Andrew Dunstan [Sun, 15 Jun 2008 21:46:02 +0000 (21:46 +0000)]
Add script to find .c and .h files that are missing CVS PostgreSQL markers
and add them. Avoids third party files or those that would cause regression
failures.
Tom Lane [Sun, 15 Jun 2008 16:29:05 +0000 (16:29 +0000)]
Make DROP INDEX lock the parent table before locking the index. This behavior
is necessary to avoid deadlock against ordinary queries, but we'd broken it
with recent changes that made the DROP machinery lock the index before
arriving at index_drop. Per intermittent buildfarm failures.
Tom Lane [Sun, 15 Jun 2008 01:25:54 +0000 (01:25 +0000)]
Rearrange ALTER TABLE syntax processing as per my recent proposal: the
grammar allows ALTER TABLE/INDEX/SEQUENCE/VIEW interchangeably for all
subforms of those commands, and then we sort out what's really legal
at execution time. This allows the ALTER SEQUENCE/VIEW reference pages
to fully document all the ALTER forms available for sequences and views
respectively, and eliminates a longstanding cause of confusion for users.
The net effect is that the following forms are allowed that weren't before:
ALTER SEQUENCE OWNER TO
ALTER VIEW ALTER COLUMN SET/DROP DEFAULT
ALTER VIEW OWNER TO
ALTER VIEW SET SCHEMA
(There's no actual functionality gain here, but formerly you had to say
ALTER TABLE instead.)
Interestingly, the grammar tables actually get smaller, probably because
there are fewer special cases to keep track of.
I did not disallow using ALTER TABLE for these operations. Perhaps we
should, but there's a backwards-compatibility issue if we do; in fact
it would break existing pg_dump scripts. I did however tighten up
ALTER SEQUENCE and ALTER VIEW to reject non-sequences and non-views
in the new cases as well as a couple of cases where they didn't before.
The patch doesn't change pg_dump to use the new syntaxes, either.
Tom Lane [Sat, 14 Jun 2008 18:04:34 +0000 (18:04 +0000)]
Refactor the handling of the various DropStmt variants so that when multiple
objects are specified, we drop them all in a single performMultipleDeletions
call. This makes the RESTRICT/CASCADE checks more relaxed: it's not counted
as a cascade if one of the later objects has a dependency on an earlier one.
NOTICE messages about such cases go away, too.
In passing, fix the permissions check for DROP CONVERSION, which for some
reason was never made role-aware, and omitted the namespace-owner exemption
too.
Tom Lane [Fri, 13 Jun 2008 02:59:47 +0000 (02:59 +0000)]
Improve the various elog messages in tuptoaster.c to report which TOAST table
the problem happened in. These are all supposedly can't-happen cases, but
when they do happen it's useful to know where.
Back-patch to 8.3, but not further because the patch doesn't apply cleanly
further back. Given the lack of response to my proposal of this, there
doesn't seem to be enough interest to justify much back-porting effort.
Refactor XLogOpenRelation() and XLogReadBuffer() in preparation for relation
forks. XLogOpenRelation() and the associated light-weight relation cache in
xlogutils.c is gone, and XLogReadBuffer() now takes a RelFileNode as argument,
instead of Relation.
For functions that still need a Relation struct during WAL replay, there's a
new function called CreateFakeRelcacheEntry() that returns a fake entry like
XLogOpenRelation() used to.
Tom Lane [Wed, 11 Jun 2008 21:53:49 +0000 (21:53 +0000)]
Improve reporting of dependencies in DROP to work like the scheme that we
devised for pg_shdepend, namely the individual dependencies are reported as
DETAIL lines rather than coming out as separate NOTICEs. The client-side
report is capped at 100 lines, but the server log always gets a full report.
Fix bug in the WAL recovery code to finish an incomplete split.
CacheInvalidateRelcache() crashes if called in WAL recovery, because the
invalidation infrastructure hasn't been initialized yet.
Neil Conway [Tue, 10 Jun 2008 20:58:19 +0000 (20:58 +0000)]
Editorialization for the text emitted by the "help" psql command.
Basically just reuse the same text that psql emitted as part of
its startup banner in prior versions, and make some whitespace
more consistent with the conventions in other psql command output.
Tom Lane [Mon, 9 Jun 2008 19:34:02 +0000 (19:34 +0000)]
Fix datetime input functions to correctly detect integer overflow when
running on a 64-bit platform ... strtol() will happily return 64-bit
output in that case. Per bug #4231 from Geoff Tolley.
