Markus Nullmeier

This interface is designed to give an access to WAL for extensions which
could implement new access method, for example. Previously it was
impossible because restoring from custom WAL would need to access system
catalog to find a redo custom function. This patch suggests generic way
to describe changes on page with standart layout.

Bump XLOG_PAGE_MAGIC because of new record type.

Author: Alexander Korotkov with a help of Petr Jelinek, Markus Nullmeier and
	minor editorization by my
Reviewers: Petr Jelinek, Alvaro Herrera, Teodor Sigaev, Jim Nasby,
	Michael Paquier

I should have remembered that we can't use INT64_MODIFIER with sscanf():
configure chooses that to work with snprintf(), but it might be for our
src/port/snprintf.c implementation and so not compatible with the
platform's sscanf().  This appears to be the explanation for buildfarm
member frogmouth's continuing unhappiness with the tzcode update.

Fortunately, in all of the places where zic is attempting to read into
an int64 variable, it's reading a year which certainly will fit just fine
into an int.  So make it read into an int with %d, and then cast or copy
as necessary.

Previously this test was relying too much on WAL replay to occur in the
exact configured interval, which was unreliable on slow or overly busy
servers.  Use a custom loop instead of poll_query_until, which is
hopefully more reliable.

Per continued failures on buildfarm member hamster (which is probably
the only one running this test suite)

Author: Michaël Paquier

Previously, the planner would reject an index-only scan if any restriction
clause for its table used a column not available from the index, even
if that restriction clause would later be dropped from the plan entirely
because it's implied by the index's predicate.  This is a fairly common
situation for partial indexes because predicates using columns not included
in the index are often the most useful kind of predicate, and we have to
duplicate (or at least imply) the predicate in the WHERE clause in order
to get the index to be considered at all.  So index-only scans were
essentially unavailable with such partial indexes.

To fix, we have to do detection of implied-by-predicate clauses much
earlier in the planner.  This patch puts it in check_index_predicates
(nee check_partial_indexes), meaning it gets done for every partial index,
whereas we previously only considered this issue at createplan time,
so that the work was only done for an index actually selected for use.
That could result in a noticeable planning slowdown for queries against
tables with many partial indexes.  However, testing suggested that there
isn't really a significant cost, especially not with reasonable numbers
of partial indexes.  We do get a small additional benefit, which is that
cost_index is more accurate since it correctly discounts the evaluation
cost of clauses that will be removed.  We can also avoid considering such
clauses as potential indexquals, which saves useless matching cycles in
the case where the predicate columns aren't in the index, and prevents
generating bogus plans that double-count the clause's selectivity when
the columns are in the index.

Tomas Vondra and Kyotaro Horiguchi, reviewed by Kevin Grittner and
Konstantin Knizhnik, and whacked around a little by me

Author: Konstantin Knizhnik

As best as I can tell, MyReplicationSlot needs to be PGDLLIMPORT in
order for the new test_slot_timelines test module to compile.

Per buildfarm

Without this, the test_slot_timelines modules fails "make installcheck"
because the required feature is not enabled in a stock server.

Per buildfarm

When decoding from a logical slot, it's necessary for xlog reading to be
able to read xlog from historical (i.e. not current) timelines;
otherwise, decoding fails after failover, because the archives are in
the historical timeline.  This is required to make "failover logical
slots" possible; it currently has no other use, although theoretically
it could be used by an extension that creates a slot on a standby and
continues to replay from the slot when the standby is promoted.

This commit includes a module in src/test/modules with functions to
manipulate the slots (which is not otherwise possible in SQL code) in
order to enable testing, and a new test in src/test/recovery to ensure
that the behavior is as expected.

Author: Craig Ringer
Reviewed-By: Oleksii Kliukin, Andres Freund, Petr Jelínek

Some minor tweaks and comment additions, for cleanliness sake and to
avoid having the upcoming timeline-following patch be polluted with
unrelated cleanup.

Extracted from a larger patch by Craig Ringer, reviewed by Andres
Freund, with some additions by myself.

Formerly, the geometric I/O routines such as box_in and point_out relied
directly on strtod() and sprintf() for conversion of the float8 component
values of their data types.  However, the behavior of those functions is
pretty platform-dependent, especially for edge-case values such as
infinities and NaNs.  This was exposed by commit acdf2a8b, which
added test cases involving boxes with infinity endpoints, and immediately
failed on Windows and AIX buildfarm members.  We solved these problems
years ago in the main float8in and float8out functions, so let's fix it
by making the geometric types use that code instead of depending directly
on the platform-supplied functions.

