GeezerVison

Build fails on macOS with Clang 12.0.0 (clang-1200.0.32.27): "illegal thread local variable reference"

building 'math' extension
gcc -Wno-unused-result -Wsign-compare -g -O0 -Wall -std=c99 -Wextra -Wno-unused-result -Wno-unused-parameter -Wno-missing-field-initializers -Wstrict-prototypes -Werror=implicit-function-declaration -fvisibility=hidden -I./Include/internal -I./Include -I. -I/usr/local/include -I/Users/runner/work/cpython/cpython/Include -I/Users/runner/work/cpython/cpython -c /Users/runner/work/cpython/cpython/Modules/mathmodule.c -o build/temp.macosx-10.15-x86_64-3.10-pydebug/Users/runner/work/cpython/cpython/Modules/mathmodule.o -DPy_BUILD_CORE_MODULE
gcc -bundle -undefined dynamic_lookup build/temp.macosx-10.15-x86_64-3.10-pydebug/Users/runner/work/cpython/cpython/Modules/mathmodule.o Modules/_math.o -L/usr/local/lib -lm -o build/lib.macosx-10.15-x86_64-3.10-pydebug/math.cpython-310d-darwin.so

ld: illegal thread local variable reference to regular symbol __Py_current_tstate for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)

On MSC, we can try to use __declspec(thread). I tried but I got an error about DLL export. Maybe it should not be exported, and so the variables should not be used when building an extension module (not built as a builtin module).

cc @markshannon who loves TLS :-)

See https://bugs.python.org/issue40522#msg383899 for the emitted assembly code.

See also https://stackoverflow.com/questions/18298280/how-to-declare-a-variable-as-thread-local-portably

"In C11, the keyword _Thread_local is used to define thread-local variables. The header <threads.h>, if supported, defines thread_local as a synonym for that keyword."
https://en.wikipedia.org/wiki/Thread-local_storage#C.2B.2B

Would it be possible to keep all the portability macros in one place by putting something like #define Py_ThreadLocal(type, varname) ... in pyport.h?

One other remark (not for this PR, but for future work):
The HotSpot JVM uses aligned stacks to be able to access its thread-local information as fast as possible by zeroing the low bits of the machine stack pointer.
It is possible to get very close to this in (mostly) portable C https://godbolt.org/z/eG64zz

I need to test -femulated-tls clang flag. See also: http://llvm.org/docs/LangRef.html#thread-local-storage-models

This PR is stale because it has been open for 30 days with no activity.

@vstinner Did you ever get anywhere with this?

@vstinner Did you ever get anywhere with this?

I got compiler issues and then I stopped working on this PR.

A compromise would be to find a way to use a C99 keyword internally, but use a function call externally.

"externally" includes stdlib modules built as shared libraries (Linux .so, Windows .dll).

A compromise would be to find a way to use a C99 keyword internally, but use a function call externally.

Or simply mandate C99? It's been >20 years...

Or simply mandate C99? It's been >20 years...

See previous comments. I got compiler errors to use this "recent feature". Sorry, the feature comes from C11, not C99. I don't call.

I failed finding time to fix compiler errors. I prefer to abandon this PR for now.