SageMath fails to build on on Fedora 28 with gcc 8.1

[Vaush]

SageMath building produces 3 different errors when run on Fedora 28 64bit with gcc 8.1, specifically when building the packages:

• python 3.6.1
• python 2.7.14
• linbox 1.5.2

The following errors are respectively reported:

• Failure to import the crypt module
• Failure to generate posix vars

• ../../linbox/matrix/densematrix/blas-transposed-matrix.h:74:8: error: too many template-parameter-lists
    class TransposedBlasMatrix< TransposedBlasMatrix< Matrix > > : public Matrix{
        ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  

The first error is due to an unspecified issue with Fedora 28 implementation of crypt(), which as reported in ​https://fedoraproject.org/wiki/Changes/Replace_glibc_libcrypt_with_libxcrypt is not the standard glibc implementation. Since I couldn't find a way to fix the crypt implementation in a reliable manner, I opted to implement the patch reported here ​https://bugs.python.org/issue28503 to allow python to use r_crypt(), which works.

The errors in the last two packages, instead, are due to more general issues with the new 8.1 version of gcc. Python 2.7.14 fails to build because the way structs are aligned in C was changed in gcc 8.1, as described in ​https://bugs.python.org/issue33374#msg315857 .

Linbox fails to build because it uses a construct previously accepted in gcc 7 that instead produces an error in gcc 8, specifically a templated template specialization done in the following way:

template<>
template<typename T>
class Class1<Type1<T>>{
...
}

For specifics on how the errors are solved, see the sage-devel discussion at ​https://groups.google.com/forum/#!msg/sage-devel/NgzlZknrizg/o-_Exw8jCAAJ

N.B: while originally this ticket fixed all 3 bugs, since the last two were solved separately in two different tickets, now the current ticket only fixes the python 3 bug, while mergin the tickets listed as dependencies to fix the others.

[Vaush]

Linbox patch

[Vaush]

Python2 patch

[Vaush]

Python3 patch - 1 of 2

[Vaush]

Python3 patch - 2 of 2

[Vaush]

I added dependencies to #25204,#25353 which respectively fix the python 2 issue and the linbox (and fflas, which is not described here) issue. The commit to be made will be based on them, and will only modify the python 3 package

[Vaush]

New commits:

​4c1223e	Fixed python 3.6.1 crypt issue on Fedora 28
​cb566ea	fixing gcc-8.1 compilation errors.
​b20fd48	Merge remote-tracking branch 'trac/u/cpernet/fflas_and_linbox_broken_with_gcc_8_1_0' into t/25391/sagemath_fails_to_build_on_on_fedora_28_with_gcc_8_1
​d14acae	Upgrade to Python 2.7.15
​473f214	Merge remote-tracking branch 'trac/u/jdemeyer/upgrade_to_python_2_7_15' into t/25391/sagemath_fails_to_build_on_on_fedora_28_with_gcc_8_1

[dimpase]

I'll be checking that this does not break on other systems.

[dimpase]

Have you run tests on your branch?: make ptest (or make ptestlong for more tests)

This is something the ticket's author should do in such cases. (It would not be surprising if something unrelated to your changes fails, as gcc 8 is pretty much untested, but good to know anyway).

[Vaush]

I haven't, but will do

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​5be7adc

Merge branch 'master' into t/25391/sagemath_fails_to_build_on_on_fedora_28_with_gcc_8_1

[dimpase]

What is happening with this on the current 8.3.beta6? Note that python2 has been updated.

[dimpase]

For Python3, it's apparently ​https://github.com/python/cpython/pull/4691 which is still under review.

[jdemeyer]

I am missing some context and documentation. The current branch only fixes Python 3. That is not consistent with the ticket description which mentions several packages and doesn't even mention the actual failure that the Python 3 patch is trying to fix.

[jdemeyer]

The patch files should also contain some description, at a minimum a link to the upstream issue.

And if you add a patch to a package, you should bump the package version.

[Vaush]

Ok, I'm sorry if the description or the execution aren't up to par, this is my first ticket, so I'm getting used to it.
The ticket was originally opened to describe the bug. After that, I succeeded in manually fixing all the bugs listed in the description, but it was brought to my attention that two of my bugs were solved in separate tickets at the same time (see the dependencies field), so I merged those tickets and added my own fix for python3, thus fixing all the issues and solving the problem stated in the ticket title, that is SageMath not compiling on Fedora 28 with gcc 8.1.
Since the commits include the fixes for the other 2 bugs, even if by merging, I thought this counted as having a fix for all 3 of them in the branch, is this wrong?
There is a description for the python 3 bug, since the bug is literally "Python tries to call crypt instead of crypt_r on Fedora, and this for some reason breaks things, so let's allow it to call crypt_r", and the reason why it breaks things is, as pointed out by dimpase, apparently that fedora has a completely different crypt implementation, as can be seen here ​https://fedoraproject.org/wiki/Changes/Replace_glibc_libcrypt_with_libxcrypt.
I'll expand on the description a bit, anyway, to make this clearer, and also to reflect the fact that this ticket only directly fixes the python3 bug, since it's only reported in the comments. I'll also quickly add a description and a link to the patch