tags: ftbfs

plfit FTBFS on s390x with test failures.

While this is not rc for plfit because plfit has never built on s390x,
it is blocking the migration of a rc bug fix in igraph to testing as
igraph recently switched from using an embedded copy of plfit to
using the plfit package..

The issue is actually discussed upstream even before this bug was filed.

As already indicated in https://github.com/ntamas/plfit/issues/38 the
problem is related to the size of long double.

The appended debdiff solved the problem on Ubuntu Jammy by forcing long
double to 64bit.

Best regards

Heinrich

close 1003395 0.9.5+ds-1
thanks

Hello.

This was fixed upstream and there is apparently no need to apply
any patch anymore.

I've written the long explanation in the github issue:

https://github.com/ntamas/plfit/issues/38

TLDR: It was fixed by the line in src/CMakeLists.txt saying this:

target_link_libraries(plfit ${MATH_LIBRARY})

Cc:ing all relevant parties.

Thanks.

reopen 1003395
retitle plfit: under-the-radar numerical issue


Hi,

I am re-opening the issue because
it has simply gone under the radar but it is still there.

The issue is a numerical issue that deserves an appropriate numerical analysis.
The upstream code currently workarounds by using at some point `long double`.
This is a very bad numerical practice: all the computation must be made with `double`.
The issue is actually not numerically trivial. I am on my way to elaborate a numerical fix.

Best wishes,
Jerome

tags 1003395 - ftbfs
tags 1003395 - patch
thanks

El 3/8/24 a las 23:31, Jerome BENOIT escribió:

Ok, I would have used a different bug for that myself, but if you want to use
the same bug for the underlying issue, so be it.

Note, however, that the original report said the package FTBFS, which does not
happen anymore, and it had a patch from Ubuntu which will probably not be applied,
so I hope you don't mind if I adjust the metadata accordingly.

Thanks.