As usual, the release notes for other branches will be made by cutting
these down, but put them up for community review first. Note that a
fair percentage of the entries apply only to prior branches because
their issue was already fixed in 12.0. I'll remove those from the 12.1
list later.
Currently, postgres_fdw does not support preparing a remote transaction
for two-phase commit even in the case where the remote transaction is
read-only, but the old error message appeared to imply that that was not
supported only if the remote transaction modified remote tables. Change
the message so as to include the case where the remote transaction is
read-only.
Also fix a comment above the message.
Also add a note about the lack of supporting PREPARE TRANSACTION to the
postgres_fdw documentation.
Reported-by: Gilles Darold
Author: Gilles Darold and Etsuro Fujita
Reviewed-by: Michael Paquier and Kyotaro Horiguchi
Backpatch-through: 9.4
Discussion: https://postgr.es/m/08600ed3-3084-be70-65ba-279ab19618a5%40darold.net
Declaring this in the client-visible header ecpglib.h was a pretty
poor decision. It's not meant to be application-callable (and if
it was, putting it outside the extern "C" { ... } wrapper means
that C++ clients would fail to call it). And the declaration would
not even compile for a client, anyway, since it would not have the
macro pg_attribute_format_arg(). Fortunately, it seems that no
clients have tried to include this header with ENABLE_NLS defined,
or we'd have gotten complaints about that. But we have no business
putting such a restriction on client code.
Move the declaration to ecpglib_extern.h, since in fact nothing
outside src/interfaces/ecpg/ecpglib/ needs to call it.
The practical effect of this is just that clients can now safely
#include ecpglib.h while having ENABLE_NLS defined, but that seems
like enough of a reason to back-patch it.
Discussion: https://postgr.es/m/20590.1573069709@sss.pgh.pa.us
SET CONSTRAINTS ... DEFERRED failed on partitioned tables, because of a
sanity check that ensures that the affected constraints have triggers.
On partitioned tables, the triggers are in the leaf partitions, not in
the partitioned relations themselves, so the sanity check fails.
Removing the sanity check solves the problem, because the code needed to
support the case is already there.
Backpatch to 11.
Note: deferred unique constraints are not affected by this bug, because
they do have triggers in the parent partitioned table. I did not add a
test for this scenario.
Discussion: https://postgr.es/m/20191105212915.GA11324@alvherre.pgsql
This patch adopts the overflow check logic introduced by commit cbdb8b4c0
into two more places. interval_mul() failed to notice if it computed a
new microseconds value that was one more than INT64_MAX, and pgbench's
double-to-int64 logic had the same sorts of edge-case problems that
cbdb8b4c0 fixed in the core code.
To make this easier to get right in future, put the guts of the checks
into new macros in c.h, and add commentary about how to use the macros
correctly.
Back-patch to all supported branches, as we did with the previous fix.
Yuya Watari
Discussion: https://postgr.es/m/CAJ2pMkbkkFw2hb9Qb1Zj8d06EhWAQXFLy73St4qWv6aX=vqnjw@mail.gmail.com
If there is the WAL page that the continuation WAL record just fits within
(i.e., the continuation record ends just at the end of the page) and
the LSN in such page is specified with -s option, previously pg_waldump
caused an assertion failure. The cause of this assertion failure was that
XLogFindNextRecord() that pg_waldump -s calls mistakenly handled
such special WAL page.
This commit changes XLogFindNextRecord() so that it can handle
such WAL page correctly.
Back-patch to all supported versions.
Author: Andrey Lepikhov
Reviewed-by: Fujii Masao, Michael Paquier
Discussion: https://postgr.es/m/99303554-5dd5-06e6-f943-b3005ccd6edd@postgrespro.ru
The docs do say which GUCs can be changed only by superusers, but we
forgot to mention this for the new log_transaction_sample_rate. This
GUC was introduced in PostgreSQL 12, so backpatch accordingly.
Author: Adrien Nayrat
Backpatch-through: 12
Avoid creating transiently-inconsistent slot states where possible,
by not setting TTS_FLAG_SHOULDFREE until after the slot actually has
a free'able tuple pointer, and by making sure that we reset tts_nvalid
and related derived state before we replace the tuple contents. This
would only matter if something were to examine the slot after we'd
suffered some kind of error (e.g. out of memory) while manipulating
the slot. We typically don't do that, so these changes might just be
cosmetic --- but even if so, it seems like good future-proofing.
