]> granicus.if.org Git - postgresql/log
postgresql
23 years agoattached is a patch that makes SysV semaphore emulation
Bruce Momjian [Wed, 17 Jan 2001 22:11:19 +0000 (22:11 +0000)]
attached is a patch that makes SysV semaphore emulation
using POSIX semaphores more robust on Darwin 1.2/Mac OS X
Public Beta.  this is for the version of 7.1 available
via anon cvs as of Jan 14 2001 14:00 PST.

since the semaphores and shared memory created by this
emulator are shared with the backends via fork(), their
persistent names are not necessary.  removing their
names with shm_unlink() and sem_unlink() after creation
obviates the need for any "ipcclean" function.  further,
without these changes, the shared memory (and, therefore,
the semaphores) will not be re-initialized/re-created after
the first execution of the postmaster, until reboot
or until some (non-existent) ipcclean function is executed.

this patch does the following:

   1) if the shared memory segment "SysV_Sem_Info" already
      existed, it is cleaned up.  it shouldn't be there anyways.

   2) the real indicator for whether the shared memory/semaphore
      emulator has been initialized is if "SemInfo" has been
      initialized.  the shared memory and semaphores must be
      initialized regardless of whether there was a garbage shared
      memory segment lying around.

   3) the shared memory segment "SysV_Sem_Info" is created with "O_EXCL"
      to catch the case where two postmasters might be starting
      simultaneously, so they don't both end up with the same shared
      memory (one will fail).  note that this can't be done with the
      semaphores because Darwin 1.2 has a bug where attempting to
      open an existing semaphore with "O_EXCL" set will ruin the
      semaphore until the next reboot.

   4) the shared memory segment "SysV_Sem_Info" is unlinked after
      it is created.  it will then exist without a name until the
      postmaster and all backend children exit.
   5) all semaphores are unlinked after they are created.  they'll
      then exist without names until the postmaster and all backend
      children exit.

-michael thornburgh, zenomt@armory.com

23 years agoChange lcons(x, NIL) to makeList(x) where appropriate.
Bruce Momjian [Wed, 17 Jan 2001 17:26:45 +0000 (17:26 +0000)]
Change lcons(x, NIL) to makeList(x) where appropriate.

23 years agoChange comparisons of tm->tm_isdst from "nonzero" to "greater than zero".
Thomas G. Lockhart [Wed, 17 Jan 2001 16:46:56 +0000 (16:46 +0000)]
Change comparisons of tm->tm_isdst from "nonzero" to "greater than zero".
 Not sure why some were this way, and others were already correct, but it
 seems to have been like this for several years.
This caused problems on a few damaged platforms like AIX and IRIX which do
 not support DST calculations for years before 1970.
Thanks to Andreas Zeugswetter <ZeugswetterA@wien.spardat.at> for finding
 the problem.

23 years agoThere are misprints in postgres doc., in :
Bruce Momjian [Wed, 17 Jan 2001 16:34:34 +0000 (16:34 +0000)]
There are misprints in postgres doc., in :
Chapter 10. PL/pgSQL - SQL Procedural Language (c40914117.htm)

Statements
...
(resulting in a PL/pgSQL internal SELECT).
But there are cases where someone isn't interested int
-----------------------------------------(have to be)-->
But there are cases where someone isn't interested in
     the functions result.

 RAISE level format''
--(have to be)-->
 RAISE level 'format'

23 years agoMove structure comments from the top block down to the line entries for
Bruce Momjian [Wed, 17 Jan 2001 06:41:31 +0000 (06:41 +0000)]
Move structure comments from the top block down to the line entries for
this file to match all the other files, and to be clearer.

23 years agoRemove bogus backslashes in sed command.
Tom Lane [Tue, 16 Jan 2001 22:48:34 +0000 (22:48 +0000)]
Remove bogus backslashes in sed command.

23 years agoOops, I had managed to break query-cancel-while-waiting-for-lock.
Tom Lane [Tue, 16 Jan 2001 20:59:34 +0000 (20:59 +0000)]
Oops, I had managed to break query-cancel-while-waiting-for-lock.

23 years agoRename fields of lock and lockholder structures to something a tad less
Tom Lane [Tue, 16 Jan 2001 06:11:34 +0000 (06:11 +0000)]
Rename fields of lock and lockholder structures to something a tad less
confusing, and clean up documentation.

