Dear Santiago,
Le 17/06/2025 à 16:06, Santiago Vila a écrit :
The ABI hash being dependent of the version of zstd used at build time
seems to be a defect for me... What do you think? Should I fix it
right now in unstable? That means another round of binNMUs...
Would make sense to change the hashing algorithm for the ABI so that
it's calculated from uncompressed data,
Yes, I opened an upstream bug report about that:
https://github.com/ocaml/ocaml/issues/14089
except that we hardcode it
to the current value if the data matches today's data? (i.e. this would
be like changing the hash algorithm but only for the next time it has
a good reason to change).
That looks too ad-hoc, intrusive and hazardous at this time in the
release cycle.
BTW: On which data is the hash calculated?
Digests are computed for compiled "interfaces" (equivalent of C's .h
files) and "approximations" (sort of ABI for .o files). These digests
are checked by the OCaml linker to guarantee binary compatibility of
separately compiled libraries.
Cheers,
--
Stéphane
