Comment 11 for bug 505913

I was wrong. The minimal example is invalid because the test calls one job after the other, and the minimal example calls only one job. This is relevant, because the first job in the test guarantees that the signal handler is installed.

What's _really_ happening is that when the timeout is really small, the signal is being raised before the message is sent to the child process to run the job. The signal handler in the child process is triggered, and raises an exception, but nothing cares. The reactor treats it as an "unhandled error" and moves on its merry way. When the message actually arrives, it is processed and the job is run.

This difference explains "why the test considers the job to have completed successfully", which was confusing me in comment 8.

Talking w/ abentley about options to fix this.