contrib/pageinspect's tuple_data_split() function thought it could get
away with opening the referenced relation with NoLock. In practice
there's no guarantee that the current session holds any lock on that
rel (even if we just read a page from it), so that this is unsafe.
Switch to using AccessShareLock. Also, postpone closing the relation,
so that we needn't copy its tupdesc. Also, fix unsafe use of
att_isnull() for attributes past the end of the tuple.
Per testing with a patch that complains if we open a relation without
holding any lock on it. I don't plan to back-patch that patch, but we
should close the holes it identifies in all supported branches.
Discussion: https://postgr.es/m/2038.1538335244@sss.pgh.pa.us
If the column being modified is referenced by a foreign key constraint
of another table, ALTER TABLE would open the other table (to re-parse
the constraint's definition) without having first obtained a lock on it.
This was evidently intentional, but that doesn't mean it's really safe.
It's especially not safe in 9.3, which pre-dates use of MVCC scans for
catalog reads, but even in current releases it doesn't seem like a good
idea.
We know we'll need AccessExclusiveLock shortly to drop the obsoleted
constraint, so just get that a little sooner to close the hole.
Per testing with a patch that complains if we open a relation without
holding any lock on it. I don't plan to back-patch that patch, but we
should close the holes it identifies in all supported branches.
Discussion: https://postgr.es/m/2038.1538335244@sss.pgh.pa.us
Add RangeTblEntry.rellockmode, which records the appropriate lock mode for
each RTE_RELATION rangetable entry (either AccessShareLock, RowShareLock,
or RowExclusiveLock depending on the RTE's role in the query).
This patch creates the field and makes all creators of RTE nodes fill it
in reasonably, but for the moment nothing much is done with it. The plan
is to replace assorted post-parser logic that re-determines the right
lockmode to use with simple uses of rte->rellockmode. For now, just add
Asserts in each of those places that the rellockmode matches what they are
computing today. (In some cases the match isn't perfect, so the Asserts
are weaker than you might expect; but this seems OK, as per discussion.)
This passes check-world for me, but it seems worth pushing in this state
to see if the buildfarm finds any problems in cases I failed to test.
catversion bump due to change of stored rules.
Amit Langote, reviewed by David Rowley and Jesper Pedersen,
and whacked around a bit more by me
Discussion: https://postgr.es/m/468c85d9-540e-66a2-1dde-fec2b741e688@lab.ntt.co.jp
The connection authorized message has quite a bit of useful information
in it, but didn't include the application_name (when provided), so let's
add that as it can be very useful.
Note that at the point where we're emitting the connection authorized
message, we haven't processed GUCs, so it's not possible to get this by
using log_line_prefix (which pulls from the GUC). There's also
something to be said for having this included in the connection
authorized message and then not needing to repeat it for every line, as
having it in log_line_prefix would do.
The GUC cleans the application name to pure-ascii, so do that here too,
but pull out the logic for cleaning up a string into its own function
in common and re-use it from those places, and check_cluster_name which
was doing the same thing.
Author: Don Seiler <don@seiler.us>
Discussion: https://postgr.es/m/CAHJZqBB_Pxv8HRfoh%2BAB4KxSQQuPVvtYCzMg7woNR3r7dfmopw%40mail.gmail.com
Build a third version of libpgcommon.a, with -fPIC and -DFRONTEND,
as commit ea53100d5 did for src/port. Use that in libpq to avoid
symlinking+rebuilding source files retail.
Also adjust ecpg to use the new src/port and src/common libraries.
Arrange to install these libraries, too, to simplify out-of-tree
builds of shared libraries that need any of these modules.
Discussion: https://postgr.es/m/13022.1538003440@sss.pgh.pa.us
Discussion: https://postgr.es/m/E1g5Y8r-0006vs-QA@gemulon.postgresql.org
Client applications should get this function, if they need it, from
libpgport.
The fact that it's exported from libpq is a hack left over from before
we set up libpgport. It's never been documented, and there's no good
reason for non-PG code to be calling it anyway, so hopefully this won't
cause any problems. Moreover, with the previous setup it was not real
clear whether our clients that use the function were getting it from
libpgport or libpq, so this might actually prevent problems.
