Bug #57391 for IPC-Run3: errors sent to STDOUT

Tue May 11 17:40:52 2010 user42 [...] zip.com.au - Ticket created

Subject:	errors sent to STDOUT
Date:	Wed, 12 May 2010 07:39:24 +1000
To:	bug-IPC-Run3 [...] rt.cpan.org
From:	Kevin Ryde <user42 [...] zip.com.au>

While nosing around IPC::Run3 0.043_02 I wondered how the handle duping would go if permuting stdin/out/err. The program errout.pl below writes 'blah' to /tmp/xxx, whereas I think I hoped it would write to the tty (ie. the parent's STDOUT). The "sh -c" is just a program which writes to its stderr. Replace it with something simpler as desired. It might be reasonably easy to dup from SAVED_STDOUT when the child's stderr is the parent's stdout. A fully general thing would allow for all three of stdin/out/err to be mixed around. It wouldn't often make much sense to put STDIN as a child's stdout or stderr, but I think it'd be worth doing correctly so as to get the right kind of error out of the child and/or simply to ensure it doesn't operate on the wrong thing.

use strict; use warnings; use IPC::Run3; print IPC::Run3->VERSION,"\n"; open MYFILE, '>/tmp/xxx' or die; run3 (['sh','-c','echo blah >&2'], \undef, \*MYFILE, \*STDOUT);

Thu May 13 07:59:33 2010 RSCHUPP [...] cpan.org - Status changed from 'new' to 'open'

Thu May 13 07:59:39 2010 RSCHUPP [...] cpan.org - Taken

Thu May 13 08:08:35 2010 RSCHUPP [...] cpan.org - Correspondence added

On Tue May 11 17:40:52 2010, user42@zip.com.au wrote: Show quoted text

> While nosing around IPC::Run3 0.043_02 I wondered how the handle duping > would go if permuting stdin/out/err.

Good catch. I'm not sure how to fix this in the general case. As a fist step, I created a new test that will call run3() with all six possible permutations of STDIN, STDOUT and STDERR (regardless how likely such a use case is). Fails miserably, of course. Cheers, Roderich

Mon May 17 20:10:54 2010 user42 [...] zip.com.au - Correspondence added

Subject:	Re: [rt.cpan.org #57391] errors sent to STDOUT
Date:	Tue, 18 May 2010 10:09:47 +1000
To:	bug-IPC-Run3 [...] rt.cpan.org
From:	Kevin Ryde <user42 [...] zip.com.au>

"RSCHUPP via RT" <bug-IPC-Run3@rt.cpan.org> writes: Show quoted text

> > I'm not sure how to fix this in the general case.

If you've made local dups saving the handles anyway then it might work to re-dup off those. Something like (untested) my @dups = (fileno(SAVED_STDIN), fileno(SAVED_STDOUT), fileno(SAVED_STDERR)); # map a target $fd through @dups if (exists $dups[$fd]) { $fd = $dups[$fd]; } Driving it from the fd (like you've got now) would I think pick up any handle which is an IO::Handle->new_from_fd referring to 0, 1 or 2 in addition to the real STDIN/OUT/ERR handles.

Tue May 18 13:09:24 2010 RSCHUPP [...] cpan.org - Correspondence added

On Mon May 17 20:10:54 2010, user42@zip.com.au wrote: Show quoted text

> If you've made local dups saving the handles anyway then it might work > to re-dup off those. Something like (untested)

Good idea. I had a much more complex solution. This passes the new "permute STD* filehandles" torture test. I'll do a developer release tonite. BTW if you have some tuits to spare, uncomment the failing t/zz_pipe*.t tests at the end of Makefile.PL. I have absolutely no clue how to fix these or what's the difference when reading from a file script.pl < file instead of from a pipe cat file | script.pl The bug here is that when script.pl calls run3() in the middle of reading STDIN, parts of the input get lost. Cheers, Roderich

Mon Jun 07 20:39:23 2010 user42 [...] zip.com.au - Correspondence added

Subject:	Re: [rt.cpan.org #57391] errors sent to STDOUT
Date:	Tue, 08 Jun 2010 10:38:37 +1000
To:	bug-IPC-Run3 [...] rt.cpan.org
From:	Kevin Ryde <user42 [...] zip.com.au>

"RSCHUPP via RT" <bug-IPC-Run3@rt.cpan.org> writes: Show quoted text

> > The bug here is that when script.pl calls run3() in the middle > of reading STDIN, parts of the input get lost.

Hmm. I see the problem. It looks like the "open STDIN <&" discards the input buffer on the STDIN handle and because a pipe is not seekable it can't re-read like on an ordinary file. I suppose the local dup is not doing enough to preserve the underlying IO object. Dunno what the right thing to do would be. Try to squirrel away the *STDIN{IO} object or something. Or try to dig out the input buffer contents and unget them back onto the handle afterwards (for non-seekable inputs). That might be unfriendly or unworkable with layers though ...

Mon Jun 07 21:16:50 2010 user42 [...] zip.com.au - Correspondence added

Subject:	Re: [rt.cpan.org #57391] errors sent to STDOUT
Date:	Tue, 08 Jun 2010 11:15:50 +1000
To:	bug-IPC-Run3 [...] rt.cpan.org
From:	Kevin Ryde <user42 [...] zip.com.au>

I wrote: Show quoted text

> > looks like the "open STDIN <&"

I suppose only the file descriptors matter to the system() call. Maybe "flushall" the same way it will and then temporarily POSIX::dup2() the desired fds to 0,1,2 as necessary.