]> granicus.if.org Git - postgresql/log
postgresql
25 years agoFix critical error noticed by Massimo: copy.c used to have a
Tom Lane [Sat, 12 Jun 1999 20:41:25 +0000 (20:41 +0000)]
Fix critical error noticed by Massimo: copy.c used to have a
special hack to ensure it would close its output file even after failure
due to elog(ERROR) partway through the copy.  This is now unnecessary
because fd.c takes care of cleaning up open files at transaction abort;
worse, after fd.c closed the file copy.c would try to do so *again* at
the start of the next COPY command.  This would result in havoc in most
implementations of stdio library.

25 years agoRemove query_planner's overhasty rejection of cases where
Tom Lane [Sat, 12 Jun 1999 19:38:30 +0000 (19:38 +0000)]
Remove query_planner's overhasty rejection of cases where
tlist and qual are NULL.  It ought to handle these the same as the cases
where tlist contains only constant expressions, ie, be willing to generate
a Result-node plan.  This did not use to matter, but it does now because
union_planner will flatten the tlist when aggregates are present.  Thus,
'select count(1) from table' now causes query_planner to be given a null
tlist, and to duplicate 6.4's behavior we need it to give back a Result
plan rather than refusing the query.  6.4 was arguably doing the Wrong
Thing for this query, but I'm not going to open a semantics issue right
before 6.5 release ... can revisit that problem later.

25 years agoPlug hole in dike: planner would coredump if query_planner
Tom Lane [Sat, 12 Jun 1999 19:27:41 +0000 (19:27 +0000)]
Plug hole in dike: planner would coredump if query_planner
returned NULL, which it will do in some cases where an elog(ERROR) would
probably be more appropriate.  For the moment, generate a not-very-
informative error message rather than proceeding to certain coredump.
Probably ought to think about making query_planner elog instead of
returning NULL, but this is at least a safe change for now.

25 years agoWhen targetlist is NULL, ExecTargetList was passing back a
Tom Lane [Sat, 12 Jun 1999 19:22:40 +0000 (19:22 +0000)]
When targetlist is NULL, ExecTargetList was passing back a
pointer to palloc'd but uninitialized memory.  This is not cool; anyone looking
at the returned 'tuple' would at best coredump and at worst behave in a
bizarre and irreproducible way.  Fix it to return a predictable value,
namely a correctly-set-up palloc'd tuple containing zero attributes.
I believe this fix is both safe and critical.

25 years agoReversed out Massimo patch.
Bruce Momjian [Sat, 12 Jun 1999 14:07:33 +0000 (14:07 +0000)]
Reversed out Massimo patch.

25 years agoI don't like last minute patches before the final freeze, but I believe that
Bruce Momjian [Sat, 12 Jun 1999 14:05:41 +0000 (14:05 +0000)]
I don't like last minute patches before the final freeze, but I believe that
this one could be useful for people experiencing out-of-memory crashes while
executing queries which retrieve or use a very large number of tuples.

The problem happens when storage is allocated for functions results used in
a large query, for example:

  select upper(name) from big_table;
  select big_table.array[1] from big_table;
  select count(upper(name)) from big_table;

This patch is a dirty hack that fixes the out-of-memory problem for the most
common cases, like the above ones. It is not the final solution for the
problem but it can work for some people, so I'm posting it.

The patch should be safe because all changes are under #ifdef. Furthermore
the feature can be enabled or disabled at runtime by the `free_tuple_memory'
options in the pg_options file. The option is disabled by default and must
be explicitly enabled at runtime to have any effect.

To enable the patch add the follwing line to Makefile.custom:

CUSTOM_COPT += -DFREE_TUPLE_MEMORY

To enable the option at runtime add the following line to pg_option:

free_tuple_memory=1

Massimo

25 years agoHi,
Bruce Momjian [Fri, 11 Jun 1999 15:01:29 +0000 (15:01 +0000)]
Hi,

please apply the included patch. It corrects the headers in src/win32 -
there are some missing #endif.

                        Dan

25 years agoadd m86 similar entries.
Bruce Momjian [Fri, 11 Jun 1999 14:34:02 +0000 (14:34 +0000)]
add m86 similar entries.