Also remove some redundant Asserts, and add a couple for consistency.
Back-patch to v12 where all this code was rewritten.
Discussion: https://postgr.es/m/16095-c3ff2e5283b8dba5@postgresql.org
Since commit d26a810eb, we've defined bool as being either _Bool from
<stdbool.h>, or "unsigned char"; but that commit overlooked the fact
that probes.d has "#define bool char". For consistency, make it say
"unsigned char" instead. This should be strictly a cosmetic change,
but it seems best to be in sync.
Formally, in the now-normal case where we're using <stdbool.h>, it'd
be better to write "#define bool _Bool". However, then we'd need
some build infrastructure to inject that configuration choice into
probes.d, and it doesn't seem worth the trouble. We only use
<stdbool.h> if sizeof(_Bool) is 1, so having DTrace think that
bool parameters are "unsigned char" should be close enough.
Back-patch to v12 where d26a810eb came in.
Discussion: https://postgr.es/m/CAA4eK1LmaKO7Du9M9Lo=kxGU8sB6aL8fa3sF6z6d5yYYVe3BuQ@mail.gmail.com
When sending data for logical decoding using the streaming replication
protocol via a WAL sender, the timestamp of the sent write message is
allocated at the beginning of the message when preparing for the write,
and actually computed when the write message is ready to be sent.
The timestamp was getting computed after sending the message. This
impacts anything using logical decoding, causing for example logical
replication to report mostly NULL for last_msg_send_time in
pg_stat_subscription.
This commit makes sure that the timestamp is computed before sending the
message. This is wrong since 5a991ef, so backpatch down to 9.4.
Author: Jeff Janes
Discussion: https://postgr.es/m/CAMkU=1z=WMn8jt7iEdC5sYNaPgAgOASb_OW5JYv-vMdYaJSL-w@mail.gmail.com
Backpatch-through: 9.4
WindowAgg will potentially store large numbers of input rows into
tuplestores to allow access to other rows in the frame. If the input
is coming via an explicit Sort node, then unneeded columns will
already have been discarded (since Sort requests a small tlist); but
there are idioms like COUNT(*) OVER () that result in the input not
being sorted at all, and cases where the input is being sorted by some
means other than a Sort; if we don't request a small tlist, then
WindowAgg's storage requirement is inflated by the unneeded columns.
Backpatch back to 9.6, where the current tlist handling was added.
(Prior to that, WindowAgg would always use a small tlist.)
Discussion: https://postgr.es/m/87a7ator8n.fsf@news-spur.riddles.org.uk
For a long time (since commit aed378e8d) we have had a policy to log
nothing about a connection if the client disconnects when challenged
for a password. This is because libpq-using clients will typically
do that, and then come back for a new connection attempt once they've
collected a password from their user, so that logging the abandoned
connection attempt will just result in log spam. However, this did
not work well for PAM authentication: the bottom-level function
pam_passwd_conv_proc() was on board with it, but we logged messages
at higher levels anyway, for lack of any reporting mechanism.
Add a flag and tweak the logic so that the case is silent, as it is
for other password-using auth mechanisms.
Per complaint from Yoann La Cancellera. It's been like this for awhile,
so back-patch to all supported branches.
Discussion: https://postgr.es/m/CACP=ajbrFFYUrLyJBLV8=q+eNCapa1xDEyvXhMoYrNphs-xqPw@mail.gmail.com
get_relkind_objtype, and hence get_object_type, failed when applied to a
toast table. This is not a good thing, because it prevents reporting of
perfectly legitimate permissions errors. (At present, these functions
are in fact *only* used to determine the ObjectType argument for
acl_error() calls.) It seems best to have them fall back to returning
OBJECT_TABLE in every case where they can't determine an object type
for a pg_class entry, so do that.
In passing, make some edits to alter.c to make it more obvious that
those calls of get_object_type() are used only for error reporting.
This might save a few cycles in the non-error code path, too.
Back-patch to v11 where this issue originated.
John Hsu, Michael Paquier, Tom Lane
Discussion: https://postgr.es/m/C652D3DF-2B0C-4128-9420-FB5379F6B1E4@amazon.com
Commit d25ea0127 got rid of what I thought were entirely unnecessary
derived child expressions in EquivalenceClasses for EC members that
mention multiple baserels. But it turns out that some of the child
expressions that code created are necessary for partitionwise joins,
else we fail to find matching pathkeys for Sort nodes. (This happens
only for certain shapes of the resulting plan; it may be that
partitionwise aggregation is also necessary to show the failure,
though I'm not sure of that.)
