Andres Freund [Mon, 29 Jan 2018 20:16:53 +0000 (12:16 -0800)]
Introduce ExecQualAndReset() helper.
It's a common task to evaluate a qual and reset the corresponding
expression context. Currently that requires storing the result of the
qual eval, resetting the context, and then reacting on the result. As
that's awkward several places only reset the context next time through
a node. That's not great, so introduce a helper that evaluates and
resets.
It's a bit ugly that it currently uses MemoryContextReset() instead of
ResetExprContext(), but that seems easier than reordering all of
executor.h.
Author: Andres Freund
Discussion: https://postgr.es/m/20180109222544.f7loxrunqh3xjl5f@alap3.anarazel.de
Tom Lane [Mon, 29 Jan 2018 20:13:07 +0000 (15:13 -0500)]
Save a few bytes by removing useless last argument to SearchCatCacheList.
There's never any value in giving a fully specified cache key to
SearchCatCacheList: you might as well call SearchCatCache instead,
since there could be only one match. So the maximum useful number of
key arguments is one less than the supported number of key columns.
We might as well remove the useless extra argument and save some few
bytes per call site, as well as a cycle or so per call.
I believe the reason it was coded like this is that originally, callers
had to write out all the dummy arguments in each call, and so it seemed
less confusing if SearchCatCache and SearchCatCacheList took the same
number of key arguments. But since commit e26c539e9, callers only write
their live arguments explicitly, making that a non-factor; and there's
surely been enough time for third-party modules to adapt to that coding
style. So this is only an ABI break not an API break for callers.
Per discussion with Oliver Ford, this might also make it less confusing
how to use SearchCatCacheList correctly.
Andres Freund [Wed, 24 Jan 2018 07:20:02 +0000 (23:20 -0800)]
Initialize unused ExprEvalStep fields.
ExecPushExprSlots didn't initialize ExprEvalStep's resvalue/resnull
steps as it didn't use them. That caused wrong valgrind warnings for
an upcoming patch, so zero-intialize.
Also zero-initialize all scratch ExprEvalStep's allocated on the
stack, to avoid issues with similar future omissions of non-critial
data.
Peter Eisentraut [Mon, 29 Jan 2018 19:26:17 +0000 (14:26 -0500)]
doc: Clarify pg_upgrade documentation
Clarify that the restriction against reg* types only applies to table
columns using these types, not to the type appearing in any other way,
for example as a function argument.
Andres Freund [Mon, 29 Jan 2018 19:02:09 +0000 (11:02 -0800)]
Prevent growth of simplehash tables when they're "too empty".
In cases where simplehash tables where filled with either a lot of
conflicting hash-values, or values that hash to consecutive
values (i.e. build "chains") the growth heuristics in d4c62a6b623d6eef88218158e9fa3cf974c6c7e5 could trigger rather
explosively.
To fix that, address some of the reasons (see previous commit) of why
the growth heuristics where needed, and only allow growth when the
table isn't too empty. While that means there's a few cases of bad
input that can be slower, that seems a lot better than running very
quickly out of memory.
Author: Tomas Vondra and Andres Freund, with additional input by
Thomas Munro, Tom Lane Todd A. Cook Reported-By: Todd A. Cook, Tomas Vondra, Thomas Munro
Discussion: https://postgr.es/m/20171127185700.1470.20362@wrigleys.postgresql.org
Backpatch: 10, where simplehash was introduced
Andres Freund [Mon, 29 Jan 2018 19:02:09 +0000 (11:02 -0800)]
Improve bit perturbation in TupleHashTableHash.
The changes in b81b5a96f424531b97cdd1dba97d9d1b9c9d372e did not fully
address the issue, because the bit-mixing of the IV into the final
hash-key didn't prevent clustering in the input-data survive in the
output data.
This didn't cause a lot of problems because of the additional growth
conditions added d4c62a6b623d6eef88218158e9fa3cf974c6c7e5. But as we
want to rein those in due to explosive growth in some edges, this
needs to be fixed.
Author: Andres Freund
Discussion: https://postgr.es/m/20171127185700.1470.20362@wrigleys.postgresql.org
Backpatch: 10, where simplehash was introduced
Tom Lane [Mon, 29 Jan 2018 17:57:09 +0000 (12:57 -0500)]
Avoid misleading psql password prompt when username is multiply specified.
When a password is needed, cases such as
psql -d "postgresql://alice@localhost/testdb" -U bob
would incorrectly prompt for "Password for user bob: ", when actually the
connection will be attempted with username alice. The priority order of
which name to use isn't that important here, but the misleading prompt is.
When we are prompting for a password after initial connection failure,
we can fix this reliably by looking at PQuser(conn) to see how libpq
interpreted the connection arguments. But when we're doing a forced
password prompt because of a -W switch, we can't use that solution.
Fortunately, because the main use of -W is for noninteractive situations,
it's less critical to produce a helpful prompt in such cases. I made
the startup prompt for -W just say "Password: " all the time, rather
than expending extra code on trying to identify which username to use.
In the case of a \c command (after -W has been given), there's already
logic in do_connect that determines whether the "dbname" is a connstring
or URI, so we can avoid lobotomizing the prompt except in cases that
are actually dubious. (We could do similarly in startup.c if anyone
complains, but for now it seems not worthwhile, especially since that
would still be only a partial solution.)
Per bug #15025 from Akos Vandra. Although this is arguably a bug fix,
it doesn't seem worth back-patching. The case where it matters seems
like a very corner-case usage, and someone might complain that we'd
changed the behavior of -W in a minor release.
Tom Lane [Sun, 28 Jan 2018 18:39:07 +0000 (13:39 -0500)]
Add stack-overflow guards in set-operation planning.
create_plan_recurse lacked any stack depth check. This is not per
our normal coding rules, but I'd supposed it was safe because earlier
planner processing is more complex and presumably should eat more
stack. But bug #15033 from Andrew Grossman shows this isn't true,
at least not for queries having the form of a many-thousand-way
INTERSECT stack.
Further testing showed that recurse_set_operations is also capable
of being crashed in this way, since it likewise will recurse to the
bottom of a parsetree before calling any support functions that
might themselves contain any stack checks. However, its stack
consumption is only perhaps a third of create_plan_recurse's.
It's possible that this particular problem with create_plan_recurse can
only manifest in 9.6 and later, since before that we didn't build a Path
tree for set operations. But having seen this example, I now have no
faith in the proposition that create_plan_recurse doesn't need a stack
check, so back-patch to all supported branches.