25 years agoChange Assert(Ptp.t_data->t_xmax == tp.t_data->t_xmin) to :
Vadim B. Mikheev [Fri, 11 Jun 1999 09:35:08 +0000 (09:35 +0000)]
Change Assert(Ptp.t_data->t_xmax == tp.t_data->t_xmin) to :
/*
 * Read above about cases when !ItemIdIsUsed(Citemid)
 * (child item is removed)... Due to the fact that
 * at the moment we don't remove unuseful part of
 * update-chain, it's possible to get too old
 * parent row here. Like as in the case which
 * caused this problem, we stop shrinking here.
 * I could try to find real parent row but want
 * not to do it because of real solution will
 * be implemented anyway, latter, and we are too
 * close to 6.5 release.        - vadim 06/11/99
 */
if (Ptp.t_data->t_xmax != tp.t_data->t_xmin)
...

25 years agoRemoved bad Assert(!buf->ri_lock) when unlocking exclusively
Vadim B. Mikheev [Fri, 11 Jun 1999 09:00:02 +0000 (09:00 +0000)]
Removed bad Assert(!buf->ri_lock) when unlocking exclusively
locked buffer.

25 years agoMore about chained mode and isolation.
Vadim B. Mikheev [Fri, 11 Jun 1999 05:40:18 +0000 (05:40 +0000)]
More about chained mode and isolation.

25 years agoMore about chained mode and isolation levels.
Vadim B. Mikheev [Fri, 11 Jun 1999 05:19:53 +0000 (05:19 +0000)]
More about chained mode and isolation levels.

25 years agoChange mdtruncate to truncate and not unlink.
Bruce Momjian [Fri, 11 Jun 1999 02:39:43 +0000 (02:39 +0000)]
Change mdtruncate to truncate and not unlink.

Hiroshi Inoue

25 years agoThis patch should enable 6.5 to build on Motorola 68000 architecture.
Bruce Momjian [Thu, 10 Jun 1999 22:59:22 +0000 (22:59 +0000)]
This patch should enable 6.5 to build on Motorola 68000 architecture.
It comes from Roman Hodek <Roman.Hodek@informatik.uni-erlangen.de>.

25 years ago*** empty log message ***
Michael Meskes [Thu, 10 Jun 1999 19:11:33 +0000 (19:11 +0000)]
*** empty log message ***

25 years agoAdded 2 regression tests for NUMERIC data type.
Jan Wieck [Thu, 10 Jun 1999 17:49:32 +0000 (17:49 +0000)]
Added 2 regression tests for NUMERIC data type.

1. Using 100 digits after decimal point on the default
   make runtest.

2. Using 1000 digits after decimal point in a new target
   make bigtest.

At the end of 'make runtest', a hint about the new bigtest is
printed.

Jan

25 years ago1. Fix for elog(ERROR, "EvalPlanQual: t_xmin is uncommitted ?!")
Vadim B. Mikheev [Thu, 10 Jun 1999 14:17:12 +0000 (14:17 +0000)]
1. Fix for elog(ERROR, "EvalPlanQual: t_xmin is uncommitted ?!")
   and possibly for other cases too:

   DO NOT cache status of transaction in unknown state
   (i.e. non-committed and non-aborted ones)

   Example:
   T1 reads row updated/inserted by running T2 and cache T2 status.
   T2 commits.
   Now T1 reads a row updated by T2 and with HEAP_XMAX_COMMITTED
   in t_infomask (so cached T2 status is not changed).
   Now T1 EvalPlanQual gets updated row version without HEAP_XMIN_COMMITTED
   -> TransactionIdDidCommit(t_xmin) and TransactionIdDidAbort(t_xmin)
   return FALSE and T2 decides that t_xmin is not committed and gets
   ERROR above.

   It's too late to find more smart way to handle such cases and so
   I just changed xact status caching and got rid TransactionIdFlushCache()
   from code.

   Changed: transam.c, xact.c, lmgr.c and transam.h - last three
   just because of TransactionIdFlushCache() is removed.

2. heapam.c:

   T1 marked a row for update. T2 waits for T1 commit/abort.
   T1 commits. T3 updates the row before T2 locks row page.
   Now T2 sees that new row t_xmax is different from xact id (T1)
   T2 was waiting for. Old code did Assert here. New one goes to
   HeapTupleSatisfiesUpdate. Obvious changes too.

