mirrors/git-annex
2
0
Fork 0
Code Issues Pull requests Projects Releases Packages 1 Wiki Activity Actions

split up quickcheck tests for hashes and macs

Browse source 
So when one fais, it's clear which one is the problem.
This commit is contained in:
Joey Hess 2020-03-02 14:25:45 -04:00
parent ddeb7bea99
commit 716e573514
No known key found for this signature in database
GPG key ID: DB12DB0FF05F8F38
3 changed files with 63 additions and 38 deletions
Download patch file Download diff file Expand all files Collapse all files

22
doc/bugs/__34__Illegal_instruction__34___when_running_certain_tests_on_AMD_Phenom_II/comment_1_3932b6954cc42d2c376f48a6cdf0f879._comment Normal file
View file

@ -0,0 +1,22 @@
[[!comment format=mdwn
username="joey"
subject="""comment 1"""
date="2020-03-02T17:48:17Z"
content="""
The failing test case is testing all the hashes that git-annex uses.
Quite likely what has happened is that the hash implementation has been
built with a CPU-specific optimisation that is not available on your AMD.
The hash implementations are in http://hackage.haskell.org/package/cryptonite
While cryptonite has build flags for CPU-specific features,
there are also, in its C code, some checks of gcc defines that indicate
specific CPU features. I found #defines checking `__SSE2__` and `__SSE4__`
and `__AVX__` (mostly in blake2) and there might be others. That seems
likely to be a bug in cryptonite, if its build flags are not fully
controlling use of things like SSE.
I've made a change to git-annex's test suite to narrow down the problem;
it now tests each hash individually, so you'll see which specific ones
fail.
"""]]