]> granicus.if.org Git - postgresql/log
postgresql
14 years agoAdd get_bit/set_bit functions for bit strings, paralleling those for bytea,
Tom Lane [Mon, 25 Jan 2010 20:55:32 +0000 (20:55 +0000)]
Add get_bit/set_bit functions for bit strings, paralleling those for bytea,
and implement OVERLAY() for bit strings and bytea.

In passing also convert text OVERLAY() to a true built-in, instead of
relying on a SQL function.

Leonardo F, reviewed by Kevin Grittner

14 years agoAdd note that PREPARE TRANSACTION is for transaction managers, not
Heikki Linnakangas [Mon, 25 Jan 2010 18:23:10 +0000 (18:23 +0000)]
Add note that PREPARE TRANSACTION is for transaction managers, not
regular applications. Also add a comment pointing out that tab-complition
for PREPARE TRANSACTION is missing on purpose.

14 years agoApply Tcl_Init() to the "hold" interpreter created by pltcl.
Tom Lane [Mon, 25 Jan 2010 01:58:14 +0000 (01:58 +0000)]
Apply Tcl_Init() to the "hold" interpreter created by pltcl.

You might think this is unnecessary since that interpreter is never used
to run code --- but it turns out that's wrong.  As of Tcl 8.5, the "clock"
command (alone among builtin Tcl commands) is partially implemented by
loaded-on-demand Tcl code, which means that it fails if there's not
unknown-command support, and also that it's impossible to run it directly
in a safe interpreter.  The way they get around the latter is that
Tcl_CreateSlave() automatically sets up an alias command that forwards any
execution of "clock" in a safe slave interpreter to its parent interpreter.
Thus, when attempting to execute "clock" in trusted pltcl, the command
actually executes in the "hold" interpreter, where it will fail if
unknown-command support hasn't been introduced by sourcing the standard
init.tcl script, which is done by Tcl_Init().  (This is a pretty dubious
design decision on the Tcl boys' part, if you ask me ... but they didn't.)

Back-patch all the way.  It's not clear that anyone would try to use ancient
versions of pltcl with a recent Tcl, but it's not clear they wouldn't, either.
Also add a regression test using "clock", in branches that have regression
test support for pltcl.

Per recent trouble report from Kyle Bateman.

14 years agoRewrite dblink_record_internal() and dblink_fetch() to use a tuplestore
Joe Conway [Sun, 24 Jan 2010 22:19:38 +0000 (22:19 +0000)]
Rewrite dblink_record_internal() and dblink_fetch() to use a tuplestore
(SFRM_Materialize mode) to return tuples. Since we don't return from the
dblink function in tuplestore mode, release the PGresult with a PG_CATCH
block on error. Also rearrange to share the same code to materialize the
tuplestore. Patch by Takahiro Itagaki.

14 years agoFix assorted core dumps and Assert failures that could occur during
Tom Lane [Sun, 24 Jan 2010 21:49:17 +0000 (21:49 +0000)]
Fix assorted core dumps and Assert failures that could occur during
AbortTransaction or AbortSubTransaction, when trying to clean up after an
error that prevented (sub)transaction start from completing:
* access to TopTransactionResourceOwner that might not exist
* assert failure in AtEOXact_GUC, if AtStart_GUC not called yet
* assert failure or core dump in AfterTriggerEndSubXact, if
  AfterTriggerBeginSubXact not called yet

Per testing by injecting elog(ERROR) at successive steps in StartTransaction
and StartSubTransaction.  It's not clear whether all of these cases could
really occur in the field, but at least one of them is easily exposed by
simple stress testing, as per my accidental discovery yesterday.

14 years agoAdd missing closing tag.
Magnus Hagander [Sun, 24 Jan 2010 14:46:01 +0000 (14:46 +0000)]
Add missing closing tag.

14 years agoInsert CHECK_FOR_INTERRUPTS calls into loops in dbsize.c, to ensure that
Tom Lane [Sat, 23 Jan 2010 21:29:00 +0000 (21:29 +0000)]
Insert CHECK_FOR_INTERRUPTS calls into loops in dbsize.c, to ensure that
the various disk-size-reporting functions will respond to query cancel
reasonably promptly even in very large databases.  Per report from
Kevin Grittner.

14 years agoAdd explanatory detail to Hot Standby cancelation error messages
Simon Riggs [Sat, 23 Jan 2010 17:04:05 +0000 (17:04 +0000)]
Add explanatory detail to Hot Standby cancelation error messages
with errdetail(). Add errhint() to suggest retry in certain cases.

14 years agoIn HS, Startup process sets SIGALRM when waiting for buffer pin. If
Simon Riggs [Sat, 23 Jan 2010 16:37:12 +0000 (16:37 +0000)]
In HS, Startup process sets SIGALRM when waiting for buffer pin. If
woken by alarm we send SIGUSR1 to all backends requesting that they
check to see if they are blocking Startup process. If so, they throw
ERROR/FATAL as for other conflict resolutions. Deadlock stop gap
removed. max_standby_delay = -1 option removed to prevent deadlock.

14 years agoRevert mention that HTML documentation has to be built first.
Bruce Momjian [Fri, 22 Jan 2010 22:50:03 +0000 (22:50 +0000)]
Revert mention that HTML documentation has to be built first.

Now require gmake 3.79.1 or later to build source/sgml.