3. Added Assert to vacuum.c
4. bufmgr.c: break
   Assert(buf->r_locks == 0 && !buf->ri_lock)
   into two Asserts.

25 years agoFix errors in SELECT ... GROUP BY ... UNION SELECT ...
Tom Lane [Thu, 10 Jun 1999 06:55:40 +0000 (06:55 +0000)]
Fix errors in SELECT ... GROUP BY ... UNION SELECT ...
ye proverbial one-line patch (not counting five lines of comment so's
maybe it won't happen again)

25 years agoAdd mention of hot backups to the release notes.
Thomas G. Lockhart [Wed, 9 Jun 1999 13:44:59 +0000 (13:44 +0000)]
Add mention of hot backups to the release notes.
Add id tags to chapter and section headings to ensure good labels
 on output html files.

25 years agoMake minor changes in wording.
Thomas G. Lockhart [Wed, 9 Jun 1999 13:43:42 +0000 (13:43 +0000)]
Make minor changes in wording.
Adjust tags to get a clean build.

25 years agoRemove -O2 for linux/alpha, make -O.
Bruce Momjian [Wed, 9 Jun 1999 12:39:46 +0000 (12:39 +0000)]
Remove -O2 for linux/alpha, make -O.

25 years agoReset evaluation plan tuple table next free slot counter to 0
Vadim B. Mikheev [Wed, 9 Jun 1999 12:23:42 +0000 (12:23 +0000)]
Reset evaluation plan tuple table next free slot counter to 0
after ExecEndNode. It must be done! Or we'll be out of free
tuple slots very soon, though slots are freed by ExecEndNode
and ready for reusing.

We didn't see this problem before because of

int         nSlots = ExecCountSlotsNode(plan);
TupleTable  tupleTable = ExecCreateTupleTable(nSlots + 10);
    /* why add ten? - jolly */

code in InitPlan - i.e. extra 10 slots. Simple select uses
3 slots and so it was possible to re-use evaluation plan
3 additional times and didn't get

elog(NOTICE, "Plan requires more slots than are available");
elog(ERROR, "send mail to your local executor guru to fix this");

Changes are obvious and shouldn't be problems with them.
Though, I added Assert(epqstate->es_tupleTable->next == 0)
before EvalPlanQual():ExecInitNode and we'll notice if
something is still wrong. Is it better to change Assert
to elog(ERROR) ?

25 years agoMVCC updates.
Vadim B. Mikheev [Wed, 9 Jun 1999 03:51:40 +0000 (03:51 +0000)]
MVCC updates.

25 years agoMVCC updation.
Vadim B. Mikheev [Tue, 8 Jun 1999 20:01:01 +0000 (20:01 +0000)]
MVCC updation.

25 years agoLinux-alpha gets -O
Bruce Momjian [Tue, 8 Jun 1999 19:19:51 +0000 (19:19 +0000)]
Linux-alpha gets -O

25 years agoRemove optimization on linux/alpha.
Bruce Momjian [Tue, 8 Jun 1999 17:34:43 +0000 (17:34 +0000)]
Remove optimization on linux/alpha.

25 years agoFix for duplicate LC_COLLATE
Bruce Momjian [Tue, 8 Jun 1999 16:42:55 +0000 (16:42 +0000)]
Fix for duplicate LC_COLLATE

25 years agoAdd mention of new pg_dump feature.
Bruce Momjian [Tue, 8 Jun 1999 16:42:10 +0000 (16:42 +0000)]
Add mention of new pg_dump feature.

25 years agoInstall proper Irix FAQ.
Bruce Momjian [Tue, 8 Jun 1999 16:27:26 +0000 (16:27 +0000)]
Install proper Irix FAQ.

25 years agoCleanup
Bruce Momjian [Tue, 8 Jun 1999 01:37:55 +0000 (01:37 +0000)]
Cleanup

25 years agoDisable pg_upgrade for 6.5.
Bruce Momjian [Tue, 8 Jun 1999 01:32:05 +0000 (01:32 +0000)]
Disable pg_upgrade for 6.5.

25 years agoI attach a patch for pg_upgrade. This does two things:
Bruce Momjian [Mon, 7 Jun 1999 22:58:31 +0000 (22:58 +0000)]
I attach a patch for pg_upgrade.  This does two things:

1. check whether the program is being executed in $PGDATA/..  This is
   necessary if the data tree is not in the standard place, as is the
   case with the Debian distribution (because of Debian policy).

2. give a clearer error message if the dumped data structure fails to
   be loaded.

