Tom Lane 647fa50054 Remove arbitrary FUNC_MAX_ARGS limit in int2vectorin and oidvectorin. ...

int2vectorin limited the number of array elements it'd take to
FUNC_MAX_ARGS, which is probably fine for the traditional use-cases.
But now that pg_publication_rel.prattrs is an int2vector, it's not
fine at all: it's easy to construct cases where that can have up to
about MaxTupleAttributeNumber entries.  Trying to replicate such
tables leads to logical-replication failures.

As long as we have to touch this code anyway, let's just remove
the a-priori limit altogether, and let it accept any size that'll
be allowed by repalloc.  (Note that since int2vector isn't toastable,
we cannot store arrays longer than about BLCKSZ/2; but there is no
good excuse for letting int2vectorin depend on that.  Perhaps we
will lift the no-toast restriction someday.)

While at it, also improve the equivalent logic in oidvectorin.
I don't know of any practical use-case for long oidvectors right
now, but doing it right actually makes the code shorter.

Per report from Erik Rijkers.  Back-patch to v15 where
pg_publication_rel.prattrs was added.

Discussion: https://postgr.es/m/668ba539-33c5-8190-ca11-def2913cb94b@xs4all.nl

2023-01-15 17:32:09 -05:00

35 KiB

Raw Blame History

View Raw