The reason for changing it now is that in the wake of commit ea53100d5,
some linkers won't export the symbol, apparently because it's coming from
a .a library instead of a .o file. We could get around that by continuing
to symlink pqsignal.c into libpq as before; but unless somebody complains
very hard, I don't want to adopt such a kluge.
Discussion: https://postgr.es/m/13022.1538003440@sss.pgh.pa.us
Discussion: https://postgr.es/m/E1g5Y8r-0006vs-QA@gemulon.postgresql.org
When the checkpointer receives a SIGHUP signal to update its configuration,
it may need to update the shared memory for full_page_writes and need to
write a WAL record for it. Now, it is quite possible that the XLOG
machinery has not been initialized by that time and it will lead to
assertion failure while doing that. Fix is to allow the initialization of
the XLOG machinery outside critical section.
This bug has been introduced by the commit 2c03216d83 which added the XLOG
machinery initialization in RecoveryInProgress code path.
Reported-by: Dilip Kumar
Author: Dilip Kumar
Reviewed-by: Michael Paquier and Amit Kapila
Backpatch-through: 9.5
Discussion: https://postgr.es/m/CAFiTN-u4BA8KXcQUWDPNgaKAjDXC=C2whnzBM8TAcv=stckYUw@mail.gmail.com
A restart point or a checkpoint recycling WAL segments treats segments
marked with neither ".done" (archiving is done) or ".ready" (segment is
ready to be archived) in archive_status the same way for archive_mode
being "on" or "always". While for a primary this is fine, a standby
running a restart point with archive_mode = on would try to mark such a
segment as ready for archiving, which is something that will never
happen except after the standby is promoted.
Note that this problem applies only to WAL segments coming from the
local pg_wal the first time archive recovery is run. Segments part of a
self-contained base backup are the most common case where this could
happen, however even in this case normally the .done markers would be
most likely part of the backup. Segments recovered from an archive are
marked as .ready or .done by the startup process, and segments finished
streaming are marked as such by the WAL receiver, so they are handled
already.
Reported-by: Haruka Takatsuka
Author: Michael Paquier
Discussion: https://postgr.es/m/15402-a453c90ed4cf88b2@postgresql.org
Backpatch-through: 9.5, where archive_mode = always has been added.
It failed if passed a nonexistent relation OID, or one that was a non-heap
relation, because of blindly applying heap_open to a user-supplied OID.
This is not OK behavior for a SQL-exposed function; we have a project
policy that we should return NULL in such cases. Moreover, since
pg_get_partition_constraintdef ought now to work on indexes, restricting
it to heaps is flat wrong anyway.
The underlying function generate_partition_qual() wasn't on board with
indexes having partition quals either, nor for that matter with rels
having relispartition set but yet null relpartbound. (One wonders
whether the person who wrote the function comment blocks claiming that
these functions allow a missing relpartbound had ever tested it.)
Fix by testing relispartition before opening the rel, and by using
relation_open not heap_open. (If any other relkinds ever grow the
ability to have relispartition set, the code will work with them
automatically.) Also, don't reject null relpartbound in
generate_partition_qual.
Back-patch to v11, and all but the null-relpartbound change to v10.
(It's not really necessary to change generate_partition_qual at all
in v10, but I thought s/heap_open/relation_open/ would be a good
idea anyway just to keep the code in sync with later branches.)
Per report from Justin Pryzby.
Discussion: https://postgr.es/m/20180927200020.GJ776@telsasoft.com
The previous commit wasn't careful enough to remove all traces of
TupleDescGetSlot().
Besides fixing the oversight of not removing TupleDescGetSlot()'s
declaration, this also removes FuncCallContext->slot. That was
documented to be for use in combination with TupleDescGetSlot(), a
cursory search over extensions finds no users, and there doesn't seem
to be convincing reasons to keep it around. If we later in the v12
release cycle find users, we can re-consider this part of the commit.
Reported-By: Michael Paquier
Discussion: https://postgr.es/m/20180926000413.GC1659@paquier.xyz
libpq and ecpg need shared-library-friendly versions of assorted src/port/
and src/common/ modules. Up to now, they got those by symlinking the
individual source files and compiling them locally. That's baroque, and a
pain to maintain, and it results in some amount of duplicated compile work.
It might've made sense when only a couple of files were needed, but the
list has grown and grown and grown :-(
It makes more sense to have the originating directory build a third variant
of libpgport.a/libpgcommon.a containing modules built with $(CFLAGS_SL),
and just link that into the shared library. Unused files won't get linked,
so the end result should be the same.
