Re: buildfarm warnings

Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> If not, I think that your quick-and-dirty fix is about right, except
> that we probably need to do it every place where we set
> progress_filename, and we should arrange to free the memory later.
> With -Ft, you won't have enough archives to matter, but with -Fp, you
> might have enough archive members to matter.

Yeah, I came to the same conclusion while out doing some errands.
There's no very convincing reason to believe that what's passed to
progress_update_filename has got adequate lifespan either, or that
that would remain true even if it's true today, or that testing
would detect such a problem (even if we had any, ahem). The file
names I was seeing in testing just now tended to contain fragments
of temporary directory names, which'd be mighty hard to tell from
random garbage in any automated way:

3/22774 kB (0%), 0/2 tablespaces (...pc1/PG_15_202202141/5/16392_init)
3/22774 kB (0%), 1/2 tablespaces (...pc1/PG_15_202202141/5/16392_init)
22785/22785 kB (100%), 1/2 tablespaces (...t_T8u0/backup1/global/pg_control)

So I think we should make progress_update_filename strdup each
input while freeing the last value, and insist that every update
go through that logic. (This is kind of a lot of cycles to spend
in support of an optional mode, but maybe we could do it only if
showprogress && verbose.) I'll go make it so.

regards, tom lane