Function ensure_token_or_cheat

redo/jobserver.py:386–421

Wait for a job token to become available, or cheat if possible. If we already have a token, we return immediately. If we have any processes running *and* we don't own any tokens, we wait for a process to finish, and then use that token. Otherwise, we're allowed to cheat. We call

(reason, cheatfunc)

def ensure_token_or_cheat(reason, cheatfunc):
    """Wait for a job token to become available, or cheat if possible.

    If we already have a token, we return immediately.  If we have any
    processes running *and* we don't own any tokens, we wait for a process
    to finish, and then use that token.

    Otherwise, we're allowed to cheat.  We call cheatfunc() occasionally
    to consider cheating; if it returns n > 0, we materialize that many
    cheater tokens and return.

    Args:
      reason: the reason (for debugging purposes) we need a token.  Usually
        the name of a target we want to build.
      cheatfunc: a function which returns n > 0 (usually 1) if we should
        cheat right now, because we're the "foreground" process that must
        be allowed to continue.
    """
    global _mytokens, _cheats
    backoff = 0.01
    while not has_token():
        while running() and not has_token():
            # If we already have a subproc running, then effectively we
            # already have a token.  Don't create a cheater token unless
            # we're completely idle.
            _ensure_token(reason, max_delay=None)
        _ensure_token(reason, max_delay=min(1.0, backoff))
        backoff *= 2
        if not has_token():
            assert _mytokens == 0
            n = cheatfunc()
            _debug('%s: %s: cheat = %d\n' % (env.v.TARGET, reason, n))
            if n > 0:
                _mytokens += n
                _cheats += n
                break


def running():