As pointed out by Naoya Anzai, my previous try at this was a few bricks
shy of a load, because I had forgotten that the initial-positioning logic
might not try to skip over nulls at the end of the index the scan will
start from.  We ought to fix that, because it represents an unnecessary
inefficiency, but first let's get the scan-stop logic back to a safe
state.  With this patch, we preserve the performance benefit requested
in bug #6278 for the case of scanning forward into NULLs (in a NULLS
LAST index), but the reverse case of scanning backward across NULLs
when there's no suitable initial-positioning qual is still inefficient.

Previously, we skipped a checkpoint if no WAL had been written since
last checkpoint, though this does not appear in user documentation.
As of now, we skip a checkpoint until we have written at least one
enough WAL to switch the next WAL file. This greatly reduces the
level of activity and number of WAL messages generated by a very
low activity server. This is safe because the purpose of a checkpoint
is to act as a starting place for a recovery, in case of crash.
This patch maintains minimal WAL volume for replay in case of crash,
thus maintaining very low crash recovery time.

Startup process now has its own dedicated file, just like all other
special/background processes. Reduces role and size of xlog.c

There was a timing window between when oldestActiveXid was derived
and when it should have been derived that only shows itself under
heavy load. Move code around to ensure correct timing of derivation.
No change to StartupSUBTRANS() code, which is where this failed.

Bug report by Chris Redekop

If the initial snapshot had overflowed then we can start whenever
the latest snapshot is empty, not overflowed or as we did already,
start when the xmin on primary was higher than xmax of our starting
snapshot, which proves we have full snapshot data.

Bug report by Chris Redekop

Patch by me, bug report by Chris Redekop, analysis by Florian Pflug

In assert-enabled builds, we assert during the shutdown sequence that
the queues have been properly emptied, and during process startup that
we are inheriting empty queues.  In non-assert enabled builds, we just
save a few cycles.

This allows us to give correct syntax error pointers when complaining
about ungrouped variables in a join query with aggregates or GROUP BY.
It's pretty much irrelevant for the planner's use of the function, though
perhaps it might aid debugging sometimes.

If a tuple in a syscache contains an out-of-line toasted field, and we
try to fetch that field shortly after some other transaction has committed
an update or deletion of the tuple, there is a race condition: vacuum
could come along and remove the toast tuples before we can fetch them.
This leads to transient failures like "missing chunk number 0 for toast
value NNNNN in pg_toast_2619", as seen in recent reports from Andrew
Hammond and Tim Uckun.

The design idea of syscache is that access to stale syscache entries
should be prevented by relation-level locks, but that fails for at least
two cases where toasted fields are possible: ANALYZE updates pg_statistic
rows without locking out sessions that might want to plan queries on the
same table, and CREATE OR REPLACE FUNCTION updates pg_proc rows without
any meaningful lock at all.

The least risky fix seems to be an idea that Heikki suggested when we
were dealing with a related problem back in August: forcibly detoast any
out-of-line fields before putting a tuple into syscache in the first place.
This avoids the problem because at the time we fetch the parent tuple from
the catalog, we should be holding an MVCC snapshot that will prevent
removal of the toast tuples, even if the parent tuple is outdated
immediately after we fetch it.  (Note: I'm not convinced that this
statement holds true at every instant where we could be fetching a syscache
entry at all, but it does appear to hold true at the times where we could
fetch an entry that could have a toasted field.  We will need to be a bit
wary of adding toast tables to low-level catalogs that don't have them
already.)  An additional benefit is that subsequent uses of the syscache
entry should be faster, since they won't have to detoast the field.

Back-patch to all supported versions.  The problem is significantly harder
to reproduce in pre-9.0 releases, because of their willingness to flush
every entry in a syscache whenever the underlying catalog is vacuumed
(cf CatalogCacheFlushRelation); but there is still a window for trouble.

These are not touched by pgindent, so clean them up a bit manually.

Noted by Fujii Masao

'postgres' database.

bgwriter is now a much less important process, responsible for page
cleaning duties only. checkpointer is now responsible for checkpoints
and so has a key role in shutdown. Later patches will correct doc
references to the now old idea that bgwriter performs checkpoints.
Has beneficial effect on performance at high write rates, but mainly
refactoring to more easily allow changes for power reduction by
simplifying previously tortuous code around required to allow page
cleaning and checkpointing to time slice in the same process.

Patch by me, Review by Dickson Guedes

The existing scan-direction-sensitive tests were overly complex, and
failed to stop the scan in cases where it's perfectly legitimate to do so.
Per bug #6278 from Maksym Boguk.

Back-patch to 8.3, which is as far back as the patch applies easily.
Doesn't seem worth sweating over a relatively minor performance issue in
8.2 at this late date.  (But note that this was a performance regression
from 8.1 and before, so 8.2 is being left as an outlier.)

