Backend C compiler comparisons

[kayhayen]

I am aiming at adding resource usage information for Scons jobs launched to the compilation reports. On Linux and macOS this will be available from rusage and once we tie it to modules, we can checkout the variance added with the C cache of course being disabled. We can then.

a) Measure how much overhead ccache is in the cold cache case.
b) Compare "--eperimental=thin-lto --clang" with "--clang" to decide what we want to make the default for LTO with Clang in Nuitka.
c) Make a page comparing C backends for performance on Linux and macOS, while for the later, really only ever system gcc should be used, we can also check Anaconda's conda-cc still there, and compare with gcc, and such for large C compilations.

[polarathene]

It might be good to compare with Zig too via ziglang package?

I am still putting together a report of findings as I troubleshoot an example. Some scenarios have proven to be less viable with Zig alone (unrelated to Nuitka AFAIK as only encountered when installing python packages via sdist instead of bdist).

[kayhayen]

Currently zig works only via the C++03 workaround, which limits its use to Python3.10 as we need full C11 support with atomics, so zig is not a supported C compiler yet outside of configurations that few people will consider, so i would skip on it. It's major strength would be cross compilation, but Nuitka itself needs to learn that properly first. So while I am a huge zig fan, unfortunately no.