To do this, refactor the float8in code so that it can be used to parse
just part of a string, and as a convenience make the guts of float8out
usable without going through DirectFunctionCall.

While at it, get rid of geo_ops.c's fairly shaky assumptions about the
maximum output string length for a double, by having it build results in
StringInfo buffers instead of fixed-length strings.

In passing, convert all the "invalid input syntax for type foo" messages
in this area of the code into "invalid input syntax for type %s" to reduce
the number of distinct translatable strings, per recent discussion.
We would have needed a fair number of the latter anyway for code-sharing
reasons, so we might as well just go whole hog.

Note: this patch is by no means intended to guarantee that the geometric
types uniformly behave sanely for infinity or NaN component values.
But any bugs we have in that line were there all along, they were just
harder to reach in a platform-independent way.

My compiler doesn't like the lack of initialization of "flag", and
I think it's right: if there were zero keys we'd have an undefined
result.  The AND of zero items is TRUE, so initialize to TRUE.

Patch implements quad-tree over boxes, naive approach of 2D quad tree will not
work for any non-point objects because splitting space on node is not
efficient. The idea of pathc is treating 2D boxes as 4D points, so,
object will not overlap (in 4D space).

The performance tests reveal that this technique especially beneficial
with too much overlapping objects, so called "spaghetti data".

Author: Alexander Lebedev with editorization by Emre Hasegeli and me

Author: Alexander Lebedev

During scan sometimes it would be very helpful to know some information about
parent node or all 	ancestor nodes. Right now reconstructedValue could be used
but it's not a right usage of it (range opclass uses that).

traversalValue is arbitrary piece of memory in separate MemoryContext while
reconstructedVale should have the same type as indexed column.

Subsequent patches for range opclass and quad4d tree will use it.

Author: Alexander Lebedev, Teodor Sigaev

Michael Paquier

Reporting the specific out-of-range input value produces platform-dependent
results.  We could skip reporting the value, but that's contrary to our
message style guidelines and unhelpful to users.  Or we could add a
separate expected-output file for Windows, but that would be a substantial
maintenance burden, and these test cases seem unlikely to be worth it.

Per buildfarm.

The server hasn't paid attention to the TZ environment variable since
commit ca4af308, but that commit missed removing this documentation
reference, as did commit d883b916 which added the reference where
it now belongs (initdb).

Back-patch to 9.2 where the behavior changed.  Also back-patch
d883b916 as needed.

Matthew Somerville

In this mode, the master waits for the transaction to be applied on
the remote side, not just written to disk.  That means that you can
count on a transaction started on the standby to see all commits
previously acknowledged by the master.

To make this work, the standby sends a reply after replaying each
commit record generated with synchronous_commit >= 'remote_apply'.
This introduces a small inefficiency: the extra replies will be sent
even by standbys that aren't the current synchronous standby.  But
previously-existing synchronous_commit levels make no attempt at all
to optimize which replies are sent based on what the primary cares
about, so this is no worse, and at least avoids any extra replies for
people not using the feature at all.

Thomas Munro, reviewed by Michael Paquier and by me.  Some additional
tweaks by me.

interval_mul() attempts to prevent its calculations from producing silly
results, but it forgot that zero times infinity yields NaN in IEEE
arithmetic.  Hence, a case like '1 second'::interval * 'infinity'::float8
produced a NaN for the months product, which didn't trigger the range
check, resulting in bogus and possibly platform-dependent output.

This isn't terribly obvious to the naked eye because if you try that
exact case, you get "interval out of range" which is what you expect
--- but if you look closer, the error is coming from interval_out not
interval_mul.  interval_mul has allowed a bogus value into the system.

Fix by adding isnan tests.

Noted while testing Vitaly Burovoy's fix for infinity input to
to_timestamp().  Given the lack of field complaints, I doubt this
is worth a back-patch.

With the original SQL-function implementation, such cases failed because
we don't support infinite intervals.  Converting the function to C lets
us bypass the interval representation, which should be a bit faster as
well as more flexible.

Vitaly Burovoy, reviewed by Anastasia Lubennikova

resulttypeLen and resulttypeByVal must be set correctly when serializing
aggregates, not just when finalizing them.  This was in David's final
patch but I downloaded the wrong version by mistake and failed to spot
the error.

David Rowley

This is necessary infrastructure for supporting parallel aggregation
for aggregates whose transition type is "internal".  Such values
can't be passed between cooperating processes, because they are
just pointers.

