http://qs321.pair.com?node_id=1140402


in reply to Can't close pipe to invalid process

I think that this behaviour is caused by the fact that Perl has to launch a shell to process a command given in a string form, and the open doesn't fail because shell does launch successfully (it's the command that fails later):

use Test::Simple tests => 3; ok !open(my $fh, '|-', 'nonexistent program'), 'string form of open fa +ils'; ok open(my $prince, '|-', 'nonexistent program 2>/dev/null'), 'but suc +ceeds if shell uses redirection'; ok !open(my $none, '|-', 'nonexistent', 'program'), 'list form of open + fails, as it should'; __END__ 1..3 ok 1 - string form of open fails ok 2 - but succeeds if shell uses redirection ok 3 - list form of open fails, as it should

But why does string form of open correctly fail in the first test? Perl is able to process a simple command without help of a shell, as strace shows:

$ strace -f -e clone,execve perl -e'open(my $fh, "|-", "nonexistent pr +ogram")' execve("/usr/bin/perl", ["perl", "-eopen(my $fh, \"|-\", \"nonexisten" +...], [/* 45 vars */]) = 0 clone(child_stack=0, flags=CLONE_CHILD_CLEARTID|CLONE_CHILD_SETTID|SIG +CHLD, child_tidptr=0xb74f7a68) = 7792 Process 7792 attached [pid 7792] execve("/home/aitap/perl5/bin/nonexistent", ["nonexistent" +, "program"], [/* 45 vars */]) = -1 ENOENT (No such file or directory +) [pid 7792] execve("/home/aitap/scripts/nonexistent", ["nonexistent", +"program"], [/* 45 vars */]) = -1 ENOENT (No such file or directory) [pid 7792] execve("/home/aitap/bin/nonexistent", ["nonexistent", "pro +gram"], [/* 45 vars */]) = -1 ENOENT (No such file or directory) [pid 7792] execve("/usr/local/bin/nonexistent", ["nonexistent", "prog +ram"], [/* 45 vars */]) = -1 ENOENT (No such file or directory) [pid 7792] execve("/usr/bin/nonexistent", ["nonexistent", "program"], + [/* 45 vars */]) = -1 ENOENT (No such file or directory) [pid 7792] execve("/bin/nonexistent", ["nonexistent", "program"], [/* + 45 vars */]) = -1 ENOENT (No such file or directory) [pid 7792] execve("/usr/local/games/nonexistent", ["nonexistent", "pr +ogram"], [/* 45 vars */]) = -1 ENOENT (No such file or directory) [pid 7792] execve("/usr/games/nonexistent", ["nonexistent", "program" +], [/* 45 vars */]) = -1 ENOENT (No such file or directory) [pid 7792] +++ exited with 1 +++ --- SIGCHLD {si_signo=SIGCHLD, si_code=CLD_EXITED, si_pid=7792, si_uid +=1000, si_status=1, si_utime=0, si_stime=0} --- +++ exited with 0 +++
Stream redirection, however, requires a shell:
$ strace -f -e clone,execve perl -e'open(my $fh, "|-", "nonexistent pr +ogram 2>/dev/null")' execve("/usr/bin/perl", ["perl", "-eopen(my $fh, \"|-\", \"nonexisten" +...], [/* 45 vars */]) = 0 clone(Process 7809 attached child_stack=0, flags=CLONE_CHILD_CLEARTID|CLONE_CHILD_SETTID|SIGCHLD, +child_tidptr=0xb753ea68) = 7809 [pid 7809] execve("/bin/sh", ["sh", "-c", "nonexistent program 2>/dev +/null"], [/* 45 vars */]) = 0 [pid 7809] +++ exited with 127 +++ --- SIGCHLD {si_signo=SIGCHLD, si_code=CLD_EXITED, si_pid=7809, si_uid +=1000, si_status=127, si_utime=0, si_stime=0} --- +++ exited with 0 +++

Edit: spelling

Replies are listed 'Best First'.
Re^2: Can't close pipe to invalid process
by BrowserUk (Patriarch) on Aug 29, 2015 at 19:18 UTC

      Correct me if I'm wrong, but OP said:

      Turns out it was because the PATH didn't exist in the Jenkins environment.
      Why would the open succeed to an unknown executable?
      As far as I understood OP, missing PATH meant that the script was unable to run executables except by full path:
      $ env LC_ALL=C PATH="" /usr/bin/perl -E'open my $fh, "-|", $_ and say +"success" or say $! for "ls", "/bin/ls"' No such file or directory success
      So the question was: how does open succeed despite inability to find the executable due to empty $PATH?

      By the way, thanks for the spelling reminder.

        The only path I see in the OP is that pointing to the file the executable is meant to process:

        /opt/humana/svn/checkouts/rosalind-repo/prod/tmp/testfile.pdf

        The executable name is provided unadorned; so either in the current working directory or to be located via the path environment variable.

        But, as I've no knowledge of what the "Jenkins environment" is; it could be me misinterpreting the question.


        With the rise and rise of 'Social' network sites: 'Computers are making people easier to use everyday'
        Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.
        "Science is about questioning the status quo. Questioning authority". I knew I was on the right track :)
        In the absence of evidence, opinion is indistinguishable from prejudice.
        I'm with torvalds on this Agile (and TDD) debunked I told'em LLVM was the way to go. But did they listen!

      It was the not finding "prince" that caused it to fail (pretty much what aitap said).

      Intuitively, I would think the open should fail if the executable being piped to can't be found, but that wasn't the case.

      I added a test for the print, and that appears to return true.

      So, what I have is:

      ok 3 - Opened pipe ok 4 - Piped content to prince not ok 5 - Closed pipe

      I'm wondering if an assumption is made internally about piping to the executable until Perl receives a return value on closing?!?

        I'm wondering if an assumption is made internally about piping to the executable until Perl receives a return value on closing?!?

        Looking back at what you're sending down the pipe; it is way less than the typical 4kb buffer size; which means nothing will be sent until you try to close, at which point it will attempt to flush the buffer through and it is only at that point that the broken pipe will be discovered.

        You could verify that conjecture by trying to print more data to the pipe:

        my $fh; ok( open($fh, "|-", "prince - 2>/dev/null $test_file"), "Open +$test_file for writing" ); print( $fh 'x' x 1024 ) or die "Print failed at $_ with $!" fo +r 1 .. 5; ok( close($fh), "Close $test_file");

        My guess is that will fail at the 4th or 5th iteration with "Broken pipe".


        With the rise and rise of 'Social' network sites: 'Computers are making people easier to use everyday'
        Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.
        "Science is about questioning the status quo. Questioning authority". I knew I was on the right track :)
        In the absence of evidence, opinion is indistinguishable from prejudice.
        I'm with torvalds on this Agile (and TDD) debunked I told'em LLVM was the way to go. But did they listen!