concurrent.sh

A concurrency "module" for /bin/sh shell. Works via messaging to a dedicated helper process. Far from being complete.

Written out of wishing for configure scripts to be parallelized. Thread-ifying the shell itself would mean reimplementing subshells as threads instead of child processes, due to restrictions imposed by pthread_atfork.

Primitives available:

• spawn process (a "job")
• query job state
• join a job (wait for it to exit)
• parallel-each
• waiting on an N-job barrier
• lock/try-to-lock/unlock a spinning lock (The locks are non-recursive, non-error-checking, and non-robust).

Relies on:

• Temporary files not to be touched by another process.
• Async subshells implemented by forking.
• Either:
  • Writes to FIFOs being uninterveined. E.g. if shell A does echo a b c >fifo, and shell B does echo d e f >fifo, then read x <fifo result should be either a b c or d e f, but not ad be f c or similar. The test_fifo.sh could be used to probabilistically check this for a given OS+libc+shell combination.
  • CONCURRENTSH_TRANSFER being set to files. The helper process would then poll temporary directory instead of reading from a pipe (FIFO).

Requires the user not to call trap ... EXIT, as well as not to assign to atexit_functions, and some variables starting with job_ and peach_. Manually created async subshells, before invoking concurrency functions, need to assign a unique value to job_self, and then call job_create_wakeup_fifo.

Takes up a file descriptor for communicating with the helper.

Tested on:

• Linux tmpfs, glibc 2.25, bash 4.3.42, dash 0.5.8.2, mksh 52.