Alvaro Herrera [Thu, 13 Feb 2014 22:30:30 +0000 (19:30 -0300)]
Separate multixact freezing parameters from xid's
Previously we were piggybacking on transaction ID parameters to freeze
multixacts; but since there isn't necessarily any relationship between
rates of Xid and multixact consumption, this turns out not to be a good
idea.
Therefore, we now have multixact-specific freezing parameters:
vacuum_multixact_freeze_min_age: when to remove multis as we come across
them in vacuum (default to 5 million, i.e. early in comparison to Xid's
default of 50 million)
vacuum_multixact_freeze_table_age: when to force whole-table scans
instead of scanning only the pages marked as not all visible in
visibility map (default to 150 million, same as for Xids). Whichever of
both which reaches the 150 million mark earlier will cause a whole-table
scan.
autovacuum_multixact_freeze_max_age: when for cause emergency,
uninterruptible whole-table scans (default to 400 million, double as
that for Xids). This means there shouldn't be more frequent emergency
vacuuming than previously, unless multixacts are being used very
rapidly.
Backpatch to 9.3 where multixacts were made to persist enough to require
freezing. To avoid an ABI break in 9.3, VacuumStmt has a couple of
fields in an unnatural place, and StdRdOptions is split in two so that
the newly added fields can go at the end.
Patch by me, reviewed by Robert Haas, with additional input from Andres
Freund and Tom Lane.
Tom Lane [Thu, 13 Feb 2014 19:24:45 +0000 (14:24 -0500)]
Fix length checking for Unicode identifiers containing escapes (U&"...").
We used the length of the input string, not the de-escaped string, as
the trigger for NAMEDATALEN truncation. AFAICS this would only result
in sometimes printing a phony truncation warning; but it's just luck
that there was no worse problem, since we were violating the API spec
for truncate_identifier(). Per bug #9204 from Joshua Yanovski.
This has been wrong since the Unicode-identifier support was added,
so back-patch to all supported branches.
Tom Lane [Thu, 13 Feb 2014 00:09:21 +0000 (19:09 -0500)]
Improve cross-references between minor version release notes.
We have a practice of providing a "bread crumb" trail between the minor
versions where the migration section actually tells you to do something.
Historically that was just plain text, eg, "see the release notes for
9.2.4"; but if you're using a browser or PDF reader, it's a lot nicer
if it's a live hyperlink. So use "<xref>" instead. Any argument against
doing this vanished with the recent decommissioning of plain-text release
notes.
Tom Lane [Wed, 12 Feb 2014 22:50:10 +0000 (17:50 -0500)]
Improve libpq's error recovery for connection loss during COPY.
In pqSendSome, if the connection is already closed at entry, discard any
queued output data before returning. There is no possibility of ever
sending the data, and anyway this corresponds to what we'd do if we'd
detected a hard error while trying to send(). This avoids possible
indefinite bloat of the output buffer if the application keeps trying
to send data (or even just keeps trying to do PQputCopyEnd, as psql
indeed will).
Because PQputCopyEnd won't transition out of PGASYNC_COPY_IN state
until it's successfully queued the COPY END message, and pqPutMsgEnd
doesn't distinguish a queuing failure from a pqSendSome failure,
this omission allowed an infinite loop in psql if the connection closure
occurred when we had at least 8K queued to send. It might be worth
refactoring so that we can make that distinction, but for the moment
the other changes made here seem to offer adequate defenses.
To guard against other variants of this scenario, do not allow
PQgetResult to return a PGRES_COPY_XXX result if the connection is
already known dead. Make sure it returns PGRES_FATAL_ERROR instead.
Per report from Stephen Frost. Back-patch to all active branches.
Tom Lane [Wed, 12 Feb 2014 19:52:20 +0000 (14:52 -0500)]
In XLogReadBufferExtended, don't assume P_NEW yields consecutive pages.
In a database that's not yet reached consistency, it's possible that some
segments of a relation are not full-size but are not the last ones either.
Because of the way smgrnblocks() works, asking for a new page with P_NEW
will fill in the last not-full-size segment --- and if that makes it full
size, the apparent EOF of the relation will increase by more than one page,
so that the next P_NEW request will yield a page past the next consecutive
one. This breaks the relation-extension logic in XLogReadBufferExtended,
possibly allowing a page update to be applied to some page far past where
it was intended to go. This appears to be the explanation for reports of
table bloat on replication slaves compared to their masters, and probably
explains some corrupted-slave reports as well.
Fix the loop to check the page number it actually got, rather than merely
Assert()'ing that dead reckoning got it to the desired place. AFAICT,
there are no other places that make assumptions about exactly which page
they'll get from P_NEW.
Problem identified by Greg Stark, though this is not the same as his
proposed patch.
It's been like this for a long time, so back-patch to all supported
branches.
Magnus Hagander [Sun, 9 Feb 2014 12:10:14 +0000 (13:10 +0100)]
Kill pg_basebackup background process when exiting
If an error occurs in the foreground (backup) process of pg_basebackup,
and we exit in a controlled way, the background process (streaming
xlog process) would stay around and keep streaming.
Tom Lane [Tue, 11 Feb 2014 01:48:12 +0000 (20:48 -0500)]
Don't generate plain-text HISTORY and src/test/regress/README anymore.
Providing this information as plain text was doubtless worth the trouble
ten years ago, but it seems likely that hardly anyone reads it in this
format anymore. And the effort required to maintain these files (in the
form of extra-complex markup rules in the relevant parts of the SGML
documentation) is significant. So, let's stop doing that and rely solely
on the other documentation formats.
Per discussion, the plain-text INSTALL instructions might still be worth
their keep, so we continue to generate that file.
Rather than remove HISTORY and src/test/regress/README from distribution
tarballs entirely, replace them with simple stub files that tell the reader
where to find the relevant documentation. This is mainly to avoid possibly
breaking packaging recipes that expect these files to exist.
Back-patch to all supported branches, because simplifying the markup
requirements for release notes won't help much unless we do it in all
branches.
Tom Lane [Tue, 4 Feb 2014 02:30:05 +0000 (21:30 -0500)]
Improve connection-failure error handling in contrib/postgres_fdw.
postgres_fdw tended to say "unknown error" if it tried to execute a command
on an already-dead connection, because some paths in libpq just return a
null PGresult for such cases. Out-of-memory might result in that, too.