23 years agoShow intarray contrib module in contrib/README and Makefile.
Tom Lane [Mon, 15 Jan 2001 22:20:00 +0000 (22:20 +0000)]
Show intarray contrib module in contrib/README and Makefile.

23 years agoAdjust file names.
Peter Eisentraut [Mon, 15 Jan 2001 21:17:27 +0000 (21:17 +0000)]
Adjust file names.

23 years agoFix problems with parentheses around sub-SELECT --- for the last time,
Tom Lane [Mon, 15 Jan 2001 20:36:36 +0000 (20:36 +0000)]
Fix problems with parentheses around sub-SELECT --- for the last time,
I hope.  I finally realized that we were going at it backwards: when
there are excess parentheses, they need to be treated as part of the
sub-SELECT, not as part of the surrounding expression.  Although either
choice yields an unambiguous grammar, only this way produces a grammar
that is LALR(1).  With the old approach we were guaranteed to fail on
either 'SELECT (((SELECT 2)) + 3)' or
'SELECT (((SELECT 2)) UNION SELECT 2)' depending on which way we
resolve the initial shift/reduce conflict.  With the new way, the same
reduction track can be followed in both cases until we have advanced
far enough to know whether we are done with the sub-SELECT or not.

23 years agoTweak heap_update/delete so that we do not hold the buffer context lock
Tom Lane [Mon, 15 Jan 2001 05:29:19 +0000 (05:29 +0000)]
Tweak heap_update/delete so that we do not hold the buffer context lock
on the old tuple's page while we are doing TOAST pushups.

23 years agoMinor coding cleanups.
Tom Lane [Sun, 14 Jan 2001 22:21:54 +0000 (22:21 +0000)]
Minor coding cleanups.

23 years agoAnother go-round on making GetRawDatabaseInfo behave as well as it can,
Tom Lane [Sun, 14 Jan 2001 22:21:05 +0000 (22:21 +0000)]
Another go-round on making GetRawDatabaseInfo behave as well as it can,
given the fundamental restriction of not looking at transaction commit
data in pg_log.  Use code that is actually based on tqual.c rather than
ad-hoc tests.  Also write the tuple fetch loop using standard access
macros rather than ad-hoc code.

23 years agopg_database's datpath column must not be marked toastable, because
Tom Lane [Sun, 14 Jan 2001 22:17:22 +0000 (22:17 +0000)]
pg_database's datpath column must not be marked toastable, because
GetRawDatabaseInfo() won't cope with a compressed path spec (much less
a moved-off one).  I'm not going to force an initdb for this change,
because it's noncritical --- we're not actually using datpath at all
right now.  But it seems a good idea to apply the fix while I'm thinking
about it.

23 years agoNeed to do BufferSync at end of DROP DATABASE as well as CREATE DATABASE.
Tom Lane [Sun, 14 Jan 2001 22:14:10 +0000 (22:14 +0000)]
Need to do BufferSync at end of DROP DATABASE as well as CREATE DATABASE.
Otherwise, newly connecting backends will still think the deleted DB is
valid, and will generate unexpected error messages.

23 years agoMake aclcontains() do something that's at least vaguely reasonable:
Tom Lane [Sun, 14 Jan 2001 19:23:27 +0000 (19:23 +0000)]
Make aclcontains() do something that's at least vaguely reasonable:
it now returns true if the aclitem argument exactly matches any one of
the elements of the aclitem[] argument.  Per complaint from Wolff 1/10/01.

23 years agoRestructure backend SIGINT/SIGTERM handling so that 'die' interrupts
Tom Lane [Sun, 14 Jan 2001 05:08:17 +0000 (05:08 +0000)]
Restructure backend SIGINT/SIGTERM handling so that 'die' interrupts
are treated more like 'cancel' interrupts: the signal handler sets a
flag that is examined at well-defined spots, rather than trying to cope
with an interrupt that might happen anywhere.  See pghackers discussion
of 1/12/01.

23 years agoTerminology cleanup: class -> table, instance -> row, attribute -> column,
Peter Eisentraut [Sat, 13 Jan 2001 23:58:55 +0000 (23:58 +0000)]
Terminology cleanup: class -> table, instance -> row, attribute -> column,
etc.

23 years agoBacked out:
Bruce Momjian [Sat, 13 Jan 2001 18:52:42 +0000 (18:52 +0000)]
Backed out:

---------------------------------------------------------------------------

Attached is a set of patches for a couple of bugs dealing with
timestamps in JDBC.

Bug#1) Incorrect timestamp stored in DB if client timezone different
than DB.