Tom Lane [Sat, 27 Jan 2018 21:42:28 +0000 (16:42 -0500)]
Update time zone data files to tzdata release 2018c.
DST law changes in Brazil, Sao Tome and Principe. Historical corrections
for Bolivia, Japan, and South Sudan. The "US/Pacific-New" zone has been
removed (it was only a link to America/Los_Angeles anyway).
Tom Lane [Sat, 27 Jan 2018 18:52:24 +0000 (13:52 -0500)]
Avoid crash during EvalPlanQual recheck of an inner indexscan.
Commit 09529a70b changed nodeIndexscan.c and nodeIndexonlyscan.c to
postpone initialization of the indexscan proper until the first tuple
fetch. It overlooked the question of mark/restore behavior, which means
that if some caller attempts to mark the scan before the first tuple fetch,
you get a null pointer dereference.
The only existing user of mark/restore is nodeMergejoin.c, which (somewhat
accidentally) will never attempt to set a mark before the first inner tuple
unless the inner child node is a Material node. Hence the case can't arise
normally, so it seems sufficient to document the assumption at both ends.
However, during an EvalPlanQual recheck, ExecScanFetch doesn't call
IndexNext but just returns the jammed-in test tuple. Therefore, if we're
doing a recheck in a plan tree with a mergejoin with inner indexscan,
it's possible to reach ExecIndexMarkPos with iss_ScanDesc still null,
as reported by Guo Xiang Tan in bug #15032.
Really, when there's a test tuple supplied during an EPQ recheck, touching
the index at all is the wrong thing: rather, the behavior of mark/restore
ought to amount to saving and restoring the es_epqScanDone flag. We can
avoid finding a place to actually save the flag, for the moment, because
given the assumption that no caller will set a mark before fetching a
tuple, es_epqScanDone must always be set by the time we try to mark.
So the actual behavior change required is just to not reach the index
access if a test tuple is supplied.
The set of plan node types that need to consider this issue are those
that support EPQ test tuples (i.e., call ExecScan()) and also support
mark/restore; which is to say, IndexScan, IndexOnlyScan, and perhaps
CustomScan. It's tempting to try to fix the problem in one place by
teaching ExecMarkPos() itself about EPQ; but ExecMarkPos supports some
plan types that aren't Scans, and also it seems risky to make assumptions
about what a CustomScan wants to do here. Also, the most likely future
change here is to decide that we do need to support marks placed before
the first tuple, which would require additional work in IndexScan and
IndexOnlyScan in any case. Hence, fix the EPQ issue in nodeIndexscan.c
and nodeIndexonlyscan.c, accepting the small amount of code duplicated
thereby, and leave it to CustomScan providers to fix this bug if they
have it.
Back-patch to v10 where commit 09529a70b came in. In earlier branches,
the index_markpos() call is a waste of cycles when EPQ is active, but
no more than that, so it doesn't seem appropriate to back-patch further.
Tom Lane [Fri, 26 Jan 2018 23:25:02 +0000 (18:25 -0500)]
Avoid unnecessary use of pg_strcasecmp for already-downcased identifiers.
We have a lot of code in which option names, which from the user's
viewpoint are logically keywords, are passed through the grammar as plain
identifiers, and then matched to string literals during command execution.
This approach avoids making words into lexer keywords unnecessarily. Some
places matched these strings using plain strcmp, some using pg_strcasecmp.
But the latter should be unnecessary since identifiers would have been
downcased on their way through the parser. Aside from any efficiency
concerns (probably not a big factor), the lack of consistency in this area
creates a hazard of subtle bugs due to different places coming to different
conclusions about whether two option names are the same or different.
Hence, standardize on using strcmp() to match any option names that are
expected to have been fed through the parser.
This does create a user-visible behavioral change, which is that while
formerly all of these would work:
alter table foo set (fillfactor = 50);
alter table foo set (FillFactor = 50);
alter table foo set ("fillfactor" = 50);
alter table foo set ("FillFactor" = 50);
now the last case will fail because that double-quoted identifier is
different from the others. However, none of our documentation says that
you can use a quoted identifier in such contexts at all, and we should
discourage doing so since it would break if we ever decide to parse such
constructs as true lexer keywords rather than poor man's substitutes.
So this shouldn't create a significant compatibility issue for users.
Daniel Gustafsson, reviewed by Michael Paquier, small changes by me
Robert Haas [Fri, 26 Jan 2018 20:03:12 +0000 (15:03 -0500)]
Factor some code out of create_grouping_paths.
This is preparatory refactoring to prepare the way for partition-wise
aggregate, which will reuse the new subroutines for child grouping
rels. It also does not seem like a bad idea on general principle,
as the function was getting pretty long.
Jeevan Chalke. The larger patch series of which this patch is a part
was reviewed and tested by Antonin Houska, Rajkumar Raghuwanshi,
Ashutosh Bapat, David Rowley, Dilip Kumar, Konstantin Knizhnik,
Pascal Legrand, and me. Some cosmetic changes by me.
Tom Lane [Fri, 26 Jan 2018 17:25:44 +0000 (12:25 -0500)]
Remove the obsolete WITH clause of CREATE FUNCTION.
This clause was superseded by SQL-standard syntax back in 7.3.
We've kept it around for backwards-compatibility purposes ever since;
but 15 years seems like long enough for that, especially seeing that
there are undocumented weirdnesses in how it interacts with the
SQL-standard syntax for specifying the same options.
Michael Paquier, per an observation by Daniel Gustafsson;
some small cosmetic adjustments to nearby code by me.
Peter Eisentraut [Thu, 25 Jan 2018 13:58:00 +0000 (08:58 -0500)]
Use abstracted SSL API in server connection log messages
The existing "connection authorized" server log messages used OpenSSL
API calls directly, even though similar abstracted API calls exist.
Change to use the latter instead.
Change the function prototype for the functions that return the TLS
version and the cipher to return const char * directly instead of
copying into a buffer. That makes them slightly easier to use.
Add bits= to the message. psql shows that, so we might as well show the
same information on the client and server.
Reviewed-by: Daniel Gustafsson <daniel@yesql.se> Reviewed-by: Michael Paquier <michael.paquier@gmail.com>
Peter Eisentraut [Tue, 23 Jan 2018 16:19:12 +0000 (11:19 -0500)]
Remove byte-masking macros for Datum conversion macros
As the comment there stated, these were needed for old-style
user-defined functions, but since we removed support for those, we don't
need this anymore.
