command (runner) & conquer

Major release v1.2.0

This release is a major refactor of the stdout capture methods employed within command_runner.
So we jump from v0.7.0 to v1.2.0 since we have a major rework, and minor fixes too.
Since we include a critical fix, everybody is encouraged to migrate to this version.

Critical fixes

• There was a quite rare but annoying issue where the process finished, but there was still output to read in the pipe reader queue, rendering partial outputs. This has been fully refactored, and a test with 2500 rounds of file reading was added

Features

• There are now two distinct methods to capture output
  • Spawning a thread to enforce timeouts, and using process.communicate() (monitor method)
  • Spawning a thread to readlines from stdout pipe to an output queue, and reading from that output queue while enforcing timeouts (polller method)
• On the fly output (live_output=True) option is now explicit (uses poller method only)
• Returns partial stdout output when timeouts are reached
• Returns partial stdout output when CTRL+C signal is received (only with poller method)

Misc

• Adds a default 16K stdout buffer
• Default command execution timeout is 3600s (1 hour)
• Highly improved tests
  • All tests are done for both capture methods
  • Timeout tests are more accurate
  • Added missing encoding tests
  • 2500 rounds of file reading and comparaison are added to detect rare queue read misses

Fixes

• Use process signals in favor of direct os.kill API to avoid potential race conditions when PID is reused too fast
• Allow full process subtree killing on Windows & Linux, hence not blocking multiple commands like echo "test" && sleep 100 && echo "done"
  • Windows does not maintain an explicit process subtree, so we runtime walk processes to establish the child processes to kill. Obviously, orphaned processes cannot be killed that way.-