package IPC::Run3;
BEGIN { require 5.006_000; } # i.e. 5.6.0
use strict;
=head1 NAME
IPC::Run3 - run a subprocess with input/ouput redirection
=head1 VERSION
version 0.045
=cut
our $VERSION = '0.045';
=head1 SYNOPSIS
use IPC::Run3; # Exports run3() by default
run3 \@cmd, \$in, \$out, \$err;
=head1 DESCRIPTION
This module allows you to run a subprocess and redirect stdin, stdout,
and/or stderr to files and perl data structures. It aims to satisfy 99% of the
need for using C<system>, C<qx>, and C<open3>
with a simple, extremely Perlish API.
Speed, simplicity, and portability are paramount. (That's speed of Perl code;
which is often much slower than the kind of buffered I/O that this module uses
to spool input to and output from the child command.)
=cut
use Exporter;
our @ISA = qw(Exporter);
our @EXPORT = qw( run3 );
our %EXPORT_TAGS = ( all => \@EXPORT );
use constant debugging => $ENV{IPCRUN3DEBUG} || $ENV{IPCRUNDEBUG} || 0;
use constant profiling => $ENV{IPCRUN3PROFILE} || $ENV{IPCRUNPROFILE} || 0;
use constant is_win32 => 0 <= index $^O, "Win32";
BEGIN {
if ( is_win32 ) {
eval "use Win32 qw( GetOSName ); 1" or die $@;
}
}
#use constant is_win2k => is_win32 && GetOSName() =~ /Win2000/i;
#use constant is_winXP => is_win32 && GetOSName() =~ /WinXP/i;
use Carp qw( croak );
use File::Temp qw( tempfile );
use POSIX qw( dup dup2 );
# We cache the handles of our temp files in order to
# keep from having to incur the (largish) overhead of File::Temp
my %fh_cache;
my $fh_cache_pid = $$;
my $profiler;
sub _profiler { $profiler } # test suite access
BEGIN {
if ( profiling ) {
eval "use Time::HiRes qw( gettimeofday ); 1" or die $@;
if ( $ENV{IPCRUN3PROFILE} =~ /\A\d+\z/ ) {
require IPC::Run3::ProfPP;
IPC::Run3::ProfPP->import;
$profiler = IPC::Run3::ProfPP->new(Level => $ENV{IPCRUN3PROFILE});
} else {
my ( $dest, undef, $class ) =
reverse split /(=)/, $ENV{IPCRUN3PROFILE}, 2;
$class = "IPC::Run3::ProfLogger"
unless defined $class && length $class;
if ( not eval "require $class" ) {
my $e = $@;
$class = "IPC::Run3::$class";
eval "require IPC::Run3::$class" or die $e;
}
$profiler = $class->new( Destination => $dest );
}
$profiler->app_call( [ $0, @ARGV ], scalar gettimeofday() );
}
}
END {
$profiler->app_exit( scalar gettimeofday() ) if profiling;
}
sub _binmode {
my ( $fh, $mode, $what ) = @_;
# if $mode is not given, then default to ":raw", except on Windows,
# where we default to ":crlf";
# otherwise if a proper layer string was given, use that,
# else use ":raw"
my $layer = !$mode
? (is_win32 ? ":crlf" : ":raw")
: ($mode =~ /^:/ ? $mode : ":raw");
warn "binmode $what, $layer\n" if debugging >= 2;
binmode $fh, ":raw" unless $layer eq ":raw"; # remove all layers first
binmode $fh, $layer or croak "binmode $layer failed: $!";
}
sub _spool_data_to_child {
my ( $type, $source, $binmode_it ) = @_;
# If undef (not \undef) passed, they want the child to inherit
# the parent's STDIN.
return undef unless defined $source;
my $fh;
if ( ! $type ) {
open $fh, "<", $source or croak "$!: $source";
_binmode($fh, $binmode_it, "STDIN");
warn "run3(): feeding file '$source' to child STDIN\n"
if debugging >= 2;
} elsif ( $type eq "FH" ) {
$fh = $source;
warn "run3(): feeding filehandle '$source' to child STDIN\n"
if debugging >= 2;
} else {
$fh = $fh_cache{in} ||= tempfile;
truncate $fh, 0;
seek $fh, 0, 0;
_binmode($fh, $binmode_it, "STDIN");
my $seekit;
