This seems to offer significantly better search performance than the
existing GiST opclass for inet/cidr, at least on data with a wide mix
of network mask lengths.  (That may suggest that the data splitting
heuristics in the GiST opclass could be improved.)

Emre Hasegeli, with mostly-cosmetic adjustments by me

Discussion: <CAE2gYzxtth9qatW_OAqdOjykS0bxq7AYHLuyAQLPgT7H9ZU0Cw@mail.gmail.com>

If you have previously pinned a segment and decide that you don't
actually want to keep it around until shutdown, this new API lets you
remove the pin.  This is pretty trivial except on Windows, where it
requires closing the duplicate handle that was used to implement the
pin.

Thomas Munro and Amit Kapila, reviewed by Amit Kapila and by me.

Kyotaro Horiguchi

Discussion: dcc4113d-1eda-4f60-d1c5-f50eee160bad@gmail.com

Author: Alexander Law <exclusion@gmail.com>

Backpatch-through: 9.6

Previously, the spgSplitTuple action could only create a new upper tuple
containing a single labeled node.  This made it useless for opclasses
that prefer to work with fixed sets of nodes (labeled or otherwise),
which meant that restrictive prefixes could not be used with such
node definitions.  Change the output field set for the choose() method
to allow it to specify any valid node set for the new upper tuple,
and to specify which of these nodes to place the modified lower tuple in.

In addition to its primary use for fixed node sets, this feature could
allow existing opclasses that use variable node sets to skip a separate
spgAddNode action when splitting a tuple, by setting up the node needed
for the incoming value as part of the spgSplitTuple action.  However, care
would have to be taken to add the extra node only when it would not make
the tuple bigger than before.  (spgAddNode can enlarge the tuple,
spgSplitTuple can't.)

This is a prerequisite for an upcoming SP-GiST inet opclass, but is
being committed separately to increase the visibility of the API change.

In passing, improve the documentation about the traverse-values feature
that was added by commit ccd6eb49.

Emre Hasegeli, with cosmetic adjustments and documentation rework by me

Discussion: <CAE2gYzxtth9qatW_OAqdOjykS0bxq7AYHLuyAQLPgT7H9ZU0Cw@mail.gmail.com>

Add a variant of txid_current() that returns NULL if no transaction ID
is assigned.  This version can be used even on a standby server,
although it will always return NULL since no transaction IDs can be
assigned during recovery.

Craig Ringer, per suggestion from Jim Nasby.  Reviewed by Petr Jelinek
and by me.

Erik Rijkers

Merge several copies of "copy an inet value and adjust the mask length"
code to create a single, conveniently C-callable function.  This function
is exported for future use by inet SPGiST support, but it's good cleanup
anyway since we had three slightly-different-for-no-good-reason copies.

(Extracted from a larger patch, to separate new code from refactoring
of old code)

Emre Hasegeli

Reported-by: Jeff Janes

Backpatch-through: 9.6

Due to an error in the abbreviated key abort logic, the most recently
processed SortTuple could be incorrectly marked NULL, resulting in an
incorrect final sort order.

In the worst case, this could result in a corrupt btree index, which
would need to be rebuild using REINDEX.  However, abbrevation doesn't
abort very often, not all data types use it, and only one tuple would
end up in the wrong place, so the practical impact of this mistake may
be somewhat limited.

Report and patch by Peter Geoghegan.

Dimitry Ivanov spotted a typo, and I added a bit of wordsmithing.

Remove the plpgsql wrapping that hides the context.  So now the test
will fail if the work doesn't actually happen in a parallel worker.  Run
the test in its own test group to ensure it won't run out of resources
for that.

More fallout from commit a00c5831.  Pointed out by Michael Paquier.

It is redundant with appendConnStrVal(), which became an extern function
in commit 41f18f02.  This changes the
handling of out-of-memory and of certain inputs for which quoting is
optional, but pg_basebackup has no need for unusual treatment thereof.

Fallout from commit a00c5831.  Per buildfarm.

The original coding here was not nearly careful enough about quoting
special characters, and it didn't get corner cases right for constructing
the pg_ctl path either.  Use join_path_components() and appendShellString()
to do it honestly, so that the string will more likely work if blindly
copied-and-pasted.

While at it, teach appendShellString() not to quote strings that clearly
don't need it, so that the output from initdb doesn't become uglier than
it was before in typical cases where quoting is not needed.

Ryan Murphy, reviewed by Michael Paquier and myself

Discussion: <CAHeEsBeAe1FeBypT3E8R1ZVZU0e8xv3A-7BHg6bEOi=jZny2Uw@mail.gmail.com>

This might have been too much of a foot-gun before 9.6, but with the
new commands-end-at-semicolons parsing rule, the only way to get an
empty query into a script is to explicitly write an extra ";".
So we may as well allow the case.

