[Debian-science-sagemath] cypari2 hangs on ppc64el when pari stack overflows

[Tobias Hansen]

On 10/06/2018 12:30 PM, Jerome BENOIT wrote:
>
> On 06/10/18 14:17, Jerome BENOIT wrote:
>> Hello,
>>
>> On 06/10/18 13:36, Tobias Hansen wrote:
>>> Hi,
>>>
>>> sagemath 8.3 currently fails to build on ppc64el due to a test that triggers a pari stack overflow sometimes hanging indefinitely. The bug can be triggered by the following Python code:
>>>
>>> import cypari2
>>> pari = cypari2.Pari()
>>> pari.allocatemem(200000)
>>> pari('x^2 + 10^100 + 1').nfinit()
>>>
>>> The last line sometimes hangs indefinitely on ppc64el instead of throwing an error. When running the equivalent code directly in gp (here one needs a smaller stack size to trigger the overflow)
>>>
>>> allocatemem(50000)
>>> p = x^2 + 10^100 + 1
>>> nfinit(p)
>>>
>>> it throws the error as on other architectures. So I think the bug should be in cysignals or cypari2. Before reporting this (which will maybe not help much since people usually don't have access to this architecture) it would be good to test with the latest version of cysignals (we still have 1.6.7). What's the status of getting that packaged?
>> I am working on cysignals:
>> the a upstream set-up machinery was changed (and badly). I have my idea on how to fix it.
>> But I need time before me. I may have time at the hand of the month to  focus on it.
> Is the issue the one reported here: https://launchpad.net/ubuntu/+source/cysignals/1.6.7+ds-4ubuntu1
>
> Jerome
>

Probably not. I found a failed Ubuntu autopkgtest of cysignals on armfh where C backtraces were missing, similar to [1]. I think that's what this patch is about.

I don't even know if cysignals 1.7.2 fixes the problem I talked about.

Maybe I'll just disable the test in sage for now.

Best,
Tobias

[1] https://github.com/sagemath/cysignals/issues/77