if ( $type eq "SCALAR" ) {
# When the run3()'s caller asks to feed an empty file
# to the child's stdin, we want to pass a live file
# descriptor to an empty file (like /dev/null) so that
# they don't get surprised by invalid fd errors and get
# normal EOF behaviors.
return $fh unless defined $$source; # \undef passed
warn "run3(): feeding SCALAR to child STDIN",
debugging >= 3
? ( ": '", $$source, "' (", length $$source, " chars)" )
: (),
"\n"
if debugging >= 2;
$seekit = length $$source;
print $fh $$source or die "$! writing to temp file";
} elsif ( $type eq "ARRAY" ) {
warn "run3(): feeding ARRAY to child STDIN",
debugging >= 3 ? ( ": '", @$source, "'" ) : (),
"\n"
if debugging >= 2;
print $fh @$source or die "$! writing to temp file";
$seekit = grep length, @$source;
} elsif ( $type eq "CODE" ) {
warn "run3(): feeding output of CODE ref '$source' to child STDIN\n"
if debugging >= 2;
my $parms = []; # TODO: get these from $options
while (1) {
my $data = $source->( @$parms );
last unless defined $data;
print $fh $data or die "$! writing to temp file";
$seekit = length $data;
}
}
seek $fh, 0, 0 or croak "$! seeking on temp file for child's stdin"
if $seekit;
}
croak "run3() can't redirect $type to child stdin"
unless defined $fh;
return $fh;
}
sub _fh_for_child_output {
my ( $what, $type, $dest, $options ) = @_;
my $fh;
if ( $type eq "SCALAR" && $dest == \undef ) {
warn "run3(): redirecting child $what to oblivion\n"
if debugging >= 2;
$fh = $fh_cache{nul} ||= do {
open $fh, ">", File::Spec->devnull;
$fh;
};
} elsif ( $type eq "FH" ) {
$fh = $dest;
warn "run3(): redirecting $what to filehandle '$dest'\n"
if debugging >= 3;
} elsif ( !$type ) {
warn "run3(): feeding child $what to file '$dest'\n"
if debugging >= 2;
open $fh, $options->{"append_$what"} ? ">>" : ">", $dest
or croak "$!: $dest";
} else {
warn "run3(): capturing child $what\n"
if debugging >= 2;
$fh = $fh_cache{$what} ||= tempfile;
seek $fh, 0, 0;
truncate $fh, 0;
}
my $binmode_it = $options->{"binmode_$what"};
_binmode($fh, $binmode_it, uc $what);
return $fh;
}
sub _read_child_output_fh {
my ( $what, $type, $dest, $fh, $options ) = @_;
return if $type eq "SCALAR" && $dest == \undef;
seek $fh, 0, 0 or croak "$! seeking on temp file for child $what";
if ( $type eq "SCALAR" ) {
warn "run3(): reading child $what to SCALAR\n"
if debugging >= 3;