14 years agoFix mismatched tags introduced by attribute options patch.
Robert Haas [Fri, 22 Jan 2010 17:30:24 +0000 (17:30 +0000)]
Fix mismatched tags introduced by attribute options patch.

Thanks to Devrim Gunduz for the report.

14 years agoFix several oversights in previous commit - attribute options patch.
Robert Haas [Fri, 22 Jan 2010 16:42:31 +0000 (16:42 +0000)]
Fix several oversights in previous commit - attribute options patch.

I failed to 'cvs add' the new files and also neglected to bump catversion.

14 years agoReplace ALTER TABLE ... SET STATISTICS DISTINCT with a more general mechanism.
Robert Haas [Fri, 22 Jan 2010 16:40:19 +0000 (16:40 +0000)]
Replace ALTER TABLE ... SET STATISTICS DISTINCT with a more general mechanism.

Attributes can now have options, just as relations and tablespaces do, and
the reloptions code is used to parse, validate, and store them.  For
simplicity and because these options are not performance critical, we store
them in a separate cache rather than the main relcache.

Thanks to Alex Hunsaker for the review.

14 years agoDocument that the HTML documentation must be built before the PDF
Bruce Momjian [Fri, 22 Jan 2010 15:49:29 +0000 (15:49 +0000)]
Document that the HTML documentation must be built before the PDF
documentation.

14 years agoPL/Python DO handler
Peter Eisentraut [Fri, 22 Jan 2010 15:45:15 +0000 (15:45 +0000)]
PL/Python DO handler

Also cleaned up some redundancies between the primary error messages and the
error context in PL/Python.

Hannu Valtonen

14 years agoMake argument const char * as per Takahiro Itagaki's suggestion.
Michael Meskes [Fri, 22 Jan 2010 14:19:27 +0000 (14:19 +0000)]
Make argument const char * as per Takahiro Itagaki's suggestion.

14 years agoApplied patch by Boszormenyi Zoltan <zb@cybertec.at> to fix problem in auto-prepare...
Michael Meskes [Fri, 22 Jan 2010 14:13:03 +0000 (14:13 +0000)]
Applied patch by Boszormenyi Zoltan <zb@cybertec.at> to fix problem in auto-prepare mode if the connection is closed and re-opened and the previously prepared query is issued again.

14 years agoFix unsafe loop test, and declare as_ident as bool rather than int.
Robert Haas [Thu, 21 Jan 2010 18:43:25 +0000 (18:43 +0000)]
Fix unsafe loop test, and declare as_ident as bool rather than int.

14 years agoAdd new escaping functions PQescapeLiteral and PQescapeIdentifier.
Robert Haas [Thu, 21 Jan 2010 14:58:53 +0000 (14:58 +0000)]
Add new escaping functions PQescapeLiteral and PQescapeIdentifier.

PQescapeLiteral is similar to PQescapeStringConn, but it relieves the
caller of the need to know how large the output buffer should be, and
it provides the appropriate quoting (in addition to escaping special
characers within the string).  PQescapeIdentifier provides similar
functionality for escaping identifiers.

Per recent discussion with Tom Lane.

14 years agoAdd missing flag reset to ensure subsequent manual cancelation gives correct reason.
Simon Riggs [Thu, 21 Jan 2010 09:30:36 +0000 (09:30 +0000)]
Add missing flag reset to ensure subsequent manual cancelation gives correct reason.

14 years agoFix bogus comments.
Heikki Linnakangas [Thu, 21 Jan 2010 08:19:57 +0000 (08:19 +0000)]
Fix bogus comments.

14 years agoAdjust psql to use pg_get_triggerdef(pretty=true) to remove extra ()'s
Itagaki Takahiro [Thu, 21 Jan 2010 06:11:46 +0000 (06:11 +0000)]
Adjust psql to use pg_get_triggerdef(pretty=true) to remove extra ()'s
from description of triggers with WHEN clause.

Thanks to Brad T. Sliger for the review.

14 years agoBetter internal documentation of locking for Hot Standby conflict resolution.
Simon Riggs [Thu, 21 Jan 2010 00:53:58 +0000 (00:53 +0000)]
Better internal documentation of locking for Hot Standby conflict resolution.
Discuss the reasons for the lock type we hold on ProcArrayLock while deriving
the conflict list. Cover the idea of false positive conflicts and seemingly
strange effects on snapshot derivation.

14 years agoWell, the systemtap guys moved the goalposts again: with the latest version,
Tom Lane [Wed, 20 Jan 2010 23:12:03 +0000 (23:12 +0000)]
Well, the systemtap guys moved the goalposts again: with the latest version,
we *must* generate probes.o or the dtrace probes don't work.  Revert our
workaround for their previous bug.  Details at
https://bugzilla.redhat.com/show_bug.cgi?id=557266

14 years agoAdd user-specific .pg_service.conf file
Peter Eisentraut [Wed, 20 Jan 2010 21:15:21 +0000 (21:15 +0000)]
Add user-specific .pg_service.conf file

This extends the existing pg_service.conf facility to first look for a
service definition file in the user's home directory.

14 years agoFix bogus subdir setting. Again. I must've unfixed it by accident while
Heikki Linnakangas [Wed, 20 Jan 2010 20:34:51 +0000 (20:34 +0000)]
Fix bogus subdir setting. Again. I must've unfixed it by accident while
moving files around.