Reverting that commit entirely would be quite painful performance-wise
for large partition sets. So instead, add code that explicitly
generates child expressions that match only partitionwise child join
rels we have actually generated.
Per report from Justin Pryzby. (Amit Langote noticed the problem
earlier, though it's not clear if he recognized then that it could
result in a planner error, not merely failure to exploit partitionwise
join, in the code as-committed.) Back-patch to v12 where commit
d25ea0127 came in.
Amit Langote, with lots of kibitzing from me
Discussion: https://postgr.es/m/CA+HiwqG2WVUGmLJqtR0tPFhniO=H=9qQ+Z3L_ZC+Y3-EVQHFGg@mail.gmail.com
Discussion: https://postgr.es/m/20191011143703.GN10470@telsasoft.com
Since 898e5e32, this command uses partially ShareUpdateExclusiveLock,
but the docs did not get the call.
Author: Justin Pryzby
Reviewed-by: Amit Langote, Álvaro Herrera, Michael Paquier
Discussion: https://postgr.es/m/20191028001207.GB23808@telsasoft.com
Backpatch-through: 12
This clarifies more how to use and how to take advantage of constraints
when attaching a new partition.
Author: Justin Pryzby
Reviewed-by: Amit Langote, Álvaro Herrera, Michael Paquier
Discussion: https://postgr.es/m/20191028001207.GB23808@telsasoft.com
Backpatch-through: 10
The code only compared two triggers' names and namespaces (the latter
being the owning table's schema). This could result in falling back
to an OID-based sort of similarly-named triggers on different tables.
We prefer to avoid that, so add a comparison of the table names too.
(The sort order is thus table namespace, trigger name, table name,
which is a bit odd, but it doesn't seem worth contorting the code
to work around that.)
Likewise for policy objects, in 9.5 and up.
Complaint and fix by Benjie Gillam. Back-patch to all supported
branches.
Discussion: https://postgr.es/m/CAMThMzEEt2mvBbPgCaZ1Ap1N-moGn=Edxmadddjq89WG4NpPtQ@mail.gmail.com
Rearrange the logic in record_image_cmp() and datum_image_eq() to
error out on unexpected typlens (either not supported there or
completely invalid due to corruption). Barring corruption, this is
not possible today but it seems more future-proof and robust to fix
this.
Reported-by: Peter Geoghegan <pg@bowt.ie>
Commit 8af1624e3 introduced a warning about possibly returning
without a value, on compilers that don't realize that ereport(ERROR)
doesn't return. Tweak the code to avoid that.
Per buildfarm. Back-patch to 9.6, like the aforesaid commit.
Using incorrect, or just mismatched, dictionary and affix files
could result in a crash, due to failure to cross-check offsets
obtained from the file. Add necessary validation, as well as
some Asserts for future-proofing.
Per bug #16050 from Alexander Lakhin. Back-patch to 9.6 where the
problem was introduced.
Arthur Zakirov, per initial investigation by Tomas Vondra
Discussion: https://postgr.es/m/16050-024ae722464ab604@postgresql.org
Discussion: https://postgr.es/m/20191013012610.2p2fp3zzpoav7jzf@development
When using CREATE TABLE for a new partition, the partitioned indexes of
the parent are created automatically in a fashion similar to LIKE
INDEXES. The new partition and its parent use a mapping for attribute
numbers for this operation, and while the mapping was correctly built,
its length was defined as the number of attributes of the newly-created
child, and not the parent. If the parent includes dropped columns, this
could cause failures.
This is wrong since 8b08f7d which has introduced the concept of
partitioned indexes, so backpatch down to 11.
Reported-by: Wyatt Alt
Author: Michael Paquier
Reviewed-by: Amit Langote
Discussion: https://postgr.es/m/CAGem3qCcRmhbs4jYMkenYNfP2kEusDXvTfw-q+eOhM0zTceG-g@mail.gmail.com
Backpatch-through: 11
When a backend exits, it gets deleted from the syncrep queue if present.
The queue was checked without SyncRepLock taken in exclusive mode, so it
would have been possible for a backend to remove itself after a WAL
sender already did the job. Fix this issue based on a suggestion from
Fujii Masao, by first checking the queue without the lock. Then, if the
backend is present in the queue, take the lock and perform an additional
lookup check before doing the element deletion.