This patch makes a down payment on that idea by having src/port/ build
such a library and making libpq use it. If the buildfarm doesn't expose
fatal problems with the approach, I'll extend it to the other cases.
Discussion: https://postgr.es/m/13022.1538003440@sss.pgh.pa.us
strerror.c now requires strlcpy() in some cases, and a couple of the
ecpg libraries did not have that at hand. Pull it in from src/port/
following the usual recipe. Per buildfarm.
Those previously used bool, which should be safe on any modern
platforms, however the C standard is clear that it is better to use
sig_atomic_t for variables manipulated in signal handlers. This commit
adds at the same time PGDLLIMPORT to ClientConnectionLost.
Author: Michael Paquier
Reviewed-by: Tom Lane, Chris Travers, Andres Freund
Discussion: https://postgr.es/m/20180925011311.GD1354@paquier.xyz
The method we've traditionally used, of redeclaring strerror_r() to
see if the compiler complains of inconsistent declarations, turns out
not to work reliably because some compilers only report a warning,
not an error. Amazingly, this has gone undetected for years, even
though it certainly breaks our detection of whether strerror_r
succeeded.
Let's instead test whether the compiler will take the result of
strerror_r() as a switch() argument. It's possible this won't
work universally either, but it's the best idea I could come up with
on the spur of the moment.
We should probably back-patch this once the dust settles, but
first let's see what the buildfarm thinks of it.
Discussion: https://postgr.es/m/10877.1537993279@sss.pgh.pa.us
We must define the macro "printf" with arguments, else it can mess
up format archetype attributes in builds where PG_PRINTF_ATTRIBUTE
is just "printf". Fortunately, that's easy to do now that we're
requiring C99; we can use __VA_ARGS__.
On the other hand, it's better not to use __VA_ARGS__ for the rest
of the *printf crew, so that one can take the addresses of those
functions without surprises.
I'd proposed doing this some time ago, but forgot to make it happen;
buildfarm failures subsequent to 96bf88d52 reminded me.
Discussion: https://postgr.es/m/22709.1535135640@sss.pgh.pa.us
Discussion: https://postgr.es/m/20180926190934.ea4xvzhkayuw7gkx@alap3.anarazel.de
When a table ownership is changed, we must apply that also to any owned
sequences. (Otherwise, it would result in a situation that cannot be
restored, because linked sequences must have the same owner as the
table.) But this was previously only applied to regular tables and
materialized views. But it should also apply to at least foreign
tables. This patch removes the relkind check altogether, because it
doesn't save very much and just introduces the possibility of similar
omissions.
Bug: #15238
Reported-by: Christoph Berg <christoph.berg@credativ.de>
I started out with the idea that we needed to detect use of %m format specs
in contexts other than elog/ereport calls, because we couldn't rely on that
working in *printf calls. But a better answer is to fix things so that it
does work. Now that we're using snprintf.c all the time, we can implement
%m in that and we've fixed the problem.
This requires also adjusting our various printf-wrapping functions so that
they ensure "errno" is preserved when they call snprintf.c.
Remove elog.c's handmade implementation of %m, and let it rely on
snprintf to support the feature. That should provide some performance
gain, though I've not attempted to measure it.
There are a lot of places where we could now simplify 'printf("%s",
strerror(errno))' into 'printf("%m")', but I'm not in any big hurry
to make that happen.
Patch by me, reviewed by Michael Paquier
Discussion: https://postgr.es/m/2975.1526862605@sss.pgh.pa.us
We've spent an awful lot of effort over the years in coping with
platform-specific vagaries of the *printf family of functions. Let's just
forget all that mess and standardize on always using src/port/snprintf.c.
This gets rid of a lot of configure logic, and it will allow a saner
approach to dealing with %m (though actually changing that is left for
a follow-on patch).
Preliminary performance testing suggests that as it stands, snprintf.c is
faster than the native printf functions for some tasks on some platforms,
and slower for other cases. A pending patch will improve that, though
cases with floating-point conversions will doubtless remain slower unless
we want to put a *lot* of effort into that. Still, we've not observed
that *printf is really a performance bottleneck for most workloads, so
I doubt this matters much.
Patch by me, reviewed by Michael Paquier
Discussion: https://postgr.es/m/2975.1526862605@sss.pgh.pa.us
This provides the features that used to exist in useful_strerror()
for users of strerror_r(), too. Also, standardize on the GNU convention
that strerror_r returns a char pointer that may not be NULL.
I notice that libpq's win32.c contains a variant version of strerror_r
that probably ought to be folded into strerror.c. But lacking a
Windows environment, I should leave that to somebody else.
Discussion: https://postgr.es/m/2975.1526862605@sss.pgh.pa.us
elog.c has long had a private strerror wrapper that handles assorted
possible failures or deficiencies of the platform's strerror. On Windows,
it also knows how to translate Winsock error codes, which the native
strerror does not. Move all this code into src/port/strerror.c and
define strerror() as a macro that invokes it, so that both our frontend
and backend code will have all of this behavior.
I believe this constitutes an actual bug fix on Windows, since AFAICS
our frontend code did not report Winsock error codes properly before this.
However, the main point is to lay the groundwork for implementing %m
in src/port/snprintf.c: the behavior we want %m to have is this one,
not the native strerror's.
Note that this throws away the prior use of src/port/strerror.c,
which was to implement strerror() on platforms lacking it. That's
been dead code for nigh twenty years now, since strerror() was
already required by C89.
We should likewise cause strerror_r to use this behavior, but
I'll tackle that separately.
Patch by me, reviewed by Michael Paquier
Discussion: https://postgr.es/m/2975.1526862605@sss.pgh.pa.us
While we are probably still far away from fully implementing
assertions, all patch proposals appear to take issue with the existing
dummy grammar CREATE/DROP ASSERTION productions, so update those a
little bit. Rename the rule, use any_name instead of name, and remove
some unused code. Also remove the production for DROP ASSERTION,
since that would most likely be handled via the generic DROP support.
extracted from a patch by Joe Wildish
This commit significantly increases test coverage of geo_ops.c, adding
tests for various issues addressed by 2e2a392de3 (which went undetected
for a long time, at least partially due to not being covered).
This also removes alternative results expecting -0 on some platforms.
Instead the functions are should return the same results everywhere,
transforming -0 to 0 if needed.
The tests are added to geometric.sql file, sorted by the left hand side
of the operators. There are many cross datatype operators, so this seems
like the best solution.
Author: Emre Hasegeli
Reviewed-by: Tomas Vondra
Discussion: https://www.postgresql.org/message-id/CAE2gYzxF7-5djV6-cEvqQu-fNsnt%3DEqbOURx7ZDg%2BVv6ZMTWbg%40mail.gmail.com
According to the source history, the internal format of line data type
has changed, but various functions working with it did were not updated
and thus were producing wrong results.
This patch addresses various such issues, in particular:
* Reject invalid specification A=B=0 on receive
* Reject same points on line_construct_pp()
* Fix perpendicular operator when negative values are involved
* Avoid division by zero on perpendicular operator
* Fix intersection and distance operators when neither A nor B are 1
* Return NULL for closest point when objects are parallel
* Check whether closest point of line segments is the intersection point
* Fix closest point of line segments being on the wrong segment
Aside from handling those issues, the patch also aims to make operators
more symmetric and less sen to precision loss. The EPSILON interferes
with even minor changes, but the least we can do is applying it to both
sides of the operators equally.
Author: Emre Hasegeli
Reviewed-by: Tomas Vondra
Discussion: https://www.postgresql.org/message-id/CAE2gYzxF7-5djV6-cEvqQu-fNsnt%3DEqbOURx7ZDg%2BVv6ZMTWbg%40mail.gmail.com
The activation and deactivation of commit timestamp tracking has not
been handled consistently for a primary or standbys at recovery. The
facility can be activated at three different moments of recovery:
- The beginning, where a primary would use the GUC value for the
decision-making, and where a standby relies on the contents of the
control file.
- When replaying a XLOG_PARAMETER_CHANGE record at redo.
- The end, where both primary and standby rely on the GUC value.
Using the GUC value for a primary at the beginning of recovery causes
problems with commit timestamp access when doing crash recovery.
Particularly, when replaying transaction commits, it could be possible
that an attempt to read commit timestamps is done for a transaction
which committed at a moment when track_commit_timestamp was disabled.
A test case is added to reproduce the failure. The test works down to
v11 as it takes advantage of transaction commits within procedures.
