Port to Windows

[wegank]

This work was done during the JNCF 2025 and is now rebased on master. I guess I won't have time to clean up everything or split it into multiple PRs before going on vacation tonight, so I'm marking it as a draft (but at least it can serve as a demo).

In short, this is achieved by providing the following headers:

• src/fglm/aligned_alloc.h, which provides a handwritten implementation of posix_memalign on Windows. I also provided posix_memalign_free for the reason in https://learn.microsoft.com/en-us/cpp/standard-library/cstdlib?view=msvc-170#remarks-6, and modified files doing free over memory allocated by posix_memalign to use posix_memalign_free.
• src/msolve/getdelim.h, which provides Gnulib's implementation of getdelim and getline on Windows.

In addition, I modified src/neogb/sort_r.h to prevent undefined reference to qsort_r (though noporpoise/sort_r#17 might be a better idea), and added b to fopen access modes so that we have LF instead of CRLF on Windows. I modified src/usolve/refine.c to use mpz_get_ull from Bitwuzla instead of mpz_get_ui to always get 64-bit integers even if long is 32-bit. I also fixed a few bugs that will be the subject of separate PRs.

All tests passed on CI and my aarch64-windows machine.

[ederc]

Commit f72bc6c is already included in #252.

[wegank]

Sure, my goal was precisely to split this into multiple PRs to make it easier to review; I included the last three commits because each of them is necessary to fix part of the tests on Windows.

[wegank]

I had almost forgot that I had this PR on Windows, and it is already post JNCF 2026, so another quick rebase; I haven't checked if everything still works on Windows (the CI should tell me anyway).

[talregev]

This PR can also handle the msvc atan2?

I check with the AI, it only the test on configure.ac, that can test with flags for configure.

[wegank]

This port targets MinGW. I assume that MSVC support won't be available until CMake builds are implemented…

[talregev]

This port targets MinGW. I assume that MSVC support won't be available until CMake builds are implemented…

That not true. In vcpkg we can compile msvc with autoconf.
I already succeeded to run the config step, now it fail on build step, and that you PR is needed.

[wegank]

Ah, that's good news. I really should try to figure out how to build with MSVC one day.

[talregev]

I can help you to do it on the ci.

currently when I compile with msvc (without your PR)
I get this error:

C:\src\vcpkg\buildtrees\msolve\src\v0.9.4-7cd9c24250.clean\src\usolve../neogb/tools.h(25): fatal error C1083: Cannot open include file: 'sys/time.h': No such file or directory

Do you want first let your PR merge, then we can deal with msvc errors?

[wegank]

I can help you to do it on the ci.

currently when I compile with msvc (without your PR) I get this error:

C:\src\vcpkg\buildtrees\msolve\src\v0.9.4-7cd9c24250.clean\src\usolve../neogb/tools.h(25): fatal error C1083: Cannot open include file: 'sys/time.h': No such file or directory

This looks like a portability issue. If you find a solution, feel free to open a new PR for that.

Do you want first let your PR merge, then we can deal with msvc errors?

I don't think this PR will be merged anytime soon. As you can see, there has been no new review for a few months, while my other PRs are processed rather quickly. It is clear that Windows support is absolutely not a priority for the devs…

[talregev]

It is clear that Windows support is absolutely not a priority for the devs…

I guess the msvc support is very low priority as well.

[talregev]

Keep your good job and split this PR to many smaller PRs that devs may be interested.