# two read()s are used instead of 1 so that the first will be
# logged even it reads 0 bytes; the second won't.
my $count = read $fh, $$dest, 10_000,
$options->{"append_$what"} ? length $$dest : 0;
while (1) {
croak "$! reading child $what from temp file"
unless defined $count;
last unless $count;
warn "run3(): read $count bytes from child $what",
debugging >= 3 ? ( ": '", substr( $$dest, -$count ), "'" ) : (),
"\n"
if debugging >= 2;
$count = read $fh, $$dest, 10_000, length $$dest;
}
} elsif ( $type eq "ARRAY" ) {
if ($options->{"append_$what"}) {
push @$dest, <$fh>;
} else {
@$dest = <$fh>;
}
if ( debugging >= 2 ) {
my $count = 0;
$count += length for @$dest;
warn
"run3(): read ",
scalar @$dest,
" records, $count bytes from child $what",
debugging >= 3 ? ( ": '", @$dest, "'" ) : (),
"\n";
}
} elsif ( $type eq "CODE" ) {
warn "run3(): capturing child $what to CODE ref\n"
if debugging >= 3;
local $_;
while ( <$fh> ) {
warn
"run3(): read ",
length,
" bytes from child $what",
debugging >= 3 ? ( ": '", $_, "'" ) : (),
"\n"
if debugging >= 2;
$dest->( $_ );
}
} else {
croak "run3() can't redirect child $what to a $type";
}
}
sub _type {
my ( $redir ) = @_;
return "FH" if eval {
local $SIG{'__DIE__'};
$redir->isa("IO::Handle")
};
my $type = ref $redir;
return $type eq "GLOB" ? "FH" : $type;
}
sub _max_fd {
my $fd = dup(0);
POSIX::close $fd;
return $fd;
}
my $run_call_time;
my $sys_call_time;
my $sys_exit_time;
sub run3 {
$run_call_time = gettimeofday() if profiling;
my $options = @_ && ref $_[-1] eq "HASH" ? pop : {};
my ( $cmd, $stdin, $stdout, $stderr ) = @_;
print STDERR "run3(): running ",
join( " ", map "'$_'", ref $cmd ? @$cmd : $cmd ),
"\n"
if debugging;
if ( ref $cmd ) {
croak "run3(): empty command" unless @$cmd;
croak "run3(): undefined command" unless defined $cmd->[0];
croak "run3(): command name ('')" unless length $cmd->[0];
} else {
croak "run3(): missing command" unless @_;
croak "run3(): undefined command" unless defined $cmd;
croak "run3(): command ('')" unless length $cmd;
}
foreach (qw/binmode_stdin binmode_stdout binmode_stderr/) {
if (my $mode = $options->{$_}) {
croak qq[option $_ must be a number or a proper layer string: "$mode"]
unless $mode =~ /^(:|\d+$)/;
}
}
my $in_type = _type $stdin;
my $out_type = _type $stdout;
my $err_type = _type $stderr;
if ($fh_cache_pid != $$) {
# fork detected, close all cached filehandles and clear the cache
close $_ foreach values %fh_cache;
%fh_cache = ();
$fh_cache_pid = $$;
}
# This routine procedes in stages so that a failure in an early
# stage prevents later stages from running, and thus from needing
# cleanup.
my $in_fh = _spool_data_to_child $in_type, $stdin,
$options->{binmode_stdin} if defined $stdin;
my $out_fh = _fh_for_child_output "stdout", $out_type, $stdout,
$options if defined $stdout;
my $tie_err_to_out =
defined $stderr && defined $stdout && $stderr eq $stdout;
my $err_fh = $tie_err_to_out
? $out_fh
: _fh_for_child_output "stderr", $err_type, $stderr,
$options if defined $stderr;
# this should make perl close these on exceptions
# local *STDIN_SAVE;
local *STDOUT_SAVE;
local *STDERR_SAVE;
my $saved_fd0 = dup( 0 ) if defined $in_fh;
# open STDIN_SAVE, "<&STDIN"# or croak "run3(): $! saving STDIN"
# if defined $in_fh;
open STDOUT_SAVE, ">&STDOUT" or croak "run3(): $! saving STDOUT"
if defined $out_fh;
open STDERR_SAVE, ">&STDERR" or croak "run3(): $! saving STDERR"
if defined $err_fh;
my $errno;
my $ok = eval {
# The open() call here seems to not force fd 0 in some cases;
# I ran in to trouble when using this in VCP, not sure why.
# the dup2() seems to work.
dup2( fileno $in_fh, 0 )
# open STDIN, "<&=" . fileno $in_fh
or croak "run3(): $! redirecting STDIN"
if defined $in_fh;
# close $in_fh or croak "$! closing STDIN temp file"
# if ref $stdin;
open STDOUT, ">&" . fileno $out_fh
or croak "run3(): $! redirecting STDOUT"
if defined $out_fh;
open STDERR, ">&" . fileno $err_fh
or croak "run3(): $! redirecting STDERR"
if defined $err_fh;
$sys_call_time = gettimeofday() if profiling;
