc and AIX in my previous mail.
It does work with the following patch applied and gcc 2.95.2 .

Use --with-template=aix_gcc to compile the whole lot with gcc.

The geometry regression test produces different precision.
With optimization I run into regression failures starting at oidjoins,
thus no -O2. Anybody else try gcc 2.95.2 and -O2 on beta4 ?

This is an important patch, since recent versions of the IBM compiler
are not for free, and thus most questions I get concern gcc.

Andreas

PS.: I am testing with beta4

24 years agoSolaris geometry results were out of sync with the input file
Peter Eisentraut [Wed, 5 Apr 2000 11:02:28 +0000 (11:02 +0000)]
Solaris geometry results were out of sync with the input file

24 years ago*** empty log message ***
Michael Meskes [Wed, 5 Apr 2000 09:05:40 +0000 (09:05 +0000)]
*** empty log message ***

24 years agoRename geometry.*bsdi to geo.*bsd. for Freebsd.
Bruce Momjian [Wed, 5 Apr 2000 02:36:41 +0000 (02:36 +0000)]
Rename geometry.*bsdi to geo.*bsd. for Freebsd.

24 years agofreebsd 4.0 is like bsdi for geometry test ...
Marc G. Fournier [Wed, 5 Apr 2000 02:25:11 +0000 (02:25 +0000)]
freebsd 4.0 is like bsdi for geometry test ...

24 years agofreebsd has *-freebsd and *-freebsdelf ... float8 passes with this
Marc G. Fournier [Wed, 5 Apr 2000 02:08:29 +0000 (02:08 +0000)]
freebsd has *-freebsd and *-freebsdelf ... float8 passes with this

24 years agoActually, that still wasn't quite right. If we skip a query because of
Tom Lane [Tue, 4 Apr 2000 23:52:50 +0000 (23:52 +0000)]
Actually, that still wasn't quite right.  If we skip a query because of
xact abort state in pg_exec_query_dest, we should continue scanning the
querytree list, on the off chance that one of the later queries in the
string is COMMIT or ROLLBACK.

24 years agoFix bug noted by Bruce: FETCH in an already-aborted transaction block
Tom Lane [Tue, 4 Apr 2000 21:44:40 +0000 (21:44 +0000)]
Fix bug noted by Bruce: FETCH in an already-aborted transaction block
would crash, due to premature invocation of SetQuerySnapshot().  Clean
up problems with handling of multiple queries by splitting
pg_parse_and_plan into two routines.  The old code would not, for
example, do the right thing with END; SELECT... submitted in one query
string when it had been in transaction abort state, because it'd decide
to skip planning the SELECT before it had executed the END.  New
arrangement is simpler and doesn't force caller to plan if only
parse+rewrite is needed.

24 years agoRemove FETCH syntax heading from MOVE manual page.
Bruce Momjian [Tue, 4 Apr 2000 19:42:36 +0000 (19:42 +0000)]
Remove FETCH syntax heading from MOVE manual page.

24 years agoRemove from FETCH manual page:
Bruce Momjian [Tue, 4 Apr 2000 19:29:18 +0000 (19:29 +0000)]
Remove from FETCH manual page:

       Once  all  rows  are  fetched,  every  other  fetch access
       returns no rows.

24 years agoNo ID file needed.
Bruce Momjian [Tue, 4 Apr 2000 13:51:38 +0000 (13:51 +0000)]
No ID file needed.

24 years agoUpdate QNX FAQ
Bruce Momjian [Tue, 4 Apr 2000 13:49:30 +0000 (13:49 +0000)]
Update QNX FAQ

24 years agoAdd a check to pg_dump to see whether backend is same version as pg_dump.
Tom Lane [Tue, 4 Apr 2000 05:22:46 +0000 (05:22 +0000)]
Add a check to pg_dump to see whether backend is same version as pg_dump.
If not, abort by default.  Abort can be prevented by using -i or
--ignore-version switch.

24 years agoUpdate regression for bsdi.
Bruce Momjian [Tue, 4 Apr 2000 04:07:53 +0000 (04:07 +0000)]
Update regression for bsdi.

24 years agoCorrect off-by-one error in strncat() usage.
Tom Lane [Tue, 4 Apr 2000 03:00:16 +0000 (03:00 +0000)]
Correct off-by-one error in strncat() usage.

24 years agoWhen rewriting an aggregate introduced into WHERE, allow agg argument to
Tom Lane [Tue, 4 Apr 2000 02:30:52 +0000 (02:30 +0000)]
When rewriting an aggregate introduced into WHERE, allow agg argument to
be an expression not just a simple Var, so long as only one table is
referenced (so that code isn't really any more difficult than before).
This whole thing is still fundamentally bogus, but at least we can accept
a few more cases than before.

24 years agoFor no good reason at all, libplpgsql was getting linked with libpq.
Tom Lane [Tue, 4 Apr 2000 01:47:59 +0000 (01:47 +0000)]
For no good reason at all, libplpgsql was getting linked with libpq.
Since plpgsql is a backend extension, it doesn't need libpq...

