Various places were testing TRIGGER_FIRED_BEFORE() where what they really
meant was !TRIGGER_FIRED_AFTER(), or vice versa.  This needs to be cleaned
up because there are about to be more than two possible states.

We might want to note this in the 9.1 release notes as something for
trigger authors to double-check.

For consistency's sake I also changed some places that assumed that
TRIGGER_FIRED_FOR_ROW and TRIGGER_FIRED_FOR_STATEMENT are necessarily
mutually exclusive; that's not in immediate danger of breaking, but
it's still sloppier than it should be.

Extracted from Dean Rasheed's patch for triggers on views.  I'm committing
this separately since it's an identifiable separate issue, and is the
only reason for the patch to touch most of these particular files.

This patch resurrects some of the information that could be logged by the
old, now-dead implementation of VACUUM FULL, in particular counts of live
and dead tuples and the time taken for the table rebuild proper.  There's
still no logging about the ensuing index rebuilds, though.

Itagaki Takahiro

Use a macro LogicalTapeReadExact() to encapsulate the error check when
we want to read an exact number of bytes from a "tape".  Per a suggestion
of Takahiro Itagaki.

... or at least, when the planner's cost estimates say it will be faster.

Leonardo Francalanci, reviewed by Itagaki Takahiro and Tom Lane

In particular, we are now more explicit about the fact that you may need
wal_sync_method=fsync_writethrough for crash-safety on some platforms,
including MaxOS X.  There's also now an explicit caution against assuming
that the default setting of wal_sync_method is either crash-safe or best
for performance.

This patch eliminates per-chunk palloc overhead for most small allocations
needed in the representation of an ispell dictionary.  This saves close to
a factor of 2 on the current Czech ispell data.  While it doesn't cover
every last small allocation in the ispell code, we are at the point of
diminishing returns, because about 95% of the allocations are covered
already.

Pavel Stehule, rather heavily revised by Tom

Add explicit initialization and cleanup functions to spell.c, and keep
all working state in the already-existing ISpellDict struct.  This lets us
get rid of a static variable along with some extremely shaky assumptions
about usage of child memory contexts.

This commit is just code beautification and has no impact on functionality
or performance, but it opens the way to a less-grotty implementation of
Pavel's memory-saving hack, which will follow shortly.

Actual behaviour did not match documented behaviour and we have agreed
that it should be the docs that change.

Spotted by Bernd Helmle

Per Tatsuhito Kasahara.

In versions 8.2 and up, the grammar allows attaching ORDER BY, LIMIT,
FOR UPDATE, or WITH to VALUES, and hence to INSERT ... VALUES.  But the
special-case code for VALUES in transformInsertStmt() wasn't expecting any
of those, and just ignored them, leading to unexpected results.  Rather
than complicate the special-case path, just ensure that the presence of any
of those clauses makes us treat the query as if it had a general SELECT.
Per report from Hitoshi Harada.

Many compilers don't complain about this, but some do, and it's certainly
wrong.  Back-patch to 8.4 where the error was introduced.

Mark Kirkwood

Actually making this case work, if the column is used in the trigger's
WHEN condition, will take some new code that probably isn't appropriate
to back-patch.  For now, just throw a FEATURE_NOT_SUPPORTED error rather
than allowing control to reach the "unexpected object" case.  Per bug #5688
from Daniel Grace.  Back-patch to 9.0 where the possibility of such a
dependency was introduced.

There are numerous methods by which a Perl or Tcl function can subvert
the behavior of another such function executed later; for example, by
redefining standard functions or operators called by the target function.
If the target function is SECURITY DEFINER, or is called by such a
function, this means that any ordinary SQL user with Perl or Tcl language
usage rights can do essentially anything with the privileges of the target
function's owner.

To close this security hole, create a separate Perl or Tcl interpreter for
each SQL userid under which plperl or pltcl functions are executed within
a session.  However, all plperlu or pltclu functions run within a session
still share a single interpreter, since they all execute at the trust
level of a database superuser anyway.

