The ALTER SYSTEM ref page hadn't been held to a very high standard, nor
was the feature well integrated into section 18.1 (parameter setting).
Also, though commit 4c4654af had improved the structure of 18.1, it also
introduced a lot of poor wording, imprecision, and outright falsehoods.
Try to clean that up.

Set release date, do a final pass of wordsmithing, improve some other
new-in-9.4 documentation.

Fabrízio de Royes Mello reviewed by Rushabh Lathia.

Commit 815d71de hadn't bothered to update the documentation to match the
behavioral change, and a lot of other text in this section was badly in
need of copy-editing.

The code for advancing through the input rows overlooked the case that we
might already be past the first row of the row pair now being considered,
in case the previous percentile also fell between the same two input rows.

Report and patch by Andrew Gierth; logic rewritten a bit for clarity by me.

Mark Dilger

The planner seems to like to do this join query as a hash join, making
the output ordering machine-dependent; worse, it's a hash on OIDs, so
that it's a bit astonishing that the result doesn't change from run to
run even on one machine.  Add an ORDER BY to get consistent results.
Per buildfarm.

I also suppressed output from the final DROP SCHEMA CASCADE, to avoid
occasional failures similar to those fixed in commit 81d815dc.
That hasn't been observed in the buildfarm yet, but it seems likely
to happen in future if we leave it as-is.

The functions are:
    to_jsonb()
    jsonb_object()
    jsonb_build_object()
    jsonb_build_array()
    jsonb_agg()
    jsonb_object_agg()

Also along the way some better logic is implemented in
json_categorize_type() to match that in the newly implemented
jsonb_categorize_type().

Andrew Dunstan, reviewed by Pavel Stehule and Alvaro Herrera.

In commit 462bd957, I changed postgres_fdw
to rely on get_plan_rowmark() instead of get_parse_rowmark().  I still
think that's a good idea in the long run, but as Etsuro Fujita pointed out,
it doesn't work today because planner.c forces PlanRowMarks to have
markType = ROW_MARK_COPY for all foreign tables.  There's no urgent reason
to change this in the back branches, so let's just revert that part of
yesterday's commit rather than trying to design a better solution under
time pressure.

Also, add a regression test case showing what postgres_fdw does with FOR
UPDATE/SHARE.  I'd blithely assumed there was one already, else I'd have
realized yesterday that this code didn't work.

The functions remove object fields, including in nested objects, that
have null as a value. In certain cases this can lead to considerably
smaller datums, with no loss of semantic information.

Andrew Dunstan, reviewed by Pavel Stehule.

This avoids duplicating the code.

Michael Paquier, reviewed by Simon Riggs and me

Otherwise the pg_ctl start and stop messages get mixed up with the TAP
output, which isn't technically valid.

Ordinarily we can omit checking of a WHERE condition that matches a partial
index's condition, when we are using an indexscan on that partial index.
However, in SELECT FOR UPDATE we must include the "redundant" filter
condition in the plan so that it gets checked properly in an EvalPlanQual
recheck.  The planner got this mostly right, but improperly omitted the
filter condition if the index in question was on an inheritance child
table.  In READ COMMITTED mode, this could result in incorrectly returning
just-updated rows that no longer satisfy the filter condition.

The cause of the error is using get_parse_rowmark() when get_plan_rowmark()
is what should be used during planning.  In 9.3 and up, also fix the same
mistake in contrib/postgres_fdw.  It's currently harmless there (for lack
of inheritance support) but wrong is wrong, and the incorrect code might
get copied to someplace where it's more significant.

Report and fix by Kyotaro Horiguchi.  Back-patch to all supported branches.

In READ COMMITTED mode, if a SELECT FOR UPDATE discovers it has to redo
WHERE-clause checking on rows that have been updated since the SELECT's
snapshot, it invokes EvalPlanQual processing to do that.  If this first
occurs within a non-first child table of an inheritance tree, the previous
coding could accidentally re-return a matching row from an earlier,
already-scanned child table.  (And, to add insult to injury, I think this
could make it miss returning a row that should have been returned, if the
updated row that this happens on should still have passed the WHERE qual.)
Per report from Kyotaro Horiguchi; the added isolation test is based on his
test case.

This has been broken for quite awhile, so back-patch to all supported
branches.

Ensure we reindex indexes built on Mat Views.
Based on patch from Micheal Paquier

Add thorough tests to check that indexes on
tables, toast tables and mat views are reindexed.

