gabrielle <gorthx@gmail.com>

implementation deficiencies.  Per discussion of bug #5592, we're not
going to change it, but these things should be documented so that if
anyone ever reimplements type tinterval, they will be more careful.

Without this patch, constraints inherited by children of a parent
table which itself has multiple inheritance parents can end up with
the wrong coninhcount.  After dropping the constraint, the children
end up with a leftover copy of the constraint that is not dumped
and cannot be dropped.  There is a similar problem with ALTER TABLE
.. ADD COLUMN, but that looks significantly more difficult to
resolve, so I'm committing this fix separately.

Back-patch to 8.4, which is the first release that has coninhcount.

Report by Hank Enting.

makeTSQuerySign.  The first of these is a live bug, on some platforms,
as per bug #5590 from John Regehr.  However the consequences seem limited
because of the relatively narrow scope of use of QTNode.sign.  The shift in
makeTSQuerySign is actually safe because TSQS_SIGLEN is unsigned, but it
seems like a good idea to insert an explicit cast rather than depend on that.

tsqueries.  CompareTSQ has to have a guard for the case rather than blindly
applying QTNodeCompare to random data past the end of the datums.  Also,
change QTNodeCompare to be a little less trusting: use an actual test rather
than just Assert'ing that the input is sane.  Problem encountered while
investigating another issue (I saw a core dump in autoanalyze on a table
containing multiple empty tsquery values).

Back-patch to all branches with tsquery support.

In HEAD, also fix some bizarre (though not outright wrong) coding in
tsq_mcontains().

Based on a patch by, and further ideas from, Alexander Korotkov.

Jeff Davis, with slight editorialization by me.

since Apple shipped a compiler that needed this switch, and there's
increasing interest in using other compilers that won't accept the switch
at all.  Better to let anybody who still needs the switch inject it via
CPPFLAGS.  Per gripe from Neil Conway.

While this hack arguably has some benefit in terms of making PL/pgsql's
line numbering match the programmer's expectations, it also makes
PL/pgsql inconsistent with the remaining PLs, making it difficult for
clients to reliably determine where the error actually is.  On balance,
it seems better to be consistent.

Pavel Stehule

from "clang".  The VERR changes make an assignment unconditional, which is
probably easier to read/understand anyway, and one can hardly argue that
it's worth shaving cycles off the case of reporting another error when
one has already been detected.  The INSIST change limits where that macro
can be used, but not in a way that creates a problem for any existing call.

interval input "invalid" was specified together with other fields.  Spotted
by Neil Conway with the help of a clang warning.  Although this has been
wrong since the interval code was written more than 10 years ago, it doesn't
affect anything beyond which error message you get for a wrong input, so not
worth back-patching very far.

indexes when the index column type (the opclass opckeytype) is different from
the expression's datatype.  When coded, this limitation wasn't worth worrying
about because we had no intelligence to speak of in stats collection for the
datatypes used by such opclasses.  However, now that there's non-toy
estimation capability for tsvector queries, it amounts to a bug that ANALYZE
fails to do this.

The fix changes struct VacAttrStats, and therefore constitutes an API break
for custom typanalyze functions.  Therefore we can't back-patch it into
released branches, but it was agreed that 9.0 isn't yet frozen hard enough
to make such a change unacceptable.  Ergo, back-patch to 9.0 but no further.
The API break had better be mentioned in 9.0 release notes.

bright, but it beats assuming that a prefix match behaves identically to an
exact match, which is what the code was doing before :-(.  Noted while
experimenting with Artur Dobrowski's example.

Although the key-combining code claimed to work correctly if its input
contained both lossy and exact pointers for a single page in a single TID
stream, in fact this did not work, and could not work without pretty
fundamental redesign.  Modify keyGetItem so that it will not return such a
stream, by handling lossy-pointer cases a bit more explicitly than we did
before.

