[pybind11] Do not support Mac for now

[uilianries]

Specify library name and version: pybind11/2.9.1

• Mac OS is no longer supported (temporary until [BUG] Fatal Python error: PyMUTEX_LOCK(gil->mutex) failed pybind/pybind11#3081 become fixed)
• Removed old releases
• Cache CMake instance with lru_cache
• Remove cmake test support. We use test_package.py only

Related to #10817

/cc @ericLemanissier
/cc @vectorsli

---

• I've read the guidelines for contributing.
• I've followed the PEP8 style guides for Python code in the recipes.
• I've used the latest Conan client version.
• I've tried at least one configuration locally with the conan-center hook activated.

[ghost]

I detected other pull requests that are modifying pybind11/all recipe:

• [pybind11/2.9.2] Add pybind11 2.9.2 #10817
• [pybind11/2.9.2] Add latest version #10515
• [DRAFT] Pybind11: Confirm is failing #9492
• pybind11 cmakedeps #9348

This message is automatically generated by https://github.com/ericLemanissier/conan-center-conflicting-prs so don't hesitate to report issues/improvements there.

[conan-center-bot]

All green in build 6 (2927b51771e6d2cf91198a0f223dbebda0a01876):

• pybind11/2.9.1@:
  All packages built successfully! (All logs)

• pybind11/2.7.1@:
  All packages built successfully! (All logs)

• pybind11/2.8.1@:
  All packages built successfully! (All logs)

[uilianries]

@SSE4 could you please review? So far, it failed with 11.0, 12.0 and 13.0, but those are the version that we have, I don't have an older compiler version to validate, so I only included those 3

[stefansli]

@uilianries Thanks for looking into this and providing a fix. The changes look good to me 👍

[SSE4]

@uilianries I've checked locally, and the same error (Fatal Python error: PyMUTEX_LOCK(_PyRuntime.ceval.gil.mutex) failed) happens for me as well.
checking the comment pybind/pybind11#3081 (comment), they say we're linking against the whole set of components. but seems we aren't, we are just using:

find_package(pybind11 REQUIRED CONFIG)

pybind11_add_module(test_package MODULE test_package.cpp)

so I think the reason it fails is different 🤔

[stefansli]

Let me jump in since we are using pybind11 and this recipe. There are already several issues in CCI regarding the bugged pybind11 recipe.

To sum it up:
Conan creates the target pybind11:pybind11 and links all "components" into this "main" target. However the Module from pybind11 also configures a pybind11:pybind11 target which is the base for all subsequent targets (such as pybind11::embed or pybind11::lto. For more details please see #6605. Now some of the target pybind11 defines and uses in their cmake macros conflict. For example bindings created with pybind11 are either used as a module in which case they are compiled to a *.pyd shared library, or they are embedded in a library/executable.

From what I understand, the recipe will never work properly with cmake_find_package and cmake_find_package_multi (unless one uses the CMake helper files generated by the build).
It might be possible to configure the CMakeDeps generator correctly but I haven't tried this yet because we are not using the CMakeDeps generator in production.

[SSE4]

@vectorsli thanks for the clarification. this is beyond my knowledge on how to set up these components and modules correctly, to closely resemble upstream behavior. maybe @czoido or @jgsogo can help here.

[stefansli]

A nice overview of the problem can be found in this comment:

#6605 (comment)

[prince-chrismc]

The changes look good to me
Let me jump in since we are using pybind11 and this recipe.

Let's get this merged and since I have no clue, @vectorsli i'll ping you for help in the future ❤️