Reviewed-by: Michael Paquier <michael.paquier@gmail.com>
Tom Lane [Thu, 25 Jan 2018 20:27:24 +0000 (15:27 -0500)]
Support --no-comments in pg_dump, pg_dumpall, pg_restore.
We have switches already to suppress other subsidiary object properties,
such as ACLs, security labels, ownership, and tablespaces, so just on
the grounds of symmetry we should allow suppressing comments as well.
Also, commit 0d4e6ed30 added a positive reason to have this feature,
i.e. to allow obtaining the old behavior of selective pg_restore should
anyone desire that.
Recent commits have removed the cases where pg_dump emitted comments on
built-in objects that the restoring user might not have privileges to
comment on, so the original primary motivation for this feature is gone,
but it still seems at least somewhat useful in its own right.
Tom Lane [Thu, 25 Jan 2018 19:26:07 +0000 (14:26 -0500)]
Clean up some aspects of pg_dump/pg_restore item-selection logic.
Ensure that CREATE DATABASE and related commands are issued when, and
only when, --create is specified. Previously there were scenarios
where using selective-dump switches would prevent --create from having
any effect. For example, it would fail to do anything in pg_restore
if the archive file had been made by a selective dump, because there
would be no TOC entry for the database.
Since we don't issue \connect either if we don't issue CREATE DATABASE,
this could result in unexpectedly restoring objects into the wrong
database.
Also fix pg_restore's selective restore logic so that when an object is
selected to be restored, we also restore its ACL, comment, and security
label if any. Previously there was no way to get the latter properties
except through tedious mucking about with a -L file. If, for some
reason, you don't want these properties, you can match the old behavior
by adding --no-acl etc.
While at it, try to make _tocEntryRequired() a little better organized
and better documented.
Alvaro Herrera [Thu, 25 Jan 2018 19:11:51 +0000 (16:11 -0300)]
Ignore partitioned indexes where appropriate
get_relation_info() was too optimistic about opening indexes in
partitioned tables, which would raise errors when any queries were
planned on such tables. Fix by ignoring any indexes of the partitioned
kind.
CLUSTER (and ALTER TABLE CLUSTER ON) had a similar problem. Fix by
disallowing these commands in partitioned tables.
Tom Lane [Thu, 25 Jan 2018 18:54:42 +0000 (13:54 -0500)]
Improve pg_dump's handling of "special" built-in objects.
We had some pretty ad-hoc handling of the public schema and the plpgsql
extension, which are both presumed to exist in template0 but might be
modified or deleted in the database being dumped.
Up to now, by default pg_dump would emit a CREATE EXTENSION IF NOT EXISTS
command as well as a COMMENT command for plpgsql. The usefulness of the
former is questionable, and the latter caused annoying errors in
non-superuser dump/restore scenarios. Let's instead install a rule that
built-in extensions (identified by having low-numbered OIDs) are not to be
dumped. We were doing it that way already in binary-upgrade mode, so this
just makes regular mode behave the same. It remains true that if someone
has installed a non-default ACL on the plpgsql language, that will get
dumped thanks to the pg_init_privs mechanism. This is more consistent with
the handling of built-in objects of other kinds.
Also, change the very ad-hoc mechanism that was used to avoid dumping
creation and comment commands for the public schema. Instead of hardwiring
a test in _printTocEntry(), make use of the DUMP_COMPONENT_ infrastructure
to mark that schema up-front about what we want to do with it. This has
the visible effect that the public schema won't be mentioned in the output
at all, except for updating its ACL if it has a non-default ACL.
Previously, while it was normally not mentioned, --clean mode would drop
and recreate it, again causing headaches for non-superuser usage. This
change likewise makes the public schema less special and more like other
built-in objects.
If plpgsql, or the public schema, has been removed entirely in the source
DB, that situation won't be reproduced in the destination ... but that
was true before.
Peter Eisentraut [Thu, 25 Jan 2018 16:14:24 +0000 (11:14 -0500)]
Update documentation to mention huge pages on other OSes
Previously, the docs implied that only Linux and Windows could use huge
pages. That's not quite true: it's just that we only know how to
request them explicitly on those OSes. Be more explicit about what
huge_pages really does and mention that some OSes may use huge pages
automatically.
Author: Thomas Munro and Catalin Iacob Reviewed-By: Justin Pryzby, Peter Eisentraut
Discussion: https://postgr.es/m/CAEepm=3qzR-hfjepymohuC4XO5phxoSoipOjm6BEhnJHjNR+jg@mail.gmail.com
Peter Eisentraut [Tue, 23 Jan 2018 15:55:40 +0000 (10:55 -0500)]
Remove use of byte-masking macros in record_image_cmp
These were introduced in 4cbb646334b3b998a29abef0d57608d42097e6c9, but
after further analysis and testing, they should not be necessary and
probably weren't the part of that commit that fixed anything.
Reviewed-by: Michael Paquier <michael.paquier@gmail.com>
Peter Eisentraut [Thu, 25 Jan 2018 14:14:24 +0000 (09:14 -0500)]
Allow spaces in connection strings in SSL tests
Connection strings can have items with spaces in them, wrapped in
quotes. The tests however ran a SELECT '$connstr' upon connection which
broke on the embedded quotes. Use dollar quotes on the connstr to
protect against this. This was hit during the development of the macOS
Secure Transport patch, but is independent of it.
Tom Lane [Wed, 24 Jan 2018 04:07:13 +0000 (23:07 -0500)]
Improve implementation of pg_attribute_always_inline.
Avoid compiler warnings on MSVC (which doesn't want to see both
__forceinline and inline) and ancient GCC (which doesn't have
__attribute__((always_inline))).
Don't force inline-ing when building at -O0, as the programmer is probably
hoping for exact source-to-object-line correspondence in that case.
(For the moment this only works for GCC; maybe we can extend it later.)
Make pg_attribute_always_inline be syntactically a drop-in replacement
for inline, rather than an additional wart.
And improve the comments.
Thomas Munro and Michail Nikolaev, small tweaks by me
Tom Lane [Tue, 23 Jan 2018 21:50:34 +0000 (16:50 -0500)]
Teach reparameterize_path() to handle AppendPaths.
If we're inside a lateral subquery, there may be no unparameterized paths
for a particular child relation of an appendrel, in which case we *must*
be able to create similarly-parameterized paths for each other child
relation, else the planner will fail with "could not devise a query plan
for the given query". This means that there are situations where we'd
better be able to reparameterize at least one path for each child.