David Rowley, reviewed by Tomas Vondra and by me.

The old documentation didn't know about the new -b flag, only about -f.

Fabien Coelho

The description of what the per-transaction log file says for skipped
transactions is just plain wrong.

Report and patch by Tomas Vondra, reviewed by Fabien Coelho and
modified by me.

This refines the previous weight range and allows a script to be "turned
off" by passing a zero weight, which is useful when scripting multiple
pgbench runs.

I did not apply the suggested warning when a script uses zero weight; we
use the principle elsewhere that if there's nothing to be done, do
nothing quietly.

Adjust docs accordingly.

Author: Jeff Janes, Fabien Coelho

You can now do the same thing via \set using the appropriate function,
either random(), random_gaussian(), or random_exponential(), depending
on the desired distribution.  This is not backward-compatible, but per
discussion, it's worth it to avoid having the old syntax hang around
forever.

Fabien Coelho, reviewed by Michael Paquier, and adjusted by me.

Whenever this function is used with the FORMAT_MESSAGE_FROM_SYSTEM flag,
it's good practice to include FORMAT_MESSAGE_IGNORE_INSERTS as well.
Otherwise, if the message contains any %n insertion markers, the function
will try to fetch argument strings to substitute --- which we are not
passing, possibly leading to a crash.  This is exactly analogous to the
rule about not giving printf() a format string you're not in control of.

Noted and patched by Christian Ullrich.
Back-patch to all supported branches.

When tsearch was implemented I did several mistakes in hostname/email
definition rules:
1) allow underscore in hostname what prohibited by RFC
2) forget to allow leading digits separated by hyphen (like 123-x.com)
   in hostname
3) do no allow underscore/hyphen after leading digits in localpart of email

Artur's patch resolves two last issues, but by the way allows hosts name like
123_x.com together with 123-x.com. RFC forbids underscore usage in hostname
but pg allows that since initial tsearch version in core, although only
for non-digits. Patch syncs support digits and nondigits in both hostname and
email.

Forbidding underscore in hostname may break existsing usage of tsearch and,
anyhow, it should be done by separate patch.

Author: Artur Zakirov
BUG: #13964

Per discussion, the new extensible node framework is thought to be
better designed than the custom path/scan/scanstate stuff we added
in PostgreSQL 9.5.  Rework the latter to be more like the former.

This is not backward-compatible, but we generally don't promise that
for C APIs, and there probably aren't many people using this yet
anyway.

KaiGai Kohei, reviewed by Petr Jelinek and me.  Some further
cosmetic changes by me.

The IANA crew seem to think that symlink() exists everywhere nowadays,
and they may well be right.  But we use #ifdef HAVE_SYMLINK elsewhere
so for consistency we should do it here too.  Noted by Michael Paquier.

The new coding of dolink() is dependent on link() returning an on-point
errno when it fails; but the quick-hack implementation of link() that
we'd put in for Windows didn't bother with setting errno.  Fix that.

Analysis and patch by Christian Ullrich.

INT64_MIN/MAX should be spelled PG_INT64_MIN/MAX, per well established
convention in our sources.  Less obviously, a symbol named DOUBLE causes
problems on Windows builds, so rename that to DOUBLE_CONST; and rename
INTEGER to INTEGER_CONST for consistency.

Also, get rid of incorrect/obsolete hand-munging of yycolumn, and fix
the grammar for float constants to handle expected cases such as ".1".

First two items by Michael Paquier, second two by me.

Commit fbe5a3fb accidentally changed
this behavior; put things back the way they were, and add some
regression tests.

Report by Andres Freund; patch by Ashutosh Bapat, with a bit of
kibitzing by me.

Up until now, we've been using timezone and tzname, but Visual Studio
2015 (for which we wish to add support) no longer declares those
symbols.  All versions since Visual Studio 2003 apparently support the
underscore-equipped names, and we don't support anything older than
Visual Studio 2005, so this should work OK everywhere.  But let's see
what the buildfarm thinks.

Michael Paquier, reviewed by Petr Jelinek

Thomas Munro

The new functions are pi(), random(), random_exponential(),
random_gaussian(), and sqrt().  I was worried that this would be
slower than before, but, if anything, it actually turns out to be
slightly faster, because we now express the built-in pgbench scripts
using fewer lines; each \setrandom can be merged into a subsequent
\set.

Fabien Coelho

Corrects an oversight in 2c83f435 where the $timed_out reference var
isn't initialized; using it would require the caller to initialize it
beforehand, which is cumbersome.

Author: Craig Ringer