On Wed, Jun 15, 2022 at 02:02:03PM -0400, Tom Lane wrote:
> Yeah, that was what was bugging me about this proposal. Removing
> one function's dependency on MAXPGPATH isn't much of a step forward.
This comes down to out-of-memory vs path length at the end. Changing
only the paths of make_outputdirs() without touching all the paths in
check.c and the one in function.c does not sound good to me, as this
increases the risk of failing pg_upgrade in the middle, and that's
what we should avoid, as said upthread.
> I note also that the patch leaks quite a lot of memory (a kilobyte or
> so per pathname, IIRC). That's probably negligible in this particular
> context, but anyplace that was called more than once per program run
> would need to be more tidy.
Surely.
--
Michael