[BUG]: trying to catch a py::type_error makes the compilation crash with Clang on Windows

[MatthieuHernandez]

Required prerequisites

• Make sure you've read the documentation. Your issue may be addressed there.
• Search the issue tracker and Discussions to verify that this hasn't already been reported. +1 or comment there if it has.
• Consider asking first in the Gitter chat room or in a Discussion.

Problem description

The compiler Clang v13.0.0 craches at the linking step on Windows when add to following line to my code :

 try {
    auto casted_values = py::cast<std::vector<int32_t>>(values); // Here values is a py::list&.
  } catch (const py::cast_error& e) {
    throw py::type_error("this method expects integers as param");
  }

The error is :

LLVM ERROR: Associative COMDAT symbol '??0builtin_exception@pybind11@@QEAA@PEBD@Z' does not exist.
PLEASE submit a bug report to https://bugs.llvm.org/ and include the crash backtrace.

The strange thing is that if I recompile the same code but keep the CMake cache, it works. 🤔
It also works perfectly under Linux with CGG.

Does anyone have an explanation for this?

EDIT : It seems that the problem comes from the try catch but I have no idea why.

[henryiii]

This looks like a compiler error - "PLEASE submit a bug report to https://bugs.llvm.org/ and include the crash backtrace.".

[MatthieuHernandez]

Yes but this bug only happens when I try to catch a pybind11 error, if I throw a std::runtime_error instead it works perfectly.

[henryiii]

Still shouldn't cause a compiler error. Nothing should cause a compiler error.

[MatthieuHernandez]

I just updated Pybind11 to v2.9.1 and that solved the problem. Sorry for the inconvenience 😅
EDIT : Actually no, I just forgot to clean my CMake cache. But I can now tell you that this problem does not happen in debug mode, only in release mode.

[henryiii]

A change to pybind11 isn't going to change anything, this is still a clang on Windows bug, I'd recommend opening it with them and linking here. A compiler should never segfault when compiling.

catch a pybind11 error, if I throw a std::runtime_error instead it works perfectly.

That just means their bug has to do with something custom in our error. I expect they test on std::runtime_error. Still their bug. If the compiler stops crashing and instead complains about something, then we can work it out.

(However, since it works the second time around, I very much doubt it will complain about anything)