Matthieu Darbois

Thanks for digging into this. Indeed if all libraries are in the wheel and RUN_PATH are correct, this seems like a bug in auditwheel. An MRE would be much appreciated.

@oraluben, thanks for digging into this. I'd keep the default level for compression to 6. While the less than 1% (around .5/.6%) improvement does not seem like a big one,...

For the " do not re-unzip wheel if possible", I'd probably move unzipping at the main_repair/main_show level not to introduce complexity with caching.

I wanted to [play a bit with this](https://github.com/mayeut/pybase64/commit/0558aa317647a915d79f9c21888b4613a8dee6e5) before reviewing (not sure when I'll have time to review it properly). I ran in kind of the same issues as @henryiii...

> I've now added a minimal implementation of that to this PR. Thanks for your detailed answer about CMake handling @mhsmith. Everything works out of the box with the toolchain...

> On Android the OS doesn't provide a C++ library, so every app needs to bundle its own copy. This required giving cibuildwheel some auditwheel-like code which adds the C++...

> I thought it was other way around and we do novel stuff here which venv may or may not choose to incorporate I thought the same. > I don't...

> There is a third type of use case employed by the manylinux container images, that of "executable wrappers" These only exist because of the different/inconsistent/buggy behaviors when using "executable...

There's no limited/stable API/ABI for free-threading yet: https://docs.python.org/3/howto/free-threading-extensions.html#limited-c-api-and-stable-abi The existing abi3 wheel will continue to work on non free-threading builds of python 3.13+. For the free-threading builds, a wheel per...

Thanks for your initial work on this @cyprienc I'll close this PR and follow-up in #1698