Re^2: wait versus close on a pipe

Part of the problem is that when you create the pipe using open(..., "-|"), calling close will always try to wait for the child to exit. Most of the time closing one end of the pipe (especially the reading end), will cause the child to terminate. However, if your child is uncooperative (for instance, ignores SIGPIPE), the close call can effectively hang. If you only have one child this is generally not a problem. However, in your case you have two child processes.

One alternative is to create the pipe using IPC::Open[23]. This will allow you to separate the closing of the pipe from the waiting/reaping of the child.

One final comment... I would close the pipes first and then reap the children. In your situation, read whatever data you need from the child , close that child's pipe and then proceed to wait for it to terminate. This also holds if you are sending data to a child via a pipe. In that case, closing your end of the pipe sends EOF to the child which is a conventional way to tell it to finish its processing and exit.

Comment on Re^2: wait versus close on a pipe Select or Download Code

Replies are listed 'Best First'.
Re^3: wait versus close on a pipe (open) by tye (Sage) on Jun 03, 2008 at 06:40 UTC
create the pipe using `IPC::Open[23]`. This will allow you to separate the closing of the pipe from the waiting/reaping of the child. Note that you can close a Perl-opened pipe w/o waiting for the child by re-opening the pipe instead of directly closing it. Re-open the file handle to `File::Spec->devnull()` and then close that. - tye	[reply] [d/l] [select]
Re^3: wait versus close on a pipe by kyle (Abbot) on Jun 03, 2008 at 04:14 UTC
A few things: Since close does a wait, there's no reason to close the pipe and try to reap the child (unless you were referring to IPC::Open3 there). If you `close` before the child terminates, the child will get a SIGPIPE only if it tries to write to the closed pipe (I think). It could ignore that, but it also might not ever trigger it, if it was done writing anyway. A `close` would send EOF to the child only if the child were reading. Since the child is writing, I don't think I'd call that EOF. Rather, it's SIGPIPE. I'm writing all this without testing it, and I'm starting to wonder if I should try some of these things before I talk further.	[reply] [d/l] [select]