Tom Lane [Sun, 8 Jun 2008 22:41:04 +0000 (22:41 +0000)]
Rewrite DROP's dependency traversal algorithm into an honest two-pass
algorithm, replacing the original intention of a one-pass search, which
had been hacked up over time to be partially two-pass in hopes of handling
various corner cases better. It still wasn't quite there, especially as
regards emitting unwanted NOTICE messages. More importantly, this approach
lets us fix a number of open bugs concerning concurrent DROP scenarios,
because we can take locks during the first pass and avoid traversing to
dependent objects that were just deleted by someone else.
There is more that can be done here, but I'll go ahead and commit the
base patch before working on the options.
Alvaro Herrera [Sun, 8 Jun 2008 22:00:48 +0000 (22:00 +0000)]
Move BufferGetPageSize and BufferGetPage from bufpage.h to bufmgr.h. It is
more logical that way, and also it reduces the amount of unnecessary includes
in bufpage.h, which is widely used.
Zdenek Kotala.
My previous patch to bufpage.h should also have credited him as author, but I
forgot (sorry about that).
Tom Lane [Sun, 8 Jun 2008 21:09:48 +0000 (21:09 +0000)]
ALTER AGGREGATE OWNER seems to have been missed by the last couple of
patches that dealt with object ownership. It wasn't updating pg_shdepend
nor adjusting the aggregate's ACL. In 8.2 and up, fix this permanently
by making it use AlterFunctionOwner_oid. In 8.1, the function code wasn't
factored that way, so just copy and paste.
Tom Lane [Fri, 6 Jun 2008 17:59:29 +0000 (17:59 +0000)]
Fix pg_get_ruledef() so that negative numeric constants are parenthesized.
This is needed because :: casting binds more tightly than minus, so for
example -1::integer is not the same as (-1)::integer, and there are cases
where the difference is important. In particular this caused a failure
in SELECT DISTINCT ... ORDER BY ... where expressions that should have
matched were seen as different by the parser; but I suspect that there
could be other cases where failure to parenthesize leads to subtler
semantic differences in reloaded rules. Per report from Alexandr Popov.
Alvaro Herrera [Thu, 5 Jun 2008 15:47:32 +0000 (15:47 +0000)]
Modify vacuum() to accept a single relation OID instead of a list (which we
always pass as a single element anyway.) In passing, fix an outdated comment.
Tom Lane [Mon, 2 Jun 2008 03:48:00 +0000 (03:48 +0000)]
Fix initdb to reject a relative path for -X (--xlogdir) argument. This
doesn't work, and the real reason why not is it's unclear where the path
is relative to (initdb's CWD, or the data directory?). We could make an
arbitrary decision, but it seems best to make the user be unambiguous.
Per gripe from Devrim.
Tom Lane [Sun, 1 Jun 2008 17:32:48 +0000 (17:32 +0000)]
Refactor SPI_cursor_open/SPI_cursor_open_with_args so that the latter sets
the PARAM_FLAG_CONST flag on the parameters that are passed into the portal,
while the former's behavior is unchanged. This should only affect the case
where the portal is executing an EXPLAIN; it will cause the generated plan to
look more like what would be generated if the portal were actually executing
the command being explained. Per gripe from Pavel.
Tom Lane [Fri, 30 May 2008 00:04:32 +0000 (00:04 +0000)]
Copy refint.so and autoinc.so into the src/test/regress directory during
"make all", and then reference them there during the actual tests. This
makes the handling of these files more parallel to that of regress.so,
and in particular simplifies use of the regression tests outside the
original build tree. The PGDG and Red Hat RPMs have been doing this via
patches for a very long time. Inclusion of the change in core was requested
by Jørgen Austvik of Sun, and I can't see any reason not to.
I attempted to fix the MSVC scripts for this too, but they may need
further tweaking ...
Tom Lane [Thu, 29 May 2008 22:02:44 +0000 (22:02 +0000)]
Tweak libpq to avoid crashing due to incorrect buffer size calculation when
we are on a 64-bit machine (ie, size_t is wider than int) and someone passes
in a query string that approaches or exceeds INT_MAX bytes. Also, just for
paranoia's sake, guard against similar overflows in sizing the input buffer.
The backend will not in the foreseeable future be prepared to send or receive
strings exceeding 1GB, so I didn't take the more invasive step of switching
all the buffer index variables from int to size_t; though someday we might
want to do that.
I have a suspicion that this is not the only such bug in libpq, but this
fix is enough to take care of the crash reported by Francisco Reyes.
Tom Lane [Thu, 29 May 2008 18:46:40 +0000 (18:46 +0000)]
Fix some bugs introduced by the 8.2-era conversion of cube functions to V1
calling convention. cube_inter and cube_distance could attempt to pfree
their input arguments, and cube_dim returned a value from a struct it
might have just pfree'd (which would only really cause a problem in a
debug build, but it's still wrong). Per bug #4208 and additional code
reading.
