The text search functions that involve parsing raw text into lexemes are
remarkably CPU-intensive, so estimating them at the same cost as most other
built-in functions seems like a mistake; moreover, doing so turns out to
discourage the optimizer from using functional indexes on these functions.
After some debate, we've agreed to raise procost from 1 to 100 for
to_tsvector(), plainto_tsvector(), to_tsquery(), ts_headline(),
ts_match_tt(), and ts_match_tq(), which are all the text search functions
that parse raw text.

Also increase procost for the 2-argument form of ts_rewrite()
(tsquery_rewrite_query); while this function doesn't do text parsing,
it does execute a user-supplied SQL query, so its previous procost of 1 is
clearly a drastic underestimate.  It seems reasonable to assign it the same
cost we assign to PL functions by default, so 100 is the number here too.

I did not bother bumping catversion for this change, since it does not
break catalog compatibility with the server executable nor result in
any regression test changes.

Per complaint from Andrew Gierth and subsequent discussion.

Otherwise, if there's another crash, some writes from after the first
crash might make it to disk while writes from before the crash fail
to make it to disk.  This could lead to data corruption.

Back-patch to all supported versions.

Abhijit Menon-Sen, reviewed by Andres Freund and slightly revised
by me.

prev_regbuf was never set, and therefore the same-rel flag was never set on
WAL records.

Report and fix by Zhanq Zq

The first bug is not releasing a tupdesc when doing an early return out
of the function. The second bug is a logic error in choosing when to do
an early return if given an empty jsonb object.

Bug reports from Pavel Stehule and Tom Lane respectively.

Backpatch to 9.4 where these were introduced.

Commit ef3f9e64 suppressed one cause of warnings here, but
recent clang on OS X is still unhappy because we're passing a "long"
to abs().  The fact that tm_gmtoff is declared as long is no doubt a
hangover from days when int might be only 16 bits; but Postgres has
never been able to run on such machines, so we can just cast it to int
with no worries.  For consistency, also cast to int in the other
uses of tm_gmtoff in this stanza.

Note: this code is still broken on machines that don't follow C99
integer-division-truncates-towards-zero rules.  Given the lack of
complaints about it, I don't feel a large desire to complicate things
enough to cope with the pre-C99 rules.

When altering the deferredness state of a foreign key constraint, we
correctly updated the catalogs and then invalidated the relcache state for
the target relation ... but that's not the only relation with relevant
triggers.  Must invalidate the other table as well, or the state change
fails to take effect promptly for operations triggered on the other table.
Per bug #13224 from Christian Ullrich.

In passing, reorganize regression test case for this feature so that it
isn't randomly injected into the middle of an unrelated test sequence.

Oversight in commit f177cbfe.  Back-patch
to 9.4 where the faulty code was added.

These modules were all missing essential Windows scaffolding, including
resources files and descriptions, and links to the relevant library
import files. This latter item means that the modules can't be built
with pgxs on Windows, as we don't install the import files. If we ever
decide to install them this restriction could probably be removed.

Also, as with plperl we need to make sure that perl's CORE directory is
last on the include list, as on Windows it appears to contain some
headers with names that clash with names of some headers we include.

By converting to using forward slashes at configure time we avoid
having to repeat the logic anywhere that this is needed, such as
in transforms modules for plpython.

This eliminates many seconds of test duration and the cause to invoke
"rm -rf", which is typically unavailable on Windows.

Michael Paquier and Noah Misch

Commit c67a86f7 caught most of these,
but this negative test escaped notice.  The test did pass, for the wrong
reason, under affected configurations.

Michael Paquier

coerce_type() has local variables named targetTypeId, baseTypeId, and
targetType.  targetType has been the Type structure for baseTypeId, so
rename it to baseType.

PORTNAME isn't set until the global makefiles have been included.

Apparently, looking for an appropriately named file doesn't work on some
older versions, so put the back the explicit platform detection.

Combine the two places that set CPPFLAGS into one.  Also, some settings
should be restricted to Windows only.  More precisely, -Wno-comment is
a GCC-only option, but Windows in a makefile implies GCC at the moment.