14 years agoWrite a WAL record whenever we perform an operation without WAL-logging
Heikki Linnakangas [Wed, 20 Jan 2010 19:43:40 +0000 (19:43 +0000)]
Write a WAL record whenever we perform an operation without WAL-logging
that would've been WAL-logged if archiving was enabled. If we encounter
such records in archive recovery anyway, we know that some data is
missing from the log. A WARNING is emitted in that case.

Original patch by Fujii Masao, with changes by me.

14 years agoNow that much of walreceiver has been pulled back into the postgres
Heikki Linnakangas [Wed, 20 Jan 2010 18:54:27 +0000 (18:54 +0000)]
Now that much of walreceiver has been pulled back into the postgres
binary, revert PGDLLIMPORT decoration of global variables. I'm not sure
if there's any real harm from unnecessary PGDLLIMPORTs, but these are all
internal variables that external modules really shouldn't be messing
with. ThisTimeLineID still needs PGDLLIMPORT.

14 years agoAdd missing "!= NULL", for the sake of consistency.
Heikki Linnakangas [Wed, 20 Jan 2010 11:58:44 +0000 (11:58 +0000)]
Add missing "!= NULL", for the sake of consistency.

Fujii Masao

14 years agoAdjust windows makefiles too, now that the walreceiver dynamic module
Heikki Linnakangas [Wed, 20 Jan 2010 09:30:07 +0000 (09:30 +0000)]
Adjust windows makefiles too, now that the walreceiver dynamic module
has been renamed.

14 years agoAdjust MSVC build script too, now that the walreceiver dynamic module
Heikki Linnakangas [Wed, 20 Jan 2010 09:22:43 +0000 (09:22 +0000)]
Adjust MSVC build script too, now that the walreceiver dynamic module
has been renamed.

14 years agoRethink the way walreceiver is linked into the backend. Instead than shoving
Heikki Linnakangas [Wed, 20 Jan 2010 09:16:24 +0000 (09:16 +0000)]
Rethink the way walreceiver is linked into the backend. Instead than shoving
walreceiver as whole into a dynamically loaded module, split the
libpq-specific parts of it into dynamically loaded module and keep the rest
in the main backend binary.

Although Tom fixed the Windows compilation problems with the old walreceiver
module already, this is a cleaner division of labour and makes the code
more readable. There's also the prospect of adding new transport methods
as pluggable modules in the future, which this patch makes easier, though for
now the API between libpqwalreceiver and walreceiver process should be
considered private.

The libpq-specific module is now in src/backend/replication/libpqwalreceiver,
and the part linked with postgres binary is in
src/backend/replication/walreceiver.c.

14 years agoBefore attempting to create a composite type, check whether a type of that
Peter Eisentraut [Wed, 20 Jan 2010 05:47:09 +0000 (05:47 +0000)]
Before attempting to create a composite type, check whether a type of that
name already exists, so we'd get an error message about a "type" instead
of about a "relation", because the composite type code shares code with
relation creation.

14 years agoFix typo in Mkvcbuild.pl
Andrew Dunstan [Wed, 20 Jan 2010 04:29:07 +0000 (04:29 +0000)]
Fix typo in Mkvcbuild.pl

14 years agoBuild plperl's new Util.xs
Andrew Dunstan [Wed, 20 Jan 2010 04:14:06 +0000 (04:14 +0000)]
Build plperl's new Util.xs

14 years agoFix doc build, which was broken by PL/perl utility functions patch.
Robert Haas [Wed, 20 Jan 2010 03:37:10 +0000 (03:37 +0000)]
Fix doc build, which was broken by PL/perl utility functions patch.

14 years agoAdd utility functions to PLPerl:
Andrew Dunstan [Wed, 20 Jan 2010 01:08:21 +0000 (01:08 +0000)]
Add utility functions to PLPerl:
    quote_literal, quote_nullable, quote_ident,
    encode_bytea, decode_bytea, looks_like_number,
    encode_array_literal, encode_array_constructor.
Split SPI.xs into two - SPI.xs now contains only SPI functions. Remainder
are in new Util.xs.
Some more code and documentation cleanup along the way, as well as
adding some CVS markers to files missing them.

Original patch from Tim Bunce, with a little editing from me.

14 years agoReformat documentation of libpq escaping functions.
Robert Haas [Wed, 20 Jan 2010 00:42:28 +0000 (00:42 +0000)]
Reformat documentation of libpq escaping functions.

Modify the "Escaping Strings for Inclusion in SQL Commands" section
to use a <variablelist> as the preceding and following sections do,
and merge the "Escaping Binary Strings for Inclusion in SQL Commands"
section into it.

This changes only the formatting of these sections, not the content.
It is intended to lay the groundwork for a follow-on patch to add
some new escaping functions, but it makes sense to commit this first,
for clarity.

14 years agoWhen doing a parallel restore, we must guard against out-of-range dependency
Tom Lane [Tue, 19 Jan 2010 18:39:19 +0000 (18:39 +0000)]
When doing a parallel restore, we must guard against out-of-range dependency
dump IDs, because the array we're using is sized according to the highest
dump ID actually defined in the archive file.  In a partial dump there could
be references to higher dump IDs that weren't dumped.  Treat these the same
as references to in-range IDs that weren't dumped.  (The whole thing is a
bit scary because the missing objects might have been part of dependency
chains, which we won't know about.  Not much we can do though --- throwing
an error is probably overreaction.)