To fix, pass the PGconn to pgfdw_report_error, and look at its
PQerrorMessage() string if we can't get anything out of the PGresult.
Also, fix the transaction-exit logic to reliably drop a dead connection.
It was attempting to do that already, but it assumed that only connection
cache entries with xact_depth > 0 needed to be examined. The folly in that
is that if we fail while issuing START TRANSACTION, we'll not have bumped
xact_depth. (At least for the case I was testing, this fix masks the
other problem; but it still seems like a good idea to have the PGconn
fallback logic.)
Per investigation of bug #9087 from Craig Lucas. Backpatch to 9.3 where
this code was introduced.
Tom Lane [Mon, 3 Feb 2014 19:46:54 +0000 (14:46 -0500)]
Fix *-qualification of named parameters in SQL-language functions.
Given a composite-type parameter named x, "$1.*" worked fine, but "x.*"
not so much. This has been broken since named parameter references were
added in commit 9bff0780cf5be2193a5bad0d3df2dbe143085264, so patch back
to 9.2. Per bug #9085 from Hardy Falk.
Tom Lane [Sat, 1 Feb 2014 23:26:58 +0000 (18:26 -0500)]
Fix some wide-character bugs in the text-search parser.
In p_isdigit and other character class test functions generated by the
p_iswhat macro, the code path for non-C locales with multibyte encodings
contained a bogus pointer cast that would accidentally fail to malfunction
if types wchar_t and wint_t have the same width. Apparently that is true
on most platforms, but not on recent Cygwin releases. Remove the cast,
as it seems completely unnecessary (I think it arose from a false analogy
to the need to cast to unsigned char when dealing with the <ctype.h>
functions). Per bug #8970 from Marco Atzeri.
In the same functions, the code path for C locale with a multibyte encoding
simply ANDed each wide character with 0xFF before passing it to the
corresponding <ctype.h> function. This could result in false positive
answers for some non-ASCII characters, so use a range test instead.
Noted by me while investigating Marco's complaint.
Also, remove some useless though not actually buggy maskings and casts
in the hand-coded p_isalnum and p_isalpha functions, which evidently
got tested a bit more carefully than the macro-generated functions.
Tom Lane [Sat, 1 Feb 2014 21:21:00 +0000 (16:21 -0500)]
Fix some more bugs in signal handlers and process shutdown logic.
WalSndKill was doing things exactly backwards: it should first clear
MyWalSnd (to stop signal handlers from touching MyWalSnd->latch),
then disown the latch, and only then mark the WalSnd struct unused by
clearing its pid field.
Also, WalRcvSigUsr1Handler and worker_spi_sighup failed to preserve
errno, which is surely a requirement for any signal handler.
Per discussion of recent buildfarm failures. Back-patch as far
as the relevant code exists.
Andrew Dunstan [Sat, 1 Feb 2014 20:16:06 +0000 (15:16 -0500)]
Copy the libpq DLL to the bin directory on Mingw and Cygwin.
This has long been done by the MSVC build system, and has caused
confusion in the past when programs like psql have failed to start
because they can't find the DLL. If it's in the same directory as it now
will be they will find it.
Robert Haas [Sat, 1 Feb 2014 02:31:08 +0000 (21:31 -0500)]
Clear MyProc and MyProcSignalState before they become invalid.
Evidence from buildfarm member crake suggests that the new test_shm_mq
module is routinely crashing the server due to the arrival of a SIGUSR1
after the shared memory segment has been unmapped. Although processes
using the new dynamic background worker facilities are more likely to
receive a SIGUSR1 around this time, the problem is also possible on older
branches, so I'm back-patching the parts of this change that apply to
older branches as far as they apply.
It's already generally the case that code checks whether these pointers
are NULL before deferencing them, so the important thing is mostly to
make sure that they do get set to NULL before they become invalid. But
in master, there's one case in procsignal_sigusr1_handler that lacks a
NULL guard, so add that.
Tom Lane [Thu, 30 Jan 2014 23:10:04 +0000 (18:10 -0500)]
Fix potential coredump on bad locale value in pg_upgrade.
Thinko in error report (and a typo in the message text, too). We're
failing anyway, but it would be good to print something useful first.
Noted while reviewing a patch to make pg_upgrade's locale code laxer.
Tom Lane [Thu, 30 Jan 2014 19:51:19 +0000 (14:51 -0500)]
Fix bogus handling of "postponed" lateral quals.
When pulling a "postponed" qual from a LATERAL subquery up into the quals
of an outer join, we must make sure that the postponed qual is included
in those seen by make_outerjoininfo(). Otherwise we might compute a
too-small min_lefthand or min_righthand for the outer join, leading to
"JOIN qualification cannot refer to other relations" failures from
distribute_qual_to_rels. Subtler errors in the created plan seem possible,
too, if the extra qual would only affect join ordering constraints.
Per bug #9041 from David Leverton. Back-patch to 9.3.
Tom Lane [Thu, 30 Jan 2014 01:04:01 +0000 (20:04 -0500)]
Fix unsafe references to errno within error messaging logic.
Various places were supposing that errno could be expected to hold still
within an ereport() nest or similar contexts. This isn't true necessarily,
though in some cases it accidentally failed to fail depending on how the
compiler chanced to order the subexpressions. This class of thinko
explains recent reports of odd failures on clang-built versions, typically
missing or inappropriate HINT fields in messages.
Problem identified by Christian Kruse, who also submitted the patch this
commit is based on. (I fixed a few issues in his patch and found a couple
of additional places with the same disease.)
Back-patch as appropriate to all supported branches.
Stephen Frost [Fri, 24 Jan 2014 20:10:08 +0000 (15:10 -0500)]
Avoid minor leak in parallel pg_dump
During parallel pg_dump, a worker process closing the connection caused
a minor memory leak (particularly minor as we are likely about to exit
anyway). Instead, free the memory in this case prior to returning NULL
to indicate connection closed.
Fujii Masao [Thu, 23 Jan 2014 14:00:30 +0000 (23:00 +0900)]
Fix bugs in PQhost().
In the platform that doesn't support Unix-domain socket, when
neither host nor hostaddr are specified, the default host
'localhost' is used to connect to the server and PQhost() must
return that, but it didn't. This patch fixes PQhost() so that
it returns the default host in that case.
Also this patch fixes PQhost() so that it doesn't return
Unix-domain socket directory path in the platform that doesn't
support Unix-domain socket.