This calls into question the assumption in reparameterize_path() that
it can just punt if it feels like it. However, the only case that is
known broken right now is where the child is itself an appendrel so that
all its paths are AppendPaths. (I think possibly I disregarded that in
the original coding on the theory that nested appendrels would get folded
together --- but that only happens *after* reparameterize_path(), so it's
not excused from handling a child AppendPath.) Given that this code's been
like this since 9.3 when LATERAL was introduced, it seems likely we'd have
heard of other cases by now if there were a larger problem.
Per report from Elvis Pranskevichus. Back-patch to 9.3.
Robert Haas [Tue, 23 Jan 2018 16:20:18 +0000 (11:20 -0500)]
Update obsolete sentence in README.parallel.
Since 9.6, heavyweight locking is not an abstract and unhandled
concern of the parallel machinery, but rather something to which
we have a specific approach.
Robert Haas [Tue, 23 Jan 2018 16:03:03 +0000 (11:03 -0500)]
Report an ERROR if a parallel worker fails to start properly.
Commit 28724fd90d2f85a0573a8107b48abad062a86d83 fixed things so that
if a background worker fails to start due to fork() failure or because
it is terminated before startup succeeds, BGWH_STOPPED will be
reported. However, that only helps if the code that uses the
background worker machinery notices the change in status, and the code
in parallel.c did not.
To fix that, do two things. First, make sure that when a worker
exits, it triggers the leader to read from error queues. That way, if
a worker which has attached to an error queue exits uncleanly, the
leader is sure to throw some error, either the contents of the
ErrorResponse sent by the worker, or "lost connection to parallel
worker" if it exited without sending one. To cover the case where
the worker never starts up in the first place or exits before
attaching to the error queue, the ParallelContext now keeps track
of which workers have sent at least one message via the error
queue. A worker which sends no messages by the time the parallel
operation finishes will be checked to see whether it exited before
attaching to the error queue; if so, a new error message, "parallel
worker failed to initialize", will be reported. If not, we'll
continue to wait until it either starts up and exits cleanly, starts
up and exits uncleanly, or fails to start, and then take the
appropriate action.
Tom Lane [Tue, 23 Jan 2018 15:55:08 +0000 (10:55 -0500)]
In pg_dump, force reconnection after issuing ALTER DATABASE SET command(s).
The folly of not doing this was exposed by the buildfarm: in some cases,
the GUC settings applied through ALTER DATABASE SET may be essential to
interpreting the reloaded data correctly. Another argument why we can't
really get away with the scheme proposed in commit b3f840120 is that it
cannot work for parallel restore: even if the parent process manages to
hang onto the previous GUC state, worker processes would see the state
post-ALTER-DATABASE. (Perhaps we could have dodged that bullet by
delaying DATABASE PROPERTIES restoration to the end of the run, but
that does nothing for the data semantics problem.)
This leaves us with no solution for the default_transaction_read_only issue
that commit 4bd371f6f intended to work around, other than "you gotta remove
such settings before dumping/upgrading". However, in view of the fact that
parallel restore broke that hack years ago and no one has noticed, it's
fair to question how many people care. I'm unexcited about adding a large
dollop of new complexity to handle that corner case.
This would be a one-liner fix, except it turns out that ReconnectToServer
tries to optimize away "redundant" reconnections. While that may have been
valuable when coded, a quick survey of current callers shows that there are
no cases where that's actually useful, so just remove that check. While at
it, remove the function's useless return value.
Peter Eisentraut [Fri, 19 Jan 2018 15:17:56 +0000 (10:17 -0500)]
Extract common bits from OpenSSL implementation
Some things in be-secure-openssl.c and fe-secure-openssl.c were not
actually specific to OpenSSL but could also be used by other
implementations. In order to avoid copy-and-pasting, move some of that
code to common files.
Peter Eisentraut [Fri, 19 Jan 2018 00:53:22 +0000 (19:53 -0500)]
Move SSL API comments to header files
Move the documentation of the SSL API calls are supposed to do into the
headers files, instead of keeping them in the files for the OpenSSL
implementation. That way, they don't have to be duplicated or be
inconsistent when other implementations are added.
Peter Eisentraut [Fri, 19 Jan 2018 00:12:05 +0000 (19:12 -0500)]
Split out documentation of SSL parameters into their own section
Split the "Authentication and Security" section into two separate
sections "Authentication" and "SSL". The latter part has gotten much
longer over time, and doesn't primarily have to do with authentication.
Also, the row_security parameter was inconsistently categorized, so
clean that up while we're here.
Tom Lane [Mon, 22 Jan 2018 19:09:09 +0000 (14:09 -0500)]
Move handling of database properties from pg_dumpall into pg_dump.
This patch rearranges the division of labor between pg_dump and pg_dumpall
so that pg_dump itself handles all properties attached to a single
database. Notably, a database's ACL (GRANT/REVOKE status) and local GUC
settings established by ALTER DATABASE SET and ALTER ROLE IN DATABASE SET
can be dumped and restored by pg_dump. This is a long-requested
improvement.
"pg_dumpall -g" will now produce only role- and tablespace-related output,
nothing about individual databases. The total output of a regular
pg_dumpall run remains the same.
pg_dump (or pg_restore) will restore database-level properties only when
creating the target database with --create. This applies not only to
ACLs and GUCs but to the other database properties it already handled,
that is database comments and security labels. This is more consistent
and useful, but does represent an incompatibility in the behavior seen
without --create.
(This change makes the proposed patch to have pg_dump use "COMMENT ON
DATABASE CURRENT_DATABASE" unnecessary, since there is no case where
the command is issued that we won't know the true name of the database.
We might still want that patch as a feature in its own right, but pg_dump
no longer needs it.)
pg_dumpall with --clean will now drop and recreate the "postgres" and
"template1" databases in the target cluster, allowing their locale and
encoding settings to be changed if necessary, and providing a cleaner
way to set nondefault tablespaces for them than we had before. This
means that such a script must now always be started in the "postgres"
database; the order of drops and reconnects will not work otherwise.
Without --clean, the script will not adjust any database-level properties
of those two databases (including their comments, ACLs, and security
labels, which it formerly would try to set).
Another minor incompatibility is that the CREATE DATABASE commands in a
pg_dumpall script will now always specify locale and encoding settings.
Formerly those would be omitted if they matched the cluster's default.
While that behavior had some usefulness in some migration scenarios,
it also posed a significant hazard of unwanted locale/encoding changes.
