*cough* ... look again.

/var/tmp/otherdir/symdir/../datadir' => '/var/tmp/otherdir/datadir'

I read it fine the first time.

/var/tmp/otherdir/symdir/../datadir
[download]

is equivalent to

/var/tmp/datadir
[download]

and not equivalent to

/var/tmp/otherdir/datadir
[download]

In "/var/tmp/otherdir/symdir/../datadir", the ".." is that of "mydir" which is a hard link to "/var/tmp".

Do like I did and test before contradicting someone.

$ cd /var/tmp
$ mkdir mydir
$ mkdir datadir
$ mkdir otherdir
$ touch mydir/in_mydir
$ touch datadir/in_datadir
$ touch otherdir/in_otherdir
$ cd otherdir
$ ln -s ../mydir symdir
$ ls /var/tmp/otherdir/symdir/../datadir
in_datadir
$ perl -le'opendir $dh, $ARGV[0] or die; print for grep /in/, readdir 
+$dh' /var/tmp/otherdir/symdir/../datadir
in_datadir
[download]

Ok, so, soon on CPAN (when I have time to write the PODs):

package File::ScriptName;

use strict;
use warnings;

use version; our $VERSION = qv('0.1');
our $AUTHORITY = 'MASSA';

use Exporter q(import);
our @EXPORT_OK = qw(myname mybase mydir);

use Carp;
use Cwd qw(realpath);
use File::Spec::Functions qw(rel2abs);
use File::Basename qw(basename dirname);

BEGIN {
  my $myname = $0 = realpath rel2abs $0;
  my $mybase = basename $0, qw(.t .pm .pl .perl);
  my $mydir = dirname $0;

  croak 'chdir() used too early' unless -f $0;  

  sub myname { $myname }
  sub mybase { $mybase }
  sub mydir  { $mydir  }
}

1
[download]

Just use $0 (or File::ScriptName::myname) to find the real script name or use File::ScriptName::mybase and File::ScriptName::mydir for whatever purposes you want; traverses all symlinks and stuff... is it enough??
Update: croaks if you chdir before executing the module (possibly in a BEGIN block). Don't do that.

UpUpdate:It's on CPAN now, as scriptname...
[]s, HTH, Massa (κς,πμ,πλ)

Awesome!

rel2abs isn't needed.
realpath rel2abs $0
is the same as
realpath $0

Is there a reason to croak anymore, now that realpath is used instead of rel2abs?

I just tried this code with an executable bound with PerlApp. The code as written works fine if the executable is called by "exename.exe". However, it dies if the executable is called by "exename".

Here's my ugly modification to your code. There's probably a nicer way to do it. Clean it up as you desire.

BEGIN {
  my $rel2abs = rel2abs $0;
  my $realpath = eval { realpath $rel2abs };
  unless ( defined $realpath ) {
    warn "$rel2abs";
    $rel2abs .= '.exe';
    $realpath = realpath $rel2abs;
  }

  my $myname = $0 = $realpath;
  #my $myname = $0 = realpath rel2abs $0;
  my $mybase = basename $0, qw(.t .pm .pl .perl .exe);
  my $mydir = dirname $0;

  die 'chdir() used too early' unless -f $0;  

  sub myname { $myname }
  sub mybase { $mybase }
  sub mydir  { $mydir  }

}
[download]

Thanks for the useful code. ++!.


TGI says moo

Do like I did and test before contradicting someone.

Huh, okay, that's unexpected, probably because the last time I did this I was dealing with a symlink to a symlink, not a symlink to a real file just in a symlinked directory. I should have used the /tmp example to begin with to avoid confusion.

Still, my main point holds: dirname(rel2abs($0)) breaks under symlinks, and even a single readlink won't get you out of nested symlinks.

Realpath works, and dirname(realpath($0)) is consistent, accurate, and clean. I withdraw all prior claims, and will be using this in the future. I don't see the need for a module for it, though I give my thanks to massa for the effort.

My original test code on rel2abs (with the major bug removed) remains after the jump for the curious.

Read more... (2 kB)

You are considering that datadir is a sibling to the symlink's directory. I, OTOH, consider that datadir will be most probably a sibling to the original script file's directory (because you can symlink it from anywhere, and considering otherwise would be a security risk *). We will have to agree on disagreeing.
* create symlink, create malicious configuration file side-by-side with symlink, ... profit!
Update: I stand corrected about the (meaning of) final destination of the datadir according to the answer by AZed below... You're welcome! And seems that we will have to agree in agreeing :-)
[]s, HTH, Massa (κς,πμ,πλ)

The fix is to use Cwd::realpath instead of rel2abs. I'd appreciate it if you could test that for me.