Stephen Frost [Wed, 22 Jan 2014 03:49:22 +0000 (22:49 -0500)]
Allow type_func_name_keywords in even more places
A while back, 2c92edad48796119c83d7dbe6c33425d1924626d allowed
type_func_name_keywords to be used in more places, including role
identifiers. Unfortunately, that commit missed out on cases where
name_list was used for lists-of-roles, eg: for DROP ROLE. This
resulted in the unfortunate situation that you could CREATE a role
with a type_func_name_keywords-allowed identifier, but not DROP it
(directly- ALTER could be used to rename it to something which
could be DROP'd).
This extends allowing type_func_name_keywords to places where role
lists can be used.
Tom Lane [Tue, 21 Jan 2014 21:34:31 +0000 (16:34 -0500)]
Tweak parse location assignment for CURRENT_DATE and related constructs.
All these constructs generate parse trees consisting of a Const and
a run-time type coercion (perhaps a FuncExpr or a CoerceViaIO). Modify
the raw parse output so that we end up with the original token's location
attached to the type coercion node while the Const has location -1;
before, it was the other way around. This makes no difference in terms
of what exprLocation() will say about the parse tree as a whole, so it
should not have any user-visible impact. The point of changing it is that
we do not want contrib/pg_stat_statements to treat these constructs as
replaceable constants. It will do the right thing if the Const has
location -1 rather than a valid location.
This is a pretty ugly hack, but then this code is ugly already; we should
someday replace this translation with special-purpose parse node(s) that
would allow ruleutils.c to reconstruct the original query text.
Robert Haas [Tue, 21 Jan 2014 16:42:37 +0000 (11:42 -0500)]
Fix inadvertent semantics change in last patch to plug memory leaks.
Commit a5bca4ef034f71175d46462963af2329d22068c2 accidentally changed
the semantics when the "skipping missing configuration file" is
emitted, because it forced OK to true instead of leaving the value
untouched.
Stephen Frost [Sat, 18 Jan 2014 23:41:52 +0000 (18:41 -0500)]
Allow SET TABLESPACE to database default
We've always allowed CREATE TABLE to create tables in the database's default
tablespace without checking for CREATE permissions on that tablespace.
Unfortunately, the original implementation of ALTER TABLE ... SET TABLESPACE
didn't pick up on that exception.
This changes ALTER TABLE ... SET TABLESPACE to allow the database's default
tablespace without checking for CREATE rights on that tablespace, just as
CREATE TABLE works today. Users could always do this through a series of
commands (CREATE TABLE ... AS SELECT * FROM ...; DROP TABLE ...; etc), so
let's fix the oversight in SET TABLESPACE's original implementation.
Fix Hot Standby feedback sending when streaming busily.
Commit 6f60fdd7015b032bf49273c99f80913d57eac284 accidentally removed a
call to XLogWalRcvSendHSFeedback() after flushing received WAL to disk.
The consequence is that when walsender is busy streaming WAL, it doesn't
send HS feedback messages. One is sent if nothing is received from the
master for 100ms, but if there's a steady stream of WAL, it never happens.
Tom Lane [Wed, 15 Jan 2014 00:27:57 +0000 (19:27 -0500)]
Improve FILES section of psql reference page.
Primarily, explain where to find the system-wide psqlrc file, per recent
gripe from John Sutton. Do some general wordsmithing and improve the
markup, too.
Also adjust psqlrc.sample so its comments about file location are somewhat
trustworthy. (Not sure why we bother with this file when it's empty,
but whatever.)
Back-patch to 9.2 where the startup file naming scheme was last changed.
Tom Lane [Tue, 14 Jan 2014 22:34:51 +0000 (17:34 -0500)]
Fix multiple bugs in index page locking during hot-standby WAL replay.
In ordinary operation, VACUUM must be careful to take a cleanup lock on
each leaf page of a btree index; this ensures that no indexscans could
still be "in flight" to heap tuples due to be deleted. (Because of
possible index-tuple motion due to concurrent page splits, it's not enough
to lock only the pages we're deleting index tuples from.) In Hot Standby,
the WAL replay process must likewise lock every leaf page. There were
several bugs in the code for that:
* The replay scan might come across unused, all-zero pages in the index.
While btree_xlog_vacuum itself did the right thing (ie, nothing) with
such pages, xlogutils.c supposed that such pages must be corrupt and
would throw an error. This accounts for various reports of replication
failures with "PANIC: WAL contains references to invalid pages". To
fix, add a ReadBufferMode value that instructs XLogReadBufferExtended
not to complain when we're doing this.
* btree_xlog_vacuum performed the extra locking if standbyState ==
STANDBY_SNAPSHOT_READY, but that's not the correct test: we won't open up
for hot standby queries until the database has reached consistency, and
we don't want to do the extra locking till then either, for fear of reading
corrupted pages (which bufmgr.c would complain about). Fix by exporting a
new function from xlog.c that will report whether we're actually in hot
standby replay mode.
* To ensure full coverage of the index in the replay scan, btvacuumscan
would emit a dummy WAL record for the last page of the index, if no
vacuuming work had been done on that page. However, if the last page
of the index is all-zero, that would result in corruption of said page,
since the functions called on it weren't prepared to handle that case.
There's no need to lock any such pages, so change the logic to target
the last normal leaf page instead.
The first two of these bugs were diagnosed by Andres Freund, the other one
by me. Fixes based on ideas from Heikki Linnakangas and myself.
This has been wrong since Hot Standby was introduced, so back-patch to 9.0.
Tom Lane [Mon, 13 Jan 2014 18:07:13 +0000 (13:07 -0500)]
Fix possible buffer overrun in contrib/pg_trgm.
Allow for the possibility that folding a string to lower case makes it
longer (due to replacing a character with a longer multibyte character).
This doesn't change the number of trigrams that will be extracted, but
it does affect the required size of an intermediate buffer in
generate_trgm(). Per bug #8821 from Ufuk Kayserilioglu.
Also install some checks that the input string length is not so large
as to cause overflow in the calculations of palloc request sizes.
Tom Lane [Sun, 12 Jan 2014 00:03:15 +0000 (19:03 -0500)]
Disallow LATERAL references to the target table of an UPDATE/DELETE.
On second thought, commit 0c051c90082da0b7e5bcaf9aabcbd4f361137cdc was
over-hasty: rather than allowing this case, we ought to reject it for now.