Also, reject parallel restore with pre-1.8 archive version (made by pre-8.0
pg_dump).  In these old versions the dependency entries are OIDs, not dump
IDs, and we don't have enough information to interpret them.

Per bug #5288 from Jon Erdman.

14 years agoFix thinko in my recent change to put an explicit argisrow field in NullTest:
Tom Lane [Tue, 19 Jan 2010 16:33:33 +0000 (16:33 +0000)]
Fix thinko in my recent change to put an explicit argisrow field in NullTest:
when the planner splits apart a ROW(...) IS NULL test, the argisrow values
of the component tests have to be determined from the component field types,
not copied from the original NullTest (in which argisrow is surely true).

14 years agoAdd pg_stat_reset_shared('bgwriter') to reset the cluster-wide shared
Magnus Hagander [Tue, 19 Jan 2010 14:11:32 +0000 (14:11 +0000)]
Add pg_stat_reset_shared('bgwriter') to reset the cluster-wide shared
statistics of the bgwriter.

Greg Smith

14 years agoAdd pg_table_size() and pg_indexes_size() to provide more user-friendly
Tom Lane [Tue, 19 Jan 2010 05:50:18 +0000 (05:50 +0000)]
Add pg_table_size() and pg_indexes_size() to provide more user-friendly
wrappers around the pg_relation_size() function.

Bernd Helmle, reviewed by Greg Smith

14 years agoAdd "USING expressions" option to plpgsql's OPEN cursor FOR EXECUTE.
Tom Lane [Tue, 19 Jan 2010 01:35:31 +0000 (01:35 +0000)]
Add "USING expressions" option to plpgsql's OPEN cursor FOR EXECUTE.

This is the last EXECUTE-like plpgsql statement that was missing
the capability of inserting parameter values via USING.

Pavel Stehule, reviewed by Itagaki Takahiro

14 years agoCast slightly abused enum to int, so that GCC 4.5 won't warn about switch
Peter Eisentraut [Mon, 18 Jan 2010 22:19:34 +0000 (22:19 +0000)]
Cast slightly abused enum to int, so that GCC 4.5 won't warn about switch
cases that are not part of the enum's definition.

14 years agoFix an oversight in convert_EXISTS_sublink_to_join: we can't convert an
Tom Lane [Mon, 18 Jan 2010 18:17:45 +0000 (18:17 +0000)]
Fix an oversight in convert_EXISTS_sublink_to_join: we can't convert an
EXISTS that contains a WITH clause.  This would usually lead to a
"could not find CTE" error later in planning, because the WITH wouldn't
get processed at all.  Noted while playing with an example from Ken Marshall.

14 years agoFix incorrect comparison of scan key in GIN. Per report from
Teodor Sigaev [Mon, 18 Jan 2010 11:50:43 +0000 (11:50 +0000)]
Fix incorrect comparison of scan key in GIN. Per report from
Vyacheslav Kalinin <vka@mgcp.com>

14 years agoFix portalmem.c to avoid keeping a dangling pointer to a cached plan list
Tom Lane [Mon, 18 Jan 2010 02:30:25 +0000 (02:30 +0000)]
Fix portalmem.c to avoid keeping a dangling pointer to a cached plan list
after it's released its reference count for the cached plan.  There are
code paths that might try to examine the plan list before noticing that
the portal is already in aborted state.  Report and diagnosis by Tatsuo
Ishii, though this isn't exactly his proposed patch.

14 years agoUpdate SET CONSTRAINTS reference page --- it failed to mention EXCLUDE
Tom Lane [Mon, 18 Jan 2010 00:32:21 +0000 (00:32 +0000)]
Update SET CONSTRAINTS reference page --- it failed to mention EXCLUDE
constraints and was lacking a couple of other interesting details.

14 years agoImprove the handling of SET CONSTRAINTS commands by having them search
Tom Lane [Sun, 17 Jan 2010 22:56:23 +0000 (22:56 +0000)]
Improve the handling of SET CONSTRAINTS commands by having them search
pg_constraint before searching pg_trigger.  This allows saner handling of
corner cases; in particular we now say "constraint is not deferrable"
rather than "constraint does not exist" when the command is applied to
a constraint that's inherently non-deferrable.  Per a gripe several months
ago from hubert depesz lubaczewski.

To make this work without breaking user-defined constraint triggers,
we have to add entries for them to pg_constraint.  However, in return
we can remove the pgconstrname column from pg_constraint, which represents
a fairly sizable space savings.  I also replaced the tgisconstraint column
with tgisinternal; the old meaning of tgisconstraint can now be had by
testing for nonzero tgconstraint, while there is no other way to get
the old meaning of nonzero tgconstraint, namely that the trigger was
internally generated rather than being user-created.

In passing, fix an old misstatement in the docs and comments, namely that
pg_trigger.tgdeferrable is exactly redundant with pg_constraint.condeferrable.
Actually, we mark RI action triggers as nondeferrable even when they belong to
a nominally deferrable FK constraint.  The SET CONSTRAINTS code now relies on
that instead of hard-coding a list of exception OIDs.

