El 20/10/24 a las 23:19, Chris Hofstaedtler escribió:
I'm just concerned because we're looking at "ln -fs" failing, and
the other involved parts seem relatively simple.
Maybe something very strange is going on, but that needs someone who
can repro the failure to investigate it.
Well, good news and sorry for not noticing before: I've checked and
this is one of the 800 packages which FTBFS (either always or randomly)
when built with make --shuffle:
https://people.debian.org/~sanvila/make-shuffle/
(Note to self: I have to announce that some day).
So this is most surely a Makefile bug, and very likely
using "make --shuffle" will "amplify" the probability
of the failure to happen.
I even have a personal theory or suggestion to deal with
those bugs: Forget for a while about reproducing them and
just try to prove that the Makefile is "correct" (in a
"mathematical sense", i.e. try to prove that it will always
do what is supposed to do) using rubber duck debugging.
Since we know the Makefile is buggy, any attempt at proving its
correctness should fail, hopefully at the point where the Makefile
bug becomes self-evident.
(Also: I've just tried dpkg-buildpackage in a chroot with make 4.4.1
at /usr/bin and GNUMAKEFLAGS=--shuffle in the environment, and the
failure rate is very high indeed. I think everybody will be able
to reproduce it that way. Just try it.)
Thanks.