That leaves the field clear for a future feature that allows the target
table to be re-specified in the FROM (or USING) clause, which will enable
left-joining the target table to something else. We can then also allow
LATERAL references to such an explicitly re-specified target table.
But allowing them right now will create ambiguities or worse for such a
feature, and it isn't something we documented 9.3 as supporting.
While at it, add a convenience subroutine to avoid having several copies
of the ereport for disalllowed-LATERAL-reference cases.
Tom Lane [Sat, 11 Jan 2014 21:35:30 +0000 (16:35 -0500)]
Fix possible crashes due to using elog/ereport too early in startup.
Per reports from Andres Freund and Luke Campbell, a server failure during
set_pglocale_pgservice results in a segfault rather than a useful error
message, because the infrastructure needed to use ereport hasn't been
initialized; specifically, MemoryContextInit hasn't been called.
One known cause of this is starting the server in a directory it
doesn't have permission to read.
We could try to prevent set_pglocale_pgservice from using anything that
depends on palloc or elog, but that would be messy, and the odds of future
breakage seem high. Moreover there are other things being called in main.c
that look likely to use palloc or elog too --- perhaps those things
shouldn't be there, but they are there today. The best solution seems to
be to move the call of MemoryContextInit to very early in the backend's
real main() function. I've verified that an elog or ereport occurring
immediately after that is now capable of sending something useful to
stderr.
I also added code to elog.c to print something intelligible rather than
just crashing if MemoryContextInit hasn't created the ErrorContext.
This could happen if MemoryContextInit itself fails (due to malloc
failure), and provides some future-proofing against someone trying to
sneak in new code even earlier in server startup.
Back-patch to all supported branches. Since we've only heard reports of
this type of failure recently, it may be that some recent change has made
it more likely to see a crash of this kind; but it sure looks like it's
broken all the way back.
Tom Lane [Sat, 11 Jan 2014 18:41:41 +0000 (13:41 -0500)]
Fix compute_scalar_stats() for case that all values exceed WIDTH_THRESHOLD.
The standard typanalyze functions skip over values whose detoasted size
exceeds WIDTH_THRESHOLD (1024 bytes), so as to limit memory bloat during
ANALYZE. However, we (I think I, actually :-() failed to consider the
possibility that *every* non-null value in a column is too wide. While
compute_minimal_stats() seems to behave reasonably anyway in such a case,
compute_scalar_stats() just fell through and generated no pg_statistic
entry at all. That's unnecessarily pessimistic: we can still produce
valid stanullfrac and stawidth values in such cases, since we do include
too-wide values in the average-width calculation. Furthermore, since the
general assumption in this code is that too-wide values are probably all
distinct from each other, it seems reasonable to set stadistinct to -1
("all distinct").
Per complaint from Kadri Raudsepp. This has been like this since roughly
neolithic times, so back-patch to all supported branches.
Alvaro Herrera [Fri, 10 Jan 2014 21:03:18 +0000 (18:03 -0300)]
Accept pg_upgraded tuples during multixact freezing
The new MultiXact freezing routines introduced by commit 8e9a16ab8f7
neglected to consider tuples that came from a pg_upgrade'd database; a
vacuum run that tried to freeze such tuples would die with an error such
as
ERROR: MultiXactId 11415437 does no longer exist -- apparent wraparound
To fix, ensure that GetMultiXactIdMembers is allowed to return empty
multis when the infomask bits are right, as is done in other callsites.
Per trouble report from F-Secure.
In passing, fix a copy&paste bug reported by Andrey Karpov from VIVA64
from their PVS-Studio static checked, that instead of setting relminmxid
to Invalid, we were setting relfrozenxid twice. Not an important
mistake because that code branch is about relations for which we don't
use the frozenxid/minmxid values at all in the first place, but seems to
warrants a fix nonetheless.
Michael Meskes [Thu, 9 Jan 2014 14:41:51 +0000 (15:41 +0100)]
Fix descriptor output in ECPG.
While working on most platforms the old way sometimes created alignment
problems. This should fix it. Also the regresion tests were updated to test for
the reported case.
Tom Lane [Thu, 9 Jan 2014 01:18:10 +0000 (20:18 -0500)]
Fix "cannot accept a set" error when only some arms of a CASE return a set.
In commit c1352052ef1d4eeb2eb1d822a207ddc2d106cb13, I implemented an
optimization that assumed that a function's argument expressions would
either always return a set (ie multiple rows), or always not. This is
wrong however: we allow CASE expressions in which some arms return a set
of some type and others just return a scalar of that type. There may be
other examples as well. To fix, replace the run-time test of whether an
argument returned a set with a static precheck (expression_returns_set).
This adds a little bit of query startup overhead, but it seems barely
measurable.
Per bug #8228 from David Johnston. This has been broken since 8.0,
so patch all supported branches.
If pause_at_recovery_target is set, recovery pauses *before* applying the
target record, even if recovery_target_inclusive is set. If you then
continue with pg_xlog_replay_resume(), it will apply the target record
before ending recovery. In other words, if you log in while it's paused
and verify that the database looks OK, ending recovery changes its state
again, possibly destroying data that you were tring to salvage with PITR.
Backpatch to 9.1, this has been broken since pause_at_recovery_target was
added.
Fix bug in determining when recovery has reached consistency.
When starting WAL replay from an online checkpoint, the last replayed WAL
record variable was initialized using the checkpoint record's location, even
though the records between the REDO location and the checkpoint record had
not been replayed yet. That was noted as "slightly confusing" but harmless
in the comment, but in some cases, it fooled CheckRecoveryConsistency to
incorrectly conclude that we had already reached a consistent state
immediately at the beginning of WAL replay. That caused the system to accept
read-only connections in hot standby mode too early, and also PANICs with
message "WAL contains references to invalid pages".
Fix by initializing the variables to the REDO location instead.
In 9.2 and above, change CheckRecoveryConsistency() to use
lastReplayedEndRecPtr variable when checking if backup end location has
been reached. It was inconsistently using EndRecPtr for that check, but
lastReplayedEndRecPtr when checking min recovery point. It made no
difference before this patch, because in all the places where
CheckRecoveryConsistency was called the two variables were the same, but
it was always an accident waiting to happen, and would have been wrong
after this patch anyway.
Report and analysis by Tomonari Katsumata, bug #8686. Backpatch to 9.0,
where hot standby was introduced.
