[Debian-science-sagemath] Overview of doctest failures

[Ximin Luo]

Ximin Luo:
> [..]
> 
>> I didn't use the new flint-arb because it FTBFS for me so I got 67 failing doctests in sage/rings/complex_arb.pyx.
>>

As you may have noticed, I pushed a workaround for this to our apt repo. The underlying bug is not in ARB itself, but is present in many FOSS projects, it's caused by omission of __volatile__ when doing __asm__ division, and I've documented it here:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82677

I also filed a bug to GMP to discuss the proper fix:

https://gmplib.org/list-archives/gmp-bugs/2017-October/thread.html#4231

I've pushed an incomplete fix to the FLINT (not ARB) Debian git repo, since that contains the code that is copied from GMP that is causing the test failure in ARB. I'm waiting on further input from GMP, and then I'll forward the final version of the patch onto FLINT (and linux, coreutils, mpfr, gcrypt, openSSL, etc etc...). Yeah, shit sucks. If anyone has expertise in assembly, please contribute to the GMP thread.

Anyway in the meantime in Debian we can upload this incompletely-fixed FLINT, and upload ARB without changes, and carry on with other things.

>> We also have some failing tests due to additional glpk output now because of dt-version-glpk-4.60-extra-hacky-fixes.patch
>> Should we just add some ellipsis to the tests to silence that?
>>
> 
> Not sure, I'll take a look in the next few days.
> 

The test failures are not caused by my patch, but caused by Sage's extra GLPK patch:

https://git.sagemath.org/sage.git/tree/build/pkgs/glpk/patches/glp_exact_verbosity.patch?h=develop

It has not yet been accepted upstream.

I tried to work around this by instead patching Debian's Sage as attached, but it doesn't work. Some of the test failures are indeed fixed, and the set_verbosity failures remain as expected. However, even more tests fail because apparently Sage's existing set_verbosity(0) does not silence everything (as my patch does), and Sage doctests are currently expecting some extra output from GLPK even on verbosity=0.

I think we should just live with these test failures for now, there are only 6. Ideally Sage would push the patches upstream, perhaps ping the 2011 thread again and explain why glp_term_out is not sufficient:

https://lists.gnu.org/archive/html/help-glpk/2011-10/msg00057.html

X


-- 
GPG: ed25519/56034877E1F87C35
GPG: rsa4096/1318EFAC5FBBDBCE
https://github.com/infinity0/pubkeys.git
-------------- next part --------------
A non-text attachment was scrubbed...
Name: glpk_sage_fake_verbosity.patch
Type: text/x-patch
Size: 1268 bytes
Desc: not available
URL: <http://lists.alioth.debian.org/pipermail/debian-science-sagemath/attachments/20171030/4509c978/attachment.bin>