Reported-by: Hailong Li
Author: Masahiko Sawasa, Michael Paquier
Reviewed-by: Kyotaro Horiguchi
Discussion: https://postgr.es/m/11224478-a782-203b-1f17-e4797b39bdf0@qunar.com
Backpatch-through: 9.5, where commit timestamps have been introduced.
TupleDescGetSlot() was kept around for backward compatibility for
user-written SRFs. With the TupleTableSlot abstraction work, that code
will need to be version specific anyway, so there's no point in
keeping the function around any longer.
Author: Ashutosh Bapat
Reviewed-By: Andres Freund
Discussion: https://postgr.es/m/20180220224318.gw4oe5jadhpmcdnm@alap3.anarazel.de
Upcoming changes introduce further types of tuple table slots, in
preparation of making table storage pluggable. New storage methods
will have different representation of tuples, therefore the slot
accessor should refer explicitly to heap tuples.
Instead of just renaming the functions, split it into one function
that accepts heap tuples not residing in buffers, and one accepting
ones in buffers. Previously one function was used for both, but that
was a bit awkward already, and splitting will allow us to represent
slot types for tuples in buffers and normal memory separately.
This is split out from the patch introducing abstract slots, as this
largely consists out of mechanical changes.
Author: Ashutosh Bapat
Reviewed-By: Andres Freund
Discussion: https://postgr.es/m/20180220224318.gw4oe5jadhpmcdnm@alap3.anarazel.de
Previously it was an int / 4 bytes. The maximum number of attributes
in a tuple is restricted by the maximum value Var->varattno, which is
an AttrNumber/int16. Hence use the same data type for
TupleTableSlot->tts_nvalid.
Author: Ashutosh Bapat
Discussion: https://postgr.es/m/20180220224318.gw4oe5jadhpmcdnm@alap3.anarazel.de
Previously, when using parallel query, EXPLAIN (ANALYZE)'s JIT
compilation timings did not include the overhead from doing so on the
workers. Fix that.
We do so by simply aggregating the cost of doing JIT compilation on
workers and the leader together. Arguably that's not quite accurate,
because the total time spend doing so is spent in parallel - but it's
hard to do much better. For additional detail, when VERBOSE is
specified, the stats for workers are displayed separately.
Author: Amit Khandekar and Andres Freund
Discussion: https://postgr.es/m/CAJ3gD9eLrz51RK_gTkod+71iDcjpB_N8eC6vU2AW-VicsAERpQ@mail.gmail.com
Backpatch: 11-
Apple's latest rearrangements of the system-supplied headers have broken
building of PL/Perl and PL/Tcl. The only practical way to fix PL/Tcl is to
start using the "-isysroot" compiler flag to point to SDK-supplied headers,
as Apple expects. We must also start distinguishing where to find Perl's
headers from where to find its shared library; but that seems like good
cleanup anyway.
Extensions that formerly did something like -I$(perl_archlibexp)/CORE
should now do -I$(perl_includedir)/CORE instead. perl_archlibexp
is still the place to look for libperl.so, though.
If for some reason you don't like the default -isysroot setting, you can
override that by setting PG_SYSROOT in configure's arguments. I don't
currently think people would need to do so, unless maybe for cross-version
build purposes.
In addition, teach configure where to find tclConfig.sh. Our traditional
method of searching $auto_path hasn't worked for the last couple of macOS
releases, and it now seems clear that Apple's not going to change that.
The workaround of manually specifying --with-tclconfig was annoying
already, but Mojave's made it a lot more so because the sysroot path now
has to be included as well. Let's just wire the knowledge into configure
instead. To avoid breaking builds against non-default Tcl installations
(e.g. MacPorts) wherein the $auto_path method probably still works,
arrange to try the additional case only after all else has failed.
Back-patch to all supported versions, since at least the buildfarm
cares about that. The changes are set up to not do anything on macOS
releases that are old enough to not have functional sysroot trees.
It's fairly silly to truncate the throttle_delay to integer when the only
math we ever do with it requires converting back to double. Furthermore,
given that people are starting to complain about restrictions like only
supporting 1K client connections, I don't think we're very far away from
situations where the precision loss matters. As the code stood, for
example, there's no difference between --rate 100001 and --rate 111111;
both get converted to throttle_delay = 9. Somebody trying to run 100
threads and have each one dispatch around 1K TPS would find this lack of
precision rather surprising, especially since the required per-thread
delays are around 1ms, well within the timing precision of modern systems.