Tom Lane [Tue, 7 Jan 2014 20:25:19 +0000 (15:25 -0500)]
Fix LATERAL references to target table of UPDATE/DELETE.
I failed to think much about UPDATE/DELETE when implementing LATERAL :-(.
The implemented behavior ended up being that subqueries in the FROM or
USING clause (respectively) could access the update/delete target table as
though it were a lateral reference; which seems fine if they said LATERAL,
but certainly ought to draw an error if they didn't. Fix it so you get a
suitable error when you omit LATERAL. Per report from Emre Hasegeli.
Magnus Hagander [Tue, 7 Jan 2014 16:47:52 +0000 (17:47 +0100)]
Move permissions check from do_pg_start_backup to pg_start_backup
And the same for do_pg_stop_backup. The code in do_pg_* is not allowed
to access the catalogs. For manual base backups, the permissions
check can be handled in the calling function, and for streaming
base backups only users with the required permissions can get past
the authentication step in the first place.
Reported by Antonin Houska, diagnosed by Andres Freund
Magnus Hagander [Tue, 7 Jan 2014 16:04:40 +0000 (17:04 +0100)]
Avoid including tablespaces inside PGDATA twice in base backups
If a tablespace was crated inside PGDATA it was backed up both as part
of the PGDATA backup and as the backup of the tablespace. Avoid this
by skipping any directory inside PGDATA that contains one of the active
tablespaces.
Tom Lane [Sat, 4 Jan 2014 21:05:20 +0000 (16:05 -0500)]
Fix translatability markings in psql, and add defenses against future bugs.
Several previous commits have added columns to various \d queries without
updating their translate_columns[] arrays, leading to potentially incorrect
translations in NLS-enabled builds. Offenders include commit 893686762
(added prosecdef to \df+), c9ac00e6e (added description to \dc+) and 3b17efdfd (added description to \dC+). Fix those cases back to 9.3 or
9.2 as appropriate.
Since this is evidently more easily missed than one would like, in HEAD
also add an Assert that the supplied array is long enough. This requires
an API change for printQuery(), so it seems inappropriate for back
branches, but presumably all future changes will be tested in HEAD anyway.
In HEAD and 9.3, also clean up a whole lot of sloppiness in the emitted
SQL for \dy (event triggers): lack of translatability due to failing to
pass words-to-be-translated through gettext_noop(), inadequate schema
qualification, and sloppy formatting resulting in unnecessarily ugly
-E output.
Peter Eisentraut and Tom Lane, per bug #8702 from Sergey Burladyan
Alvaro Herrera [Thu, 2 Jan 2014 21:17:29 +0000 (18:17 -0300)]
Handle 5-char filenames in SlruScanDirectory
Original users of slru.c were all producing 4-digit filenames, so that
was all that that code was prepared to handle. Changes to multixact.c
in the course of commit 0ac5ad5134f made pg_multixact/members create
5-digit filenames once a certain threshold was reached, which
SlruScanDirectory wasn't prepared to deal with; in particular,
5-digit-name files were not removed during truncation. Change that
routine to make it aware of those files, and have it process them just
like any others.
Right now, some pg_multixact/members directories will contain a mixture
of 4-char and 5-char filenames. A future commit is expected fix things
so that each slru.c user declares the correct maximum width for the
files it produces, to avoid such unsightly mixtures.
Noticed while investigating bug #8673 reported by Serge Negodyuck.
Alvaro Herrera [Thu, 2 Jan 2014 21:17:07 +0000 (18:17 -0300)]
Wrap multixact/members correctly during extension
In the 9.2 code for extending multixact/members, the logic was very
simple because the number of entries in a members page was a proper
divisor of 2^32, and thus at 2^32 wraparound the logic for page switch
was identical than at any other page boundary. In commit 0ac5ad5134f I
failed to realize this and introduced code that was not able to go over
the 2^32 boundary. Fix that by ensuring that when we reach the last
page of the last segment we correctly zero the initial page of the
initial segment, using correct uint32-wraparound-safe arithmetic.
Noticed while investigating bug #8673 reported by Serge Negodyuck, as
diagnosed by Andres Freund.
Alvaro Herrera [Thu, 2 Jan 2014 21:16:54 +0000 (18:16 -0300)]
Handle wraparound during truncation in multixact/members
In pg_multixact/members, relying on modulo-2^32 arithmetic for
wraparound handling doesn't work all that well. Because we don't
explicitely track wraparound of the allocation counter for members, it
is possible that the "live" area exceeds 2^31 entries; trying to remove
SLRU segments that are "old" according to the original logic might lead
to removal of segments still in use. To fix, have the truncation
routine use a tailored SlruScanDirectory callback that keeps track of
the live area in actual use; that way, when the live range exceeds 2^31
entries, the oldest segments still live will not get removed untimely.
This new SlruScanDir callback needs to take care not to remove segments
that are "in the future": if new SLRU segments appear while the
truncation is ongoing, make sure we don't remove them. This requires
examination of shared memory state to recheck for false positives, but
testing suggests that this doesn't cause a problem. The original coding
didn't suffer from this pitfall because segments created when truncation
is running are never considered to be removable.
Per Andres Freund's investigation of bug #8673 reported by Serge
Negodyuck.
Tom Lane [Mon, 30 Dec 2013 19:00:05 +0000 (14:00 -0500)]
Fix broken support for event triggers as extension members.
CREATE EVENT TRIGGER forgot to mark the event trigger as a member of its
extension, and pg_dump didn't pay any attention anyway when deciding
whether to dump the event trigger. Per report from Moshe Jacobson.
Given the obvious lack of testing here, it's rather astonishing that
ALTER EXTENSION ADD/DROP EVENT TRIGGER work, but they seem to.
Kevin Grittner [Sun, 29 Dec 2013 21:18:22 +0000 (15:18 -0600)]
Don't attempt to limit target database for pg_restore.
There was an apparent attempt to limit the target database for
pg_restore to version 7.1.0 or later. Due to a leading zero this
was interpreted as an octal number, which allowed targets with
version numbers down to 2.87.36. The lowest actual release above
that was 6.0.0, so that was effectively the limit.
Since the success of the restore attempt will depend primarily on
on what statements were generated by the dump run, we don't want
pg_restore trying to guess whether a given target should be allowed
based on version number. Allow a connection to any version. Since
it is very unlikely that anyone would be using a recent version of
pg_restore to restore to a pre-6.0 database, this has little to no
practical impact, but it makes the code less confusing to read.