Also, since -Wno-comment is more properly a preprocessor option, move it
to CPPFLAGS to simplify things a bit.

For building PL/Perl, PL/Python, and PL/Tcl, we need a shared library of
libperl, libpython, and libtcl, respectively.  Previously, this was
checked in the makefiles, skipping the PL build with a warning if no
shared library was available.  Now this is checked in configure, with an
error if no shared library is available.

The previous situation arose because in the olden days, the configure
options --with-perl, --with-python, and --with-tcl controlled whether
frontend interfaces for those languages would be built.  The procedural
languages were added later, and shared libraries were often not
available in the beginning.  So it was decided skip the builds of the
procedural languages in those cases.  The frontend interfaces have since
been removed from the tree, and shared libraries are now available most
of the time, so that setup makes much less sense now.

Also, the new setup allows contrib modules and pgxs users to rely on the
respective PLs being available based on configure flags.

This involves moving perl's CORE library to the end of the include list,
and adding other compilation settings that plperl uses. This won't
completely fix the breakage currently being seen by gcc builds on
Windows, but it will let the build get further, and should be wholly
benign, if not beneficial, on *nix.

pg_dump turns tables into views using a method that was not setting
pg_class.relreplident properly.

Patch by Marko Tiikkaja

Backpatch through 9.4

Tom Lane pointed out that this wasn't done, and asked whether that was
intentional.  Subsequent discussion was in favor of making the change,
so here we go.

Foreign data wrappers can use this capability for so-called "join
pushdown"; that is, instead of executing two separate foreign scans
and then joining the results locally, they can generate a path which
performs the join on the remote server and then is scanned locally.
This commit does not extend postgres_fdw to take advantage of this
capability; it just provides the infrastructure.

Custom scan providers can use this in a similar way.  Previously,
it was only possible for a custom scan provider to scan a single
relation.  Now, it can scan an entire join tree, provided of course
that it knows how to produce the same results that the join would
have produced if executed normally.

KaiGai Kohei, reviewed by Shigeru Hanada, Ashutosh Bapat, and me.

Michael Paquier and myself.

ParseCommitRecord() accessed xl_xact_origin directly. But the chunks in
the commit record's data only have 4 byte alignment, whereas
xl_xact_origin's members require 8 byte alignment on some
platforms. Update comments to make not of that and copy the record to
stack local storage before reading.

With help from Stefan Kaltenbrunner in pinning down the buildfarm and
verifying the fix.

pg_rewind looks at the control file to determine the server's timeline. If
the standby performs a "fast promotion", the timeline ID in the control
file is not updated until the next checkpoint. The startup process requests
a checkpoint immediately after promotion, so this is unlikely to be an
issue in the real world, but the regression suite ran pg_rewind so quickly
after promotion that the checkpoint had not yet completed.

Reported by Stephen Frost

In commit 31eae602, some documents were not updated to show the new
capability; fix that.  Also, the error message you get when CURRENT_USER
and SESSION_USER are used in a context that doesn't accept them could be
clearer about it being a problem only in those contexts; so add the
word "here".

Author: Kyotaro HORIGUCHI

His patch submission also included changes to GRANT/REVOKE, but those
seemed more controversial, so I left them out.  We can reconsider these
changes later.

This does four basic things.  First, it provides convenience routines
to coordinate the startup and shutdown of parallel workers.  Second,
it synchronizes various pieces of state (e.g. GUCs, combo CID
mappings, transaction snapshot) from the parallel group leader to the
worker processes.  Third, it prohibits various operations that would
result in unsafe changes to that state while parallelism is active.
Finally, it propagates events that would result in an ErrorResponse,
NoticeResponse, or NotifyResponse message being sent to the client
from the parallel workers back to the master, from which they can then
be sent on to the client.

Robert Haas, Amit Kapila, Noah Misch, Rushabh Lathia, Jeevan Chalke.
Suggestions and review from Andres Freund, Heikki Linnakangas, Noah
Misch, Simon Riggs, Euler Taveira, and Jim Nasby.