my $r = ref $cmd
? system { $cmd->[0] }
is_win32
? map {
# Probably need to offer a win32 escaping
# option, every command may be different.
( my $s = $_ ) =~ s/"/"""/g;
$s = qq{"$s"};
$s;
} @$cmd
: @$cmd
: system $cmd;
$errno = $!; # save $!, because later failures will overwrite it
$sys_exit_time = gettimeofday() if profiling;
if ( debugging ) {
my $err_fh = defined $err_fh ? \*STDERR_SAVE : \*STDERR;
if ( defined $r && $r != -1 ) {
print $err_fh "run3(): \$? is $?\n";
} else {
print $err_fh "run3(): \$? is $?, \$! is $errno\n";
}
}
die $! if defined $r && $r == -1 && !$options->{return_if_system_error};
1;
};
my $x = $@;
my @errs;
if ( defined $saved_fd0 ) {
dup2( $saved_fd0, 0 );
POSIX::close( $saved_fd0 );
}
# open STDIN, "<&STDIN_SAVE"# or push @errs, "run3(): $! restoring STDIN"
# if defined $in_fh;
open STDOUT, ">&STDOUT_SAVE" or push @errs, "run3(): $! restoring STDOUT"
if defined $out_fh;
open STDERR, ">&STDERR_SAVE" or push @errs, "run3(): $! restoring STDERR"
if defined $err_fh;
croak join ", ", @errs if @errs;
die $x unless $ok;
_read_child_output_fh "stdout", $out_type, $stdout, $out_fh, $options
if defined $out_fh && $out_type && $out_type ne "FH";
_read_child_output_fh "stderr", $err_type, $stderr, $err_fh, $options
if defined $err_fh && $err_type && $err_type ne "FH" && !$tie_err_to_out;
$profiler->run_exit(
$cmd,
$run_call_time,
$sys_call_time,
$sys_exit_time,
scalar gettimeofday()
) if profiling;
$! = $errno; # restore $! from system()
return 1;
}
1;
__END__
=head2 C<< run3($cmd, $stdin, $stdout, $stderr, \%options) >>
All parameters after C<$cmd> are optional.
The parameters C<$stdin>, C<$stdout> and C<$stderr> indicate how the child's
corresponding filehandle (C<STDIN>, C<STDOUT> and C<STDERR>, resp.) will be
redirected. Because the redirects come last, this allows C<STDOUT> and
C<STDERR> to default to the parent's by just not specifying them -- a common
use case.
C<run3> throws an exception if the wrapped C<system> call returned -1 or
anything went wrong with C<run3>'s processing of filehandles. Otherwise it
returns true. It leaves C<$?> intact for inspection of exit and wait status.
Note that a true return value from C<run3> doesn't mean that the command had a
successful exit code. Hence you should always check C<$?>.
See L</%options> for an option to handle the case of C<system> returning -1
yourself.
=head3 C<$cmd>
Usually C<$cmd> will be an ARRAY reference and the child is invoked via
system @$cmd;
But C<$cmd> may also be a string in which case the child is invoked via
system $cmd;
(cf. L<perlfunc/system> for the difference and the pitfalls of using
the latter form).
=head3 C<$stdin>, C<$stdout>, C<$stderr>
The parameters C<$stdin>, C<$stdout> and C<$stderr> can take one of the
following forms:
=over 4
=item C<undef> (or not specified at all)
The child inherits the corresponding filehandle from the parent.
run3 \@cmd, $stdin; # child writes to same STDOUT and STDERR as parent
run3 \@cmd, undef, $stdout, $stderr; # child reads from same STDIN as parent
=item C<\undef>
The child's filehandle is redirected from or to the local equivalent of
C</dev/null> (as returned by C<< File::Spec->devnull() >>).
run3 \@cmd, \undef, $stdout, $stderr; # child reads from /dev/null
=item a simple scalar
The parameter is taken to be the name of a file to read from
or write to. In the latter case, the file will be opened via
open FH, ">", ...
i.e. it is created if it doesn't exist and truncated otherwise.
