mirror of
https://github.com/pytorch/pytorch.git
synced 2025-10-20 12:54:11 +08:00
17567e5b29
Summary: - Importing torch on Windows can cause a crash within python. - The problem was introduced by the change in `Module.cpp` from https://github.com/pytorch/pytorch/pull/94927 - The cause is that a call to `PyObject* initModule(void)` declared with a `__declspec(dllimport)` specifier can lead to a crash if the definition doesn't include the `__declspec(dllexport)` counterpart. - To mitigate the problem without introducing customized macros and changing the build system (note, `#include <c10/macros/Export.h>` doesn't work in `stub.c`) is to simply remove the `__declspec(dllimport)` specifier. - According to https://web.archive.org/web/20140808231508/http://blogs.msdn.com/b/russellk/archive/2005/03/20/399465.aspx and other sources, `__declspec(dllimport)` only leads to some code optimizations, and since `initModule()` is only called once at startup, this is marginal. - Note: the `stub_with_flatbuffer.c` file counterpart wasn't affected, therefore, not touched. Differential Revision: D44236183 Pull Request resolved: https://github.com/pytorch/pytorch/pull/97193 Approved by: https://github.com/ezyang
16 lines
241 B
C
16 lines
241 B
C
#include <Python.h>
|
|
|
|
extern PyObject* initModule(void);
|
|
|
|
#ifndef _WIN32
|
|
#ifdef __cplusplus
|
|
extern "C"
|
|
#endif
|
|
__attribute__((visibility("default"))) PyObject* PyInit__C(void);
|
|
#endif
|
|
|
|
PyMODINIT_FUNC PyInit__C(void)
|
|
{
|
|
return initModule();
|
|
}
|