Issue reported and initial patch suggestion from Joel Jacobson
based on an article by Andrey Karpov reporting on issues found by
PVS-Studio static code analyzer. Final patch based on analysis by
Tom Lane. Back-patch to all supported branches.
Andrew Dunstan [Fri, 27 Dec 2013 22:21:04 +0000 (17:21 -0500)]
Properly detect invalid JSON numbers when generating JSON.
Instead of looking for characters that aren't valid in JSON numbers, we
simply pass the output string through the JSON number parser, and if it
fails the string is quoted. This means among other things that money and
domains over money will be quoted correctly and generate valid JSON.
Fixes bug #8676 reported by Anderson Cristian da Silva.
Backpatched to 9.2 where JSON generation was introduced.
Kevin Grittner [Fri, 27 Dec 2013 21:40:51 +0000 (15:40 -0600)]
Fix misplaced right paren bugs in pgstatfuncs.c.
The bug would only show up if the C sockaddr structure contained
zero in the first byte for a valid address; otherwise it would
fail to fail, which is probably why it went unnoticed for so long.
Patch submitted by Joel Jacobson after seeing an article by Andrey
Karpov in which he reports finding this through static code
analysis using PVS-Studio. While I was at it I moved a definition
of a local variable referenced in the buggy code to a more local
context.
Tom Lane [Tue, 24 Dec 2013 03:18:12 +0000 (22:18 -0500)]
Fix ANALYZE failure on a column that's a domain over a range.
Most other range operations seem to work all right on domains,
but this one not so much, at least not since commit 918eee0c.
Per bug #8684 from Brett Neumeier.
Alvaro Herrera [Fri, 20 Dec 2013 15:37:30 +0000 (12:37 -0300)]
Avoid useless palloc during transaction commit
We can allocate the initial relations-to-drop array when first needed,
instead of at function entry; this avoids allocating it when the
function is not going to do anything, which is most of the time.
Backpatch to 9.3, where this behavior was introduced by commit 279628a0a7cf5.
There's more that could be done here, such as possible reworking of the
code to avoid having to palloc anything, but that doesn't sound as
backpatchable as this relatively minor change.
Per complaint from Noah Misch in 20131031145234.GA621493@tornado.leadboat.com
Alvaro Herrera [Thu, 19 Dec 2013 19:39:59 +0000 (16:39 -0300)]
Optimize updating a row that's locked by same xid
Updating or locking a row that was already locked by the same
transaction under the same Xid caused a MultiXact to be created; but
this is unnecessary, because there's no usefulness in being able to
differentiate two locks by the same transaction. In particular, if a
transaction executed SELECT FOR UPDATE followed by an UPDATE that didn't
modify columns of the key, we would dutifully represent the resulting
combination as a multixact -- even though a single key-update is
sufficient.
Optimize the case so that only the strongest of both locks/updates is
represented in Xmax. This can save some Xmax's from becoming
MultiXacts, which can be a significant optimization.
This missed optimization opportunity was spotted by Andres Freund while
investigating a bug reported by Oliver Seemann in message
CANCipfpfzoYnOz5jj=UZ70_R=CwDHv36dqWSpwsi27vpm1z5sA@mail.gmail.com
and also directly as a performance regression reported by Dong Ye in
message d54b8387.000012d8.00000010@YED-DEVD1.vmware.com
Reportedly, this patch fixes the performance regression.
Since the missing optimization was reported as a significant performance
regression from 9.2, backpatch to 9.3.
Alvaro Herrera [Wed, 18 Dec 2013 16:31:27 +0000 (13:31 -0300)]
Don't ignore tuple locks propagated by our updates
If a tuple was locked by transaction A, and transaction B updated it,
the new version of the tuple created by B would be locked by A, yet
visible only to B; due to an oversight in HeapTupleSatisfiesUpdate, the
lock held by A wouldn't get checked if transaction B later deleted (or
key-updated) the new version of the tuple. This might cause referential
integrity checks to give false positives (that is, allow deletes that
should have been rejected).
This is an easy oversight to have made, because prior to improved tuple
locks in commit 0ac5ad5134f it wasn't possible to have tuples created by
our own transaction that were also locked by remote transactions, and so
locks weren't even considered in that code path.
It is recommended that foreign keys be rechecked manually in bulk after
installing this update, in case some referenced rows are missing with
some referencing row remaining.
Per bug reported by Daniel Wood in
CAPweHKe5QQ1747X2c0tA=5zf4YnS2xcvGf13Opd-1Mq24rF1cQ@mail.gmail.com
Alvaro Herrera [Mon, 16 Dec 2013 14:29:51 +0000 (11:29 -0300)]
Rework tuple freezing protocol
Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by
Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted
update might cause locks in the multi to go ignored by later
transactions. This is because the code depended on a multixact above
their cutoff point not having any lock-only member older than the cutoff
point for Xids, which is easily defeated in READ COMMITTED transactions.
The fix for this involves creating a new MultiXactId when necessary.
But this cannot be done during WAL replay, and moreover multixact
examination requires using CLOG access routines which are not supposed
to be used during WAL replay either; so tuple freezing cannot be done
with the old freeze WAL record. Therefore, separate the freezing
computation from its execution, and change the WAL record to carry all
necessary information. At WAL replay time, it's easy to re-execute
freezing because we don't need to re-compute the new infomask/Xmax
values but just take them from the WAL record.
While at it, restructure the coding to ensure all page changes occur in
a single critical section without much room for failures. The previous
coding wasn't using a critical section, without any explanation as to
why this was acceptable.
In replication scenarios using the 9.3 branch, standby servers must be
upgraded before their master, so that they are prepared to deal with the
new WAL record once the master is upgraded; failure to do so will cause
WAL replay to die with a PANIC message. Later upgrade of the standby
will allow the process to continue where it left off, so there's no
disruption of the data in the standby in any case. Standbys know how to
deal with the old WAL record, so it's okay to keep the master running
the old code for a while.
In master, the old freeze WAL record is gone, for cleanliness' sake;
there's no compatibility concern there.
Backpatch to 9.3, where the original bug was introduced and where the
previous fix was backpatched.
Tatsuo Ishii [Sun, 15 Dec 2013 01:33:06 +0000 (10:33 +0900)]
Add "SHIFT_JIS" as an accepted encoding name for locale checking.