Note that the file is opened by the parent which will L<croak|Carp/croak>
in case of failure.
run3 \@cmd, \undef, "out.txt"; # child writes to file "out.txt"
=item a filehandle (either a reference to a GLOB or an C<IO::Handle>)
The filehandle is inherited by the child.
open my $fh, ">", "out.txt";
print $fh "prologue\n";
...
run3 \@cmd, \undef, $fh; # child writes to $fh
...
print $fh "epilogue\n";
close $fh;
=item a SCALAR reference
The referenced scalar is treated as a string to be read from or
written to. In the latter case, the previous content of the string
is overwritten.
my $out;
run3 \@cmd, \undef, \$out; # child writes into string
run3 \@cmd, \<<EOF; # child reads from string (can use "here" notation)
Input
to
child
EOF
=item an ARRAY reference
For C<$stdin>, the elements of C<@$stdin> are simply spooled to the child.
For C<$stdout> or C<$stderr>, the child's corresponding file descriptor
is read line by line (as determined by the current setting of C<$/>)
into C<@$stdout> or C<@$stderr>, resp. The previous content of the array
is overwritten.
my @lines;
run3 \@cmd, \undef, \@lines; # child writes into array
=item a CODE reference
For C<$stdin>, C<&$stdin> will be called repeatedly (with no arguments) and
the return values are spooled to the child. C<&$stdin> must signal the end of
input by returning C<undef>.
For C<$stdout> or C<$stderr>, the child's corresponding file descriptor
is read line by line (as determined by the current setting of C<$/>)
and C<&$stdout> or C<&$stderr>, resp., is called with the contents of the line.
Note that there's no end-of-file indication.
my $i = 0;
sub producer {
return $i < 10 ? "line".$i++."\n" : undef;
}
run3 \@cmd, \&producer; # child reads 10 lines
Note that this form of redirecting the child's I/O doesn't imply
any form of concurrency between parent and child - run3()'s method of
operation is the same no matter which form of redirection you specify.
=back
If the same value is passed for C<$stdout> and C<$stderr>, then the child
will write both C<STDOUT> and C<STDERR> to the same filehandle.
In general, this means that
run3 \@cmd, \undef, "foo.txt", "foo.txt";
run3 \@cmd, \undef, \$both, \$both;
will DWIM and pass a single file handle to the child for both C<STDOUT> and
C<STDERR>, collecting all into file "foo.txt" or C<$both>.
=head3 C<\%options>
The last parameter, C<\%options>, must be a hash reference if present.
Currently the following keys are supported:
=over 4
=item C<binmode_stdin>, C<binmode_stdout>, C<binmode_stderr>
The value must a "layer" as described in L<perlfunc/binmode>. If specified the
corresponding parameter C<$stdin>, C<$stdout> or C<$stderr>, resp., operates
with the given layer.
For backward compatibility, a true value that doesn't start with ":"
(e.g. a number) is interpreted as ":raw". If the value is false
or not specified, the default is ":crlf" on Windows and ":raw" otherwise.
Don't expect that values other than the built-in layers ":raw", ":crlf",
and (on newer Perls) ":bytes", ":utf8", ":encoding(...)" will work.
=item C<append_stdout>, C<append_stderr>
If their value is true then the corresponding parameter C<$stdout> or
C<$stderr>, resp., will append the child's output to the existing "contents" of
the redirector. This only makes sense if the redirector is a simple scalar (the
corresponding file is opened in append mode), a SCALAR reference (the output is
appended to the previous contents of the string) or an ARRAY reference (the
output is C<push>ed onto the previous contents of the array).
=item C<return_if_system_error>
If this is true C<run3> does B<not> throw an exception if C<system> returns -1
(cf. L<perlfunc/system> for possible failure scenarios.), but returns true
instead. In this case C<$?> has the value -1 and C<$!> contains the errno of
the failing C<system> call.
=back
=head1 HOW IT WORKS
=over 4
=item (1)
For each redirector C<$stdin>, C<$stdout>, and C<$stderr>, C<run3()> furnishes
a filehandle:
=over 4
=item *
if the redirector already specifies a filehandle it just uses that
=item *
if the redirector specifies a filename, C<run3()> opens the file
in the appropriate mode
=item *
in all other cases, C<run3()> opens a temporary file (using
L<tempfile|Temp/tempfile>)
=back
=item (2)
If C<run3()> opened a temporary file for C<$stdin> in step (1),
it writes the data using the specified method (either
from a string, an array or returnd by a function) to the temporary file and rewinds it.