23 years agoAdd information about bit types. Adjust some other things to promote
Peter Eisentraut [Sat, 13 Jan 2001 18:34:51 +0000 (18:34 +0000)]
Add information about bit types.  Adjust some other things to promote
SQL type names over internal type names.

23 years agoAttached is a set of patches for a couple of bugs dealing with
Bruce Momjian [Sat, 13 Jan 2001 05:18:05 +0000 (05:18 +0000)]
Attached is a set of patches for a couple of bugs dealing with
timestamps in JDBC.

Bug#1) Incorrect timestamp stored in DB if client timezone different
than DB.

The buggy implementation of setTimestamp() in PreparedStatement simply
used the toString() method of the java.sql.Timestamp object to convert
to a string to send to the database.  The format of this is yyyy-MM-dd
hh:mm:ss.SSS which doesn't include any timezone information.  Therefore
the DB assumes its timezone since none is specified.  That is OK if the
timezone of the client and server are the same, however if they are
different the wrong timestamp is received by the server.  For example if
the client is running in timezone GMT and wants to send the timestamp
for noon to a server running in PST (GMT-8 hours), then the server will
receive 2000-01-12 12:00:00.0 and interprete it as 2000-01-12
12:00:00-08 which is 2000-01-12 04:00:00 in GMT.  The fix is to send a
format to the server that includes the timezone offset.  For simplicity
sake the fix uses a SimpleDateFormat object with its timezone set to GMT
so that '+00' can be used as the timezone for postgresql.  This is done
as SimpleDateFormat doesn't support formating timezones in the way
postgresql expects.

Bug#2) Incorrect handling of partial seconds in getting timestamps from
the DB

When the SimpleDateFormat object parses a string with a format like
yyyy-MM-dd hh:mm:ss.SS it expects the fractional seconds to be three
decimal places (time precision in java is miliseconds = three decimal
places).  This seems like a bug in java to me, but it is unlikely to be
fixed anytime soon, so the postgresql code needed modification to
support the java behaviour.  So for example a string of '2000-01-12
12:00:00.12-08' coming from the database was being converted to a
timestamp object with a value of 2000-01-12 12:00:00.012GMT-08:00.  The
fix was to check for a '.' in the string and if one is found append on
an extra zero to the fractional seconds part.

Bug#3) Performance problems

In fixing the above two bugs, I noticed some things that could be
improved.  In PreparedStatement.setTimestamp(),
PreparedStatement.setDate(), ResultSet.getTimestamp(), and
ResultSet.getDate() these methods were creating a new SimpleDateFormat
object everytime they were called.  To avoid this unnecessary object
creation overhead, I changed the code to use static variables for
keeping a single instance of the needed formating objects.
Also the code used the + operator for string concatenation.  As everyone
should know this is very inefficient and the use of StringBuffers is
prefered.

I also did some cleanup in ResultSet.getTimestamp().  This method has
had multiple patches applied some of which resulted in code that was no
longer needed.  For example the ISO timestamp format that postgresql
uses specifies the timezone as an offset like '-08'.  Code was added at
one point to convert the postgresql format to the java one which is
GMT-08:00, however the old code was left around which did nothing.  So
there was code that looked for yyyy-MM-dd hh:mm:sszzzzzzzzz and
yyyy-MM-dd hh:mm:sszzz.  This second format would never be encountered
because zzz (i.e. -08) would be converted into the former (also note
that the SimpleDateFormat object treats zzzzzzzzz and zzz the same, the
number of z's does not matter).

There was another problem/fix mentioned on the email lists today by
mcannon@internet.com which is also fixed by this patch:

Bug#4) Fractional seconds lost when getting timestamp from the DB
A patch by Jan Thomea handled the case of yyyy-MM-dd hh:mm:sszzzzzzzzz
but not the fractional seconds version yyyy-MM-dd hh:mm:ss.SSzzzzzzzzz.

The code is fixed to handle this case as well.

Barry Lind

23 years agoMore cleanup.
Bruce Momjian [Sat, 13 Jan 2001 04:58:53 +0000 (04:58 +0000)]
More cleanup.

23 years agoRelax test on typmod matching between a table and its proposed ON SELECT
Tom Lane [Sat, 13 Jan 2001 03:58:28 +0000 (03:58 +0000)]
Relax test on typmod matching between a table and its proposed ON SELECT
rule.  Needed to avoid failure when reloading a 7.0 pg_dump of a view
that has a NUMERIC column.