In HEAD and 8.3, I also made a batch of cosmetic changes to bring these
functions into line with the preferred coding style for V1 functions,
ie declare and fetch all the arguments at the top so readers can easily
see what they are.
Bruce Momjian [Thu, 29 May 2008 02:04:15 +0000 (02:04 +0000)]
Add description to:
* Add deferred trigger queue file
< This item involves dumping large queues into files.
> This item involves dumping large queues into files, or doing some
> kind of join to process all the triggers, or some bulk operation.
Magnus Hagander [Wed, 28 May 2008 09:04:06 +0000 (09:04 +0000)]
Add a field to guc enums to allow hiding of values from display while
still accepting them as input, used to allow alternate syntax for the
same setting.
Tom Lane [Wed, 28 May 2008 00:45:40 +0000 (00:45 +0000)]
Improve GRANT documentation to point out that UPDATE and DELETE typically
require SELECT privilege as well, since you normally need to read existing
column values within such commands. This behavior is according to spec,
but we'd never documented it before. Per gripe from Volkan Yazici.
Tom Lane [Tue, 27 May 2008 22:18:04 +0000 (22:18 +0000)]
Require bind_textdomain_codeset() not just gettext() to enable NLS support.
GNU gettext before 0.10.36 does not have that function, and is generally too
incomplete to be usable.
Magnus Hagander [Tue, 27 May 2008 12:24:42 +0000 (12:24 +0000)]
Explicitly bind gettext() to the UTF8 locale when in use.
This is required on Windows due to the special locale
handling for UTF8 that doesn't change the full environment.
Fixes crash with translated error messages per bugs 4180
and 4196.
Tom Lane [Tue, 27 May 2008 00:13:09 +0000 (00:13 +0000)]
Alter the xxx_pattern_ops opclasses to use the regular equality operator of
the associated datatype as their equality member. This means that these
opclasses can now support plain equality comparisons along with LIKE tests,
thus avoiding the need for an extra index in some applications. This
optimization was not possible when the pattern opclasses were first introduced,
because we didn't insist that text equality meant bitwise equality; but we
do now, so there is no semantic difference between regular and pattern
equality operators.
I removed the name_pattern_ops opclass altogether, since it's really useless:
name's regular comparisons are just strcmp() and are unlikely to become
something different. Instead teach indxpath.c that btree name_ops can be
used for LIKE whether or not the locale is C. This might lead to a useful
speedup in LIKE queries on the system catalogs in non-C locales.
The ~=~ and ~<>~ operators are gone altogether. (It would have been nice to
keep them for backward compatibility's sake, but since the pg_amop structure
doesn't allow multiple equality operators per opclass, there's no way.)
A not-immediately-obvious incompatibility is that the sort order within
bpchar_pattern_ops indexes changes --- it had been identical to plain
strcmp, but is now trailing-blank-insensitive. This will impact
in-place upgrades, if those ever happen.
Tom Lane [Mon, 26 May 2008 18:54:29 +0000 (18:54 +0000)]
Fix an old corner-case bug in set_config_option: push_old_value has to be
called before, not after, calling the assign_hook if any. This is because
push_old_value might fail (due to palloc out-of-memory), and in that case
there would be no stack entry to tell transaction abort to undo the GUC
assignment. Of course the actual assignment to the GUC variable hasn't
happened yet --- but the assign_hook might have altered subsidiary state.
Without a stack entry we won't call it again to make it undo such actions.
So this is necessary to make the world safe for assign_hooks with side
effects. Per a discussion a couple weeks ago with Magnus.
Back-patch to 8.0. 7.x did not have the problem because it did not have
allocatable stacks of GUC values.
Tom Lane [Sun, 25 May 2008 21:51:00 +0000 (21:51 +0000)]
Adjust timestamp regression tests to prevent two low-probability failure
cases. Recent buildfarm experience shows that it is sometimes possible
to execute several SQL commands in less time than the granularity of
Windows' not-very-high-resolution gettimeofday(), leading to a failure
because the tests expect the value of now() to change and it doesn't.
Also, it was recognized some time ago that the same area of the tests
could fail if local midnight passes between the insertion and the checking
of the values for 'yesterday', 'tomorrow', etc. Clean all this up per
ideas from myself and Greg Stark.
There remains a window for failure if the transaction block is entered
exactly at local midnight (so that 'now' and 'today' have the same value),
but that seems low-probability enough to live with.