The POSIX spec defines locale fields for controlling the ordering of the
value, sign, and currency symbol in monetary output, but cash_out only
supported a small subset of these options.  Fully implement p/n_sign_posn,
p/n_cs_precedes, and p/n_sep_by_space per spec.  Fix up cash_in so that
it will accept all these format variants.

Also, make sure that thousands_sep is only inserted to the left of the
decimal point, as required by spec.

Per bug #6144 from Eduard Kracmar and discussion of bug #6277.  This patch
includes some ideas from Alexander Lakhin's proposed patch, though it is
very different in detail.

Make sure that it considers all the possibilities that the old code did,
instead of trying only one possibility per character position.  To keep the
runtime in bounds, instead tweak the character incrementers to not try
every possible multibyte character code.  Remove unnecessary logic to
restore the old character value on failure.  Additional comment and
formatting cleanup.

Recent work on index-only scans left this somewhat out of date.

cash_out failed to handle multiple-byte thousands separators, as per bug
#6277 from Alexander Law.  In addition, cash_in didn't handle that either,
nor could it handle multiple-byte positive_sign.  Both routines failed to
support multiple-byte mon_decimal_point, which I did not think was worth
changing, but at least now they check for the possibility and fall back to
using '.' rather than emitting invalid output.  Also, make cash_in handle
trailing negative signs, which formerly it would reject.  Since cash_out
generates trailing negative signs whenever the locale tells it to, this
last omission represents a fail-to-reload-dumped-data bug.  IMO that
justifies patching this all the way back.

This infrastructure doesn't in any way guarantee that the character
we produce will sort before the one we incremented; but it does at least
make it much more likely that we'll end up with something that is a valid
character, which improves our chances.

Kyotaro Horiguchi, with various adjustments by me.

new cluster.   vacuumdb, used by pg_upgrade, still has this dependency.

We need not wait until the commit record is durably on disk, because
in the event of a crash the page we're updating with hint bits will
be gone anyway.  Per off-list report from Heikki Linnakangas, this
can significantly degrade the performance of unlogged tables; I was
able to show a 2x speedup from this patch on a pgbench run with scale
factor 15.  In practice, this will mostly help small, heavily updated
tables, because on larger tables you're unlikely to run into the same
row again before the commit record makes it out to disk.

Testing reveals that this macro is a hot-spot for index-only-scans.
Per discussion with Tom Lane.

This doesn't appear to accompish anything useful, and does make the
restore fail if the postgres database happens to have been dropped.

Make sure ecpg/include/ is rebuilt before the other subdirectories,
so that ecpg_config.h is up to date.  This is not likely to matter
during production builds, only development, so no back-patch.

Kevin Grittner

Report by Vik Reykja, patch by Kevin Grittner.

in the old cluster.

The recent unpleasantness with copyrights has accelerated a move that
was already in planning.

one lock per backend or auxiliary process - the need for a lock for each
aux processes was not accounted for in NumLWLocks(). No-one noticed,
because the three locks needed for the three aux processes fit into the
few extra lwlocks we allocate for 3rd party modules that don't call
RequestAddinLWLocks() (NUM_USER_DEFINED_LWLOCKS, 4 by default).

The original implementation of ELSIF in plpgsql converted the construct
into nested simple IF statements.  This was prone to stack overflow with
long ELSIF lists, in two different ways.  First, it's difficult to generate
the parsetree without using right-recursion in the bison grammar, and
that's prone to parser stack overflow since nothing can be reduced until
the whole list has been read.  Second, we'd recurse during execution, thus
creating an unnecessary risk of execution-time stack overflow.  Rewrite
so that the ELSIF list is represented as a flat list, scanned via iteration
not recursion, and generated through left-recursion in the grammar.
Per a gripe from Håvard Kongsgård.

We should generally use left-recursion not right-recursion to parse lists.
Bison hasn't got any built-in way to check for this type of inefficiency,
and I didn't find anything on the net in a quick search, so I wrote a
little Perl script to do it.  Add to src/tools/ so we don't have to
re-invent this wheel next time we wonder if we're doing anything stupid.

Currently, the only place that seems to need fixing is plpgsql's stmt_else
production, so the problem doesn't appear to be common enough to warrant
trying to include such a test in our standard build process.  If we did
want to do that, we'd need a way to ignore some false positives, such as
a_expr := '-' a_expr

expect -> except, noted by Andrew Dunstan.  Also, "cannot" seems more
readable here than "can not", per David Wheeler.

If the right-hand side of a semijoin is unique, then we can treat it like a
normal join (or another way to say that is: we don't need to explicitly
unique-ify the data before doing it as a normal join).  We were recognizing
such cases when the RHS was a sub-query with appropriate DISTINCT or GROUP
BY decoration, but there's another way: if the RHS is a plain relation with
unique indexes, we can check if any of the indexes prove the output is
unique.  Most of the infrastructure for that was there already in the join
removal code, though I had to rearrange it a bit.  Per reflection about a
recent example in pgsql-performance.