To migrate to another locale/encoding, it's now necessary to use pg_dump
without --create to restore into a database with the desired settings.
Commit 4bd371f6f's hack to emit "SET default_transaction_read_only = off"
is gone: we now dodge that problem by the expedient of not issuing ALTER
DATABASE SET commands until after reconnecting to the target database.
Therefore, such settings won't apply during the restore session.
In passing, improve some shaky grammar in the docs, and add a note pointing
out that pg_dumpall's output can't be expected to load without any errors.
(Someday we might want to fix that, but this is not that patch.)
Haribabu Kommi, reviewed at various times by Andreas Karlsson,
Vaishnavi Prabakaran, and Robert Haas; further hacking by me.
Tom Lane [Mon, 22 Jan 2018 17:37:11 +0000 (12:37 -0500)]
Reorder code in pg_dump to dump comments etc in a uniform order.
Most of the code in pg_dump dumps an object's comment, security label,
and ACL auxiliary TOC entries, in that order, immediately after the
object's main TOC entry, and at least dumpComment's API spec says this
isn't optional. dumpDatabase was significantly violating that when
in binary-upgrade mode, by inserting totally unrelated stuff between.
Also, dumpForeignDataWrapper and dumpForeignServer were being randomly
inconsistent. Reorder code so everybody does it the same.
This may be future-proofing us against some code growing a requirement for
such auxiliary entries to be adjacent to their main entry. But for now
it's just neatnik-ism, so I see no need for back-patch.
Peter Eisentraut [Mon, 22 Jan 2018 17:09:52 +0000 (12:09 -0500)]
PL/Python: Fix tests for older Python versions
Commit 8561e4840c81f7e345be2df170839846814fa004 neglected to handle
older Python versions that don't support the "with" statement. So write
the tests in a way that older versions can handle as well.
Tom Lane [Mon, 22 Jan 2018 17:06:18 +0000 (12:06 -0500)]
Make pg_dump's ACL, sec label, and comment entries reliably identifiable.
_tocEntryRequired() expects that it can identify ACL, SECURITY LABEL,
and COMMENT TOC entries that are for large objects by seeing whether
the tag for them starts with "LARGE OBJECT ". While that works fine
for actual large objects, which are indeed tagged that way, it's
subject to false positives unless every such entry's tag starts with an
appropriate type ID. And in fact it does not work for ACLs, because
up to now we customarily tagged those entries with just the bare name
of the object. This means that an ACL for an object named
"LARGE OBJECT something" would be misclassified as data not schema,
with undesirable results in a schema-only or data-only dump ---
although pg_upgrade seems unaffected, due to the special case for
binary-upgrade mode further down in _tocEntryRequired().
We can fix this by changing all the dumpACL calls to use the label
strings already in use for comments and security labels, which do
follow the convention of starting with an object type indicator.
Well, mostly they follow it. dumpDatabase() got it wrong, using
just the bare database name for those purposes, so that a database
named "LARGE OBJECT something" would similarly be subject to having
its comment or security label dropped or included when not wanted.
Bring that into line too. (Note that up to now, database ACLs have
not been processed by pg_dump, so that this issue doesn't affect them.)
_tocEntryRequired() itself is not free of fault: it was overly liberal
about matching object tags to "LARGE OBJECT " in binary-upgrade mode.
This looks like it is probably harmless because there would be no data
component to strip anyway in that mode, but at best it's trouble
waiting to happen, so tighten that up too.
The possible misclassification of SECURITY LABEL entries for databases is
in principle a security problem, but the opportunities for actual exploits
seem too narrow to be interesting. The other cases seem like just bugs,
since an object owner can change its ACL or comment for himself, he needn't
try to trick someone else into doing it by choosing a strange name.
This has been broken since per-large-object TOC entries were introduced
in 9.0, so back-patch to all supported branches.
Peter Eisentraut [Mon, 22 Jan 2018 13:30:16 +0000 (08:30 -0500)]
Transaction control in PL procedures
In each of the supplied procedural languages (PL/pgSQL, PL/Perl,
PL/Python, PL/Tcl), add language-specific commit and rollback
functions/commands to control transactions in procedures in that
language. Add similar underlying functions to SPI. Some additional
cleanup so that transaction commit or abort doesn't blow away data
structures still used by the procedure call. Add execution context
tracking to CALL and DO statements so that transaction control commands
can only be issued in top-level procedure and block calls, not function
calls or other procedure or block calls.
- SPI
Add a new function SPI_connect_ext() that is like SPI_connect() but
allows passing option flags. The only option flag right now is
SPI_OPT_NONATOMIC. A nonatomic SPI connection can execute transaction
control commands, otherwise it's not allowed. This is meant to be
passed down from CALL and DO statements which themselves know in which
context they are called. A nonatomic SPI connection uses different
memory management. A normal SPI connection allocates its memory in
TopTransactionContext. For nonatomic connections we use PortalContext
instead. As the comment in SPI_connect_ext() (previously SPI_connect())
indicates, one could potentially use PortalContext in all cases, but it
seems safest to leave the existing uses alone, because this stuff is
complicated enough already.
SPI also gets new functions SPI_start_transaction(), SPI_commit(), and
SPI_rollback(), which can be used by PLs to implement their transaction
control logic.
- portalmem.c
Some adjustments were made in the code that cleans up portals at
transaction abort. The portal code could already handle a command
*committing* a transaction and continuing (e.g., VACUUM), but it was not
quite prepared for a command *aborting* a transaction and continuing.
In AtAbort_Portals(), remove the code that marks an active portal as
failed. As the comment there already predicted, this doesn't work if
the running command wants to keep running after transaction abort. And
it's actually not necessary, because pquery.c is careful to run all
portal code in a PG_TRY block and explicitly runs MarkPortalFailed() if
there is an exception. So the code in AtAbort_Portals() is never used
anyway.
In AtAbort_Portals() and AtCleanup_Portals(), we need to be careful not
to clean up active portals too much. This mirrors similar code in
PreCommit_Portals().
- PL/Perl
Gets new functions spi_commit() and spi_rollback()
- PL/pgSQL
Gets new commands COMMIT and ROLLBACK.
Update the PL/SQL porting example in the documentation to reflect that
transactions are now possible in procedures.
- PL/Python
Gets new functions plpy.commit and plpy.rollback.
- PL/Tcl
Gets new commands commit and rollback.
