New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
test_lll_lll fails on 32-bit platforms #112
Comments
The failure persists even after I apply b1004c2 |
Strangely, the failure is only triggered if I run all of the preceding tests in the same run:
Removing any one of these will give a different error in
|
The The |
The |
(However I'm not sure if that's the best fix, since it doesn't explain why the |
TL;DR summary: GOOD Running test_lll.py by itself:
(expected OverflowError for m n > 32) BAD Running test_{bkz,bkz_python,cvp,gso,lll}.py in sequence:
GOOD Running test_{bkz,bkz_python,cvp,gso,lll}.py in sequence, with dimensions without 50,50 and 60,60:
GOOD Running test_{bkz,bkz_python,cvp,gso,lll}.py in sequence, with test_lll_init disabled/removed:
(expected OverflowError for m n > 32) |
I can reproduce this in a 32-bit VM. Not sure when I'll get around to fixing it, though, unfortunately. |
Do you think it's fine to just "fix" the test by removing those higher dimensions? I would have done it already, but the failure mode is so weird I thought it might indicate some deeper problem (that might also affect 64 bit platforms). |
I applied this work around, like this: https://anonscm.debian.org/cgit/python-modules/packages/fpylll.git/tree/debian/patches/workaround-test-lll-lll-32-bit.patch?id=9307be893b783c97d91e65bec415e55c24f355e4 Sage seems to be unaffected, I have no extra test failures on x86 relating to fpylll, beyond the failures already observed on x86-64. |
Hi, I've merged your patch. You patch is actually more than a workaround, it is the fix that's needed. Thank you! Would you like us to cut a new release for easier packaging? |
Hey, thanks for merging. A new release is not necessary for Debian since I've already applied the patch, but might make things slightly easier for other distributions. I thought it was only a "work around" because those larger dimensions cause |
AFAIK the global state is simply the global RNG state, leading to different lattices being presented to the test depending on how the thing is run. We should definitely fix that. |
See https://buildd.debian.org/status/package.php?p=fpylll&suite=experimental failing on armhf armel i386 mips mipsel powerpc x32
I think 32-bit is the common factor here.
The text was updated successfully, but these errors were encountered: