[typedrat]

With dynamic linking, a joinable thread posts CMD_MARK_AS_FINISHED to the main thread as it exits, so that dlopen()'s sync skips it. But a join cleans the thread up as soon as it returns, possibly before that message is handled, and a new thread created in the meantime can get the same pthread_t. The late message then marks the new, running thread as finished, so later dlopen() calls skip it, and its calls into newly loaded modules trap with "table index is out of bounds".

The fix only marks a thread if it is still known and has actually exited (detach_state is DT_EXITING or DT_EXITED). The check lives in markAsFinished, which reads detach_state directly via C_STRUCTS.pthread.

Test: core.test_pthread_dlsync_reused joins a thread on the main thread, creates another one that reuses its pthread_t, then dlopen()s a side module and calls into it from the new thread. It fails on main with the trap and passes with the fix, in core0–core3 and wasm64. It assumes the stale message is handled within the 100 ms before dlopen(); if it weren't, the test could pass without the fix, but it can't fail with it.

Fixes: #27841

[clementperon]

@sbc100 I reach to a similar patch.

And I repro on 6.0.9 with Kodi-style thread churn + dlopen(): 62/140 runs trap without this change, 0/140 with it.

Needs a rebase after #27999.