23 years agoWindows wants shared libraries in PATH.
Peter Eisentraut [Sat, 13 Jan 2001 03:25:48 +0000 (03:25 +0000)]
Windows wants shared libraries in PATH.

23 years agocheck one last time for any erros ...
Marc G. Fournier [Sat, 13 Jan 2001 03:17:05 +0000 (03:17 +0000)]
check one last time for any erros ...

23 years agoNew shell for the to be written CHECKPOINT documentation, so the summary
Peter Eisentraut [Sat, 13 Jan 2001 03:11:12 +0000 (03:11 +0000)]
New shell for the to be written CHECKPOINT documentation, so the summary
shows up in psql now.

23 years agoignore his too
Marc G. Fournier [Sat, 13 Jan 2001 03:11:07 +0000 (03:11 +0000)]
ignore his too

23 years agoand this time?
Marc G. Fournier [Sat, 13 Jan 2001 03:09:43 +0000 (03:09 +0000)]
and this time?

23 years agotry this again ...
Marc G. Fournier [Sat, 13 Jan 2001 03:08:02 +0000 (03:08 +0000)]
try this again ...

23 years agoClean up garbage.
Peter Eisentraut [Sat, 13 Jan 2001 02:18:31 +0000 (02:18 +0000)]
Clean up garbage.

23 years agoSurely one README file is enough.
Peter Eisentraut [Sat, 13 Jan 2001 02:14:32 +0000 (02:14 +0000)]
Surely one README file is enough.

23 years agoUpdate pgcvslog to fix problem with duplicate narratives.
Bruce Momjian [Sat, 13 Jan 2001 01:49:35 +0000 (01:49 +0000)]
Update pgcvslog to fix problem with duplicate narratives.

23 years agookay, this appears to work ...
Marc G. Fournier [Sat, 13 Jan 2001 00:57:06 +0000 (00:57 +0000)]
okay, this appears to work ...

onlly changes aer adding some white space ...

23 years agoSurely we don't need a 7.0.3 makefile in 7.1.
Peter Eisentraut [Fri, 12 Jan 2001 22:36:57 +0000 (22:36 +0000)]
Surely we don't need a 7.0.3 makefile in 7.1.

23 years agoUpdate information about compiling extension modules.
Peter Eisentraut [Fri, 12 Jan 2001 22:15:32 +0000 (22:15 +0000)]
Update information about compiling extension modules.

23 years agoAdd more critical-section calls: all code sections that hold spinlocks
Tom Lane [Fri, 12 Jan 2001 21:54:01 +0000 (21:54 +0000)]
Add more critical-section calls: all code sections that hold spinlocks
are now critical sections, so as to ensure die() won't interrupt us while
we are munging shared-memory data structures.  Avoid insecure intermediate
states in some code that proc_exit will call, like palloc/pfree.  Rename
START/END_CRIT_CODE to START/END_CRIT_SECTION, since that seems to be
what people tend to call them anyway, and make them be called with () like
a function call, in hopes of not confusing pg_indent.
I doubt that this is sufficient to make SIGTERM safe anywhere; there's
just too much code that could get invoked during proc_exit().

23 years agoUpdate TODO list.
Bruce Momjian [Fri, 12 Jan 2001 17:57:57 +0000 (17:57 +0000)]
Update TODO list.

23 years agoUpdate TODO list.
Bruce Momjian [Fri, 12 Jan 2001 17:48:18 +0000 (17:48 +0000)]
Update TODO list.

23 years agoFixed handling of renamed columns in PK constraints
Philip Warner [Fri, 12 Jan 2001 15:41:29 +0000 (15:41 +0000)]
Fixed handling of renamed columns in PK constraints

23 years agoAdd to DROP todo.
Bruce Momjian [Fri, 12 Jan 2001 05:37:37 +0000 (05:37 +0000)]
Add to DROP todo.

23 years agoUpdate TODO list.
Bruce Momjian [Fri, 12 Jan 2001 05:32:38 +0000 (05:32 +0000)]
Update TODO list.

23 years agoUpdate TODO list.
Bruce Momjian [Fri, 12 Jan 2001 05:23:10 +0000 (05:23 +0000)]
Update TODO list.

23 years agoBring CREATE TABLE syntax synopsis into line with reality; update a
Tom Lane [Fri, 12 Jan 2001 05:06:40 +0000 (05:06 +0000)]
Bring CREATE TABLE syntax synopsis into line with reality; update a
bunch of old or poorly-worded documentation.