Per followup investigation of a gripe from Artur Dabrowski.
An example of a query that failed given his data set is
select count(*) from search_tab where
(to_tsvector('german', keywords ) @@ to_tsquery('german', 'ee:* | dd:*')) and
(to_tsvector('german', keywords ) @@ to_tsquery('german', 'aa:*'));

Back-patch to 8.4 where the lossy pointer code was introduced.

struct representing a tree entry, rather than being a separately allocated
piece of storage.  This API is at least as clean as the old one (if not
more so --- there were some bizarre choices in there) and it permits a
very substantial memory savings, on the order of 2X in ginbulk.c's usage.

Also, fix minor memory leaks in code called by ginEntryInsert, in
particular in ginInsertValue and entryFillRoot, as well as ginEntryInsert
itself.  These leaks resulted in the GIN index build context continuing
to bloat even after we'd filled it to maintenance_work_mem and started
to dump data out to the index.

In combination these fixes restore the GIN index build code to honoring
the maintenance_work_mem limit about as well as it did in 8.4.  Speed
seems on par with 8.4 too, maybe even a bit faster, for a non-pathological
case in which HEAD was formerly slower.

Back-patch to 9.0 so we don't have a performance regression from 8.4.

Josh Kupershmidt.  Reviewing and kibitzing by Kevin Grittner and me.

possible (ie, whenever the tsquery is a constant), even when no statistics
are available for the tsvector.  For example, foo @@ 'a & b'::tsquery
can be expected to be more selective than foo @@ 'a'::tsquery, whether
or not we know anything about foo.  We use DEFAULT_TS_MATCH_SEL as the assumed
selectivity of individual query terms when no stats are available, then
combine the terms according to the query's AND/OR structure as usual.

Per experimentation with Artur Dabrowski's example.  (The fact that there
are no stats available in that example is a problem in itself, but
nonetheless tsmatchsel should be smarter about the case.)

Back-patch to 8.4 to keep all versions of tsmatchsel() in sync.

routines to make them behave better in the presence of "lossy" index pointers.
The previous coding was outright incorrect for some cases, as recently
reported by Artur Dabrowski: scanGetItem would fail to return index entries in
cases where one index key had multiple exact pointers on the same page as
another key had a lossy pointer.  Also, keyGetItem was extremely inefficient
for cases where a single index key generates multiple "entry" streams, such as
an @@ operator with a multiple-clause tsquery.  The presence of a lossy page
pointer in any one stream defeated its ability to use the opclass
consistentFn, resulting in probing many heap pages that didn't really need to
be visited.  In Artur's example case, a query like
	WHERE tsvector @@ to_tsquery('a & b')
was about 50X slower than the theoretically equivalent
	WHERE tsvector @@ to_tsquery('a') AND tsvector @@ to_tsquery('b')
The way that I chose to fix this was to have GIN call the consistentFn
twice with both TRUE and FALSE values for the in-doubt entry stream,
returning a hit if either call produces TRUE, but not if they both return
FALSE.  The code handles this for the case of a single in-doubt entry stream,
but punts (falling back to the stupid behavior) if there's more than one lossy
reference to the same page.  The idea could be scaled up to deal with multiple
lossy references, but I think that would probably be wasted complexity.  At
least to judge by Artur's example, such cases don't occur often enough to be
worth trying to optimize.

Back-patch to 8.4.  8.3 did not have lossy GIN index pointers, so not
subject to these problems.

Review by Tom Lane.

look through join alias Vars to avoid breaking join queries, and
move the test to someplace where it will catch more possible ways
of calling a function.  We still ought to throw away the whole thing
in favor of a data-type-based solution, but that's not feasible in
the back branches.

This needs to be back-patched further than 9.0, but I don't have time
to do so today.  Committing now so that the fix gets into 9.0beta4.

Transaction aborts now record their LSN to avoid corner case
behaviour in SR/HS, hence change of name of variables and functions.
As pointed out by Fujii Masao. Cosmetic changes only.

Operating directly on the underlying varlena saves palloc and memcpy
overhead, which testing shows to be significant.

