pthread_key_create does not initialize thread local variable with null

[AndreasReich]

The following code fails with emscripten right now:

pthread_key_t k;
pthread_key_create(&k, NULL);
assert(pthread_getspecific(k) == 0); // fine

pthread_setspecific(k, (void*)1);
pthread_key_delete(k);

pthread_key_create(&k, NULL);
assert(pthread_getspecific(k) == 0); // fails

According to POSIX specification, after creation of a thread local key, the value should be zero on key creation on all threads.
See: http://pubs.opengroup.org/onlinepubs/009695399/functions/pthread_key_create.html

[VirtualTim]

@abujalski is working on similar bugs.

[abujalski]

It looks like bug in musl library.

Function int pthread_setspecific(pthread_key_t k, const void *x) (emscripten/system/lib/libc/musl/src/thread/pthread_setspecific.c) sets struct pthread.tsd[k] array element.

Function int __pthread_key_create(pthread_key_t *k, void (*dtor)(void *)) (emscripten/system/lib/libc/musl/src/thread/pthread_key_create.c) sets only dtor associated with given pthread_key_t and doesn't set pthread.tsd[k] to NULL.

[AndreasReich]

Yes, musl's implementation is wrong. Probably should have filed a bug there?
I don't think it is enough to set the value on create to null: Any other thread that might read its value for the same key will still not see null.
I checked the libc implemention, there they use a generation counter which is compared on every access. If a thread's local generation for a given key is lower than a globally known location for the key, then the local value is set to null on access.

[stale]

This issue has been automatically marked as stale because there has been no activity in the past year. It will be closed automatically if no further activity occurs in the next 7 days. Feel free to re-open at any time if this issue is still relevant.

[AndreasReich]

I believe this is still unchanged. Need to re-test against newest version though to be fair

[stale]

This issue has been automatically marked as stale because there has been no activity in the past year. It will be closed automatically if no further activity occurs in the next 30 days. Feel free to re-open at any time if this issue is still relevant.

[sbc100]

ping

[sbc100]

This bug looks like its fixed in the latest musl code: https://git.musl-libc.org/cgit/musl/tree/src/thread/pthread_key_create.c#n57

So it should be resolved once we land: #13006 (hopefully some time in the next decade!)