GitLab

TakeNamedLock has a "waiting" mode that works like this:

  for (int i = 0; i < 10; i++)
    if (TryLock())
      break;
    usleep(i * 100);

which means that in "waiting" mode, if the lock is unavailable
for the duration, this function will block the calling thread
(with the message loop not spinning) for up to 4.5 seconds. However,
there are two problems with this design:

* The caller has no way to know whether "waiting" is appropriate or not
* The caller has to handle failure to acquire the lock either way

If retries are needed to deal with external conditions changing that might
have freed the lock up (another process exiting, e.g.) those retries should
be done at higher levels with more context.

More concretely, the only actual production use of this function is in
CloudPrintServiceProcessMain:

    std::unique_ptr<ServiceProcessState> state(new ServiceProcessState);
    if (!state->Initialize())
      return 0;

This change has TakeNamedLock make a single non-blocking attempt to
take the lock, and moves the retry logic up into
CloudPrintServiceProcessMain.

Since I was here anyway, I converted the API to use unique_ptr, since it
returns ownership of the lock if it succeeds.

Bug: 1052430
Change-Id: Ib87b23cc821fe59e69b76ffb3875917027bfdc4e
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2057530Reviewed-by: Nico Weber <thakis@chromium.org>
Commit-Queue: Elly Fong-Jones <ellyjones@chromium.org>
Cr-Commit-Position: refs/heads/master@{#744283}