Reviewed-by: Andrew Dunstan <andrew.dunstan@2ndquadrant.com>
Bruce Momjian [Sun, 21 Jan 2018 02:47:02 +0000 (21:47 -0500)]
doc: update intermediate certificate instructions
Document how to properly create root and intermediate certificates using
v3_ca extensions and where to place intermediate certificates so they
are properly transferred to the remote side with the leaf certificate to
link to the remote root certificate. This corrects docs that used to
say that intermediate certificates must be stored with the root
certificate.
Also add instructions on how to create root, intermediate, and leaf
certificates.
Peter Eisentraut [Sat, 20 Jan 2018 13:02:01 +0000 (08:02 -0500)]
Improve type conversion of SPI_processed in Python
The previous code converted SPI_processed to a Python float if it didn't
fit into a Python int. But Python longs have unlimited precision, so
use that instead in all cases.
Apparently, Peter's compiler has faith that the switch test values here
could never not be valid values of their enums. Mine does not, and
I tend to agree with it.
Peter Eisentraut [Fri, 19 Jan 2018 22:22:38 +0000 (17:22 -0500)]
PL/Python: Simplify PLyLong_FromInt64
We don't actually need two code paths, one for 32 bits and one for 64
bits. Since the existing code already assumed that "long long" is
available, we can just use PyLong_FromLongLong() for 64 bits as well.
In Python 2.5 and later, PyLong_FromLong() and PyLong_FromLongLong() use
the same code, so there will be no difference for 64-bit platforms. In
Python 2.4, the code is different, but performance testing showed no
noticeable difference in PL/Python, and that Python version is ancient
anyway.
Robert Haas [Fri, 19 Jan 2018 20:33:06 +0000 (15:33 -0500)]
Allow UPDATE to move rows between partitions.
When an UPDATE causes a row to no longer match the partition
constraint, try to move it to a different partition where it does
match the partition constraint. In essence, the UPDATE is split into
a DELETE from the old partition and an INSERT into the new one. This
can lead to surprising behavior in concurrency scenarios because
EvalPlanQual rechecks won't work as they normally did; the known
problems are documented. (There is a pending patch to improve the
situation further, but it needs more review.)
Amit Khandekar, reviewed and tested by Amit Langote, David Rowley,
Rajkumar Raghuwanshi, Dilip Kumar, Amul Sul, Thomas Munro, Álvaro
Herrera, Amit Kapila, and me. A few final revisions by me.
Alvaro Herrera [Fri, 19 Jan 2018 19:34:44 +0000 (16:34 -0300)]
Fix CompareIndexInfo's attnum comparisons
When an index column is an expression, it makes no sense to compare its
attribute numbers.
This seems to account for remaining buildfarm fallout from 8b08f7d4820f.
At least, it solves the issue in my local 32bit VM -- let's see what the
rest thinks.
AclObjectKind was basically just another enumeration for object types,
and we already have a preferred one for that. It's only used in
aclcheck_error. By using ObjectType instead, we can also give some more
precise error messages, for example "index" instead of "relation".
Reviewed-by: Michael Paquier <michael.paquier@gmail.com>
Peter Eisentraut [Wed, 11 Oct 2017 22:35:19 +0000 (18:35 -0400)]
Replace GrantObjectType with ObjectType
There used to be a lot of different *Type and *Kind symbol groups to
address objects within different commands, most of which have been
replaced by ObjectType, starting with b256f2426433c56b4bea3a8102757749885b81ba. But this conversion was never
done for the ACL commands until now.
This change ends up being just a plain replacement of the types and
symbols, without any code restructuring needed, except deleting some now
redundant code.
Reviewed-by: Michael Paquier <michael.paquier@gmail.com> Reviewed-by: Stephen Frost <sfrost@snowman.net>
Alvaro Herrera [Fri, 19 Jan 2018 14:49:22 +0000 (11:49 -0300)]
Local partitioned indexes
When CREATE INDEX is run on a partitioned table, create catalog entries
for an index on the partitioned table (which is just a placeholder since
the table proper has no data of its own), and recurse to create actual
indexes on the existing partitions; create them in future partitions
also.
As a convenience gadget, if the new index definition matches some
existing index in partitions, these are picked up and used instead of
creating new ones. Whichever way these indexes come about, they become
attached to the index on the parent table and are dropped alongside it,
and cannot be dropped on isolation unless they are detached first.
To support pg_dump'ing these indexes, add commands
CREATE INDEX ON ONLY <table>
(which creates the index on the parent partitioned table, without
recursing) and
ALTER INDEX ATTACH PARTITION
(which is used after the indexes have been created individually on each
partition, to attach them to the parent index). These reconstruct prior
database state exactly.
Reviewed-by: (in alphabetical order) Peter Eisentraut, Robert Haas, Amit
Langote, Jesper Pedersen, Simon Riggs, David Rowley
Discussion: https://postgr.es/m/20171113170646.gzweigyrgg6pwsg4@alvherre.pgsql
Alvaro Herrera [Fri, 19 Jan 2018 13:15:08 +0000 (10:15 -0300)]
Fix StoreCatalogInheritance1 to use 32bit inhseqno
For no apparent reason, this function was using a 16bit-wide inhseqno
value, rather than the correct 32 bit width which is what is stored in
the pg_inherits catalog. This becomes evident if you try to create a
table with more than 65535 parents, because this error appears:
Needless to say, having so many parents is an uncommon situations, which
explains why this error has never been reported despite being having
been introduced with the Postgres95 1.01 sources in commit d31084e9d111:
https://git.postgresql.org/gitweb/?p=postgresql.git;a=blob;f=src/backend/commands/creatinh.c;hb=d31084e9d111#l349
Backpatch all the way back.
David Rowley noticed this while reviewing a patch of mine.
Discussion: https://postgr.es/m/CAKJS1f8Dn7swSEhOWwzZzssW7747YB=2Hi+T7uGud40dur69-g@mail.gmail.com
Robert Haas [Fri, 19 Jan 2018 12:48:44 +0000 (07:48 -0500)]
Transfer state pertaining to pending REINDEX operations to workers.
This will allow the pending patch for parallel CREATE INDEX to work
on system catalogs, and to provide the same level of protection
against use of user indexes while they are being rebuilt that we
have for non-parallel CREATE INDEX.
Peter Eisentraut [Thu, 18 Jan 2018 16:24:07 +0000 (11:24 -0500)]
Add tests for session_replication_role
This was hardly tested at all. The trigger case was lightly tested by
the logical replication tests, but rules and event triggers were not
tested at all.