23 years ago- Check ntuples == 1 for various SELECT statements.
Philip Warner [Fri, 12 Jan 2001 04:32:07 +0000 (04:32 +0000)]
- Check ntuples == 1 for various SELECT statements.
- Fix handling of --tables=* (multiple tables never worked properly, AFAICT)
- strdup() the current user in DB routines
- Check results of IO routines more carefully.
- Check results of PQ routines more carefully.

Have not fixed index output yet.

23 years agoPreserve constraints and column defaults during CLUSTER.
Tom Lane [Fri, 12 Jan 2001 01:22:21 +0000 (01:22 +0000)]
Preserve constraints and column defaults during CLUSTER.
Wish they were all this easy ...

23 years agocommit Oleg and Teodor's RD-tree implementation ... this provides the
Marc G. Fournier [Fri, 12 Jan 2001 00:16:26 +0000 (00:16 +0000)]
commit Oleg and Teodor's RD-tree implementation ... this provides the
regression tests for the GiST changes ... this should be integrated into
the regular regression tests similar to Vadim's SPI contrib stuff ...

23 years agoNew feature:
Marc G. Fournier [Fri, 12 Jan 2001 00:12:58 +0000 (00:12 +0000)]
New feature:
   1. Support of variable size keys - new algorithm of insertion to tree
      (GLI - gist layrered insertion). Previous algorithm was implemented
      as described in paper by Joseph M. Hellerstein et.al
      "Generalized Search Trees for Database Systems".  This (old)
      algorithm was not suitable for variable size keys and could be
      not effective ( walking up-down ) in case of multiple levels split
Bug fixed:
   1. fixed bug in gistPageAddItem - key values were written to disk
      uncompressed. This caused failure if decompression function
      does real job.
   2. NULLs handling - we keep NULLs in tree. Right way is to remove them,
      but we don't know how to inform vacuum about index statistics. This is
      just cosmetic warning message (like in case with R-Tree),
      but I'm not sure how to recognize real problem if we remove NULLs
      and suppress this warning as Tom suggested.
   3. various memory leaks

This work was done by Teodor Sigaev (teodor@stack.net) and
Oleg Bartunov (oleg@sai.msu.su).

23 years ago#ifdef out entire file for newer Cygwin versions.
Peter Eisentraut [Thu, 11 Jan 2001 23:32:03 +0000 (23:32 +0000)]
#ifdef out entire file for newer Cygwin versions.

23 years agoAdd DLLIMPORT to TransactionCommandContext.
Peter Eisentraut [Thu, 11 Jan 2001 23:28:34 +0000 (23:28 +0000)]
Add DLLIMPORT to TransactionCommandContext.

23 years agoRemove useless DLLIMPORT (only needed in header files).
Peter Eisentraut [Thu, 11 Jan 2001 23:27:05 +0000 (23:27 +0000)]
Remove useless DLLIMPORT (only needed in header files).

23 years agoUpdate TODO list.
Bruce Momjian [Thu, 11 Jan 2001 21:07:07 +0000 (21:07 +0000)]
Update TODO list.

23 years agoreduce HISTORY changes from this list too
Marc G. Fournier [Thu, 11 Jan 2001 04:13:30 +0000 (04:13 +0000)]
reduce HISTORY changes from this list too

23 years agoremove all the TODO updates, which drops of 12k from the file ...
Marc G. Fournier [Thu, 11 Jan 2001 04:12:32 +0000 (04:12 +0000)]
remove all the TODO updates, which drops of 12k from the file ...

23 years agousing cvs2cl.pl, generate a list of changes from beta1->beta3 ...
Marc G. Fournier [Thu, 11 Jan 2001 04:00:59 +0000 (04:00 +0000)]
using cvs2cl.pl, generate a list of changes from beta1->beta3 ...

23 years agoMake checks for global variables (sys_nerr, timezone) safe against getting
Peter Eisentraut [Wed, 10 Jan 2001 17:07:18 +0000 (17:07 +0000)]
Make checks for global variables (sys_nerr, timezone) safe against getting
optimized away completely.

23 years agoRemoved a no longer needed SetWaitingForLock() call in
Hiroshi Inoue [Wed, 10 Jan 2001 01:24:19 +0000 (01:24 +0000)]
Removed a no longer needed SetWaitingForLock() call in
DeadLockCheck().