Since the point of this change is mostly to eliminate buildfarm noise,
back-patch to all versions we are still actively testing.
Remove arbitrary 10MB limit on two-phase state file size. It's not that hard
to go beoynd 10MB, as demonstrated by Gavin Sharry's example of dropping a
schema with ~25000 objects. The really bogus thing about the limit was that
it was enforced when a state file file was read in, not when it was written,
so you would end up with a prepared transaction that you can't commit or
abort, and the only recourse was to shut down the server and remove the file
by hand.
Raise the limit to MaxAllocSize, and enforce it also when a state file is
written. We could've removed the limit altogether, but reading in a file
larger than MaxAllocSize would fail anyway because we read it into a
palloc'd buffer.
Backpatch down to 8.1, where 2PC and this issue was introduced.
Tom Lane [Mon, 19 May 2008 04:14:24 +0000 (04:14 +0000)]
Coercion sanity check in ri_HashCompareOp failed to allow for enums, as per
example from Rod Taylor. On reflection the correct test here is for any
polymorphic type, not specifically ANYARRAY as in the original coding.
Tom Lane [Sun, 18 May 2008 20:13:12 +0000 (20:13 +0000)]
Make another try at using -Wl,--as-needed to suppress linking of unnecessary
shared libraries. We've tried this before and had problems with libreadline
not linking properly on some platforms, but that seems to be a libreadline
bug that may have been fixed by now. In any case, it's early enough in the
8.4 devel cycle that we can afford to have some transient breakage while
we work out any portability problems.
On Darwin, we try -Wl,-dead_strip_dylibs, which seems to be the equivalent
incantation there.
Andrew Dunstan [Sun, 18 May 2008 06:50:08 +0000 (06:50 +0000)]
Remove old kluge put in to allow Windows regression tests to succeed, and now
found to have been made necessary by our skipping tty detection on Windows. Now
that we are doing tty detection on Windows the kluge is unnecessary and wrong.
Tom Lane [Sat, 17 May 2008 23:36:27 +0000 (23:36 +0000)]
Rewrite the warning about non-transaction-safety of TRUNCATE ... RESTART
IDENTITY to be more explicit about the possible hazards. Per gripe from Neil
and subsequent discussion. Eventually we may be able to get rid of this
warning, but for now it had better be there.
Tom Lane [Sat, 17 May 2008 17:52:14 +0000 (17:52 +0000)]
Fix utterly-bogus code for computing row heights. Per crashes on
spoonbill, though one wonders why it didn't misbehave everywhere.
In passing remove some unnecessary modulo calculations.
Tom Lane [Sat, 17 May 2008 17:24:57 +0000 (17:24 +0000)]
Fix a subtle bug exposed by recent wal_sync_method rearrangements.
Formerly, the default value of wal_sync_method was determined inside xlog.c,
but now it is determined inside guc.c. guc.c was reading xlogdefs.h
without having read <fcntl.h>, leading to wrong determination of
DEFAULT_SYNC_METHOD. Obviously xlogdefs.h needs to include <fcntl.h>
for itself to ensure stable results.
Tom Lane [Sat, 17 May 2008 01:20:39 +0000 (01:20 +0000)]
Allow ALTER SEQUENCE START WITH to change the recorded start_value of a
sequence. This seems an obvious extension to the recent patch, and it
makes the code noticeably cleaner and more orthogonal.
Tom Lane [Fri, 16 May 2008 23:36:05 +0000 (23:36 +0000)]
Add a RESTART (without parameter) option to ALTER SEQUENCE, allowing a
sequence to be reset to its original starting value. This requires adding the
original start value to the set of parameters (columns) of a sequence object,
which is a user-visible change with potential compatibility implications;
it also forces initdb.
Also add hopefully-SQL-compatible RESTART/CONTINUE IDENTITY options to
TRUNCATE TABLE. RESTART IDENTITY executes ALTER SEQUENCE RESTART for all
sequences "owned by" any of the truncated relations. CONTINUE IDENTITY is
a no-op option.
Tom Lane [Fri, 16 May 2008 18:34:51 +0000 (18:34 +0000)]
Suppress a possibly-uninitialized-variable warning. (I'm only seeing it
on Apple's gcc and not my other machines, but still it seems worth
getting rid of.)
Magnus Hagander [Fri, 16 May 2008 18:30:53 +0000 (18:30 +0000)]
Implement error checking for pthreads calls in thread-safe mode. They really
should always succeed, but in the likely event of a failure we would
previously fall through *without locking* - the new code will exit(1).
Printing the error message on stderr will not work for all applications, but
it's better than nothing at all - and our API doesn't provide a way to return
the error to the caller.