Backpatch through 9.4

Backpatch to 9.4

By using curly braces, the template had specified that one of
"NOT DEFERRABLE", "INITIALLY IMMEDIATE", or "INITIALLY DEFERRED"
was required on any CREATE TRIGGER statement, which is not
accurate.  Change to square brackets makes that optional.

Backpatch to 9.1, where the error was introduced.

The catcache code is effectively assuming this already, so let's insist
that the catalog and index are actually declared that way.

Having done that, the comments in indexing.h about non-unique indexes
not being used for catcaches are completely redundant not just mostly so;
and we didn't have such a comment for every such index anyway.  So let's
get rid of them.

Per discussion of whether we should identify primary keys for catalogs.
We might or might not take that further step, but this change in itself
will allow quicker detection of misdeclared catcaches, so it seems worth
doing in any case.

dblink uses a short-lived data conversion memory context. However it
was not deleted when no longer needed, leading to a noticeable memory
leak under some circumstances. Plug the hole, along with minor
refactoring. Backpatch to 9.2 where the leak was introduced.

Report and initial patch by MauMau. Reviewed/modified slightly by
Tom Lane and me.

Since fdf9e211 lazy_vacuum_page() rechecks the all-visible status
of pages in the second pass over the heap. It does so inside a
critical section, but both visibilitymap_test() and
heap_page_is_all_visible() perform operations that should not happen
inside one. The former potentially performs IO and both potentially do
memory allocations.

To fix, simply move all the all-visible handling outside the critical
section. Doing so means that the PD_ALL_VISIBLE on the page won't be
included in the full page image of the HEAP2_CLEAN record anymore. But
that's fine, the flag will be set by the HEAP2_VISIBLE logged later.

Backpatch to 9.3 where the problem was introduced. The bug only came
to light due to the assertion added in 4a170ee9 and isn't likely to
cause problems in production scenarios. The worst outcome is a
avoidable PANIC restart.

This also gets rid of the difference in the order of operations
between master and standby mentioned in 2a8e1ac5.

Per reports from David Leverton and Keith Fiske in bug #10533.

The existance of the assert_enabled variable (backing the
debug_assertions GUC) reduced the amount of knowledge some static code
checkers (like coverity and various compilers) could infer from the
existance of the assertion. That could have been solved by optionally
removing the assertion_enabled variable from the Assert() et al macros
at compile time when some special macro is defined, but the resulting
complication doesn't seem to be worth the gain from having
debug_assertions. Recompiling is fast enough.

The debug_assertions GUC is still available, but readonly, as it's
useful when diagnosing problems. The commandline/client startup option
-A, which previously also allowed to enable/disable assertions, has
been removed as it doesn't serve a purpose anymore.

While at it, reduce code duplication in bufmgr.c and localbuf.c
assertions checking for spurious buffer pins. That code had to be
reindented anyway to cope with the assert_enabled removal.

ExecMakeTableFunctionResult evaluated the arguments for a function-in-FROM
in the query-lifespan memory context.  This is insignificant in simple
cases where the function relation is scanned only once; but if the function
is in a sub-SELECT or is on the inside of a nested loop, any memory
consumed during argument evaluation can add up quickly.  (The potential for
trouble here had been foreseen long ago, per existing comments; but we'd
not previously seen a complaint from the field about it.)  To fix, create
an additional temporary context just for this purpose.

Per an example from MauMau.  Back-patch to all active branches.

Most of the necessary quoting was in place; this catches the exceptions.

Give passwords to each user created in support of an ECPG connection
test case.  Use SET SESSION AUTHORIZATION, not a fresh connection, to
reduce privileges during a dblink test case.

To test against such a server, both the "make installcheck-world"
environment and the postmaster environment must provide the default
user's password; $PGPASSFILE is the principal way to do so.  (The
postmaster environment needs it for dblink and postgres_fdw tests.)

Haribabu Kommi, somewhat rewritten by me

Commit ea9df812 failed to include
NUM_BUFFER_PARTITIONS in this offset, resulting in a bad offset.
Ultimately this threw off NUM_FIXED_LWLOCKS which is based on
earlier offsets, leading to memory allocation problems.  It seems
likely to have also caused increased LWLOCK contention when
serializable transactions were used, because lightweight locks used
for that overlapped others.

Reported by Amit Kapila with analysis and fix.
Backpatch to 9.4, where the bug was introduced.

data_directory could be set both in postgresql.conf and postgresql.auto.conf so far.
This could cause some problematic situations like circular definition. To avoid such
situations, this commit forbids a user to set data_directory in postgresql.auto.conf.