23 years agoDo The Right Thing (tm) if asked to cluster a temp table. Previous
Tom Lane [Wed, 10 Jan 2001 01:12:28 +0000 (01:12 +0000)]
Do The Right Thing (tm) if asked to cluster a temp table.  Previous
code would cluster, but table would magically lose its tempness.

23 years agoSynced preproc.y with gram.y.
Michael Meskes [Tue, 9 Jan 2001 19:46:05 +0000 (19:46 +0000)]
Synced preproc.y with gram.y.

23 years agoRemove -L$(libdir) from DLLLIBS to prevent linking with an old version
Peter Eisentraut [Tue, 9 Jan 2001 18:45:41 +0000 (18:45 +0000)]
Remove -L$(libdir) from DLLLIBS to prevent linking with an old version
(i.e., 7.0.3) of libpostgres.a.  From Jason Tishler <jt@dothill.com>.

23 years agoAdd configure check for sys_nerr, to end all discussions.
Peter Eisentraut [Tue, 9 Jan 2001 18:40:15 +0000 (18:40 +0000)]
Add configure check for sys_nerr, to end all discussions.

23 years agoUpdate TODO list.
Bruce Momjian [Tue, 9 Jan 2001 18:06:07 +0000 (18:06 +0000)]
Update TODO list.

23 years agoThe KAME files md5.* and sha1.* have the following changelog
Bruce Momjian [Tue, 9 Jan 2001 16:07:14 +0000 (16:07 +0000)]
The KAME files md5.* and sha1.* have the following changelog
entry:

----------------------------
revision 1.2
date: 2000/12/04 01:20:38;  author: tgl;  state: Exp;  lines:
+18 -18
Eliminate some of the more blatant platform-dependencies ... it
builds here now, anyway ...
----------------------------

Which basically changes u_int*_t -> uint*_t, so now it does not
compile neither under Debian 2.2 nor under NetBSD 1.5 which
is platform independent<B8> all right.  Also it replaces $KAME$
with $Id$ which is Bad Thing. PostgreSQL Id should be added as a
separate line so the file history could be seen.

So here is patch:

* changes uint*_t -> uint*.  I guess that was the original
  intention
* adds uint64 type to include/c.h because its needed
  [somebody should check if I did it right]
* adds back KAME Id, because KAME is the master repository
* removes stupid c++ comments in pgcrypto.c
* removes <sys/types.h> from the code, its not needed

--
marko

Marko Kreen

23 years agoApply proper sql.sgml change.
Bruce Momjian [Tue, 9 Jan 2001 16:05:21 +0000 (16:05 +0000)]
Apply proper sql.sgml change.

23 years agoAttached is a doc patch for doc/src/sgml/sql.sgml.
Bruce Momjian [Tue, 9 Jan 2001 15:48:18 +0000 (15:48 +0000)]
Attached is a doc patch for doc/src/sgml/sql.sgml.

It adds information about SQL JOIN that is implemented in 7.1.

--
-------- Robert B. Easter

23 years agoUpdate TODO list.
Bruce Momjian [Tue, 9 Jan 2001 15:37:39 +0000 (15:37 +0000)]
Update TODO list.

23 years agoA patch for doc/src/sgml/plsql.sgml to add a little more info about PL/pgSQL
Bruce Momjian [Tue, 9 Jan 2001 15:26:16 +0000 (15:26 +0000)]
A patch for doc/src/sgml/plsql.sgml to add a little more info about PL/pgSQL
EXECUTE.

--
-------- Robert B. Easter

23 years agoApproaching the current documentation from a position of ignorance, I
Bruce Momjian [Tue, 9 Jan 2001 14:23:40 +0000 (14:23 +0000)]
Approaching the current documentation from a position of ignorance, I
find it ambiguous.  I propose something along the lines of the
following patch to clarify it.  Thanks.

(Alternatively, perhaps the code could maintain a count of nested
calls to SPI_connect/SPI_finish.  But I didn't try to write that
patch.)

Ian Lance Taylor

23 years agoUpdate TODO list.
Bruce Momjian [Tue, 9 Jan 2001 14:12:34 +0000 (14:12 +0000)]
Update TODO list.

23 years agoUpdate TODO list.
Bruce Momjian [Tue, 9 Jan 2001 13:52:02 +0000 (13:52 +0000)]
Update TODO list.

23 years agoUpdate TODO list.
Bruce Momjian [Tue, 9 Jan 2001 13:48:03 +0000 (13:48 +0000)]
Update TODO list.