Tom Lane [Thu, 18 Jan 2018 16:09:44 +0000 (11:09 -0500)]
Extend configure's __int128 test to check for a known gcc bug.
On Sparc64, use of __attribute__(aligned(8)) with __int128 causes faulty
code generation in gcc versions at least through 5.5.0. We can work around
that by disabling use of __int128, so teach configure to test for the bug.
This solution doesn't fix things for the case of cross-compiling with a
buggy compiler; to support that nicely, we'd need to add a manual disable
switch. Unless more such cases turn up, it doesn't seem worth the work.
Affected users could always edit pg_config.h manually.
In passing, fix some typos in the existing configure test for __int128.
They're harmless because we only compile that code not run it, but
they're still confusing for anyone looking at it closely.
This is needed in support of commit 751804998, so back-patch to 9.5
as that was.
Robert Haas [Wed, 17 Jan 2018 21:18:39 +0000 (16:18 -0500)]
postgres_fdw: Avoid 'outer pathkeys do not match mergeclauses' error.
When pushing down a join to a foreign server, postgres_fdw constructs
an alternative plan to be used for any EvalPlanQual rechecks that
prove to be necessary. This plan is stored as the outer subplan of
the Foreign Scan implementing the pushed-down join. Previously, this
alternative plan could have a different nominal sort ordering than its
parent, which seemed OK since there will only be one tuple per base
table anyway in the case of an EvalPlanQual recheck. Actually,
though, it caused a problem if that path was used as a building block
for the EvalPlanQual recheck plan of a higher-level foreign join,
because we could end up with a merge join one of whose inputs was not
labelled with the correct sort order. Repair by injecting an extra
Sort node into the EvalPlanQual recheck plan whenever it would
otherwise fail to be sorted at least as well as its parent Foreign
Scan.
Report by Jeff Janes. Patch by me, reviewed by Tom Lane, who also
provided the test case and comment text.
Tom Lane [Wed, 17 Jan 2018 19:44:15 +0000 (14:44 -0500)]
Remove useless lookup of root partitioned rel in ExecInitModifyTable().
node->partitioned_rels is only set in UPDATE/DELETE cases, but
ExecInitModifyTable only uses its "rel" variable in INSERT cases,
so the extra logic to find the root rel is just a waste of complexity
and cycles.
Simon Riggs [Wed, 17 Jan 2018 11:38:34 +0000 (11:38 +0000)]
Ability to advance replication slots
Ability to advance both physical and logical replication slots using a
new user function pg_replication_slot_advance().
For logical advance that means records are consumed as fast as possible
and changes are not given to output plugin for sending. Makes 2nd phase
(after we reached SNAPBUILD_FULL_SNAPSHOT) of replication slot creation
faster, especially when there are big transactions as the reorder buffer
does not have to deal with data changes and does not have to spill to
disk.
Andrew Dunstan [Wed, 17 Jan 2018 00:07:13 +0000 (19:07 -0500)]
Centralize json and jsonb handling of datetime types
The creates a single function JsonEncodeDateTime which will format these
data types in an efficient and consistent manner. This will be all the
more important when we come to jsonpath so we don't have to implement yet
more code doing the same thing in two more places.
This also extends the code to handle time and timetz types which were
not previously handled specially. This requires exposing the time2tm and
timetz2tm functions.
Tom Lane [Fri, 12 Jan 2018 21:52:49 +0000 (16:52 -0500)]
Fix postgres_fdw to cope with duplicate GROUP BY entries.
Commit 7012b132d, which added the ability to push down aggregates and
grouping to the remote server, wasn't careful to ensure that the remote
server would have the same idea we do about which columns are the grouping
columns, in cases where there are textually identical GROUP BY expressions.
Such cases typically led to "targetlist item has multiple sortgroupref
labels" errors.
To fix this reliably, switch over to using "GROUP BY column-number" syntax
rather than "GROUP BY expression" in transmitted queries, and adjust
foreign_grouping_ok() to be more careful about duplicating the sortgroupref
labeling of the local pathtarget.
Per bug #14890 from Sean Johnston. Back-patch to v10 where the buggy code
was introduced.
Tom Lane [Fri, 12 Jan 2018 20:46:37 +0000 (15:46 -0500)]
Avoid unnecessary failure in SELECT concurrent with ALTER NO INHERIT.
If a query against an inheritance tree runs concurrently with an ALTER
TABLE that's disinheriting one of the tree members, it's possible to get
a "could not find inherited attribute" error because after obtaining lock
on the removed member, make_inh_translation_list sees that its columns
have attinhcount=0 and decides they aren't the columns it's looking for.
An ideal fix, perhaps, would avoid including such a just-removed member
table in the query at all; but there seems no way to accomplish that
without adding expensive catalog rechecks or creating a likelihood of
deadlocks. Instead, let's just drop the check on attinhcount. In this
way, a query that's included a just-disinherited child will still
succeed, which is not a completely unreasonable behavior.
This problem has existed for a long time, so back-patch to all supported
branches. Also add an isolation test verifying related behaviors.
Patch by me; the new isolation test is based on Kyotaro Horiguchi's work.
Tom Lane [Fri, 12 Jan 2018 17:24:50 +0000 (12:24 -0500)]
Fix incorrect handling of subquery pullup in the presence of grouping sets.
If we flatten a subquery whose target list contains constants or
expressions, when those output columns are used in GROUPING SET columns,
the planner was capable of doing the wrong thing by merging a pulled-up
expression into the surrounding expression during const-simplification.
Then the late processing that attempts to match subexpressions to grouping
sets would fail to match those subexpressions to grouping sets, with the
effect that they'd not go to null when expected.
To fix, wrap such subquery outputs in PlaceHolderVars, ensuring that
they preserve their separate identity throughout the planner's expression
processing. This is a bit of a band-aid, because the wrapper defeats
const-simplification even in places where it would be safe to allow.
But a nicer fix would likely be too invasive to back-patch, and the
consequences of the missed optimizations probably aren't large in most
cases.
Back-patch to 9.5 where grouping sets were introduced.
Heikki Linnakangas, with small mods and better test cases by me;
additional review by Andrew Gierth
Alvaro Herrera [Fri, 12 Jan 2018 14:21:42 +0000 (11:21 -0300)]
Remove hard-coded schema knowledge about pg_attribute from genbki.pl
Add the ability to label a column's default value in the catalog header,
and implement this for pg_attribute. A new function in Catalog.pm is
used to fill in a tuple with defaults. The build process will complain
loudly if a catalog entry is incomplete,
Commit 8137f2c3232 labeled variable length columns for the C preprocessor.
