Follow up to #30596: missing symbol in linking on gentoo with gcc10

[kiwifb]

As per #30596 comment:4, #30596 leads to some missing symbols in linking on Gentoo with gcc-10.

The first consequence of it, is failing to build the documentation:

Traceback (most recent call last):
  File "sage_setup/docbuild/__main__.py", line 1, in <module>
    from sage_setup.docbuild import main
  File "/dev/shm/portage/sci-mathematics/sage-9999/work/sage-9999/src-python3_8/sage_setup/docbuild/__init__.py", line 58, in <module>
    import sage.all
  File "/dev/shm/portage/sci-mathematics/sage-9999/work/sage-9999/src-python3_8/build/lib/sage/all.py", line 129, in <module>
    from sage.data_structures.all import *
  File "/dev/shm/portage/sci-mathematics/sage-9999/work/sage-9999/src-python3_8/build/lib/sage/data_structures/all.py", line 3, in <module>
    from .bitset import Bitset, FrozenBitset
ImportError: /dev/shm/portage/sci-mathematics/sage-9999/work/sage-9999/src-python3_8/build/lib/sage/data_structures/bitset.cpython-38-x86_64-linux-gnu.so: undefined symbol: _bitset_issubset

Only 3 symbols from bitset_intrinsics.h are missing

ldd -r /dev/shm/portage/sci-mathematics/sage-9999/work/sage-9999/src-python3_8/build/lib/sage/data_structures/bitset.cpython-38-x86_64-linux-gnu.so | grep _bit
undefined symbol: _bitset_issubset      (/dev/shm/portage/sci-mathematics/sage-9999/work/sage-9999/src-python3_8/build/lib/sage/data_structures/bitset.cpython-38-x86_64-linux-gnu.so)
undefined symbol: _bitset_isempty       (/dev/shm/portage/sci-mathematics/sage-9999/work/sage-9999/src-python3_8/build/lib/sage/data_structures/bitset.cpython-38-x86_64-linux-gnu.so)
undefined symbol: _bitset_eq    (/dev/shm/portage/sci-mathematics/sage-9999/work/sage-9999/src-python3_8/build/lib/sage/data_structures/bitset.cpython-38-x86_64-linux-gnu.so)

Declaring those three inline functions as static fix the problem. This probably implies that the inline keyword semantic from C99+ is involved.

CC: @tscrim

Component: cython

Keywords: bitset

Author: François Bissey

Branch/Commit: 2cf0ebe

Reviewer: Jonathan Kliem

Issue created by migration from https://trac.sagemath.org/ticket/30675

[kiwifb]

Author: François Bissey

[kiwifb]

Commit: 7edd1b6

[kiwifb]

Branch: u/fbissey/bitset_static

[kiwifb]

Changed dependencies from 30596 to #30596

[kiwifb]

comment:1

@tscrim would you be so good as to check that it doesn't create problems for you?

---

New commits:

887e1f3	outsource functions that can be optimized by intrinsics
bcfa7f1	add file
49a95f3	use same type declarations as in bitset.pxi
7edd1b6	make problematic inline functions static.

[kiwifb]

comment:2

My changes are only the last commit of course.

[kliem]

comment:3

Ok, I think I understand now.

This helped me:

http://gudok.xyz/inline/

It seems that there are three solutions to this:

• declare the function as static, as you did
• add a file bitset_intrinsics.c and declare the functions in there extern inline ...; I don't know how well this plays with cython
• go for C++ and should be fine

[kiwifb]

comment:4

Yes, that would sum it up. And I think that option 2 would be problematic but I wouldn't know before trying for real. I hadn't thought about going full C++ (because I hate C++) but that would be a viable option.

[kliem]

comment:5

But then all of the functions should be made static, shouldn't they.

This problem will also show up, when you compile a debug version, because then nothing will be inlined and nothing will work anymore.

[kliem]

comment:6

At the moment, I don't see a need of going C++. It might arise at some point however, in particular because C++ supports templates. So far, I'm handling the templates via fused types in cython, that also works.