Oliver Elphick

25 years agoOn AIX 4.3.2 the third line in template/aix_42:
Bruce Momjian [Mon, 7 Jun 1999 17:20:24 +0000 (17:20 +0000)]
On AIX 4.3.2 the third line in template/aix_42:
SHARED_LIB:

needs to be changed to:
SHARED_LIB:-lc

I think this was also needed on AIX 4.2. Comments Please !!

If nobody objects, I suggest to make this change, since it cannot
break AIX 4.2 and is necessary on AIX 4.3

Andreas

25 years agoConcurrency... Highest one...
Vadim B. Mikheev [Mon, 7 Jun 1999 15:14:54 +0000 (15:14 +0000)]
Concurrency... Highest one...
DO NOT EVEN TRY TO DO PageGetMaxOffsetNumber BEFORE
LockBuffer!
-:)

25 years ago> Here is a small patch that should only affect win32 building
Bruce Momjian [Mon, 7 Jun 1999 14:29:20 +0000 (14:29 +0000)]
> Here is a small patch that should only affect win32 building
> (native win32, not cygnus).
> It does the following:
> Patches two win32.mak files to DEFINE HAVE_VSNPRINTF and
> HAVE_STRDUP. This is required to build at all.
> Bumps the version number on libpq.dll from 6.4 to 6.5.
> Required for install programs to work.
> Adds defintions for BLCKSZ and MAXIMUM_ALIGN to "win32.h" in
> the client-side libpiq directory.
>
> All these files are only used when building on native win32,
> so it should be safe I think.
>
> Again, really sorry to throw this in so late, but I would
> hate to do the same thing as with 6.4 (which required 6.4.1
> to at all compile on Win32).
>
> Thanks,
>
>   //Magnus

25 years agoRepair recently-introduced error in makeIndexable for LIKE:
Tom Lane [Mon, 7 Jun 1999 14:28:26 +0000 (14:28 +0000)]
Repair recently-introduced error in makeIndexable for LIKE:
a non-leading % would be put into the >=/<= patterns.  Also, repair
longstanding confusion about whether %% means a literal %%.  The SQL92
doesn't say any such thing, and textlike() knows that, but gram.y didn't.

25 years agoHave to release meta page before reading root one!
Vadim B. Mikheev [Mon, 7 Jun 1999 14:28:22 +0000 (14:28 +0000)]
Have to release meta page before reading root one!
< 6.5 versions were just not affected by this bug due to locking.

25 years agoUpdate TODO.
Bruce Momjian [Mon, 7 Jun 1999 02:42:07 +0000 (02:42 +0000)]
Update TODO.

25 years ago1. xact.c: update comments about changing MyProc->xid and MyProc->xmin.
Vadim B. Mikheev [Sun, 6 Jun 1999 20:19:35 +0000 (20:19 +0000)]
1. xact.c: update comments about changing MyProc->xid and MyProc->xmin.
2. varsup.c:ReadNewTransactionId(): don't read nextXid from disk -
   this func doesn't allocate next xid, so ShmemVariableCache->nextXid
   may be used (but GetNewTransactionId() must be called first).
3. vacuum.c: change elog(ERROR, "Child item....") to elog(NOTICE) -
   this is not ERROR, proper handling is just not implemented, yet.
4. s_lock.c: increase S_MAX_BUSY by 2 times.
5. shmem.c:GetSnapshotData(): have to call ReadNewTransactionId()
   _after_ SpinAcquire(ShmemIndexLock).

25 years agoequal() needs a case for Aggref nodes, as shown by:
Tom Lane [Sun, 6 Jun 1999 17:46:40 +0000 (17:46 +0000)]
equal() needs a case for Aggref nodes, as shown by:
regression=> select sum(q1) from int8_tbl group by q2 order by sum(q1);
NOTICE:  equal: don't know whether nodes of type 107 are equal

25 years agoAdd a regression test case to catch breakage of GROUP BY
Tom Lane [Sun, 6 Jun 1999 17:41:01 +0000 (17:41 +0000)]
Add a regression test case to catch breakage of GROUP BY
and aggregate functions in queries that use inheritance.

25 years agoFix problems with grouping/aggregation in queries that use
Tom Lane [Sun, 6 Jun 1999 17:38:11 +0000 (17:38 +0000)]
Fix problems with grouping/aggregation in queries that use
inheritance ... basically it was completely busted :-(

25 years agoI used bad style of comments and ... commented out some code in
Vadim B. Mikheev [Sun, 6 Jun 1999 15:14:40 +0000 (15:14 +0000)]
I used bad style of comments and ... commented out some code in
EvalPlanQualNext() when implemented it... -:)
Uncommented...

