Shell Programming and Scripting

Sorry .. i didn't see the post until today. Hopefully you will still remember this.

As STDOUT and STDERR are duped filehandles of FILE, you can treat them as two separate pipes to the output file despite they end up being connected to the same output file, and the two channels are not necessarily synchronized.

Recall that Perl IO is buffered if you use the print() function. Data written to filehandles using these functions are not written immediately but buffered independently. So although the write is sequential, the resulting sequence may be in any interleaving order.

The only way I can think of to force the ordering is to wrap the logging routine inside an exclusive lock with flock(). When the lock is released, Perl will be forced to flush the filehandle, so the correct order will always be preserved by blocking out-of-order writes:

Code:

#!/usr/bin/perl

sub toLog {
 local *HANDLE = shift;
 my $msg = shift;
 flock HANDLE, Fcntl::LOCK_EX;
 print HANDLE $msg;
 flock HANDLE, Fcntl::LOCK_UN;
}

{
 local *STDOUT;
 local *STDERR;
 local *FILE;

 open FILE, ">log.txt" || die "Cannot write to log file!";
 open STDOUT, ">&FILE" || die "Cannot redirect STDOUT";
 open STDERR, ">&FILE" || die "Cannot redirect STDOUT";

 toLog(STDOUT, "A\n");
 toLog(STDERR, "B\n");
}
print STDOUT "C\n";
print STDERR "D\n";

which on my system will now give the correct order instead of 'B', 'A'. I may be wrong in this regard given my limited knowledge in I/O in general and the way it is implemented in Perl. So if you have less clumsy ways to achieve this or would like to make corrections, do it.

Thanks cbkihong. I shall try the options that you have provided. I probably will not understand your explanation until I have actually tried the various forms that you have suggessted. Thanks Jerardfay.

If the previous method looks bizarre to you, here's another method discussed in the Perl FAQ:

Code:

#!/usr/bin/perl

use IO::Handle;

{
 local *STDOUT;
 local *STDERR;
 local *FILE;

 open FILE, ">log.txt" || die "Cannot write to log file!";
 open STDOUT, ">&FILE" || die "Cannot redirect STDOUT";
 open STDERR, ">&FILE" || die "Cannot redirect STDOUT";

 STDOUT->autoflush(1);
 STDERR->autoflush(1);

 print STDOUT "A\n";
 print STDERR "B\n";
}
print STDOUT "C\n";
print STDERR "D\n";

I have problems with this autoflush() on socket I/O so I tend to avoid it most of the time, but in this case it just seems to work quite well.

As always, there are a lot of ways to do the same thing in Perl.