24 years agoFix extremely nasty little bug observed when a sub-SELECT appears in
Tom Lane [Tue, 4 Apr 2000 01:21:48 +0000 (01:21 +0000)]
Fix extremely nasty little bug observed when a sub-SELECT appears in
WHERE in a place where it can be part of a nestloop inner indexqual.
As the code stood, it put the same physical sub-Plan node into both
indxqual and indxqualorig of the IndexScan plan node.  That confused
later processing in the optimizer (which expected that tracing the
subPlan list would visit each subplan node exactly once), and would
probably have blown up in the executor if the planner hadn't choked first.
Fix by making the 'fixed' indexqual be a complete deep copy of the
original indexqual, rather than trying to share nodes below the topmost
operator node.  This had further ramifications though, because we were
making the aforesaid list of sub-Plan nodes during SS_process_sublinks
which is run before construction of the 'fixed' indexqual, meaning that
the copy of the sub-Plan didn't show up in that list.  Fix by rearranging
logic so that the sub-Plan list is built by the final set_plan_references
pass, not in SS_process_sublinks.  This may sound like a mess, but it's
actually a good deal cleaner now than it was before, because we are no
longer dependent on the assumption that planning will never make a copy
of a sub-Plan node.

24 years agoAugment the function call map logic with code from Tom Lane.
Thomas G. Lockhart [Tue, 4 Apr 2000 01:07:54 +0000 (01:07 +0000)]
Augment the function call map logic with code from Tom Lane.
 Should be more robust to overflows.
 Pass through an unmapped function unchanged, rather than rejecting it.
Add a few more functions, but comment out those which can go through as-is.
Can be used with contrib/odbc/ package, though that isn't committed yet.

24 years agoHi,
Bruce Momjian [Mon, 3 Apr 2000 20:56:40 +0000 (20:56 +0000)]
Hi,

    here is an updated version of the bit type with a bugfix and all the necessa
ry
SQL functions defined. This should replace what is currently in contrib. I'd
appreciate any comments on what is there.

Kind regards,

Adriaan

24 years ago*** empty log message ***
Michael Meskes [Mon, 3 Apr 2000 19:34:26 +0000 (19:34 +0000)]
*** empty log message ***

24 years agoLooks like Peter forgot to commit configure with configure.in.
Tom Lane [Mon, 3 Apr 2000 00:33:54 +0000 (00:33 +0000)]
Looks like Peter forgot to commit configure with configure.in.

24 years agoCopy-editing, mostly.
Tom Lane [Sun, 2 Apr 2000 22:59:40 +0000 (22:59 +0000)]
Copy-editing, mostly.

24 years agoMention DST boundaries as a possible cause of regress test failure.
Tom Lane [Sun, 2 Apr 2000 22:28:54 +0000 (22:28 +0000)]
Mention DST boundaries as a possible cause of regress test failure.

24 years ago*** empty log message ***
Michael Meskes [Sun, 2 Apr 2000 08:51:44 +0000 (08:51 +0000)]
*** empty log message ***

24 years agoAdjusted for new (again!) config.guess output under QNX
Peter Eisentraut [Fri, 31 Mar 2000 22:03:52 +0000 (22:03 +0000)]
Adjusted for new (again!) config.guess output under QNX

24 years agoFix low-probability bug in relcache startup: write_irels wrote the
Tom Lane [Fri, 31 Mar 2000 19:39:22 +0000 (19:39 +0000)]
Fix low-probability bug in relcache startup: write_irels wrote the
pg_internal.init file in-place, which meant that if another backend
started at about the same time, it might read the incomplete file.
init_irels tries to guard against that, but I have now seen a crash
due to reading bad data from a partly-written file.  (This may indicate
a kernel bug on my platform?  Not sure.)  Anyway, clearly the safest
course is to write the new pg_internal.init file under a unique temporary
filename, and rename it into place only after it's all written.

24 years agoCorrect typo.
Tom Lane [Fri, 31 Mar 2000 17:45:32 +0000 (17:45 +0000)]
Correct typo.

24 years agoImprove wording a little bit.
Tom Lane [Fri, 31 Mar 2000 17:45:00 +0000 (17:45 +0000)]
Improve wording a little bit.

24 years agoUpdate index cost estimation docs to final 7.0 scheme.
Tom Lane [Fri, 31 Mar 2000 17:18:26 +0000 (17:18 +0000)]
Update index cost estimation docs to final 7.0 scheme.

24 years agoMinor copy-editing.
Tom Lane [Fri, 31 Mar 2000 15:00:14 +0000 (15:00 +0000)]
Minor copy-editing.

24 years agoCREATE AGGREGATE ref page claimed that a noise-word AS was allowed in
Tom Lane [Fri, 31 Mar 2000 14:57:05 +0000 (14:57 +0000)]
CREATE AGGREGATE ref page claimed that a noise-word AS was allowed in
the syntax ... but gram.y doesn't think so.