14 years agoAdd include directory for dblink to find fmgroids.h on VS2008 builds.
Magnus Hagander [Sun, 17 Jan 2010 13:21:50 +0000 (13:21 +0000)]
Add include directory for dblink to find fmgroids.h on VS2008 builds.

14 years agoSilence compiler warning, noted by Stefan Kaltenbrunner
Magnus Hagander [Sun, 17 Jan 2010 12:11:25 +0000 (12:11 +0000)]
Silence compiler warning, noted by Stefan Kaltenbrunner

14 years agoRemove unnecessary, inconsistent flag resets in ProcessInterrupts.
Tom Lane [Sun, 17 Jan 2010 04:27:54 +0000 (04:27 +0000)]
Remove unnecessary, inconsistent flag resets in ProcessInterrupts.

14 years agoFix spelling error, noticed by Thomas Shinnick
Magnus Hagander [Sat, 16 Jan 2010 20:38:54 +0000 (20:38 +0000)]
Fix spelling error, noticed by Thomas Shinnick

14 years agoRe-order configure tests to reflect the fact that the code generated for
Tom Lane [Sat, 16 Jan 2010 19:50:26 +0000 (19:50 +0000)]
Re-order configure tests to reflect the fact that the code generated for
posix_fadvise and other file-related functions can depend on _LARGEFILE_SOURCE
and/or _FILE_OFFSET_BITS.  Per report from Robert Treat.

Back-patch to 8.4.  This has been wrong all along, but we weren't really using
posix_fadvise in anger before, and AC_FUNC_FSEEKO seems to mask the issue well
enough for that function.

14 years agoFix unportable use of isxdigit() with char (rather than unsigned char)
Tom Lane [Sat, 16 Jan 2010 17:39:55 +0000 (17:39 +0000)]
Fix unportable use of isxdigit() with char (rather than unsigned char)
argument, per warnings from buildfarm member pika.  Also clean up code
formatting a trifle.

14 years agoFix bogus initialization of KnownAssignedXids shared memory state ---
Tom Lane [Sat, 16 Jan 2010 17:17:26 +0000 (17:17 +0000)]
Fix bogus initialization of KnownAssignedXids shared memory state ---
didn't work in EXEC_BACKEND case.

14 years agoLock database while running drop database in Hot Standby to protect
Simon Riggs [Sat, 16 Jan 2010 14:16:31 +0000 (14:16 +0000)]
Lock database while running drop database in Hot Standby to protect
against concurrent reconnection. Failure during testing showed issue
was possible, even though earlier analysis seemed to indicate it
would not be required. Use LockSharedObjectForSession() before
ResolveRecoveryConflictWithDatabase() and hold lock until end of
processing for that WAL record. Simple approach to avoid introducing
further bugs at this stage of development on an improbable issue.

14 years agoBuild the walreceiver library on MSVC as well.
Magnus Hagander [Sat, 16 Jan 2010 11:55:38 +0000 (11:55 +0000)]
Build the walreceiver library on MSVC as well.

14 years agoImproved printing of Python exceptions in PL/Python
Peter Eisentraut [Sat, 16 Jan 2010 11:03:51 +0000 (11:03 +0000)]
Improved printing of Python exceptions in PL/Python

Mimic the Python interpreter's own logic for printing exceptions instead
of just using the straight str() call, so that
you get

    plpy.SPIError

instead of

    <class 'plpy.SPIError'>

and for built-in exceptions merely

    UnicodeEncodeError

Besides looking better this cuts down on the endless version differences
in the regression test expected files.

14 years agoMessage mentions msec when it should be seconds, so use s instead of ms.
Simon Riggs [Sat, 16 Jan 2010 10:13:04 +0000 (10:13 +0000)]
Message mentions msec when it should be seconds, so use s instead of ms.

Noticed by Andres Freund

14 years agoTeach standby conflict resolution to use SIGUSR1
Simon Riggs [Sat, 16 Jan 2010 10:05:59 +0000 (10:05 +0000)]
Teach standby conflict resolution to use SIGUSR1
Conflict reason is passed through directly to the backend, so we can
take decisions about the effect of the conflict based upon the local
state. No specific changes, as yet, though this prepares for later work.
CancelVirtualTransaction() sends signals while holding ProcArrayLock.
Introduce errdetail_abort() to give message detail explaining that the
abort was caused by conflict processing. Remove CONFLICT_MODE states
in favour of using PROCSIG_RECOVERY_CONFLICT states directly, for clarity.

14 years agoHuh, apparently on cygwin we HAVE_SIGPROCMASK, so both variants of
Tom Lane [Sat, 16 Jan 2010 05:52:29 +0000 (05:52 +0000)]
Huh, apparently on cygwin we HAVE_SIGPROCMASK, so both variants of
the BlockSig/UnBlockSig declaration have to be PGDLLIMPORT'ified.
Per buildfarm results.

14 years agoAdd #include <sys/time.h> for struct timeval definition on BSD/OS.
Bruce Momjian [Sat, 16 Jan 2010 01:55:28 +0000 (01:55 +0000)]
Add #include <sys/time.h> for struct timeval definition on BSD/OS.

14 years agoPGDLLIMPORT-ize the remaining variables needed by walreceiver.
Tom Lane [Sat, 16 Jan 2010 00:04:41 +0000 (00:04 +0000)]
PGDLLIMPORT-ize the remaining variables needed by walreceiver.

