Thread recycling makes Celeste crash #68
Description
Crash during Celeste intro. Error message and stack trace of the thread below:
* Assertion at ../../../mono/utils/hazard-pointer.c:149, condition `mono_bitset_test_fast (small_id_table, id)' not met
Thread 4 (Thread 0x7fffb96ce700 (LWP 15511)):
#0 0x00007ffff7b2e24a in __waitpid (pid=15740, stat_loc=0x7fffb96cc9fc, options=0) at ../sysdeps/unix/sysv/linux/waitpid.c:30
#1 0x00000000006b4f4e in mono_handle_native_crash ()
#2 <signal handler called>
#3 __GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux/raise.c:51
#4 0x00007ffff797a2f1 in __GI_abort () at abort.c:79
#5 0x00000000006e0294 in mono_log_write_logfile ()
#6 0x00000000005ff640 in monoeg_g_logv ()
#7 0x00000000005ff796 in monoeg_assertion_message ()
#8 0x00000000005e521a in mono_thread_small_id_free ()
#9 0x00000000005f3ea3 in unregister_thread ()
#10 0x00000000005f5a39 in mono_thread_info_exit ()
#11 0x000000000057e692 in start_wrapper ()
#12 0x00007ffff7d8fa0f in libtas::pthread_start (arg=0x8bd7e50) at /home/clement/libtas-1.3.0/src/libTAS/threadwrappers.cpp:102
#13 0x00007ffff7b23f2a in start_thread (arg=0x7fffb96ce700) at pthread_create.c:463
#14 0x00007ffff7a3aedf in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95
Some threads in Celeste exit by calling pthread_exit. Because we recycle threads, we replace that call by some code (throwing a c++ exception or using setjmp/longjmp are both possible) to return to our wrapper that initially called the game function, so that we can accept a new thread function to execute without actually spawning a new thread.
During the intro of Celeste, a thread exits using pthread_exit, then the game spawns a new thread so this one is recycled. And the crash occurs when this thread is exiting again, before calling pthread_exit, as another thread is already joining this one. I'm unsure what is causing this, maybe mono does not like to have twice the same thread ids. mono_thread_small_id_free may be the relevant function to look into.