[rocketraman]

git timeouts never fired. spawnNow read stdout to EOF and only then called waitFor(timeout), so a git that hangs, and never closes stdout, blocks forever. dave08's 1.5.0 log shows a git fetch that ran for 2h 11m against its 90s NETWORK_TIMEOUT. Some of that may be the Mac sleeping, but nothing in the code could have stopped it. Every version is affected.

Change

• A timeout coroutine starts next to the read, under one runBlocking parent. When the timeout passes, it kills git and all of git's child processes. If the read finishes first, the coroutine is cancelled. It runs on Dispatchers.Default, because runBlocking's own thread is blocked in the read.
• Child processes are listed and killed before git itself. Once git dies its children are reparented, and they would keep the pipe open with nothing left to find them. A fetch's ssh and the shell behind an alias both hold git's stdout.
• Reads still happen on the calling thread, and git commands still run in parallel.
• A timed-out call still returns Result(-1, out, "timed out"), as before.

Testing

• New GitTimeout suite:
  • a !sleep 60 alias with a 1s timeout now returns failure in about 1s; before the fix it would have blocked for the full 60s;
  • a normal command is unaffected.
• Unit tests 291/291, Konture 10/10, detekt clean.