23 years agojump version to beta3 ... beta2 was created and pulled due to a couple of REL7_1_BETA3
PostgreSQL Daemon [Tue, 9 Jan 2001 13:11:32 +0000 (13:11 +0000)]
jump version to beta3 ... beta2 was created and pulled due to a couple of
large-ish bugs that Tom and Vadim were able to fix, but to avoid any
confusion, beta2 was removed ... and for tag'ng purposes, beta3 is being
created ...

23 years agoAdd a README file for multi-byte. This file is contributed by
Tatsuo Ishii [Tue, 9 Jan 2001 09:54:11 +0000 (09:54 +0000)]
Add a README file for multi-byte. This file is contributed by
Chih-Chang Hsieh <cch@cc.kmu.edu.tw>, written in traditional Chinese
(Big5).

23 years agoDisable query cancel during HandleDeadLock().
Hiroshi Inoue [Tue, 9 Jan 2001 09:38:57 +0000 (09:38 +0000)]
Disable query cancel during HandleDeadLock().

23 years ago1. Checkpoint.undo may be after checkpoint itself:
Vadim B. Mikheev [Tue, 9 Jan 2001 06:24:33 +0000 (06:24 +0000)]
1. Checkpoint.undo may be after checkpoint itself:
   - no more elog(STOP) in StartupXLOG();
   - both checkpoint' undo & redo are used to define
     oldest on-line log file.
2. Ability to pre-allocate a few log files at checkpoint time
   (wal_files option). Off by default.

23 years agoUpdate TODO list.
Bruce Momjian [Tue, 9 Jan 2001 06:16:23 +0000 (06:16 +0000)]
Update TODO list.

23 years agoREADME.mb has been unified into SGML documents.
Tatsuo Ishii [Tue, 9 Jan 2001 04:40:31 +0000 (04:40 +0000)]
README.mb has been unified into SGML documents.

23 years agoFix oversight in planning of GROUP queries: when an expression is used
Tom Lane [Tue, 9 Jan 2001 03:48:51 +0000 (03:48 +0000)]
Fix oversight in planning of GROUP queries: when an expression is used
as both a GROUP BY item and an output expression, the top-level Group
node should just copy up the evaluated expression value from its input,
rather than re-evaluating the expression.  Aside from any performance
benefit this might offer, this avoids a crash when there is a sub-SELECT
in said expression.

23 years agoFix small but critical typo ...
Tom Lane [Tue, 9 Jan 2001 02:15:16 +0000 (02:15 +0000)]
Fix small but critical typo ...

23 years agoPrevent vacuumdb from trying to vacuum template0.
Tom Lane [Mon, 8 Jan 2001 23:02:36 +0000 (23:02 +0000)]
Prevent vacuumdb from trying to vacuum template0.

23 years agoDocument the system attributes ctid and tableoid, which for some reason
Tom Lane [Mon, 8 Jan 2001 22:07:47 +0000 (22:07 +0000)]
Document the system attributes ctid and tableoid, which for some reason
were never yet mentioned anywhere in our documentation.  Improve
explanations of the other system attributes, too.

23 years agoDocument that we don't support ORDER BY with general expressions on
Tom Lane [Mon, 8 Jan 2001 21:30:37 +0000 (21:30 +0000)]
Document that we don't support ORDER BY with general expressions on
the output of UNION/INTERSECT/EXCEPT.

23 years agoAdd rudimentary section about controlling kernel's file and process limits.
Peter Eisentraut [Mon, 8 Jan 2001 21:01:54 +0000 (21:01 +0000)]
Add rudimentary section about controlling kernel's file and process limits.

23 years agoRemove compiler warning about uninitialized warnings.
Bruce Momjian [Mon, 8 Jan 2001 20:54:24 +0000 (20:54 +0000)]
Remove compiler warning about uninitialized warnings.

23 years agocheck for failure after vacuuming each DB, not only the last one.
Tom Lane [Mon, 8 Jan 2001 20:32:27 +0000 (20:32 +0000)]
check for failure after vacuuming each DB, not only the last one.

23 years agoAdd some debugging support code (ifdef'd out in normal use).
Tom Lane [Mon, 8 Jan 2001 18:34:44 +0000 (18:34 +0000)]
Add some debugging support code (ifdef'd out in normal use).

23 years agoLockBuffer should not elog while holding buffer's cntx_lock.
Tom Lane [Mon, 8 Jan 2001 18:31:49 +0000 (18:31 +0000)]
LockBuffer should not elog while holding buffer's cntx_lock.