25 years agoInstead of failing when the constructed name for a sequence,
Tom Lane [Sat, 5 Jun 1999 20:22:30 +0000 (20:22 +0000)]
Instead of failing when the constructed name for a sequence,
index, etc is too long, truncate until it fits.

25 years agoUpdate from Massimo
Bruce Momjian [Sat, 5 Jun 1999 20:07:54 +0000 (20:07 +0000)]
Update from Massimo

25 years agoHi,
Bruce Momjian [Sat, 5 Jun 1999 19:09:48 +0000 (19:09 +0000)]
Hi,

I have updated my contrib code for version 6.5. In the attachment you will
find the directories array, datetime, miscutil, string, tools and userlocks
which replace the corresponding directories under contrib.

In contrib/tools you will find some developement scripts which I use while
hacking the sources. I hope they will be useful for some other people.

I have also added a contrib/Makefile which tries to compile and install all
the contribs. Unfortunately many of them don't have a Makefile or don't
compile cleanly.

--
Massimo Dal Zotto

25 years agoNew FAQ with commerical support mention.
Bruce Momjian [Sat, 5 Jun 1999 18:30:43 +0000 (18:30 +0000)]
New FAQ with commerical support mention.

25 years agoUpdate faq and hpux faq.
Bruce Momjian [Sat, 5 Jun 1999 18:24:55 +0000 (18:24 +0000)]
Update faq and hpux faq.

25 years agoRemove redeclarations of default parameter values from
Tom Lane [Sat, 5 Jun 1999 18:05:17 +0000 (18:05 +0000)]
Remove redeclarations of default parameter values from
PgDatabase::DisplayTuples and PgDatabase::PrintTuples.  This is incorrect
according to strict interpretation of the C++ spec, and some compilers
will reject it.  Also silence g++ warning about unused parameter.

25 years agoa fix for Win32 support provided by yutaka@marin.or.jp.
Tatsuo Ishii [Sat, 5 Jun 1999 10:27:31 +0000 (10:27 +0000)]
a fix for Win32 support provided by yutaka@marin.or.jp.
The patch will avoid to add .exe suffix to the pg_encoding
binary.

25 years agotrace.patch (compilation error)
Marc G. Fournier [Sat, 5 Jun 1999 04:18:09 +0000 (04:18 +0000)]
trace.patch (compilation error)

        the gettimeofday doesn't compile under Linux with glibc2 because
        the DST_NONE constant is no more defined. It seems that this code
        (written by me) has always be wrong but for some reason working.

From: Massimo Dal Zotto <dz@cs.unitn.it>

25 years agoman.patch (make inconsistency)
Marc G. Fournier [Sat, 5 Jun 1999 04:15:09 +0000 (04:15 +0000)]
man.patch (make inconsistency)

        the default target is 'install' instead of 'all'. So if you do a
        make without target you actually do a make install, which is not
        what one normally expects from a standard makefile.

From: Massimo Dal Zotto <dz@cs.unitn.it>

25 years agoecpg.patch (wrong makefile expansion in some cases)
Marc G. Fournier [Sat, 5 Jun 1999 04:13:21 +0000 (04:13 +0000)]
ecpg.patch (wrong makefile expansion in some cases)

        the ecpg Makefiles use a variable DESTDIR which is never defined
        except by debian/rules makefile, in which case the ecpg makefiles
        expand wrong pathnames. If we want to support a DESTDIR root it
        must be done consistently in all the makefiles, not just in ecpg.

From: Massimo Dal Zotto <dz@cs.unitn.it>

25 years agoUpdate FAQ's for release.
Bruce Momjian [Sat, 5 Jun 1999 03:43:07 +0000 (03:43 +0000)]
Update FAQ's for release.

25 years agoUpdate with more TODO items.
Bruce Momjian [Sat, 5 Jun 1999 02:47:13 +0000 (02:47 +0000)]
Update with more TODO items.