Fabien Coelho

Patch: <alpine.DEB.2.20.1607090922170.3412@sto>

In particular, left join to pg_authid so that rows in pg_stat_activity
don't disappear if the session's owning user has been dropped.
Also convert a few joins to pg_database to left joins, in the same spirit,
though that case might be harder to hit.  We were doing this in other
views already, so it was a bit inconsistent that these views didn't.

Oskari Saarenmaa, with some further tweaking by me

Discussion: <56E87CD8.60007@ohmu.fi>

Obvious brain fade in set_rel_consider_parallel().  Noticed it while
adjusting the adjacent RTE_FUNCTION case.

In 9.6, also make the code look more like what I just did in HEAD
by removing the unnecessary function_rte_parallel_ok subroutine
(it does nothing that expression_tree_walker wouldn't do).

We need to scan the whole parse tree for parallel-unsafe functions.
If there are none, we'll later need to determine whether particular
subtrees contain any parallel-restricted functions.  The previous coding
retained no knowledge from the first scan, even though this is very
wasteful in the common case where the query contains only parallel-safe
functions.  We can bypass all of the later scans by remembering that fact.
This provides a small but measurable speed improvement when the case
applies, and shouldn't cost anything when it doesn't.

Patch by me, reviewed by Robert Haas

Discussion: <3740.1471538387@sss.pgh.pa.us>

Clobbering errno during cleanup after an error is an oft-repeated, easy
to make mistake.  Deal with it here as everywhere else, by saving it
aside and restoring after cleanup, before ereport'ing.

In passing, add a missing errcode declaration in another ereport() call
in the same file, which I noticed while skimming the file looking for
similar problems.

Backpatch to 9.4, where this code was introduced.

pcolor was used to represent function arguments that are nominally of
type color, but when using a pre-ANSI C compiler would be passed as the
promoted integer type.  We really don't need that anymore.

The regex library used to have a notion of a "collating element" that was
distinct from a "character", but Henry Spencer never actually implemented
his planned support for multi-character collating elements, and the Tcl
crew ripped out most of the stubs for that years ago.  The only thing left
that distinguished the "celt" typedef from the "chr" typedef was that
"celt" was supposed to also be able to hold the not-a-character "NOCELT"
value.  However, NOCELT was not used anywhere after the MCCE stub removal
changes, which means there's no need for celt to be different from chr.
Removing the separate typedef simplifies matters and also removes a trap
for the unwary, in that celt is signed while chr may not be, so comparisons
could mean different things.  There's no bug there today because we
restrict CHR_MAX to be less than INT_MAX, but I think there may have been
such bugs before we did that, and there could be again if anyone ever
decides to fool with the range of chr.

This patch also removes assorted unnecessary casts to "chr" of values
that are already chrs.  Many of these seem to be leftover from days when
the code was compatible with pre-ANSI C.

We already mentioned unix_socket_directories as an option.

Reported-by: https://www.postgresql.org/message-id/45016837-6cf3-3136-f959-763d06a28076%402ndquadrant.com

Backpatch-through: 9.6

On some buildfarm animals the isolationtest added in 07ef0351 failed, as
the order in which processes are run after unlocking is not
guaranteed. Add an alternative output for that.

Discussion: <7969.1471484738@sss.pgh.pa.us>
Backpatch: 9.6, like the test in the aforementioned commit

Per comment in the file, this was fixed around OS X 10.2.

As usual, we've been pretty awful about maintaining these counts.
They're not all that critical, perhaps, but let's get them right
at release time.  Also fix 9.5, which I notice is just as bad.
It's probably wrong further back, but the lack of --help=foo
options before 9.5 makes it too painful to count.

These types are storage-compatible with real arrays, but they don't support
toasting, so of course they can't support expansion either.

Per bug #14289 from Michael Overmeyer.  Back-patch to 9.5 where expanded
arrays were introduced.

Report: <20160818174414.1529.37913@wrigleys.postgresql.org>

The upstream XSLT stylesheets use some very general XPath expressions in
some places that end up being very slow.  We can optimize them with
knowledge about the DocBook document structure and our particular use
thereof.  For example, when counting preceding chapters to get a number
for the current chapter, we only need to count preceding sibling
nodes (more or less) instead of searching through the entire node tree
for chapter elements.

This change attacks the slowest pieces as identified by xsltproc
--profile.  This makes the HTML build roughly 10 times faster, resulting
in the new total build time being about the same as the old DSSSL-based
build.  Some of the non-HTML build targets (especially FO) will also
benefit a bit, but they have not been specifically analyzed.

With this, also remove the pg.fast parameter, which was previously a
hack to get the build to a manageable speed.

Alexander Lakhin <a.lakhin@postgrespro.ru>, with some additional
tweaking by me

Offer a list of database names; formerly no help was offered.

