scripts/bootstrap-prefix: make darwin bootstrap with clang

[plxty]

In Darwin there's plenty of tools that now use macOS infra such as nodejs, many rust libraries... For GCC lacks of macOS builtins and the blocks extension, building many tools require a lot of hack, switching to clang will reduce some pain.

This is still a proof-of-concept style, it works on my macOS 26, but may not be a clean way, and may not work for other macOS version.

Changes explaination:

1. Darwin ships with builtin libc++, for Gentoo clang with isolated libc++ to work, changes required: compiler flag -stdlib++-isystem to use Gentoo headers, and building libcxx requires -DLIBCXX_ABI_NAMESPACE= to avoid conflicting the system one.
2. So in stage 2 we still use macOS libcxx, then in stage3 we shall emerge the libcxx and switching toolchain like ninja and rest to use the Gentoo lib.
3. Uniform STAGE and BOOTSTRAP_STAGE, but I'm not sure if that's the correct way :/
4. Adding bash requirement check for array stuff like ( xxx ) to work

Still this is a proof-of-concept, managed to get bootstrap but may encounter other issues.

[thesamesam]

It looks good to me.

[grobian]

what in particular does this fix? does it accidentially match a wrong package?

[plxty]

It's a little fix for re-stage, for example when first attempt of stage 2 is failed for some package, when I retry it will ignore some already merged packages and then re-emerge again.

The particular example is llvm-runtimes/clang-runtime, which in here -runtime will get stripped and leaving final evdb as llvm-runtimes/clang which is a non-existance package.

Safe to revert, as it just cost a little more re-bootstrap time for debugging :)

[grobian]

ack

[grobian]

We should, but this is the same for GCC. Are you trying to avoid a rebuild or something? If the package didn't update, and it's good, it should stay where it is, or is there another reason?

[grobian]

so I now understand where this mess comes from, I just cleaned it up in da92e6e and we can use BOOTSTRAP_STAGE always/everywhere now

[plxty]

With newer clang released and BOOTSTRAP_STAGE unified, the script now has little problems which prevents bootstrapping again. Still fixing and may take a while for fully setup.

Current issues:

• New LLVM has no prebuilt docs and thus requires sphinx to build it live, which is a little burden, so for now I choose to pin down the LLVM version, may not be elegant, but still I don't want to add extra dependencies.
• The bootstrap profile in stage3 is missing /clang variant. In previous bootstrap_profile it relies BOOTSTRAP_STAGE to make the Darwin choose /clang profile for ${ROOT}/etc, and non-variant profile for ${ROOT}/tmp/etc, but after exporting the BOOTSTRAP_STAGE, the profile doesn't valid any more and need to be adjusted. I've been swapped the location but it's still under testing.
• Emerging new clang-23 (upgrading) causes config issue for llvm-runtimes/compiler-rt, which requires a fully usable -L/-isysroot before compiling, but at that time the new clang-23 has no suitable config other than the linker-config, I may try moving the Darwin prefix part inside clang-common into linker-config, but may not be ideal