From: Tom Lane <tgl@sss.pgh.pa.us>
To: Alexander Lakhin <exclusion@gmail.com>
cc: pgsql-bugs@lists.postgresql.org
Subject: Re: BUG #17912: Invalid memory access when converting plpython' array
 containing empty array
In-reply-to: <5ebae5e4-d401-fadf-8585-ac3eaf53219c@gmail.com>
References: <17912-82ceed78731d9cdc@postgresql.org>
 <1392974.1682694846@sss.pgh.pa.us>
 <bf33fe9b-f2f8-5885-aae3-537cd4591386@gmail.com>
 <1440309.1682705916@sss.pgh.pa.us> <1483429.1682709429@sss.pgh.pa.us>
 <5ebae5e4-d401-fadf-8585-ac3eaf53219c@gmail.com>
Comments: In-reply-to Alexander Lakhin <exclusion@gmail.com>
	message dated "Sat, 29 Apr 2023 16:00:00 +0300"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <1805877.1682788590.1@sss.pgh.pa.us>
Content-Transfer-Encoding: quoted-printable
Date: Sat, 29 Apr 2023 13:16:30 -0400
Message-ID: <1805878.1682788590@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/1805878.1682788590%40sss.pgh.pa.us>
Precedence: bulk

Alexander Lakhin <exclusion@gmail.com> writes:
> Thanks for fixing this!
> Now python's handling of arrays is much nicer and is aligned with the pl=
perl's behaviour:

Well, mostly.  As you noticed, it's a bit weird about zero-length
sub-lists, treating those as scalars.  I had been intending to keep
the existing behavior there, but now that I see that plperl does it
the other way (that is, you end up with an overall empty output array)
I think we ought to make plpython do likewise.

> While playing with plperl, I found that it handles arrays not perfectly =
too:
> CREATE OR REPLACE FUNCTION test_pl() RETURNS text[] AS $$return [1, []];=
$$ LANGUAGE plperl;
> SELECT * FROM test_pl();

Ugh.  I pushed a fix for that.

On the whole, I like plperl's implementation of this better, that is
it scans the data structure just once and uses an ArrayBuildState to
accumulate the datums.  So now I'm thinking about throwing out
the code in PLySequence_ToArray[_recurse] altogether and borrowing
plperl's logic.  I think that would make it easier to deal with
zero-length sublists correctly.  Haven't written the patch yet though.

> I suppose that waiting for v17 is preferable if the patch is considered =
as
> bringing a new feature (it's not the case) or could require extra time t=
o
> stabilize. But I'm afraid that anomalies, that would require additional
> fixes for the change stabilization, could be related to the existing
> code, and thus that extra time will be invested in improving v16- too.

I'm a little uncomfortable with changing the semantics of non-failing
cases in the back branches.

			regards, tom lane