Closed folkertdev closed 3 months ago
fuzzing like this
cargo +nightly fuzz run -s=leak end_to_end --target s390x-unknown-linux-gnu
with
[target.s390x-unknown-linux-gnu]
runner = "qemu-s390x -L /usr/s390x-linux-gnu"
linker = "s390x-linux-gnu-gcc"
finds examples that produce different output quickly.
https://github.com/memorysafety/zlib-rs/pull/155
fixes the final currently-known discrepancies.
We observe that
zlib-ng
generates different output between x86 and aarch64. We (currently, anyway) generate the same output on both platforms. We don't know at this time why the difference is there.The differences are only observed for larger inputs. At the time of writing, these tests fail on aarch64. The inputs are all in the tens of kilobytes, and we check that they encode to a specific sequence of bytes. We always do, but zlib-ng produces a different output (that we presume is valid and correct, just not byte-for-byte the same).