Le Wed, 25 Mar 2020 16:58:33 -0600,
Jerry James <loganjerry(a)gmail.com> a écrit :
Either I got lucky when I launched the Rawhide build, or something
fundamentally different between F32 and Rawhide. Clearly something
nondeterministic is at work. I have been unable to reproduce this
failure in mock so far, but will keep trying.
If anybody has any ideas, I am all ears. Thank you,
Well... It ring a bell for me, because I kind of seeing what seems to
be weird bugs in F32... I am kind of suspecting a gcc bug giving some
weird things... but I am not a developer, know a bit of assembler but
not enough to make the link between C code and assembler.
Anyway, one hour a go, I got a segment violation error while doing
updates with dnfdragora... and it seems that the code just jump to a
low address outside of the program:
rh bug #1817269
The other bug that make me suspect something might be wrong with gcc
was a bug that make a segment violation (seems to access a bit lower
address than a .png font file) in a game package that have not been
updated in many years as far as I know:
rh bug #1816471
but in that case, frankly the C code seems a bit hard for me to take
time to try to understand it.
What puzzle me in this case, is why a very old code would have stop to
work... that's why I came to the idea that maybe this is the generated
code itself that changed.
... yeah, I know, might have no link at all with your Coq build... I