When locale is "ja_JP.SJIS", nl_langinfo(CODESET) returns "SHIFT_JIS"
on some platforms, at least on RedHat Linux. So the encoding/locale
match table (encoding_match_list) needs the entry. Otherwise client
encoding is set to SQL_ASCII.
Tom Lane [Sat, 14 Dec 2013 22:33:56 +0000 (17:33 -0500)]
Fix inherited UPDATE/DELETE with UNION ALL subqueries.
Fix an oversight in commit b3aaf9081a1a95c245fd605dcf02c91b3a5c3a29: we do
indeed need to process the planner's append_rel_list when copying RTE
subqueries, because if any of them were flattenable UNION ALL subqueries,
the append_rel_list shows which subquery RTEs were pulled up out of which
other ones. Without this, UNION ALL subqueries aren't correctly inserted
into the update plans for inheritance child tables after the first one,
typically resulting in no update happening for those child table(s).
Per report from Victor Yegorov.
Experimentation with this case also exposed a fault in commit a7b965382cf0cb30aeacb112572718045e6d4be7: if an inherited UPDATE/DELETE
was proven totally dummy by constraint exclusion, we might arrive at
add_rtes_to_flat_rtable with root->simple_rel_array being NULL. This
should be interpreted as not having any RelOptInfos. I chose to code
the guard as a check against simple_rel_array_size, so as to also
provide some protection against indexing off the end of the array.
Back-patch to 9.2 where the faulty code was added.
Alvaro Herrera [Fri, 13 Dec 2013 20:16:25 +0000 (17:16 -0300)]
Rework MultiXactId cache code
The original performs too poorly; in some scenarios it shows way too
high while profiling. Try to make it a bit smarter to avoid excessive
cosst. In particular, make it have a maximum size, and have entries be
sorted in LRU order; once the max size is reached, evict the oldest
entry to avoid it from growing too large.
Per complaint from Andres Freund in connection with new tuple freezing
code.
Tom Lane [Fri, 13 Dec 2013 19:05:16 +0000 (14:05 -0500)]
Add HOLD/RESUME_INTERRUPTS in HandleCatchupInterrupt/HandleNotifyInterrupt.
This prevents a possible longjmp out of the signal handler if a timeout
or SIGINT occurs while something within the handler has transiently set
ImmediateInterruptOK. For safety we must hold off the timeout or cancel
error until we're back in mainline, or at least till we reach the end of
the signal handler when ImmediateInterruptOK was true at entry. This
syncs these functions with the logic now present in handle_sig_alarm.
AFAICT there is no live bug here in 9.0 and up, because I don't think we
currently can wait for any heavyweight lock inside these functions, and
there is no other code (except read-from-client) that will turn on
ImmediateInterruptOK. However, that was not true pre-9.0: in older
branches ProcessIncomingNotify might block trying to lock pg_listener, and
then a SIGINT could lead to undesirable control flow. It might be all
right anyway given the relatively narrow code ranges in which NOTIFY
interrupts are enabled, but for safety's sake I'm back-patching this.
Tom Lane [Fri, 13 Dec 2013 16:50:15 +0000 (11:50 -0500)]
Don't let timeout interrupts happen unless ImmediateInterruptOK is set.
Serious oversight in commit 16e1b7a1b7f7ffd8a18713e83c8cd72c9ce48e07:
we should not allow an interrupt to take control away from mainline code
except when ImmediateInterruptOK is set. Just to be safe, let's adopt
the same save-clear-restore dance that's been used for many years in
HandleCatchupInterrupt and HandleNotifyInterrupt, so that nothing bad
happens if a timeout handler invokes code that tests or even manipulates
ImmediateInterruptOK.
Per report of "stuck spinlock" failures from Christophe Pettus, though
many other symptoms are possible. Diagnosis by Andres Freund.
Fix WAL-logging of setting the visibility map bit.
The operation that removes the remaining dead tuples from the page must
be WAL-logged before the setting of the VM bit. Otherwise, if you replay
the WAL to between those two records, you end up with the VM bit set, but
the dead tuples are still there.
Tatsuo Ishii [Thu, 12 Dec 2013 10:01:01 +0000 (19:01 +0900)]
Fix progress logging when scale factor is large.
Integer overflow showed minus percent and minus remaining time something like this. 239300000 of 3800000000 tuples (-48%) done (elapsed 226.86 s, remaining -696.10 s).
Tom Lane [Wed, 11 Dec 2013 20:08:35 +0000 (15:08 -0500)]
Tweak placement of explicit ANALYZE commands in the regression tests.
Make the COPY test, which loads most of the large static tables used in
the tests, also explicitly ANALYZE those tables. This allows us to get
rid of various ad-hoc, and rather redundant, ANALYZE commands that had
gotten stuck into various test scripts over time to ensure we got
consistent plan choices. (We could have done a database-wide ANALYZE,
but that would cause stats to get attached to the small static tables
too, which results in plan changes compared to the historical behavior.
I'm not sure that's a good idea, so not going that far for now.)
Back-patch to 9.0, since 9.0 and 9.1 are currently sometimes failing
regression tests for lack of an "ANALYZE tenk1" in the subselect test.
There's no need for this in 8.4 since we didn't print any plans back
then.
Tom Lane [Tue, 10 Dec 2013 21:10:20 +0000 (16:10 -0500)]
Fix possible crash with nested SubLinks.
An expression such as WHERE (... x IN (SELECT ...) ...) IN (SELECT ...)
could produce an invalid plan that results in a crash at execution time,
if the planner attempts to flatten the outer IN into a semi-join.
This happens because convert_testexpr() was not expecting any nested
SubLinks and would wrongly replace any PARAM_SUBLINK Params belonging
to the inner SubLink. (I think the comment denying that this case could
happen was wrong when written; it's certainly been wrong for quite a long
time, since very early versions of the semijoin flattening logic.)
Per report from Teodor Sigaev. Back-patch to all supported branches.
Joe Conway [Sun, 8 Dec 2013 01:00:10 +0000 (17:00 -0800)]
Fix performance regression in dblink connection speed.
Previous commit e5de601267d98c5d60df6de8d436685c7105d149 modified dblink
to ensure client encoding matched the server. However the added
PQsetClientEncoding() call added significant overhead. Restore original
performance in the common case where client encoding already matches
server encoding by doing nothing in that case. Applies to all active
branches.