25 years agoSuppress indexes on large objects from psql's \d listing;
Tom Lane [Fri, 4 Jun 1999 21:21:13 +0000 (21:21 +0000)]
Suppress indexes on large objects from psql's \d listing;
they were confusing because the large object tables themselves are not
shown.  (Besides, if you've got hundreds or thousands of large objects,
you really don't want to see 'em at all.)
Also, suppress all indexes from the \z ACL listing, since indexes have
no meaningful protection information.

25 years agoAdd startup-time check that -B is not unreasonably small for
Tom Lane [Fri, 4 Jun 1999 21:14:46 +0000 (21:14 +0000)]
Add startup-time check that -B is not unreasonably small for
given number of backends (-N), per recent discussion in pghackers list.

25 years agoAdd configurable option controlling security checks in LO functions.
Tom Lane [Fri, 4 Jun 1999 21:13:38 +0000 (21:13 +0000)]
Add configurable option controlling security checks in LO functions.

25 years agoTurns out OIDNAMELEN wasn't really being used at all!
Tom Lane [Fri, 4 Jun 1999 21:12:07 +0000 (21:12 +0000)]
Turns out OIDNAMELEN wasn't really being used at all!
Get rid of it to make customization of NAMEDATALEN easier.

25 years agoAdd blurb on new docs in release notes.
Thomas G. Lockhart [Fri, 4 Jun 1999 15:51:42 +0000 (15:51 +0000)]
Add blurb on new docs in release notes.
Make minor info and markup changes for v6.5 release.

25 years agoAdd input parameters for LOCK TABLE. Still needs explanation from Vadim.
Thomas G. Lockhart [Fri, 4 Jun 1999 15:50:44 +0000 (15:50 +0000)]
Add input parameters for LOCK TABLE. Still needs explanation from Vadim.
Fix markup.

25 years agoMay as well keep an ApplixWare version of this file too.
Thomas G. Lockhart [Fri, 4 Jun 1999 06:31:14 +0000 (06:31 +0000)]
May as well keep an ApplixWare version of this file too.

25 years agoHardcopy for v6.5 release.
Thomas G. Lockhart [Fri, 4 Jun 1999 06:24:02 +0000 (06:24 +0000)]
Hardcopy for v6.5 release.

25 years agoAdd a mention of doc changes in the release notes.
Thomas G. Lockhart [Fri, 4 Jun 1999 06:23:27 +0000 (06:23 +0000)]
Add a mention of doc changes in the release notes.

25 years agoRelease notes from sgml for v6.5 release.
Thomas G. Lockhart [Fri, 4 Jun 1999 06:23:00 +0000 (06:23 +0000)]
Release notes from sgml for v6.5 release.

25 years agoAdd mention of SHARE ROW EXCLUSIVE mode.
Bruce Momjian [Fri, 4 Jun 1999 04:28:54 +0000 (04:28 +0000)]
Add mention of SHARE ROW EXCLUSIVE mode.

25 years agoman page cleanup for vadim.
Bruce Momjian [Fri, 4 Jun 1999 04:16:11 +0000 (04:16 +0000)]
man page cleanup for vadim.

25 years agoAdd description for SET TRANSACTION.
Bruce Momjian [Fri, 4 Jun 1999 03:44:42 +0000 (03:44 +0000)]
Add description for SET TRANSACTION.

25 years agoFix for failure to clean SysCache entry when a relation is deleted
Tom Lane [Fri, 4 Jun 1999 02:19:47 +0000 (02:19 +0000)]
Fix for failure to clean SysCache entry when a relation is deleted
in the same transaction that created it.

25 years agocleanup
Bruce Momjian [Thu, 3 Jun 1999 20:45:08 +0000 (20:45 +0000)]
cleanup

25 years agoSET TRANSACTION ISOLATION LEVEL SERIALIZABLE;
Bruce Momjian [Thu, 3 Jun 1999 20:44:28 +0000 (20:44 +0000)]
SET TRANSACTION ISOLATION LEVEL SERIALIZABLE;
SET TRANSACTION ISOLATION LEVEL READ COMMITTED;
                                ^^^^ required

Also note that SET TRANSACTION ISOLATION LEVEL change
isolevel for _current_ transaction, in accordance with
standard, not for session (ALTER SESSION is used in Oracle,
but it's not implemented currently).

And I would don't mention SET XACTISOLEVEL TO ...
form at all.

