Windows: leaking thread handles

[luben]

On windows, when using workers for compression it looks it's leaking thread-handles (not threads). This issue got reported by user of zstd-jni but it looks that's issue with the underlying native code: luben/zstd-jni#203 - he was able to reproduce by using only the zstd CLI.

I don't have Windows system to confirm or test a fix, but it looks we need to call https://docs.microsoft.com/en-us/windows/win32/api/handleapi/nf-handleapi-closehandle after joining the thread:

zstd/lib/common/threading.c

Lines 23 to 77 in 12c045f

	#if defined(ZSTD_MULTITHREAD) && defined(_WIN32)
	/**
	* Windows minimalist Pthread Wrapper, based on :
	* http://www.cse.wustl.edu/~schmidt/win32-cv-1.html
	*/
	/* === Dependencies === */
	#include <process.h>
	#include <errno.h>
	/* === Implementation === */
	static unsigned __stdcall worker(void *arg)
	{
	ZSTD_pthread_t* const thread = (ZSTD_pthread_t*) arg;
	thread->arg = thread->start_routine(thread->arg);
	return 0;
	}
	int ZSTD_pthread_create(ZSTD_pthread_t* thread, const void* unused,
	void* (*start_routine) (void*), void* arg)
	{
	(void)unused;
	thread->arg = arg;
	thread->start_routine = start_routine;
	thread->handle = (HANDLE) _beginthreadex(NULL, 0, worker, thread, 0, NULL);
	if (!thread->handle)
	return errno;
	else
	return 0;
	}
	int ZSTD_pthread_join(ZSTD_pthread_t thread, void **value_ptr)
	{
	DWORD result;
	if (!thread.handle) return 0;
	result = WaitForSingleObject(thread.handle, INFINITE);
	switch (result) {
	case WAIT_OBJECT_0:
	if (value_ptr) *value_ptr = thread.arg;
	return 0;
	case WAIT_ABANDONED:
	return EINVAL;
	default:
	return GetLastError();
	}
	}
	#endif /* ZSTD_MULTITHREAD */

[Cyan4973]

Thanks for the report @luben !

The issue description and the proposed fix look both correct to me.

An indirectly related issue is : how could we detect such Window-specific leaks in CI ?
We already have leak-detection tests, but they only run on Linux,
hence this code, which is only enabled on Windows, isn't tested for leaks.

Since our experience of Windows is limited, we would gladly accept any pointer that would help creating such a test for Windows CI.

CC @vnair81 , the author of #2263.