| Subject: | suggest option for blocking output pipe |
| Date: | Sun, 08 Feb 2015 15:16:15 +1100 |
| To: | bug-IPC-Run [...] rt.cpan.org |
| From: | Kevin Ryde <user42_kevin [...] yahoo.com.au> |
I wanted to start a subprocess and send it data through a pipe like the
program below, but found run() "<pipe" handle is non-blocking, when
printing a big block or faster than the child reads (EWOULDBLOCK or
EAGAIN).
Error from print to pipe: Resource temporarily unavailable at foo.pl line 5.
It could be good if the docs of "<pipe" described the non-blocking,
under the bit "The <pipe operator opens the write half of a pipe ...".
I presume the motivation is to allow simultaneous read and write or for
a master process to do other things when some child is not ready.
It could also be good to have an option for a blocking output pipe.
I see there's some deliberate fcntl() to turn off blocking, so the
option would not do that.
I think this would be useful in various cases where the perl program is
setting itself up as the start of a pipeline and so should write at the
pace of the child.
use strict;
use IPC::Run;
my $harness = IPC::Run::start(['wc'], '<pipe', \*WC)
or die "Cannot run cat";
print WC "blah blah\n" x 32768
or die "Error from print to pipe: $!";
close WC
or die "Error closing pipe";
$harness->finish;