Author: Dongming Liu
Reviewed-by: Kyotaro Horiguchi, Fujii Masao, Michael Paquier
Discussion: https://postgr.es/m/a0806273-8bbb-43b3-bbe1-c45a58f6ae21.lingce.ldm@alibaba-inc.com
Backpatch-through: 9.4
The additional newline seems to have accidentally been introduced in
2c03216d83, in 9.5. The newline is only issued when an FPW is
present for the block reference.
While there could be an argument that removing the newlines in the
back branches could cause a problem for somebody parsing the
pg_waldump output, the likelihood of that seems small enough. It seems
at least equally likely that the randomness of when newlines are
issued causes problems.
Author: Andres Freund
Discussion: https://postgr.es/m/20191029233341.4gnyau7e5v2lh5sc@alap3.anarazel.de
Backpatch: 9.5, like 2c03216d83.
When cancelling REINDEX CONCURRENTLY after swapping the old and new
indexes (for example interruption at step 5), the old index remains
around and is marked as invalid. The old index should also be manually
droppable to clean up the parent relation from any invalid indexes still
remaining. For a partition index reindexed, pg_class.relispartition was
not getting updated, causing the index to not be droppable as DROP INDEX
would look for dependencies in a partition tree, which do not exist
anymore after the swap phase is done.
The fix here is simple: when swapping the old and new indexes, make sure
that pg_class.relispartition is correctly switched, similarly to what is
done for the index name.
Reported-by: Justin Pryzby
Author: Michael Paquier
Discussion: https://postgr.es/m/20191015164047.GA22729@telsasoft.com
Backpatch-through: 12
Commit 9556aa01c rearranged the innards of text_position() in a way
that would make it not work for empty search strings. Which is fine,
because all callers of that code special-case an empty pattern in
some way. However, the primary use-case (text_position itself) got
special-cased incorrectly: historically it's returned 1 not 0 for
an empty search string. Restore the historical behavior.
Per complaint from Austin Drenski (via Shay Rojansky).
Back-patch to v12 where it got broken.
Discussion: https://postgr.es/m/CADT4RqAz7oN4vkPir86Kg1_mQBmBxCp-L_=9vRpgSNPJf0KRkw@mail.gmail.com
When swapping the dependencies of the old and new indexes, the code has
been correctly switching all links in pg_depend from the old to the new
index for both referencing and referenced entries. However it forgot
the fact that the new index may itself have existing entries in
pg_depend, like references to the parent table attributes. This
resulted in duplicated entries in pg_depend after running REINDEX
CONCURRENTLY.
Fix this problem by removing any existing entries in pg_depend for the
new index before switching the dependencies of the old index to the new
one. More regression tests are added to check the consistency of
entries in pg_depend for indexes, including partitions.
Author: Michael Paquier
Discussion: https://postgr.es/m/20191025064318.GF8671@paquier.xyz
Backpatch-through: 12
9155580 has changed the value of the first fake LSN for unlogged
relations from 1 to FirstNormalUnloggedLSN (aka 1000), GiST requiring a
non-zero LSN on some pages to allow an interlocking logic to work, but
its value was still initialized to 1 at the beginning of recovery or
after running pg_resetwal. This fixes the initialization for both code
paths.
Author: Takayuki Tsunakawa
Reviewed-by: Dilip Kumar, Kyotaro Horiguchi, Michael Paquier
Discussion: https://postgr.es/m/OSBPR01MB2503CE851940C17DE44AE3D9FE6F0@OSBPR01MB2503.jpnprd01.prod.outlook.com
Backpatch-through: 12
When we added the GUC units feature, we didn't make any great effort
to adjust the documentation of individual GUCs; they tended to still
say things like "this is the number of milliseconds that ...", even
though users might prefer to write some other units, and SHOW might
even show the value in other units. Commit 6c9fb69f2 made an effort
to improve this situation, but I thought it made things less readable
by injecting units information in mid-sentence. It also wasn't very
consistent, and did not touch all the GUCs that have units.
To improve matters, standardize on the phrasing "If this value is
specified without units, it is taken as <units>". Also, try to
standardize where this is mentioned, right before the specification
of the default. (In a couple of places, doing that would've required
more rewriting than seemed justified, so I wasn't 100% consistent
about that.) I also tried to use the phrases "amount of time",
"amount of memory", etc rather than describing the contents of GUCs
in other ways, as those were the majority usage in places that weren't
overcommitting to a particular unit. (I left "length of time" alone
in a couple of places, though.)
