Horton

Thread

Commits

  1. Add bool to btree_gist documentation

  2. Fix gist_bool_ops to use gbtreekey2

  3. Add bool GiST opclass to btree_gist

  1. GiST operator class for bool

    Emre Hasegeli <emre@hasegeli.com> — 2021-06-08T10:48:10Z

    It could be useful to use bool in exclusion constraints, but it's
currently not nicely supported.  The attached patch adds support for
bool to the btree_gist extension, so we can do this.

I am adding this to the commitfest 2021-07.

  2. Re: GiST operator class for bool

    x4mmm@yandex-team.ru — 2021-06-08T11:07:32Z

    Hi!

> 8 июня 2021 г., в 13:48, Emre Hasegeli <emre@hasegeli.com> написал(а):
> 
> It could be useful to use bool in exclusion constraints, but it's
> currently not nicely supported.  The attached patch adds support for
> bool to the btree_gist extension, so we can do this.
> 
> I am adding this to the commitfest 2021-07.
> <0001-btree_gist-Support-bool.patch>

It definitely makes sense to include bool into list of supported types.
But patch that you propose does not support sorting build added in PG14.
Or we can add this functionality later in https://commitfest.postgresql.org/31/2824/ ...

Thanks!

Best regards, Andrey Borodin.

  3. Re: GiST operator class for bool

    Emre Hasegeli <emre@hasegeli.com> — 2021-06-08T14:53:02Z

    > But patch that you propose does not support sorting build added in PG14.

It looks like the change to btree_gist is not committed yet.  I'll
rebase my patch once it's committed.

It was a long thread.  I couldn't read all of it.  Though, the last
patches felt to me like a part of what's already been committed.
Shouldn't they also be committed to version 14?

  4. Re: GiST operator class for bool

    x4mmm@yandex-team.ru — 2021-06-14T09:33:19Z

    
> 8 июня 2021 г., в 19:53, Emre Hasegeli <emre@hasegeli.com> написал(а):
> 
>> But patch that you propose does not support sorting build added in PG14.
> 
> It looks like the change to btree_gist is not committed yet. I'll
> rebase my patch once it's committed.
Changes to GiST are committed. There will be no need to rebase anyway :)

> 
> It was a long thread.  I couldn't read all of it.  Though, the last
> patches felt to me like a part of what's already been committed.
> Shouldn't they also be committed to version 14?

Well, yeah, it could would be cool to have gist build and gist_btree support it together, but there was many parts and we could did not finish it before feature freeze.

Best regards, Andrey Borodin.

  5. Re: GiST operator class for bool

    Tomas Vondra <tomas.vondra@enterprisedb.com> — 2021-11-03T15:18:42Z

    Hi,

I looked at this patch today - it's pretty simple and in pretty good
shape, I can't think of anything that'd need fixing. Perhaps the test
might also do EXPLAIN like for other types, to verify the new index is
actually used. But that's minor enough to handle during commit.


I've marked this as RFC and will get it committed in a day or two.

-- 
Tomas Vondra
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

  6. Re: GiST operator class for bool

    Tomas Vondra <tomas.vondra@enterprisedb.com> — 2021-11-06T16:09:05Z

    On 11/3/21 16:18, Tomas Vondra wrote:
> Hi,
> 
> I looked at this patch today - it's pretty simple and in pretty good
> shape, I can't think of anything that'd need fixing. Perhaps the test
> might also do EXPLAIN like for other types, to verify the new index is
> actually used. But that's minor enough to handle during commit.
> 
> 
> I've marked this as RFC and will get it committed in a day or two.
> 

Pushed, after adding some simple EXPLAIN to the regression test.

Thanks for the patch!


regards

-- 
Tomas Vondra
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

  7. Re: GiST operator class for bool

    Tom Lane <tgl@sss.pgh.pa.us> — 2021-11-07T16:44:10Z

    Tomas Vondra <tomas.vondra@enterprisedb.com> writes:
> Pushed, after adding some simple EXPLAIN to the regression test.

skink is reporting that this has some valgrind issues [1].
I suspect sloppy conversion between bool and Datum, but
didn't go looking.