23 years agoKeep relations open until they are no longer needed.
Hiroshi Inoue [Mon, 8 Jan 2001 03:14:58 +0000 (03:14 +0000)]
Keep relations open until they are no longer needed.

23 years agoMake outfuncs/readfuncs treat OIDs properly as unsigned values. Clean up
Tom Lane [Mon, 8 Jan 2001 00:31:43 +0000 (00:31 +0000)]
Make outfuncs/readfuncs treat OIDs properly as unsigned values.  Clean up
inconsistent coding practices for handling Index values and booleans,
too.

23 years agoCorrect nasty error in heap_update: it was releasing the buffer refcount
Tom Lane [Sun, 7 Jan 2001 22:14:31 +0000 (22:14 +0000)]
Correct nasty error in heap_update: it was releasing the buffer refcount
before calling RelationInvalidateHeapTuple(), which is bad because the
latter needs to look at the tuple data, which is in the shared disk
buffer.  If another backend manages to recycle the buffer while this
is going on, we will compute the wrong hashindex for the tuple or
maybe even crash outright.  Must hold buffer refcount until afterwards.
(This bug is not in 7.0.*; seems to be have introduced during WAL changes.)

23 years agoClear QueryCancel and ProcDiePending at start of proc_exit, to ensure
Tom Lane [Sun, 7 Jan 2001 04:30:41 +0000 (04:30 +0000)]
Clear QueryCancel and ProcDiePending at start of proc_exit, to ensure
that leftover cancel/die requests cannot interfere with exit activities.

23 years agoFix recent breakage of query-cancel logic, see my pghackers message
Tom Lane [Sun, 7 Jan 2001 04:17:29 +0000 (04:17 +0000)]
Fix recent breakage of query-cancel logic, see my pghackers message
of 6 Jan 2001 21:55.

23 years agotag configure as beta2 .. REL7_1_BETA2
PostgreSQL Daemon [Sun, 7 Jan 2001 02:23:25 +0000 (02:23 +0000)]
tag configure as beta2 ..

23 years agoResultmap updates for OpenBSD, per report from bpalmer@crimelabs.net.
Tom Lane [Sun, 7 Jan 2001 01:14:35 +0000 (01:14 +0000)]
Resultmap updates for OpenBSD, per report from bpalmer@crimelabs.net.

23 years agoModify readfuncs so that recursive use of stringToNode will not crash
Tom Lane [Sun, 7 Jan 2001 01:08:48 +0000 (01:08 +0000)]
Modify readfuncs so that recursive use of stringToNode will not crash
and burn.  Just for added luck, change reading of CONST nodes so that
we do not need to consult pg_type rows while reading them; this means
that no database access occurs during stringToNode.  This requires
changing the order in which const-node fields are written, which means
an initdb is forced.

23 years agoClean up checking of relkind for ALTER TABLE and LOCK TABLE commands.
Tom Lane [Sun, 7 Jan 2001 00:05:22 +0000 (00:05 +0000)]
Clean up checking of relkind for ALTER TABLE and LOCK TABLE commands.
Disallow cases like adding constraints to sequences :-(, and eliminate
now-unnecessary search of pg_rewrite to decide if a relation is a view.

23 years agoLog memory context stats to stderr when reporting a 'Memory exhausted'
Tom Lane [Sat, 6 Jan 2001 21:59:39 +0000 (21:59 +0000)]
Log memory context stats to stderr when reporting a 'Memory exhausted'
error, so as to provide a starting point for debugging.

23 years agoFix memory leak in relcache handling of rules: allocate rule parsetrees
Tom Lane [Sat, 6 Jan 2001 21:53:18 +0000 (21:53 +0000)]
Fix memory leak in relcache handling of rules: allocate rule parsetrees
in per-entry sub-memory-context, where they were supposed to go, rather
than in CacheMemoryContext where the code was putting them.  Must've
suffered a severe brain fade when I wrote this :-(

23 years agoUse more portable syntax for 'find'.
Peter Eisentraut [Sat, 6 Jan 2001 21:24:01 +0000 (21:24 +0000)]
Use more portable syntax for 'find'.

23 years agoPolish help output. Allow --help to work with BSD getopts.
Peter Eisentraut [Sat, 6 Jan 2001 20:57:26 +0000 (20:57 +0000)]
Polish help output.  Allow --help to work with BSD getopts.