I failed to resist the temptation to copy-edit some awkward text, too.
Backpatch to v12, like 6c9fb69f2, mainly because v12 hasn't diverged
much from HEAD yet.
Discussion: https://postgr.es/m/15882.1571942223@sss.pgh.pa.us
It appears that libxml2 doesn't bother to set the "children" field of
an XML_NAMESPACE_DECL node to null; that field just contains garbage.
In v10 and v11, this can result in a crash in XMLTABLE(). The rewrite
done in commit 251cf2e27 fixed this, somewhat accidentally, in v12.
We're not going to back-patch 251cf2e27, however. The case apparently
doesn't have wide use, so rather than risk introducing other problems,
just add a safety check to throw an error.
Even though no bug manifests in v12/HEAD, add the relevant test case
there too, to prevent future regressions.
Chapman Flack (per private report)
pgtypeslib_extern.h contained fallback definitions of "bool", "FALSE",
and "TRUE". The latter two are just plain unused, and have been for
awhile. The former came into play only if there wasn't a macro
definition of "bool", which is true only if we aren't using <stdbool.h>.
However, it then defined bool as "char"; since commit d26a810eb that
conflicts with c.h's desire to use "unsigned char". We'd missed seeing
any bad effects of that due to accidental header inclusion order choices,
but dddf4cdc3 exposed that it was problematic.
To fix, let's just get rid of these definitions. They should not be
needed because everyplace in Postgres should be relying on c.h to
provide a definition for type bool. (Note that despite its name,
pgtypeslib_extern.h isn't exposed to any outside code; we don't
install it.)
This doesn't fully resolve the issue, because ecpglib.h is doing
similar things, but that seems to require more thought to fix.
Back-patch to v12 where d26a810eb came in, to forestall any unpleasant
surprises from future back-patched bug fixes.
Discussion: https://postgr.es/m/CAA4eK1LmaKO7Du9M9Lo=kxGU8sB6aL8fa3sF6z6d5yYYVe3BuQ@mail.gmail.com
Phases 2 (building the new index) and 3 (validating the new index)
checked for interrupts outside a transaction context, having as
consequence to not release session-level locks taken on the parent
relation and the old and new indexes processed. This could for example
be triggered with statement_timeout and a bad timing, and would issue
confusing error messages when shutting down the session still holding
the locks (note that an assertion failure would be triggered first), on
top of more issues with concurrent sessions trying to take a lock that
would interfere with the SHARE UPDATE EXCLUSIVE locks hold here.
This moves all the interruption checks inside a transaction context.
Note that I have manually tested all interruptions to make sure that
invalid indexes can be cleaned up properly. Partition indexes still
have issues on their own with some missing dependency handling, which
will be dealt with in a follow-up patch.
Reported-by: Justin Pryzby
Author: Michael Paquier
Discussion: https://postgr.es/m/20191013025145.GC4475@telsasoft.com
Backpatch-through: 12
In the first transaction run for REINDEX CONCURRENTLY, a thinko in the
existing logic caused two session locks to be taken on the old index,
causing the session lock on the newly-created index to be missed. This
made possible concurrent DDL commands (like ALTER INDEX) on the new
index while REINDEX CONCURRENTLY was processing from the point where the
first internal transaction committed.
This issue has been discovered while digging into another bug.
Author: Michael Paquier
Discussion: https://postgr.es/m/20191021074323.GB1869@paquier.xyz
Backpatch-through: 12
A check was redundant. While on it, add an assertion to make sure that
the parsing routine is never called with a NULL input. All the code
paths currently calling the parsing routine are careful with NULL inputs
already, but future callers may forget that.
Reported-by: Peter Eisentraut, Lars Kanis
Discussion: https://postgr.es/m/ec64956b-4597-56b6-c3db-457d15250fe4@2ndquadrant.com
Backpatch-through: 12
Any callback set would have no meaning in the context of an exception.
As an autovacuum worker exits quickly in this context, this could be
only an issue within EmitErrorReport(), where the elog hook is for
example called. That's unlikely to going to be a problem, but let's be
clean and consistent with other code paths handling exceptions. This is
present since 2909419, which introduced autovacuum.