=item (3)
C<run3()> saves the parent's C<STDIN>, C<STDOUT> and C<STDERR> by duplicating
them to new filehandles. It duplicates the filehandles from step (1)
to C<STDIN>, C<STDOUT> and C<STDERR>, resp.
=item (4)
C<run3()> runs the child by invoking L<system|perlfunc/system> with C<$cmd> as
specified above.
=item (5)
C<run3()> restores the parent's C<STDIN>, C<STDOUT> and C<STDERR> saved in step (3).
=item (6)
If C<run3()> opened a temporary file for C<$stdout> or C<$stderr> in step (1),
it rewinds it and reads back its contents using the specified method (either to
a string, an array or by calling a function).
=item (7)
C<run3()> closes all filehandles that it opened explicitly in step (1).
=back
Note that when using temporary files, C<run3()> tries to amortize the overhead
by reusing them (i.e. it keeps them open and rewinds and truncates them
before the next operation).
=head1 LIMITATIONS
Often uses intermediate files (determined by File::Temp, and thus by the
File::Spec defaults and the TMPDIR env. variable) for speed, portability and
simplicity.
Use extrem caution when using C<run3> in a threaded environment if concurrent
calls of C<run3> are possible. Most likely, I/O from different invocations will
get mixed up. The reason is that in most thread implementations all threads in
a process share the same STDIN/STDOUT/STDERR. Known failures are Perl ithreads
on Linux and Win32. Note that C<fork> on Win32 is emulated via Win32 threads
and hence I/O mix up is possible between forked children here (C<run3> is "fork
safe" on Unix, though).
=head1 DEBUGGING
To enable debugging use the IPCRUN3DEBUG environment variable to
a non-zero integer value:
$ IPCRUN3DEBUG=1 myapp
=head1 PROFILING
To enable profiling, set IPCRUN3PROFILE to a number to enable emitting profile
information to STDERR (1 to get timestamps, 2 to get a summary report at the
END of the program, 3 to get mini reports after each run) or to a filename to
emit raw data to a file for later analysis.
=head1 COMPARISON
Here's how it stacks up to existing APIs:
=head2 compared to C<system()>, C<qx''>, C<open "...|">, C<open "|...">
=over
=item *
better: redirects more than one file descriptor
=item *
better: returns TRUE on success, FALSE on failure
=item *
better: throws an error if problems occur in the parent process (or the
pre-exec child)
=item *
better: allows a very perlish interface to Perl data structures and subroutines
=item *
better: allows 1 word invocations to avoid the shell easily:
run3 ["foo"]; # does not invoke shell
=item *
worse: does not return the exit code, leaves it in $?
=back
=head2 compared to C<open2()>, C<open3()>
=over
=item *
better: no lengthy, error prone polling/select loop needed
=item *
better: hides OS dependencies
=item *
better: allows SCALAR, ARRAY, and CODE references to source and sink I/O
=item *
better: I/O parameter order is like C<open3()> (not like C<open2()>).
=item *
worse: does not allow interaction with the subprocess
=back
=head2 compared to L<IPC::Run::run()|IPC::Run/run>
=over
=item *
better: smaller, lower overhead, simpler, more portable
=item *
better: no select() loop portability issues
=item *
better: does not fall prey to Perl closure leaks
=item *
worse: does not allow interaction with the subprocess (which IPC::Run::run()
allows by redirecting subroutines)
=item *
worse: lacks many features of C<IPC::Run::run()> (filters, pipes, redirects,
pty support)
=back
=head1 COPYRIGHT
Copyright 2003, R. Barrie Slaymaker, Jr., All Rights Reserved
=head1 LICENSE
You may use this module under the terms of the BSD, Artistic, or GPL licenses,
any version.
=head1 AUTHOR
Barrie Slaymaker E<lt>C<barries@slaysys.com>E<gt>
Ricardo SIGNES E<lt>C<rjbs@cpan.org>E<gt> performed routine maintenance since
2010, thanks to help from the following ticket and/or patch submitters: Jody
Belka, Roderich Schupp, David Morel, Jeff Lavallee, and anonymous others.
=cut