Simon Riggs

Leaving global objects like roles hanging around is bad practice.

Aside from not testing the case it claimed to test (namely a permissions
failure), it left a login-capable role lying around, which quite aside
from possibly being a security hole would cause subsequent regression runs
to fail since the role would already exist.

In passing, also make some debugging elog's in pgstat.c a bit more
consistently worded.

Back-patch as far as applicable (9.3 or 9.4; none of these mistakes are
really old).

Mark Dilger identified and patched the type violations; the message
rewordings are mine.

It's an OID. WRITE_UINT_FIELD is identical to WRITE_OID_FIELD, but let's
be tidy.

Mark Dilger

Author: Fabrízio de Royes Mello <fabriziomello@gmail.com>

The amount of space to reserve for the value's varlena header is
VARHDRSZ, not sizeof(VARHDRSZ).  The latter coding accidentally
failed to fail because of the way the VARHDRSZ macro is currently
defined; but if we ever change it to return size_t (as one might
reasonably expect it to do), convertToJsonb() would have failed.

Spotted by Mark Dilger.

It's not run by the global "check" or "installcheck" targets, because the
temporary installation it creates accepts TCP connections from any user
the same host, which is insecure.

Author: Michael Paquier

Previously REINDEX DATABASE and REINDEX SCHEMA
produced a stream of NOTICE messages. Removing that
since it is inconsistent for such a command to
produce output without a VERBOSE option.

Some requests count as two tests.

Add new SCHEMA option to REINDEX and reindexdb.

Sawada Masahiko

Reviewed by Michael Paquier and Fabrízio de Royes Mello

PostgreSQL on Windows 8 or Windows Server 2012 will now
get high-resolution timestamps by dynamically loading the
GetSystemTimePreciseAsFileTime function. It'll fall back to
to GetSystemTimeAsFileTime if the higher precision variant
isn't found, so the same binaries without problems on older
Windows releases.

No attempt is made to detect the Windows version.  Only the
presence or absence of the desired function is considered.

Craig Ringer

PostgreSQL was calling GetSystemTime followed by SystemTimeToFileTime in the
win32 port gettimeofday function. This is not necessary and limits the reported
precision to the 1ms granularity that the SYSTEMTIME struct can represent. By
using GetSystemTimeAsFileTime we avoid unnecessary conversions and capture
timestamps at 100ns granularity, which is then rounded to 1µs granularity for
storage in a PostgreSQL timestamp.

On most Windows systems this change will actually have no significant effect on
timestamp resolution as the system timer tick is typically between 1ms and 15ms
depending on what timer resolution currently running applications have
requested. You can check this with clockres.exe from sysinternals. Despite the
platform limiation this change still permits capture of finer timestamps where
the system is capable of producing them and it gets rid of an unnecessary
syscall.

The higher resolution GetSystemTimePreciseAsFileTime call available on Windows
8 and Windows Server 2012 has the same interface as GetSystemTimeAsFileTime, so
switching to GetSystemTimeAsFileTime makes it easier to use the Precise variant
later.

Craig Ringer, reviewed by David Rowley

This was broken in 618c9430.

From Michael Paquier

No need to set tuple tableOid twice

Jim Nasby

Generate a table_rewrite event when ALTER TABLE
attempts to rewrite a table. Provide helper
functions to identify table and reason.

Intended use case is to help assess or to react
to schema changes that might hold exclusive locks
for long periods.

Dimitri Fontaine, triggering an edit by Simon Riggs

Reviewed in detail by Michael Paquier

Rename parameter action_at_recovery_target to
recovery_target_action suggested by Christoph Berg.

Place into recovery.conf suggested by Fujii Masao,
replacing (deprecating) earlier parameters, per
Michael Paquier.

Used to say just "could not read password from file "...": Success", which
isn't very informative.

Mats Erik Andersson. Backpatch to all supported versions.

The "file mode" bits in the tar file header is not supposed to include the
file type bits, e.g. S_IFREG or S_IFDIR. The file type is stored in a
separate field. This isn't a problem in practice, all tar programs ignore
the extra bits, but let's be tidy.

This came up in a discussion around bug #11949, reported by Hendrik Grewe,
although this doesn't fix the issue with tar --append. That turned out to be
a bug in GNU tar. Schilly's tartest program revealed this defect in the tar
created by pg_basebackup.

This problem goes as far as we we've had pg_basebackup, but since this
hasn't caused any problems in practice, let's be conservative and fix in
master only.