Author: Ashwin Agrawal
Reviewed-by: Tom Lane, Michael Paquier
Discussion: https://postgr.es/m/CALfoeisM+_+dgmAdAOHAu0k-ZpEHHqSSG=GRf3pKJGm8OqWX0w@mail.gmail.com
Backpatch-through: 9.4
It emerges that recent versions of Windows (at least 2016 Standard)
spell this locale name as "Norwegian Bokmål_Norway.1252", defeating
our mapping code that translates "Norwegian (Bokmål)_Norway" to
something that's all-ASCII (cf commits db29620d4 and aa1d2fc5e).
Add another mapping entry to handle this spelling.
Per bug #16068 from Robert Ford. Like the previous patches,
back-patch to all supported branches.
Discussion: https://postgr.es/m/16068-4cb6eeaa7eb46d93@postgresql.org
On recent Red Hat platforms (at least RHEL 8 and Fedora 30, maybe older),
configure's probe for libperl failed if the user forces CFLAGS to be -O0.
This is because some code in perl's inline.h fails to be optimized away
at -O0, and said code doesn't work if compiled without -fPIC.
To fix, add CFLAGS_SL to the compile flags used during the libperl probe.
This is a better simulation of the way that plperl is built, anyway,
so it might forestall other issues in future.
Per gripe from Kyotaro Horiguchi. Back-patch to all supported branches,
since people might want to build older branches on these platforms.
Discussion: https://postgr.es/m/20191010.144533.263180400.horikyota.ntt@gmail.com
Move the platform-dependent logic that sets CFLAGS_SL from
src/makefiles/Makefile.foo to src/template/foo, so that the value
is determined at configure time and thus is available while running
configure's tests.
On a couple of platforms this might save a few microseconds of build
time by eliminating a test that make otherwise has to do over and over.
Otherwise it's pretty much a wash for build purposes; in particular,
this makes no difference to anyone who might be overriding CFLAGS_SL
via a make option.
This patch in itself does nothing with the value and thus should not
change any behavior, though you'll probably have to re-run configure
to get a correctly updated Makefile.global. We'll use the new
configure variable in a follow-on patch.
Per gripe from Kyotaro Horiguchi. Back-patch to all supported branches,
because the follow-on patch is a portability bug fix.
Discussion: https://postgr.es/m/20191010.144533.263180400.horikyota.ntt@gmail.com
We memorize all internal and empty leaf pages in the 1st vacuum stage for
gist indexes. They are used in the 2nd stage, to delete all the empty
pages. There was a memory context page_set_context for this purpose, but
we never used it.
Reported-by: Amit Kapila
Author: Dilip Kumar
Reviewed-by: Amit Kapila
Backpatch-through: 12, where it got introduced
Discussion: https://postgr.es/m/CAA4eK1LGr+MN0xHZpJ2dfS8QNQ1a_aROKowZB+MPNep8FVtwAA@mail.gmail.com
The logic was correctly detecting a parsing failure, but the parsing
error did not get reported back to the client properly.
Reported-by: Ed Morley
Author: Lars Kanis
Reviewed-by: Michael Paquier
Discussion: https://postgr.es/m/a9b4cbd7-4ecb-06b2-ebd7-1739bbff3217@greiz-reinsdorf.de
Backpatch-through: 12
Commit e7a2217 has introduced stricter checks for integer values in
connection parameters for libpq. However this failed to correctly check
after trailing whitespaces, while leading whitespaces were discarded per
the use of strtol(3). This fixes and refactors the parsing logic to
handle both cases consistently. Note that trying to restrict the use of
trailing whitespaces can easily break connection strings like in ECPG
regression tests (these have allowed me to catch the parsing bug with
connect_timeout).
Author: Michael Paquier
Reviewed-by: Lars Kanis
Discussion: https://postgr.es/m/a9b4cbd7-4ecb-06b2-ebd7-1739bbff3217@greiz-reinsdorf.de
Backpatch-through: 12
Without "b", a variant of the tas() code miscompiles on macOS 10.4.
This may also fix a compilation failure involving macOS 10.1. Today's
compilers have been allocating acceptable registers with or without this
change, but this future-proofs the code by precisely conveying the
acceptable registers. Back-patch to 9.4 (all supported versions).
Reviewed by Tom Lane.
Discussion: https://postgr.es/m/20191009063900.GA4066266@rfd.leadboat.com