Please update set.sgml - I failed to understand all these

SET TIME ZONE { '<REPLACEABLE CLASS="PARAMETER">

now.

for Vadim

25 years agoThanks for this, but SET TRANSACTION ISOLATION LEVEL
Bruce Momjian [Thu, 3 Jun 1999 20:42:18 +0000 (20:42 +0000)]
Thanks for this, but SET TRANSACTION ISOLATION LEVEL
should be described like SET TIME ZONE: TO is not allowed, as well
as DEFAULT option. There are only two forms:

for Vadim.

25 years agoUpdate of release notes with new items.
Bruce Momjian [Thu, 3 Jun 1999 20:40:50 +0000 (20:40 +0000)]
Update of release notes with new items.

25 years agoAdd mention of FOR UPDATE and LIMIT/OFFSET>
Bruce Momjian [Thu, 3 Jun 1999 19:52:09 +0000 (19:52 +0000)]
Add mention of FOR UPDATE and LIMIT/OFFSET>

25 years agoUpdate lock sgml/man/psql help pages.
Bruce Momjian [Thu, 3 Jun 1999 19:18:00 +0000 (19:18 +0000)]
Update lock sgml/man/psql help pages.

25 years agoAdd SET TRANSACTION ISOLATION LEVEL and SET NAMES mention.
Bruce Momjian [Thu, 3 Jun 1999 18:38:00 +0000 (18:38 +0000)]
Add SET TRANSACTION ISOLATION LEVEL and SET NAMES mention.

25 years agoUpdate psql help so it looks better.
Bruce Momjian [Thu, 3 Jun 1999 18:25:27 +0000 (18:25 +0000)]
Update psql help so it looks better.

25 years agoUpdate SET commands for psql \h, man pages, and sgml.
Bruce Momjian [Thu, 3 Jun 1999 18:18:43 +0000 (18:18 +0000)]
Update SET commands for psql \h, man pages, and sgml.

25 years agoMinor markup fixes.
Thomas G. Lockhart [Thu, 3 Jun 1999 16:13:54 +0000 (16:13 +0000)]
Minor markup fixes.

25 years agoMake separate subsection for Vadim's MVCC notes.
Thomas G. Lockhart [Thu, 3 Jun 1999 16:13:35 +0000 (16:13 +0000)]
Make separate subsection for Vadim's MVCC notes.
Add timing info for v6.5 on my linux box.

25 years agoFix markup and formatting.
Thomas G. Lockhart [Thu, 3 Jun 1999 16:11:58 +0000 (16:11 +0000)]
Fix markup and formatting.

25 years agoInstallation notes for v6.5.
Thomas G. Lockhart [Thu, 3 Jun 1999 16:05:38 +0000 (16:05 +0000)]
Installation notes for v6.5.
Generated from install.sgml and installation.sgml.

25 years ago1. Additional fix against ERROR: Child itemid marked as unused
Vadim B. Mikheev [Thu, 3 Jun 1999 13:33:13 +0000 (13:33 +0000)]
1. Additional fix against ERROR:  Child itemid marked as unused
   in CommitTransaction().
2. Changes in GetSnapshotData().

25 years agoUnuseful tuple.t_data->t_infomask & HEAP_XMIN_COMMITTED in
Vadim B. Mikheev [Thu, 3 Jun 1999 13:25:54 +0000 (13:25 +0000)]
Unuseful tuple.t_data->t_infomask & HEAP_XMIN_COMMITTED in
vc_scanheap().

25 years agoNotes in Migration to v6.5 section.
Vadim B. Mikheev [Thu, 3 Jun 1999 07:11:50 +0000 (07:11 +0000)]
Notes in Migration to v6.5 section.

25 years ago1. MyProc->xid assignment is moved to GetNewTransactionId so newer
Vadim B. Mikheev [Thu, 3 Jun 1999 04:41:41 +0000 (04:41 +0000)]
1. MyProc->xid assignment is moved to GetNewTransactionId so newer
   transactions will not assume that MyProc transaction was committed
   before snapshot calculations. With old MyProc->xid assignment
   (in xact.c:StartTransaction()) there was ability to see the same
   row twice (I used gdb for this)!...
2. Assignments of InvalidTransactionId to MyProc->xid and MyProc->xmin
   are moved from xact.c:CommitTransaction() to
   xact.c:RecordTransactionCommit() - this invalidation must be done
   before releasing transaction locks or bad (too high) XmaxRecent value
   might be used by vacuum ("ERROR:  Child itemid marked as unused"
   reported by "Hiroshi Inoue" <Inoue@tpf.co.jp>; once again, gdb
   allowed me reproduce this error).