14 years agoDo parse analysis of an EXPLAIN's contained statement during the normal
Tom Lane [Fri, 15 Jan 2010 22:36:35 +0000 (22:36 +0000)]
Do parse analysis of an EXPLAIN's contained statement during the normal
parse analysis phase, rather than at execution time.  This makes parameter
handling work the same as it does in ordinary plannable queries, and in
particular fixes the incompatibility that Pavel pointed out with plpgsql's
new handling of variable references.  plancache.c gets a little bit
grottier, but the alternatives seem worse.

14 years agoSecond try: walreceiver should not be built with -DBUILDING_DLL
Tom Lane [Fri, 15 Jan 2010 21:43:21 +0000 (21:43 +0000)]
Second try: walreceiver should not be built with -DBUILDING_DLL

14 years agoNo, scratch that, it was getting added twice.
Tom Lane [Fri, 15 Jan 2010 21:06:26 +0000 (21:06 +0000)]
No, scratch that, it was getting added twice.

14 years agoActually, I'll bet the mingw problem is lack of $(BE_DLLLIBS) ...
Tom Lane [Fri, 15 Jan 2010 20:45:42 +0000 (20:45 +0000)]
Actually, I'll bet the mingw problem is lack of $(BE_DLLLIBS) ...

14 years agoFix bogus subdir setting ... wonder just what that affects ...
Tom Lane [Fri, 15 Jan 2010 20:34:11 +0000 (20:34 +0000)]
Fix bogus subdir setting ... wonder just what that affects ...

14 years agoMove build of src/backend/replication/walreceiver/ later in the build
Heikki Linnakangas [Fri, 15 Jan 2010 17:01:06 +0000 (17:01 +0000)]
Move build of src/backend/replication/walreceiver/ later in the build
process, after src/interfaces, because it depends on libpq. Also add
missing lines for clean etc. targets

Report from Boszormenyi Zoltan.

14 years agoAdded correct error handling in DESCRIBE statement processing by Boszormenyi Zoltan...
Michael Meskes [Fri, 15 Jan 2010 13:19:12 +0000 (13:19 +0000)]
Added correct error handling in DESCRIBE statement processing by Boszormenyi Zoltan <zb@cybertec.at>.

14 years agoRemove unused (in non-assertion-enabled build) variable.
Heikki Linnakangas [Fri, 15 Jan 2010 11:47:15 +0000 (11:47 +0000)]
Remove unused (in non-assertion-enabled build) variable.

14 years agoApplied patch by Boszormenyi Zoltan <zb@cybertec.at> to add DESCRIBE [OUTPUT] stateme...
Michael Meskes [Fri, 15 Jan 2010 10:44:39 +0000 (10:44 +0000)]
Applied patch by Boszormenyi Zoltan <zb@cybertec.at> to add DESCRIBE [OUTPUT] statement to ecpg.

14 years agoIntroduce Streaming Replication.
Heikki Linnakangas [Fri, 15 Jan 2010 09:19:10 +0000 (09:19 +0000)]
Introduce Streaming Replication.

This includes two new kinds of postmaster processes, walsenders and
walreceiver. Walreceiver is responsible for connecting to the primary server
and streaming WAL to disk, while walsender runs in the primary server and
streams WAL from disk to the client.

Documentation still needs work, but the basics are there. We will probably
pull the replication section to a new chapter later on, as well as the
sections describing file-based replication. But let's do that as a separate
patch, so that it's easier to see what has been added/changed. This patch
also adds a new section to the chapter about FE/BE protocol, documenting the
protocol used by walsender/walreceivxer.

Bump catalog version because of two new functions,
pg_last_xlog_receive_location() and pg_last_xlog_replay_location(), for
monitoring the progress of replication.

Fujii Masao, with additional hacking by me

14 years agoAdd point_ops opclass for GiST.
Teodor Sigaev [Thu, 14 Jan 2010 16:31:09 +0000 (16:31 +0000)]
Add point_ops opclass for GiST.

14 years agoFirst part of refactoring of code for ResolveRecoveryConflict. Purposes
Simon Riggs [Thu, 14 Jan 2010 11:08:02 +0000 (11:08 +0000)]
First part of refactoring of code for ResolveRecoveryConflict. Purposes
of this are to centralise the conflict code to allow further change,
as well as to allow passing through the full reason for the conflict
through to the conflicting backends. Backend state alters how we
can handle different types of conflict so this is now required.
As originally suggested by Heikki, no longer optional.

14 years agoSimplify validate_exec() by using access(2) to check file permissions,
Tom Lane [Thu, 14 Jan 2010 00:14:06 +0000 (00:14 +0000)]
Simplify validate_exec() by using access(2) to check file permissions,
rather than trying to implement the equivalent logic by hand.  The motivation
for the original coding appears to have been to check with the effective uid's
permissions not the real uid's; but there is no longer any difference, because
we don't run the postmaster setuid (indeed, main.c enforces that they're the
same).  Using access() means we will get it right in situations the original
coding failed to handle, such as ACL-based permissions.  Besides it's a lot
shorter, cleaner, and more thread-safe.  Per bug #5275 from James Bellinger.

