search

signalapp / libsignal-protocol-c

GNU General Public License v3.0
1.41k stars 295 forks source link

test_generate_signed_pre_key fails on big-endian architectures #124

Closed etam closed 5 years ago

etam commented 5 years ago

I have:

Bug description

On big-endian architectures (s390x and ppc64) one test fails with the following error:

libsignal-protocol-c-2.3.2/tests/test_key_helper.c:248:F:case:test_generate_signed_pre_key:0: Assertion 'memcmp(signedPreKey, data, len) == 0' failed: memcmp(signedPreKey, data, len) == 1, 0 == 0

Device info

Device: Virtual machine run by openSUSE Build Service libsignal-protocol-c version: 2.3.2

Link to logs

Build log for s390x: https://build.opensuse.org/package/live_build_log/devel:libraries:c_c++/libsignal-protocol-c/SLE_12_SP4/s390x Build log for ppc64: https://build.opensuse.org/package/live_build_log/devel:libraries:c_c++/libsignal-protocol-c/openSUSE_Factory_PowerPC/ppc64

dkonigsberg commented 5 years ago

Will investigate further, as soon as I finish getting an OpenSUSE/ppc64 install going in qemu. From what little info has been provided, its unclear if this is an endian issue or an architecture/platform issue. However, I'll report my findings once I've dug deeper.

dkonigsberg commented 5 years ago

I managed to get OpenSUSE/ppc64 up and running inside qemu, and I've successfully reproduced the test failure. I'll update the issue once I've investigated the cause.

dkonigsberg commented 5 years ago

Okay, your initial assumption was correct. Its an endian issue, specifically inside the protobuf-c code. There's a flag that code uses to know whether it has to swap byte order for certain types, yet there's nothing in the build configuration process to detect endian-ness and set that flag. If I manually set the flag, all tests pass. Hopefully the fix for this will just involve a change to the CMake configuration files.