Issue reported and work sponsored by Zonar Systems.
Alvaro Herrera [Thu, 5 Dec 2013 20:47:51 +0000 (17:47 -0300)]
Fix improper abort during update chain locking
In 247c76a98909, I added some code to do fine-grained checking of
MultiXact status of locking/updating transactions when traversing an
update chain. There was a thinko in that patch which would have the
traversing abort, that is return HeapTupleUpdated, when the other
transaction is a committed lock-only. In this case we should ignore it
and return success instead. Of course, in the case where there is a
committed update, HeapTupleUpdated is the correct return value.
A user-visible symptom of this bug is that in REPEATABLE READ and
SERIALIZABLE transaction isolation modes spurious serializability errors
can occur:
ERROR: could not serialize access due to concurrent update
In order for this to happen, there needs to be a tuple that's key-share-
locked and also updated, and the update must abort; a subsequent
transaction trying to acquire a new lock on that tuple would abort with
the above error. The reason is that the initial FOR KEY SHARE is seen
as committed by the new locking transaction, which triggers this bug.
(If the UPDATE commits, then the serialization error is correctly
reported.)
When running a query in READ COMMITTED mode, what happens is that the
locking is aborted by the HeapTupleUpdated return value, then
EvalPlanQual fetches the newest version of the tuple, which is then the
only version that gets locked. (The second time the tuple is checked
there is no misbehavior on the committed lock-only, because it's not
checked by the code that traverses update chains; so no bug.) Only the
newest version of the tuple is locked, not older ones, but this is
harmless.
The isolation test added by this commit illustrates the desired
behavior, including the proper serialization errors that get thrown.
Tom Lane [Thu, 5 Dec 2013 17:48:31 +0000 (12:48 -0500)]
Clear retry flags properly in replacement OpenSSL sock_write function.
Current OpenSSL code includes a BIO_clear_retry_flags() step in the
sock_write() function. Either we failed to copy the code correctly, or
they added this since we copied it. In any case, lack of the clear step
appears to be the cause of the server lockup after connection loss reported
in bug #8647 from Valentine Gogichashvili. Assume that this is correct
coding for all OpenSSL versions, and hence back-patch to all supported
branches.
Alvaro Herrera [Thu, 5 Dec 2013 15:21:55 +0000 (12:21 -0300)]
Avoid resetting Xmax when it's a multi with an aborted update
HeapTupleSatisfiesUpdate can very easily "forget" tuple locks while
checking the contents of a multixact and finding it contains an aborted
update, by setting the HEAP_XMAX_INVALID bit. This would lead to
concurrent transactions not noticing any previous locks held by
transactions that might still be running, and thus being able to acquire
subsequent locks they wouldn't be normally able to acquire.
This bug was introduced in commit 1ce150b7bb; backpatch this fix to 9.3,
like that commit.
This change reverts the change to the delete-abort-savept isolation test
in 1ce150b7bb, because that behavior change was caused by this bug.
Noticed by Andres Freund while investigating a different issue reported
by Noah Misch.
Insertion to a non-leaf GIN page didn't make a full-page image of the page,
which is wrong. The code used to do it correctly, but was changed (commit 853d1c3103fa961ae6219f0281885b345593d101) because the redo-routine didn't
track incomplete splits correctly when the page was restored from a full
page image. Of course, that was not right way to fix it, the redo routine
should've been fixed instead. The redo-routine was surreptitiously fixed
in 2010 (commit 4016bdef8aded77b4903c457050622a5a1815c16), so all we need
to do now is revert the code that creates the record to its original form.
Kevin Grittner [Sat, 30 Nov 2013 17:33:04 +0000 (11:33 -0600)]
Fix pg_dumpall to work for databases flagged as read-only.
pg_dumpall's charter is to be able to recreate a database cluster's
contents in a virgin installation, but it was failing to honor that
contract if the cluster had any ALTER DATABASE SET
default_transaction_read_only settings. By including a SET command
for the connection for each connection opened by pg_dumpall output,
errors are avoided and the source cluster is successfully
recreated.
There was discussion of whether to also set this for the connection
applying pg_dump output, but it was felt that it was both less
appropriate in that context, and far easier to work around.
Alvaro Herrera [Thu, 28 Nov 2013 22:17:21 +0000 (19:17 -0300)]
Fix a couple of bugs in MultiXactId freezing
Both heap_freeze_tuple() and heap_tuple_needs_freeze() neglected to look
into a multixact to check the members against cutoff_xid. This means
that a very old Xid could survive hidden within a multi, possibly
outliving its CLOG storage. In the distant future, this would cause
clog lookup failures:
ERROR: could not access status of transaction 3883960912
DETAIL: Could not open file "pg_clog/0E78": No such file or directory.
This mostly was problematic when the updating transaction aborted, since
in that case the row wouldn't get pruned away earlier in vacuum and the
multixact could possibly survive for a long time. In many cases, data
that is inaccessible for this reason way can be brought back
heuristically.
As a second bug, heap_freeze_tuple() didn't properly handle multixacts
that need to be frozen according to cutoff_multi, but whose updater xid
is still alive. Instead of preserving the update Xid, it just set Xmax
invalid, which leads to both old and new tuple versions becoming
visible. This is pretty rare in practice, but a real threat
nonetheless. Existing corrupted rows, unfortunately, cannot be repaired
in an automated fashion.
Existing physical replicas might have already incorrectly frozen tuples
because of different behavior than in master, which might only become
apparent in the future once pg_multixact/ is truncated; it is
recommended that all clones be rebuilt after upgrading.
Following code analysis caused by bug report by J Smith in message
CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com
and privately by F-Secure.
Backpatch to 9.3, where freezing of MultiXactIds was introduced.
Analysis and patch by Andres Freund, with some tweaks by Álvaro.
Alvaro Herrera [Fri, 29 Nov 2013 19:08:06 +0000 (16:08 -0300)]
Don't TransactionIdDidAbort in HeapTupleGetUpdateXid
It is dangerous to do so, because some code expects to be able to see what's
the true Xmax even if it is aborted (particularly while traversing HOT
chains). So don't do it, and instead rely on the callers to verify for
abortedness, if necessary.
Several race conditions and bugs fixed in the process. One isolation test
changes the expected output due to these.
This also reverts commit c235a6a589b, which is no longer necessary.
Backpatch to 9.3, where this function was introduced.