25 years agoMarkup changes for v6.5 release.
Thomas G. Lockhart [Thu, 3 Jun 1999 04:21:51 +0000 (04:21 +0000)]
Markup changes for v6.5 release.
Clean out duplicate stuff in odbc.sgml resulting from a faulty patch.

25 years agoCGM format from Hannu. Seem to have lost the .ag original :(
Thomas G. Lockhart [Thu, 3 Jun 1999 04:20:32 +0000 (04:20 +0000)]
CGM format from Hannu. Seem to have lost the .ag original :(

25 years agoNew versions of hardcopy for 6.5 release.
Thomas G. Lockhart [Thu, 3 Jun 1999 04:18:58 +0000 (04:18 +0000)]
New versions of hardcopy for 6.5 release.

25 years agopg_dump failed on any database containing large objects,
Tom Lane [Thu, 3 Jun 1999 04:01:16 +0000 (04:01 +0000)]
pg_dump failed on any database containing large objects,
because it ignored the LOs themselves but failed to ignore the indexes
on the LOs.  How long has this been broken??

25 years agoEnsure consistent results when FormSortKeys fails to find
Tom Lane [Thu, 3 Jun 1999 03:17:37 +0000 (03:17 +0000)]
Ensure consistent results when FormSortKeys fails to find
all the expected keys (it was returning uninitialized memory).

25 years agoOoops ... dllist.c can't use Assert() when it is compiled into
Tom Lane [Thu, 3 Jun 1999 01:28:24 +0000 (01:28 +0000)]
Ooops ... dllist.c can't use Assert() when it is compiled into
libpq ...

25 years agoChanged "current." into "old." in rule string backparsing
Jan Wieck [Wed, 2 Jun 1999 11:52:29 +0000 (11:52 +0000)]
Changed "current." into "old." in rule string backparsing

Jan

25 years agoThe INET and CIDR types mistakenly compared 198.68.123.0/24 and
Bruce Momjian [Wed, 2 Jun 1999 03:37:15 +0000 (03:37 +0000)]
The INET and CIDR types mistakenly compared 198.68.123.0/24 and
198.68.123.0/27 the same when indexing them.

D'Arcy

25 years agoMake sure that only one intro is included in the integrated doc.
Thomas G. Lockhart [Tue, 1 Jun 1999 17:26:18 +0000 (17:26 +0000)]
Make sure that only one intro is included in the integrated doc.
Multiple intros cause trouble since they have some section elements
 (e.g. "y2k.sgml") in common leading to duplicate labels.
Include emacs formatting hints in the intro*.sgml sources.

25 years agotypedef struct LTAG
Vadim B. Mikheev [Tue, 1 Jun 1999 09:35:39 +0000 (09:35 +0000)]
typedef struct LTAG
{
    Oid             relId;
    Oid             dbId;
    union
    {
        BlockNumber     blkno;
        TransactionId   xid;
    }               objId;
>
> Added:
>    /*
>     * offnum should be part of objId.tupleId above, but would increase
>     * sizeof(LOCKTAG) and so moved here; currently used by userlocks only.
>     */
>    OffsetNumber    offnum;
    uint16          lockmethod;     /* needed by userlocks */
} LOCKTAG;

gmake clean required...
User locks are ready for 6.5 release...

25 years agoforgot to attach the patch. :(
Bruce Momjian [Tue, 1 Jun 1999 02:43:37 +0000 (02:43 +0000)]
forgot to attach the patch.  :(

Vince.

25 years agocommented out PgConnection& references for now. May be using them
Bruce Momjian [Tue, 1 Jun 1999 02:37:33 +0000 (02:37 +0000)]
commented out PgConnection& references for now.  May be using them
later.

Vince.

25 years agoFix some latent bugs in dllist.c (carelessness about setting
Tom Lane [Mon, 31 May 1999 23:48:04 +0000 (23:48 +0000)]
Fix some latent bugs in dllist.c (carelessness about setting
all fields that should be set).  Add a MoveToFront primitive to speed up
one of the hotspots in SearchSysCache.

25 years agoClean up memory leaks in LO operations by freeing LO's private
Tom Lane [Mon, 31 May 1999 22:53:59 +0000 (22:53 +0000)]
Clean up memory leaks in LO operations by freeing LO's private
memory context at transaction commit or abort.