Backpatch this to 9.4 where ALTER SYSTEM command was introduced.

Amit Kapila, reviewed by Abhijit Menon-Sen, with minor adjustments by me.

Arrange for postmaster child processes to respond to two environment
variables, PG_OOM_ADJUST_FILE and PG_OOM_ADJUST_VALUE, to determine whether
they reset their OOM score adjustments and if so to what.  This is superior
to the previous design involving #ifdef's in several ways.  The behavior is
now available in a default build, and both ends of the adjustment --- the
original adjustment of the postmaster's level and the subsequent
readjustment by child processes --- can now be controlled in one place,
namely the postmaster launch script.  So it's no longer necessary for the
launch script to act on faith that the server was compiled with the
appropriate options.  In addition, if someone wants to use an OOM score
other than zero for the child processes, that doesn't take a recompile
anymore; and we no longer have to cater separately to the two different
historical kernel APIs for this adjustment.

Gurjeet Singh, somewhat revised by me

The check was confusing and is a condition that should never in fact
happen.

Per gripe from Dmitry Dolgov.

Seems to have been introduced in 1a3458b6.

This SQL-standard feature allows a sub-SELECT yielding multiple columns
(but only one row) to be used to compute the new values of several columns
to be updated.  While the same results can be had with an independent
sub-SELECT per column, such a workaround can require a great deal of
duplicated computation.

The standard actually says that the source for a multi-column assignment
could be any row-valued expression.  The implementation used here is
tightly tied to our existing sub-SELECT support and can't handle other
cases; the Bison grammar would have some issues with them too.  However,
I don't feel too bad about this since other cases can be converted into
sub-SELECTs.  For instance, "SET (a,b,c) = row_valued_function(x)" could
be written "SET (a,b,c) = (SELECT * FROM row_valued_function(x))".

Catch up with commit b8cc8f94's
introduction of the HAVE_UUID_OSSP symbol to the principal build
process.  Back-patch to 9.4, where that commit appeared.

Report by Peter Geoghegan

Oops.

Since most of the system thinks AND and OR are N-argument expressions
anyway, let's have the grammar generate a representation of that form when
dealing with input like "x AND y AND z AND ...", rather than generating
a deeply-nested binary tree that just has to be flattened later by the
planner.  This avoids stack overflow in parse analysis when dealing with
queries having more than a few thousand such clauses; and in any case it
removes some rather unsightly inconsistencies, since some parts of parse
analysis were generating N-argument ANDs/ORs already.

It's still possible to get a stack overflow with weirdly parenthesized
input, such as "x AND (y AND (z AND ( ... )))", but such cases are not
mainstream usage.  The maximum depth of parenthesization is already
limited by Bison's stack in such cases, anyway, so that the limit is
probably fairly platform-independent.

Patch originally by Gurjeet Singh, heavily revised by me

This prevents several compiler warnings on Windows.

Any OS user able to access the socket can connect as the bootstrap
superuser and proceed to execute arbitrary code as the OS user running
the test.  Protect against that by placing the socket in a temporary,
mode-0700 subdirectory of /tmp.  The pg_regress-based test suites and
the pg_upgrade test suite were vulnerable; the $(prove_check)-based test
suites were already secure.  Back-patch to 8.4 (all supported versions).
The hazard remains wherever the temporary cluster accepts TCP
connections, notably on Windows.

As a convenient side effect, this lets testing proceed smoothly in
builds that override DEFAULT_PGSOCKET_DIR.  Popular non-default values
like /var/run/postgresql are often unwritable to the build user.

Security: CVE-2014-0067

This function is pervasive on free software operating systems; import
NetBSD's implementation.  Back-patch to 8.4, like the commit that will
harness it.

Just feels more natural, and is more consistent with rm_redo.

Per buildfarm member prairiedog.  Back-patch to 9.4, where the test was
introduced.

Reviewed by Tom Lane.

Back-patch to 9.4.

Back-patch to 9.4, where .dir-locals.el was introduced.

We have for a long time been able to prove implications and refutations
between clauses structured like "expr op const" with the same subexpression
and btree-related operators; for example that "x < 4" implies "x <= 5".
The implication machinery is needed to detect usability of partial indexes,
and the refutation machinery is needed to implement constraint exclusion.