96e1cb4 has added support for --no-publications in pg_dump, pg_dumpall
and pg_restore, but forgot the fact that publication tables also need to
be ignored when this option is used.
Author: Gilles Darold
Reviewed-by: Michael Paquier
Discussion: https://postgr.es/m/3f48e812-b0fa-388e-2043-9a176bdee27e@dalibo.com
Backpatch-through: 10, where publications have been added.
Commit 25fff40 has granted execute permission of the function
pg_stat_statements_reset() to default role "pg_read_all_stats", but this
role is meant to read statistics, and not to reset them. The
permissions on this function are revoked from "pg_read_all_stats". The
version of pg_stat_statements is bumped up in consequence.
Author: Haribabu Kommi
Reviewed-by: Michael Paquier, Amit Kapila
Discussion: https://postgr.es/m/CAJrrPGf5fCnKqXObpwGN9nMyD--tzOf-7LFCJiz59Z1wJ5qj9A@mail.gmail.com
We haven't touched these since text search functionality landed in core
in 2007 :-(. While the upstream project isn't a beehive of activity,
they do make additions and bug fixes from time to time. Update our
copies of these files.
Also update our documentation about how to keep things in sync, since
they're not making distribution tarballs these days. Fortunately,
their source code turns out to be a breeze to build.
Notable changes:
* The non-UTF8 version of the hungarian stemmer now works in LATIN2
not LATIN1.
* New stemmers have appeared for arabic, indonesian, irish, lithuanian,
nepali, and tamil. These all work in UTF8, and the indonesian and
irish ones also work in LATIN1.
(There are some new stemmers that I did not incorporate, mainly because
their names don't match the underlying languages, suggesting that they're
not to be considered mainstream.)
Worth noting: the upstream Nepali dictionary was contributed by
Arthur Zakirov.
initdb forced because the contents of snowball_create.sql have
changed.
Still TODO: see about updating the stopword lists.
Arthur Zakirov, minor mods and doc work by me
Discussion: https://postgr.es/m/20180626122025.GA12647@zakirov.localdomain
Discussion: https://postgr.es/m/20180219140849.GA9050@zakirov.localdomain
A discussion about also reporting JIT compilation overhead on workers
brought unhappiness with the verbosity of the current explain format
to light. Make the text format more dense, and restructure the
structured output to mirror that more closely.
As we're re-jiggering the output format anyway: The denser format
allows us to report all flags for JIT compilation (now also reporting
PGJIT_EXPR and PGJIT_DEFORM), and report the total time in addition to
the individual times.
Per complaint from Tom Lane.
Author: Andres Freund
Discussion: https://postgr.es/m/27812.1537221015@sss.pgh.pa.us
Backpatch: 11-, where JIT compilation was introduced
Ensure that triggers get properly filled in tuples for the OLD value.
Also fix the logic of detecting missing null values. The previous logic
failed to detect a missing null column before the first missing column
with a default. Fixing this has simplified the logic a bit.
Regression tests are added to test changes. This should ensure better
coverage of expand_tuple().
Original bug reports, and some code and test scripts from Tomas Vondra
Backpatch to release 11.
Previously, pgbench always used select(2) for this purpose, but that's
problematic for very high client counts, because select() can't deal
with file descriptor numbers larger than FD_SETSIZE. It's pretty common
for that to be only 1024 or so, whereas modern OSes can allow many more
open files than that. Using poll(2) would surmount that problem, but it
creates another one: poll()'s timeout resolution is only 1ms, which is
poor enough to cause problems with --rate specifications approaching or
exceeding 1K TPS.
On platforms that have ppoll(2), which includes Linux and recent
FreeBSD, we can use that to avoid the FD_SETSIZE problem without any
loss of timeout resolution. Hence, add configure logic to test for
ppoll(), and use it if available.
This patch introduces an abstraction layer into pgbench that could
be extended to support other kernel event-wait APIs such as kevents.
But actually adding such support is a matter for some future patch.
Doug Rady, reviewed by Robert Haas and Fabien Coelho, and whacked around
a good bit more by me
Discussion: https://postgr.es/m/23D017C9-81B7-484D-8490-FD94DEC4DF59@amazon.com
Tom Lane