Note: this change results in a functionality loss when libperl has been
built without the "multiplicity" option: it's no longer possible to call
plperl functions under different userids in one session, since such a
libperl can't support multiple interpreters in one process.  However, such
a libperl already failed to support concurrent use of plperl and plperlu,
so it's likely that few people use such versions with Postgres.

Security: CVE-2010-3433

Erik Rijkers

8.0.26, and 7.4.30.

Satoshi Nagayasu

from Tom.

rather than atol().

Per report from Brian Hirt

resource leak.  Of course, any such failure aborts pg_upgrade, but might
as well be clean about it.

Per patch from Grzegorz Ja?kiewicz.

Need a "return false" to prevent tests from continuing after we've moved
the "query" pointer.  As it stood, it'd accept "DROP DISCARD ALL" as a
match.

directory.

Per report from Harald Armin Massa.

Per report from Harald Armin Massa

The point of a PlaceHolderVar is to allow a non-strict expression to be
evaluated below an outer join, after which its value bubbles up like a Var
and can be forced to NULL when the outer join's semantics require that.
However, there was a serious design oversight in that, namely that we
didn't ensure that there was actually a correct place in the plan tree
to evaluate the placeholder :-(.  It may be necessary to delay evaluation
of an outer join to ensure that a placeholder that should be evaluated
below the join can be evaluated there.  Per recent bug report from Kirill
Simonov.

Back-patch to 8.4 where the PlaceHolderVar mechanism was introduced.

title, per suggestion from Ian Barwick.

Report by Shigeru Hanada.

We allowes DISCARD PLANS and TEMP in a transaction.

Backpatch to 8.3.

Reported by Sergey Burladyan.

This is intended as infrastructure to support integration with label-based
mandatory access control systems such as SE-Linux. Further changes (mostly
hooks) will be needed, but this is a big chunk of it.

KaiGai Kohei and Robert Haas

Devrim GÜNDÜZ

1. Resurrect the behavior where old commits on master will have Branch:
labels for branches sprouted after the commit was made.  I'm still
dubious about this mode, but if you want it, say --post-date or -p.

2. Annotate the Branch: labels with the release or branch in which the
commit was publicly released.  For example, on a release branch you could
see
Branch: REL8_3_STABLE Release: REL8_3_2 [92c3a8004] 2008-03-29 00:15:37 +0000
showing that the fix was released in 8.3.2.  Commits on master will
usually instead have notes like
Branch: master Release: REL8_4_BR [6fc9d427] 2008-03-29 00:15:28 +0000
showing that this commit is ancestral to release branches 8.4 and later.
If no Release: marker appears, the commit hasn't yet made it into any
release.

3. Add support for release branches older than 7.4.

4. The implementation is improved by running git log on each branch only
back to where the branch sprouts from master.  This saves a good deal
of time (about 50% of the runtime when generating the complete history).
We generate the post-date-mode tags via a direct understanding that
they should be applied to master commits made before the branch sprouted,
rather than backing into them via matching (which isn't any too
reliable when people used identical log messages for successive commits).

Like with tables, this also requires allowing the existence of
composite types with zero attributes.

reviewed by KaiGai Kohei

1. Don't assume there's only one candidate match; check them all and use the
one with the closest timestamp.  Avoids funny output when someone makes
several successive commits with the same log message, as certain people
have been known to do.

2. When the same commit (with the same SHA1) is reachable from multiple
branch tips, don't report it for all the branches; instead report it only
for the first such branch.  Given our development practices, this case
arises only for commits that occurred before a given branch split off from
master.  The original coding blamed old commits on *all* the branches,
which isn't terribly useful; the new coding blames such a commit only on
master.

1. Don't forget the last (oldest) commit on the oldest branch.

2. When considering which commit to print next, if two alternatives have
the same "distortion" score (which is actually the normal case, since
generally the "distortion" is 0), then choose the later timestamp to
print first.  I don't know where Robert got the idea to ignore timestamps
and sort by branch age, but it wasn't a good idea: the resulting ordering
of commits was just plain bizarre anywhere that some branches had many
fewer commits than others, which is the typical situation for us.