14 years agoWhen loading critical system indexes into the relcache, ensure we lock the
Tom Lane [Wed, 13 Jan 2010 23:07:08 +0000 (23:07 +0000)]
When loading critical system indexes into the relcache, ensure we lock the
underlying catalog not only the index itself.  Otherwise, if the cache
load process touches the catalog (which will happen for many though not
all of these indexes), we are locking index before parent table, which can
result in a deadlock against processes that are trying to lock them in the
normal order.  Per today's failure on buildfarm member gothic_moth; it's
surprising the problem hadn't been identified before.

Back-patch to 8.2.  Earlier releases didn't have the issue because they
didn't try to lock these indexes during load (instead assuming that they
couldn't change schema at all during multiuser operation).

14 years agoFix bug #5269: ResetPlanCache mustn't invalidate cached utility statements,
Tom Lane [Wed, 13 Jan 2010 16:56:56 +0000 (16:56 +0000)]
Fix bug #5269: ResetPlanCache mustn't invalidate cached utility statements,
especially not ROLLBACK.  ROLLBACK might need to be executed in an already
aborted transaction, when there is no safe way to revalidate the plan.  But
in general there's no point in marking utility statements invalid, since
they have no plans in the normal sense of the word; so we might as well
work a bit harder here to avoid future revalidation cycles.

Back-patch to 8.4, where the bug was introduced.

14 years agoApplied Zoltan's patch to make char the default sqlda type.
Michael Meskes [Wed, 13 Jan 2010 09:06:51 +0000 (09:06 +0000)]
Applied Zoltan's patch to make char the default sqlda type.

Given that undefined types are handled as character strings anyway the type
translation function can simply return the correcponding ECPGt_char type.

14 years agoFix SQL3 type return value.
Michael Meskes [Wed, 13 Jan 2010 08:41:50 +0000 (08:41 +0000)]
Fix SQL3 type return value.

For non-SQL3 types ecpg used to return -Oid. This will break if there are
enough Oids to fill the namespace. Therefore we play it safe and return 0 if
there is no Oid->SQL3 tyoe mapping available.

14 years agoMake fixed_paramref_hook behave properly when there are 'unused' slots
Tom Lane [Wed, 13 Jan 2010 01:17:07 +0000 (01:17 +0000)]
Make fixed_paramref_hook behave properly when there are 'unused' slots
in the parameter array.  Noted while experimenting with an example
from Pavel.  This wouldn't come up in normal use, but it ought to honor
the specification that a parameter array can have unused slots.

14 years agoUpdate MSVC build instructions.
Magnus Hagander [Tue, 12 Jan 2010 20:13:32 +0000 (20:13 +0000)]
Update MSVC build instructions.

Updated to reflect building with Visual Studio 2008, with just the
Platform SDK, and for 64-bit Windows.

14 years agoFix relcache reload mechanism to be more robust in the face of errors
Tom Lane [Tue, 12 Jan 2010 18:12:18 +0000 (18:12 +0000)]
Fix relcache reload mechanism to be more robust in the face of errors
occurring during a reload, such as query-cancel.  Instead of zeroing out
an existing relcache entry and rebuilding it in place, build a new relcache
entry, then swap its contents with the old one, then free the new entry.
This avoids problems with code believing that a previously obtained pointer
to a cache entry must still reference a valid entry, as seen in recent
failures on buildfarm member jaguar.  (jaguar is using CLOBBER_CACHE_ALWAYS
which raises the probability of failure substantially, but the problem
could occur in the field without that.)  The previous design was okay
when it was made, but subtransactions and the ResourceOwner mechanism
make it unsafe now.

Also, make more use of the already existing rd_isvalid flag, so that we
remember that the entry requires rebuilding even if the first attempt fails.

Back-patch as far as 8.2.  Prior versions have enough issues around relcache
reload anyway (due to inadequate locking) that fixing this one doesn't seem
worthwhile.

14 years agoRemoved ecpg Changelog file that was only there for historical reasons and isn't...
Michael Meskes [Tue, 12 Jan 2010 12:41:46 +0000 (12:41 +0000)]
Removed ecpg Changelog file that was only there for historical reasons and isn't actively used anymore.

14 years agoPlease tablespace directories in their own subdirectory so pg_migrator
Bruce Momjian [Tue, 12 Jan 2010 02:42:52 +0000 (02:42 +0000)]
Please tablespace directories in their own subdirectory so pg_migrator
can upgrade clusters without renaming the tablespace directories.  New
directory structure format is, e.g.:

$PGDATA/pg_tblspc/20981/PG_8.5_201001061/719849/83292814

14 years agoFix aclexplode to not explode on a zero-entry ACL array.
Tom Lane [Tue, 12 Jan 2010 02:39:22 +0000 (02:39 +0000)]
Fix aclexplode to not explode on a zero-entry ACL array.
(An Assert is no substitute for thinking clearly :-()

Minor style kibitzing too.

Per report from Robert Treat.

14 years agoRemove tabs in SGML file.
Bruce Momjian [Tue, 12 Jan 2010 02:33:45 +0000 (02:33 +0000)]
Remove tabs in SGML file.

14 years agoAdd some simple support and documentation for using process-specific oom_adj
Tom Lane [Mon, 11 Jan 2010 18:39:32 +0000 (18:39 +0000)]
Add some simple support and documentation for using process-specific oom_adj
settings to prevent the postmaster from being OOM-killed on Linux systems.