Ian Barwick, reviewed by Gerdan Santos

Patch: <5724132E.1030804@2ndquadrant.com>

Offer a list of available versions for that extension.  Formerly, since
there was no special support for this, it triggered off the UPDATE
keyword and offered a list of table names --- not too helpful.

Jeff Janes, reviewed by Gerdan Santos

Patch: <CAMkU=1z0gxEOLg2BWa69P4X4Ot8xBxipGUiGkXe_tC+raj79-Q@mail.gmail.com>

Emre Hasegeli

Patch: <CAE2gYzzF24ZHWqkMukkHwqa0otbES9Rex22LrjQUNbi=oKziNQ@mail.gmail.com>

This file had some unusual comment layout.  Most of the comments
introducing structs ended up to the right of the screen and following
the start of the struct.  Some comments for struct members ended up
after the member definition.

Fix that by moving comments consistently before what they are
describing.  Also add missing struct tags where missing so that it is
easier to tell what the struct is.

All we need is 4 bytes at the moment, for MD5 authentication. But in
upcomint patches for SCRAM authentication, SCRAM will need a salt of
different length. It's less scary for the caller to pass the buffer
length anyway, than assume a certain-sized output buffer.

Author: Michael Paquier
Discussion: <CAB7nPqQvO4sxLFeS9D+NM3wpy08ieZdAj_6e117MQHZAfxBFsg@mail.gmail.com>

This adds a couple of new timezones that are present in the newer
versions of Windows. It also updates comments to reference UTC rather
than GMT, as this change has been made in Windows.

Michael Paquier

This way sendAuthRequest doesn't need to know the details of all the
different authentication methods. This is in preparation for adding SCRAM
authentication, which will add yet another authentication request message
type, with different payload.

Reviewed-By: Michael Paquier
Discussion: <CAB7nPqQvO4sxLFeS9D+NM3wpy08ieZdAj_6e117MQHZAfxBFsg@mail.gmail.com>

INSERT ..  ON CONFLICT runs a pre-check of the possible conflicting
constraints before performing the actual speculative insertion.  In case
the inserted tuple included TOASTed columns the ON CONFLICT condition
would be handled correctly in case the conflict was caught by the
pre-check, but if two transactions entered the speculative insertion
phase at the same time, one would have to re-try, and the code for
aborting a speculative insertion did not handle deleting the
speculatively inserted TOAST datums correctly.

TOAST deletion would fail with "ERROR: attempted to delete invisible
tuple" as we attempted to remove the TOAST tuples using
simple_heap_delete which reasoned that the given tuples should not be
visible to the command that wrote them.

This commit updates the heap_abort_speculative() function which aborts
the conflicting tuple to use itself, via toast_delete, for deleting
associated TOAST datums.  Like before, the inserted toast rows are not
marked as being speculative.

This commit also adds a isolationtester spec test, exercising the
relevant code path. Unfortunately 9.5 cannot handle two waiting
sessions, and thus cannot execute this test.

Reported-By: Viren Negi, Oskari Saarenmaa
Author: Oskari Saarenmaa, edited a bit by me
Bug: #14150
Discussion: <20160519123338.12513.20271@wrigleys.postgresql.org>
Backpatch: 9.5, where ON CONFLICT was introduced

regexp_match() is like regexp_matches(), but it disallows the 'g' flag
and in consequence does not need to return a set.  Instead, it returns
a simple text array value, or NULL if there's no match.  Previously people
usually got that behavior with a sub-select, but this way is considerably
more efficient.

Documentation adjusted so that regexp_match() is presented first and then
regexp_matches() is introduced as a more complicated version.  This is
a bit historically revisionist but seems pedagogically better.

Still TODO: extend contrib/citext to support this function.

Emre Hasegeli, reviewed by David Johnston

Discussion: <CAE2gYzy42sna2ME_e3y1KLQ-4UBrB-eVF0SWn8QG39sQSeVhEw@mail.gmail.com>

Slot creation did not clear all fields upon creation. After start the
memory is zeroed, but when a physical replication slot was created in
the shared memory of a previously existing logical slot, catalog_xmin
would not be cleared. That in turn would prevent vacuum from doing its
duties.

To fix initialize all the fields. To make similar future bugs less
likely, zero all of ReplicationSlotPersistentData, and re-order the
rest of the initialization to be in struct member order.

Analysis: Andrew Gierth
Reported-By: md@chewy.com
Author: Michael Paquier
Discussion: <20160705173502.1398.70934@wrigleys.postgresql.org>
Backpatch: 9.4, where replication slots were introduced

As implemented, -e ran an EXPLAIN but then discarded the output, which
certainly seems pointless.  Make it print to stdout instead.  It's been
like that forever, so back-patch to all supported branches.

Daniel Gustafsson, reviewed by Andreas Scherbaum

Patch: <B97BDCB7-A3B3-4734-90B5-EDD586941629@yesql.se>