This patch extends that machinery to make proofs for operator expressions
involving the same two immutable-but-not-necessarily-just-Const input
expressions, ie does "expr1 op1 expr2" prove or refute "expr1 op2 expr2" or
"expr2 op2 expr1"?  An important example is that we can now prove "x = y"
given "y = x", which formerly the code could not deduce unless x or y was a
constant.  We can make use of the system's knowledge of operator commutator
and negator pairs, and can also make use of btree opclass relationships,
for example "x < y" implies "x <= y" and refutes "x > y" (notice that
neither of these could be proven just from commutator or negator links).

Inspired by a gripe from Brian Dunavant.  This seems more like a new
feature than a bug fix, though, so no back-patch.

Prior to 9.0, pg_dump handled comments on large objects by dumping a bunch
of COMMENT commands into a single BLOB COMMENTS archive object.  With
sufficiently many such comments, some of the commands would likely get
split across bufferloads when restoring, causing failures in
direct-to-database restores (though no problem would be evident in text
output).  This is the same type of issue we have with table data dumped as
INSERT commands, and it can be fixed in the same way, by using a mini SQL
lexer to figure out where the command boundaries are.  Fortunately, the
COMMENT commands are no more complex to lex than INSERTs, so we can just
re-use the existing lexer for INSERTs.

Per bug #10611 from Jacek Zalewski.  Back-patch to all active branches.

We should report the errno when we get a failure from functions like
BufFileWrite.  "ERROR: write failed" is unreasonably taciturn for a
case that's well within the realm of possibility; I've seen it a
couple times in the buildfarm recently, in situations that were
probably out-of-disk-space, but it'd be good to see the errno
to confirm it.

I think this code was originally written without assuming that
the buffile.c functions would return useful errno; but most other
callers *are* assuming that, and a quick look at the buffile code
gives no reason to suppose otherwise.

Also, a couple of the old messages were phrased on the assumption
that a short read might indicate a logic bug in tuplestore itself;
but that code's pretty well tested by now, so a filesystem-level
problem seems much more likely.

Since we commonly test pg_dump/pg_restore by seeing whether they can dump
and restore the regression test database, it behooves us to include some
large objects in that test scenario.

I tried to include a comment on one of these large objects to improve
the test scenario further ... but it turns out that pg_upgrade fails to
preserve comments on large objects, and its regression test notices
the discrepancy.  So uncommenting that COMMENT is a TODO for later.

I thought I could get away with hardcoded int4 here, but the buildfarm
says differently.

Robert Frost is no longer with us, but his copyrights still are, so
let's stop using "Stopping by Woods on a Snowy Evening" as test data
before somebody decides to sue us.  Wordsworth is more safely dead.

Memorialize the expected output of the query that libpq has been using for
many years to get the OIDs of large-object support functions.  Although
we really ought to change the way libpq does this, we must expect that
this query will remain in use in the field for the foreseeable future,
so until we're ready to break compatibility with old libpq versions
we'd better check the results stay the same.  See the recent lo_create()
fiasco.

The previous naming broke the query that libpq's lo_initialize() uses
to collect the OIDs of the server-side functions it requires, because
that query effectively assumes that there is only one function named
lo_create in the pg_catalog schema (and likewise only one lo_open, etc).

While we should certainly make libpq more robust about this, the naive
query will remain in use in the field for the foreseeable future, so it
seems the only workable choice is to use a different name for the new
function.  lo_from_bytea() won a small straw poll.

Back-patch into 9.4 where the new function was introduced.

If a sub-select-in-FROM gets flattened into the upper query, then we
naturally get rid of any output columns that are defined in the sub-select
text but not actually used in the upper query.  However, this doesn't
happen when it's not possible to flatten the subquery, for example because
it contains GROUP BY, LIMIT, etc.  Allowing the subquery to compute useless
output columns is often fairly harmless, but sometimes it has significant
performance cost: the unused output might be an expensive expression,
or it might be a Var from a relation that we could remove entirely (via
the join-removal logic) if only we realized that we didn't really need
that Var.  Situations like this are common when expanding views, so it
seems worth taking the trouble to detect and remove unused outputs.

Because the upper query's Var numbering for subquery references depends on
positions in the subquery targetlist, we don't want to renumber the items
we leave behind.  Instead, we can implement "removal" by replacing the
unwanted expressions with simple NULL constants.  This wastes a few cycles
at runtime, but not enough to justify more work in the planner.

Change the order of checks in similar functions to be the same; remove
a parameter that's not needed anymore; rename a memory context and
expand a couple of comments.

Per review comments from Amit Kapila