treat SIGPIPE as normal termination by default?

[StefanKarpinski]

See this comment from John Cowan:

In general, Julia's command execution is far more careful (aka finicky) when it comes to termination status of subprocesses than the shell or other programming languages.

I very much agree that this is the Right Thing in general, but I also strongly believe that SIGPIPE should be treated as normal termination. Pipes are essentially a form of lazy data stream connecting two coroutines, in which the consumer resumes the producer when it needs more input. If the consumer chooses not to read from the producer, this ought not to be treated as a fault in the producer, which is what any other sort of signal would mean. The reason for SIGPIPE to happen at all is to prevent useless processes from hanging about with nowhere to send their output to.

In particular, using ignorestatus on any occasion when the consumer might terminate spontaneously without reading all of its input ("sed q" is just one such case), which in principle can mean any process that writes to a pipe, will wind up masking actual errors (the best is the enemy of the good). It should be possible to determin that a SIGPIPE has occurred, of course, but it should not propagate an error into the controlling process.

[Keno]

I believe treating SIGPIPE as normal termination is right. Any objections to me implementing that?

[StefanKarpinski]

No, I think it's the right thing to do.