Expose that label to genbki.pl so we can exclude those columns from schema
macros in a general fashion. Also, format schema macro entries according
to their types.
This means slightly less code maintenance, but more importantly it's a
proving ground for mechanisms intended to be used in later commits.
While at it, I (Álvaro) couldn't resist making some changes in
genbki.pl: rename some functions to actually indicate their purpose
instead of actively misleading onlookers; and don't iterate on the whole
of pg_type to find the entry for each catalog row, using a hash instead
of an array.
Author: John Naylor, some changes by Álvaro Herrera
Discussion: https://postgr.es/m/CAJVSVGVJHwD8sfDfZW9TbCHWKf=C1YDRM-rF=2JenRU_y+VcFg@mail.gmail.com
Refactor subscription tests to use PostgresNode's wait_for_catchup
This was nearly the same code. Extend wait_for_catchup to allow waiting
for pg_current_wal_lsn() and use that in the subscription tests. Also
change one use in the pg_rewind tests to use this.
Also remove some broken code in wait_for_catchup and
wait_for_slot_catchup. The error message in case the waiting failed
wanted to show the current LSN, but the way it was written never
worked. So since nobody ever cared, just remove it.
Reviewed-by: Michael Paquier <michael.paquier@gmail.com>
Tom Lane [Thu, 11 Jan 2018 17:16:18 +0000 (12:16 -0500)]
Add QueryEnvironment to ExplainOneQuery_hook's parameter list.
This should have been done in commit 18ce3a4ab, which added that parameter
to ExplainOneQuery, but it was overlooked. This makes it impossible for
a user of the hook to pass the queryEnv down to ExplainOnePlan.
It's too late to change this API in v10, I suppose, but fortunately
passing NULL to ExplainOnePlan will work in nearly all interesting
cases in v10. That might not be true forever, so we'd better fix it.
Teodor Sigaev [Thu, 11 Jan 2018 11:49:36 +0000 (14:49 +0300)]
llow negative coordinate for ~> (cube, int) operator
~> (cube, int) operator was especially designed for knn-gist search.
However, knn-gist supports only ascending ordering of results. Nevertheless
it would be useful to support descending ordering by ~> (cube, int) operator.
We provide workaround for that: negative coordinate give us inversed value
of corresponding cube bound. Therefore, knn search using negative coordinate
gives us an effect of descending ordering by cube bound.
Author: Alexander Korotkov
Reviewed by: Tomas Vondra, Andrey Borodin
Discussion: https://www.postgresql.org/message-id/flat/a9657f6a-b497-36ff-e56-482a2c7e3292@2ndquadrant.com
Teodor Sigaev [Thu, 11 Jan 2018 11:41:14 +0000 (14:41 +0300)]
Fix behavior of ~> (cube, int) operator
~> (cube, int) operator was especially designed for knn-gist search.
However, it appears that knn-gist search can't work correctly with current
behavior of this operator when dataset contains cubes of variable
dimensionality. In this case, the same value of second operator argument
can point to different dimension depending on dimensionality of particular cube.
Such behavior is incompatible with gist indexing of cubes, and knn-gist doesn't
work correctly for it.
This patch changes behavior of ~> (cube, int) operator by introducing dimension
numbering where value of second argument unambiguously identifies number of
dimension. With new behavior, this operator can be correctly supported by
knn-gist. Relevant changes to cube operator class are also included.
Backpatch to v9.6 where operator was introduced.
Since behavior of ~> (cube, int) operator is changed, depending entities
must be refreshed after upgrade. Such as, expression indexes using this
operator must be reindexed, materialized views must be rebuilt, stored
procedures and client code must be revised to correctly use new behavior.
That should be mentioned in release notes.
Noticed by: Tomas Vondra
Author: Alexander Korotkov
Reviewed by: Tomas Vondra, Andrey Borodin
Discussion: https://www.postgresql.org/message-id/flat/a9657f6a-b497-36ff-e56-482a2c7e3292@2ndquadrant.com
Tom Lane [Wed, 10 Jan 2018 22:13:29 +0000 (17:13 -0500)]
Fix sample INSTR() functions in the plpgsql documentation.
These functions are stated to be Oracle-compatible, but they weren't.
Yugo Nagata noticed that while our code returns zero for a zero or
negative fourth parameter (occur_index), Oracle throws an error.
Further testing by me showed that there was also a discrepancy in the
interpretation of a negative third parameter (beg_index): Oracle thinks
that a negative beg_index indicates the last place where the target
substring can *begin*, whereas our code thinks it is the last place
where the target can *end*.
Adjust the sample code to behave like Oracle in both these respects.
Also change it to be a CDATA[] section, simplifying copying-and-pasting
out of the documentation source file. And fix minor problems in the
introductory comment, which wasn't very complete or accurate.
Back-patch to all supported branches. Although this patch only touches
documentation, we should probably call it out as a bug fix in the next
minor release notes, since users who have adopted the functions will
likely want to update their versions.
Peter Eisentraut [Tue, 12 Dec 2017 15:26:47 +0000 (10:26 -0500)]
Use portal pinning in PL/Perl and PL/Python
PL/pgSQL "pins" internally generated portals so that user code cannot
close them by guessing their names. Add this functionality to PL/Perl
and PL/Python as well, preventing users from manually closing cursors
created by spi_query and plpy.cursor, respectively. (PL/Tcl does not
currently offer any cursor functionality.)
Tom Lane [Wed, 10 Jan 2018 20:50:54 +0000 (15:50 -0500)]
Remove dubious micro-optimization in ckpt_buforder_comparator().
It seems incorrect to assume that the list of CkptSortItems can never
contain duplicate page numbers: concurrent activity could result in some
page getting dropped from a low-numbered buffer and later loaded into a
high-numbered buffer while BufferSync is scanning the buffer pool.
If that happened, the comparator would give self-inconsistent results,
potentially confusing qsort(). Saving one comparison step is not worth
possibly getting the sort wrong.
So far as I can tell, nothing would actually go wrong given our current
implementation of qsort(). It might get a bit slower than expected
if there were a large number of duplicates of one value, but that's
surely a probability-epsilon case. Still, the comment is wrong,
and if we ever switched to another sort implementation it might be
less forgiving.
In passing, avoid casting away const-ness of the argument pointers;
I've not seen any compiler complaints from that, but it seems likely
that some compilers would not like it.
Back-patch to 9.6 where this code came in, just in case I've underestimated
the possible consequences.