==1805451== VALGRINDERROR-BEGIN
==1805451== Uninitialised byte(s) found during client check request
==1805451==    at 0x59EFEA: PageAddItemExtended (bufpage.c:346)
==1805451==    by 0x2100B9: gistfillbuffer (gistutil.c:46)
==1805451==    by 0x2050F9: gistplacetopage (gist.c:562)
==1805451==    by 0x20546B: gistinserttuples (gist.c:1277)
==1805451==    by 0x205BB5: gistinserttuple (gist.c:1230)
==1805451==    by 0x206067: gistdoinsert (gist.c:885)
==1805451==    by 0x2084FB: gistBuildCallback (gistbuild.c:829)
==1805451==    by 0x23B572: heapam_index_build_range_scan (heapam_handler.c:1694)
==1805451==    by 0x208E7D: table_index_build_scan (tableam.h:1756)
==1805451==    by 0x208E7D: gistbuild (gistbuild.c:309)
==1805451==    by 0x2D10C8: index_build (index.c:2983)
==1805451==    by 0x2D2A7D: index_create (index.c:1232)
==1805451==    by 0x383E67: DefineIndex (indexcmds.c:1128)
==1805451==  Address 0x10cab1e4 is 12 bytes inside a block of size 16 client-defined
==1805451==    at 0x712AC5: palloc0 (mcxt.c:1118)
==1805451==    by 0x1E0A07: index_form_tuple (indextuple.c:146)
==1805451==    by 0x210BA8: gistFormTuple (gistutil.c:582)
==1805451==    by 0x2084C2: gistBuildCallback (gistbuild.c:813)
==1805451==    by 0x23B572: heapam_index_build_range_scan (heapam_handler.c:1694)
==1805451==    by 0x208E7D: table_index_build_scan (tableam.h:1756)
==1805451==    by 0x208E7D: gistbuild (gistbuild.c:309)
==1805451==    by 0x2D10C8: index_build (index.c:2983)
==1805451==    by 0x2D2A7D: index_create (index.c:1232)
==1805451==    by 0x383E67: DefineIndex (indexcmds.c:1128)
==1805451==    by 0x5AED2E: ProcessUtilitySlow (utility.c:1535)
==1805451==    by 0x5AE262: standard_ProcessUtility (utility.c:1066)
==1805451==    by 0x5AE33A: ProcessUtility (utility.c:527)
==1805451== 
==1805451== VALGRINDERROR-END

			regards, tom lane

[1] https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=skink&dt=2021-11-06%2023%3A56%3A57

  8. Re: GiST operator class for bool

    Tomas Vondra <tomas.vondra@enterprisedb.com> — 2021-11-07T19:53:21Z

    Hi,

On 11/7/21 17:44, Tom Lane wrote:
> Tomas Vondra <tomas.vondra@enterprisedb.com> writes:
>> Pushed, after adding some simple EXPLAIN to the regression test.
> 
> skink is reporting that this has some valgrind issues [1].
> I suspect sloppy conversion between bool and Datum, but
> didn't go looking.
> 

It's actually a bit worse than that :-( The opclass is somewhat confused 
about the type it should use for storage. The gbtree_ninfo struct says 
it's using gbtreekey4, the SQL script claims the params are gbtreekey8, 
and it should actually use gbtreekey2. Sorry for not noticing that.

The attached patch fixes the valgrind error for me.


regards

-- 
Tomas Vondra
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

  9. Re: GiST operator class for bool

    Tomas Vondra <tomas.vondra@enterprisedb.com> — 2021-11-08T01:24:22Z

    On 11/7/21 20:53, Tomas Vondra wrote:
> Hi,
> 
> On 11/7/21 17:44, Tom Lane wrote:
>> Tomas Vondra <tomas.vondra@enterprisedb.com> writes:
>>> Pushed, after adding some simple EXPLAIN to the regression test.
>>
>> skink is reporting that this has some valgrind issues [1].
>> I suspect sloppy conversion between bool and Datum, but
>> didn't go looking.
>>
> 
> It's actually a bit worse than that :-( The opclass is somewhat confused 
> about the type it should use for storage. The gbtree_ninfo struct says 
> it's using gbtreekey4, the SQL script claims the params are gbtreekey8, 
> and it should actually use gbtreekey2. Sorry for not noticing that.
> 
> The attached patch fixes the valgrind error for me.
> 

I've pushed the fix, hopefully that'll make skink happy.

What surprised me a bit is that the opclass used gbtreekey4 storage, the 
equality support proc was defined as using gbtreekey8

     FUNCTION    7    gbt_bool_same (gbtreekey8, gbtreekey8, internal),

yet the gistvalidate() did not report this. Turns out this is because

     ok = check_amproc_signature(procform->amproc, INTERNALOID, false,
                                 3, 3, opckeytype, opckeytype,
                                 INTERNALOID);

i.e. with exact=false, so these type differences are ignored. Changing 
it to true reports the issue (and no other issues in check-world).

But maybe there are reasons to keep using false?

regards

-- 
Tomas Vondra
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

  10. Re: GiST operator class for bool

    Pavel Luzanov <p.luzanov@postgrespro.ru> — 2021-12-06T21:35:58Z

    Hello,

I don't see any changes in the documentation.[1]

Should bool appear in the looong list of supported operator classes?

[1] https://www.postgresql.org/docs/devel/btree-gist.html

-- 
Pavel Luzanov
Postgres Professional: https://postgrespro.com
The Russian Postgres Company

  11. Re: GiST operator class for bool

    Tomas Vondra <tomas.vondra@enterprisedb.com> — 2021-12-11T04:04:00Z

    On 12/6/21 22:35, Pavel Luzanov wrote:
> Hello,
> 
> I don't see any changes in the documentation.[1]
> 
> Should bool appear in the looong list of supported operator classes?
> 

You're right, I forgot to update the list of data types in the docs. 
Fixed, thanks for the report.


regards

-- 
Tomas Vondra
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company