Add linux builds to releases

[AltTheAi]

It is incredibly annoying to build cuda projects on Arch Linux because cuda is really funny when it comes to compatibility issues with gcc. Unless you have older versions of gcc, you are screwed because current cuda is not compatible with latest gcc. Downgrading gcc is not really easy since many packages rely on it being up to date. So would it be possible to add linux builds to releases?

[mooskagh]

What build error do you have in the latest GCC?

Usually the latest gcc is not a problem because some devs actually use arch for development so incompatibilities are noticed and fixed quickly.

[SolsticeSpectrum]

You can't use gcc 14 to compile it, nvcc supports only gcc up to 13.2. So I had to compile gcc 13.2 because I installed it with yay -S gcc13. This step took me 4 hours to compile it on my machine. And I don't even mention many packages I had to uninstall first because yay would throw errors that dependency would break. After all that I was able to use the flag -Dnvcc_ccbin=g++-13. All this could have been skipped only if linux build was in releases.

https://forums.developer.nvidia.com/t/cuda-12-4-nvcc-and-gcc-14-1-incompatibility/293295

Using older toolchains on rolling distros is annoying.

[mooskagh]

But gcc13 is a package in arch (i.e. it's not in AUR or something), and cuda (which provides nvcc) does depend on gcc13. You install cuda package, and then Lc0 just builds.

I agree that NVIDIA tools not supporting the newest compilers is annoying, but having pre-built linux binaries doesn't sound like the right way to fix it.

[SolsticeSpectrum]

maybe im doing something wrong but gcc13 is not available on the regular repo for me

[mooskagh]

Here https://archlinux.org/packages/extra/x86_64/gcc13/ it suggests it's part of the extra/ which is where other compilers (e.g. latest clang, https://archlinux.org/packages/extra/x86_64/clang/) are located.