We need to create the pg_multixact/offsets file deleted by pg_upgrade
much earlier than we originally were: it was in TrimMultiXact(), which
runs after we exit recovery, but it actually needs to run earlier than
the first call to SetMultiXactIdLimit (before recovery), because that
routine already wants to read the first offset segment.

Per pg_upgrade trouble report from Jeff Janes.

While at it, silence a compiler warning about a pointless assert that an
unsigned variable was being tested non-negative.  This was a signed
constant in Thomas Munro's patch which I changed to unsigned before
commit.  Pointed out by Andres Freund.

Amit Langote

The "check" target no longer needs to depend on "all", because it now
runs "install" directly, which in turn depends on "all".  Doing both
will cause problems with parallel make, because two builds will run next
to each other.

Also remove the redirection of the temp-install output into a log file.
This was appropriate when this was done from within pg_regress, but now
it's just a regular make run, and especially with the above changes this
will now take the place of running the "all" target before the test
suites.

problem report by Jeff Janes, patch in part by Michael Paquier

We can't rely on UINT16_MAX being present, which is why we introduced
PG_UINT16_MAX...

Buildfarm animal bowerbird via Andrew Gierth.

This file isn't entirely consistent about whether "on" and "off"
should be marked up with <literal>, but it doesn't make much sense
to be inconsistent within a single sentence.

Etsuro Fujita

When this code was written, catalog scans were normally performed using
SnapshotNow, making special handling necessary here.  Now, however, all
catalog scans use MVCC snapshots, so we can change these cases to look
more like what we do for catalog scans elsewhere in the code.

Per discussion with Tom Lane and a reminder from Bruce Momjian.

Currently regression tests for python 3 are disabled on MSVC, and these
tests fail with python 3, too, so we have some work to do to enable
both. Meanwhile, all the buildfarm hosts seem to be building with python
2 anyway, so this at least gets us some coverage.

Original patch from Michael Paquier, significantly modified by me.

When implementing a replication solution ontop of logical decoding, two
related problems exist:
* How to safely keep track of replication progress
* How to change replication behavior, based on the origin of a row;
  e.g. to avoid loops in bi-directional replication setups

The solution to these problems, as implemented here, consist out of
three parts:

1) 'replication origins', which identify nodes in a replication setup.
2) 'replication progress tracking', which remembers, for each
   replication origin, how far replay has progressed in a efficient and
   crash safe manner.
3) The ability to filter out changes performed on the behest of a
   replication origin during logical decoding; this allows complex
   replication topologies. E.g. by filtering all replayed changes out.

Most of this could also be implemented in "userspace", e.g. by inserting
additional rows contain origin information, but that ends up being much
less efficient and more complicated.  We don't want to require various
replication solutions to reimplement logic for this independently. The
infrastructure is intended to be generic enough to be reusable.

This infrastructure also replaces the 'nodeid' infrastructure of commit
timestamps. It is intended to provide all the former capabilities,
except that there's only 2^16 different origins; but now they integrate
with logical decoding. Additionally more functionality is accessible via
SQL.  Since the commit timestamp infrastructure has also been introduced
in 9.5 (commit 73c986ad) changing the API is not a problem.

For now the number of origins for which the replication progress can be
tracked simultaneously is determined by the max_replication_slots
GUC. That GUC is not a perfect match to configure this, but there
doesn't seem to be sufficient reason to introduce a separate new one.

Bumps both catversion and wal page magic.

Author: Andres Freund, with contributions from Petr Jelinek and Craig Ringer
Reviewed-By: Heikki Linnakangas, Petr Jelinek, Robert Haas, Steve Singer
Discussion: 20150216002155.GI15326@awork2.anarazel.de,
    20140923182422.GA15776@alap3.anarazel.de,
    20131114172632.GE7522@alap2.anarazel.de

Etsuro Fujita

Previously both hours and minutes displayed as negative.

Report by David Pozsar

Patch by Craig Ringer

This avoids problems with double-slash-specified paths.

Patch by Ian Barwick