Extracted from a larger patch by Alexander Korotkov.

related functions.  Per today's discussion, we will henceforth assume
that datatype I/O functions are either stable or immutable, never volatile.
(This implies in particular that domain CHECK constraint expressions shouldn't
be volatile, since domain_in executes them.)  In turn, functions that execute
the I/O functions of arbitrary datatypes should always be labeled stable.
This affects the labeling of array_to_string, which was unsafely marked
immutable, and record_in, record_out, record_recv, record_send,
domain_in, domain_recv, which were over-conservatively marked volatile.
The array I/O functions were already marked stable, which is correct
per this policy but would have been wrong if we maintained domain_in
as volatile.

Back-patch to 9.0, along with an earlier fix to correctly mark cash_in
and cash_out as stable not immutable (since they depend on lc_monetary).

No catversion bump --- the implications of this are not currently
severe enough to justify a forced initdb.

Block elements with verbatim formatting (literallayout, programlisting,
screen, synopsis) should be aligned at column 0 independent of the surrounding
SGML, because whitespace is significant, and indenting them creates erratic
whitespace in the output.  The CSS stylesheets already take care of indenting
the output.

Assorted markup improvements to go along with it.

assuming that a local char[] array would be aligned on at least a word
boundary.  There are architectures on which that is pretty much guaranteed to
NOT be the case ... and those arches also don't like non-aligned memory
accesses, meaning that log_newpage() would crash if it ever got invoked.
Even on Intel-ish machines there's a potential for a large performance penalty
from doing I/O to an inadequately aligned buffer.  So palloc it instead.

Backpatch to 8.0 --- 7.4 doesn't have this code.

table with a <variablelist> carrying the same information.  Previously the
9.0 documentation was failing to build as a US-size PDF file.  It's quite
obscure what the real problem is or why this avoids it, but we need a hack
now so we can build docs for beta4.

In passing do a bit of editing in the AIX installation docs, in particular
remove a long-obsolete claim that the regression tests are likely to fail.

If a zeroed page is present in the heap, ALTER TABLE .. SET TABLESPACE will
set the LSN and TLI while copying it, which is wrong, and heap_xlog_newpage()
will do the same thing during replay, so the corruption propagates to any
standby.  Note, however, that the bug can't be demonstrated unless archiving
is enabled, since in that case we skip WAL logging altogether, and the LSN/TLI
are not set.

Back-patch to 8.0; prior releases do not have tablespaces.

Analysis and patch by Jeff Davis.  Adjustments for back-branches and minor
wordsmithing by me.

Use this to catch a couple of lock level assignments that slipped
through manual testing, per Peter Eisentraut.

list in ExecLockRows() forgot to allow for the possibility that some of the
rowmarks are for child tables that aren't relevant to the current row.
Per report from Kenichiro Tanaka.

Avoid hard-coding lockmode used for many altering DDL commands, allowing easier
future changes of lock levels. Implementation of initial analysis on DDL
sub-commands, so that many lock levels are now at ShareUpdateExclusiveLock or
ShareRowExclusiveLock, allowing certain DDL not to block reads/writes.
First of number of planned changes in this area; additional docs required
when full project complete.

a pass-by-reference datatype with a nontrivial projection step.
We were using the same memory context for the projection operation as for
the temporary context used by the hashtable routines in execGrouping.c.
However, the hashtable routines feel free to reset their temp context at
any time, which'd lead to destroying input data that was still needed.
Report and diagnosis by Tao Ma.

Back-patch to 8.1, where the problem was introduced by the changes that
allowed us to work with "virtual" tuples instead of materializing intermediate
tuple values everywhere.  The earlier code looks quite similar, but it doesn't
suffer the problem because the data gets copied into another context as a
result of having to materialize ExecProject's output tuple.

These two settings ought to be independent of each other.

Review by Dave Page.

Backpatch to 8.4.

Marc Cousin.  Review by Kevin Grittner.

We used to be consistent about this, but my recent patch to add a
restart_after_crash GUC failed to follow the existing convention.

Report and patch from Fujii Masao.

Author: Alex Hunsaker