lab-threads-1

- finish the sleep -> join (thread num in argv[1]) task
- write a simple program that creates couple of threads, each doing a number arithm ops
  - after random number of the ops, the thread SIGSEGV's, observe the result
    - compile with debug flags, see if gdb can display stacks of all threads at the moment of crash
- write simple program with 2 thread types - one sleeping in syscall, another looping and printing to stdout
  - create more than 1 thread of each type
  - the threads doing syscals can e.g. pick a random file from /etc directory, open it and read it
  - each thread will write its own thread ID to stdout once running
  - goal: see if the other threads run while another sleeps in syscall
  - see if truss/strace can display thread IDs
    - need to use strace -f + print the thread ID with gettid()
- thread glob vars
  - create N threads
  - each thread will have its own "global" variable with dynamically allocated content,
    i.e. have to specify a destructor in pthread_key_create() and call free() in it
  - see if the destructor is called when thread a) returns b) pthread_exit() c) is cancelled
- check that errno is different in each thread
- pthread_once()
  - find good use case, e.g. shared connection in a library (to LDAP server say)
- implement thread pool
  - each thread will perform arithmetic operations in a cycle and will exit after random number of iterations
  - the main thread will notice the exit and create new thread
  - need condvars or signals+pthread_kill() (in both cases use pthread_detach())
  - use pthread_cleanup_push() (always call pthread_exit() when the thread is done) to perform the signalling

----

- thread synchro using signals
  - block given signal (e.g. USR1) and use sigwait() in producer/consumer
- parallel connect using threads
  - cannot use pthread_join() since it waits on specific thread
  - really needs condvars