Alex Hunsaker and Tom Lane

14 years agoImprove ExecEvalVar's handling of whole-row variables in cases where the
Tom Lane [Mon, 11 Jan 2010 15:31:04 +0000 (15:31 +0000)]
Improve ExecEvalVar's handling of whole-row variables in cases where the
rowtype contains dropped columns.  Sometimes the input tuple will be formed
from a select targetlist in which dropped columns are filled with a NULL
of an arbitrary type (the planner typically uses INT4, since it can't tell
what type the dropped column really was).  So we need to relax the rowtype
compatibility check to not insist on physical compatibility if the actual
column value is NULL.

In principle we might need to do this for functions returning composite
types, too (see tupledesc_match()).  In practice there doesn't seem to be
a bug there, probably because the function will be using the same cached
rowtype descriptor as the caller.  Fixing that code path would require
significant rearrangement, so I left it alone for now.

Per complaint from Filip Rembialkowski.

14 years agoremove use of temp file for perlchunks.h, as in recent change to plperl GNUmakefile
Andrew Dunstan [Mon, 11 Jan 2010 14:16:18 +0000 (14:16 +0000)]
remove use of temp file for perlchunks.h, as in recent change to plperl GNUmakefile

14 years agoSome trivial adjustments in comments for struct RelationData.
Tom Lane [Sun, 10 Jan 2010 22:19:17 +0000 (22:19 +0000)]
Some trivial adjustments in comments for struct RelationData.

14 years agoRemove unnecessary use of temp file. Tim Bunce
Tom Lane [Sun, 10 Jan 2010 18:10:03 +0000 (18:10 +0000)]
Remove unnecessary use of temp file.  Tim Bunce

14 years agoImprove plpgsql parsing to report "foo is not a known variable", rather than a
Tom Lane [Sun, 10 Jan 2010 17:56:50 +0000 (17:56 +0000)]
Improve plpgsql parsing to report "foo is not a known variable", rather than a
generic syntax error, when seeing "foo := something" and foo isn't recognized.
This buys back most of the helpfulness discarded in my previous patch by not
throwing errors when a qualified name appears to match a row variable but the
last component doesn't match any field of the row.  It covers other cases
where our error messages left something to be desired, too.

14 years agoImprove plpgsql's handling of record field references by forcing all potential
Tom Lane [Sun, 10 Jan 2010 17:15:18 +0000 (17:15 +0000)]
Improve plpgsql's handling of record field references by forcing all potential
field references in SQL expressions to have RECFIELD datum-array entries at
parse time.  If it turns out that the reference is actually to a SQL column,
the RECFIELD entry is useless, but it costs little.  This allows us to get rid
of the previous use of FieldSelect applied to a whole-row Param for the record
variable; which was not only slower than a direct RECFIELD reference, but
failed for references to system columns of a trigger's NEW or OLD record.
Per report and fix suggestion from Dean Rasheed.

14 years agoDocs for behaviour change of drop database during Hot Standby
Simon Riggs [Sun, 10 Jan 2010 16:27:27 +0000 (16:27 +0000)]
Docs for behaviour change of drop database during Hot Standby

Adding missing docs for previous Hot Standby patch.

14 years agoUpdate Windows installation notes.
Magnus Hagander [Sun, 10 Jan 2010 15:54:11 +0000 (15:54 +0000)]
Update Windows installation notes.

pginstaller isn't used anymore, in favor of the one-click installers.
Make it clear that we support Windows 2000 and newer with the native
port, instead of first saying we support NT4 and then saying we don't.

14 years agoDuring Hot Standby, fix drop database when sessions idle.
Simon Riggs [Sun, 10 Jan 2010 15:44:28 +0000 (15:44 +0000)]
During Hot Standby, fix drop database when sessions idle.
Previously we only cancelled sessions that were in-transaction.

Simple fix is to just cancel all sessions without waiting. Doing
it this way avoids complicating common code paths, which would
not be worth the trouble to cover this rare case.

Problem report and fix by Andres Freund, edited somewhat by me

14 years agoCreate typedef pgsocket for storing socket descriptors.
Magnus Hagander [Sun, 10 Jan 2010 14:16:08 +0000 (14:16 +0000)]
Create typedef pgsocket for storing socket descriptors.

This silences some warnings on Win64. Not using the proper SOCKET datatype
was actually wrong on Win32 as well, but didn't cause any warnings there.

Also create define PGINVALID_SOCKET to indicate an invalid/non-existing
socket, instead of using a hardcoded -1 value.

14 years agoRemove partial, broken support for NULL pointers when fetching attributes.
Robert Haas [Sun, 10 Jan 2010 04:26:36 +0000 (04:26 +0000)]
Remove partial, broken support for NULL pointers when fetching attributes.

Previously, fastgetattr() and heap_getattr() tested their fourth argument
against a null pointer, but any attempt to use them with a literal-NULL
fourth argument evaluated to *(void *)0, resulting in a compiler error.
Remove these NULL tests to avoid leading future readers of this code to
believe that this has a chance of working.  Also clean up related legacy
code in nocachegetattr(), heap_getsysattr(), and nocache_index_getattr().

The new coding standard is that any code which calls a getattr-type
function or macro which takes an isnull argument MUST pass a valid
boolean pointer.  Per discussion with Bruce Momjian, Tom Lane, Alvaro
Herrera.