I read up on some of the documentation on END and I think the answer to your problem is pretty simple. According to perlmod: END {} blocks are executed in reverse order. The last END created is the first END that gets executed. It seems all you need to do is define your END {} at the very beginning of your program before the other ones get defined. That should allow you to set your $? variable to whatever you want since by the time its seen all of the modifications to it should have already happened.

use is really just a BEGIN {} block and BEGIN executes in the order its seen. You could create a module that has your final end block and use it the very first. Tell it to redefine $? to some other value. use all of your other modules then at the end of all that use another module that creates an END { } that saves the original exit value from the program. Something like this:

#!/usr/bin/perl
use strict;
use vars qw($myexitvalue);
use MyLastEnd;
use Foo;
use Bar;
use SomeOtherModule;
use MyFirstEnd;

# rest of script

#MyLastEnd.pm
END {
     $? = $myexitvalue;
}

#MyFirstEnd.pm
END {
    $myexitvalue = $?;
}
[download]

All of this is untested of course, but if I'm reading the docs right it *should* work. Hope that helps!

Can you say "DOH!"?

Of course the "last" END block has to be compiled before the use statements get processed -- it's end blocks in use-d items that are giving me the headache. So you need:

use vars qw($exitval);
END {$? = $exitval}

use LotsOStuff;
use MoreStuff;
use TheOtherGuy'sStuff;
use StuffYouProbablyDontNeed;

# your code here

END {$exitval = $?}
[download]

my $rv = 2;  # assume fatal.
END { $? = $rv; }

use ...
use ...

{
   ...

   if (...) { $rv = 1; exit(); }

   ...
   
   $rv = 0;
}
[download]

But we're using 5.6.1 and Manip::END requires 5.8.0.

I think you're right (I can't remember) and I never tested it on anything else as it was more a proof of concept than a serious module. That said, it does work.

A new version should be on CPAN by the time you read this. It removes the 5.8 dependency and also fixes a small bug in the error reporting, was it you who sent me the patch on RT?

As for the original question, the best way to add an END block is by writing an END block! If you want it to be triggered before all the others, put it at the end of your program, if you want it to be triggered after all the others put it at the start, before any use or require . A combination of the two can solve your problem.

my $saved_value;
END { $? = $saved_value } # this will be the final END block

#### your program goes here
#### using whatever modules you want

END { $saved_value = $? # } # this will be the first END block
[download]

The only thing that could go wrong with this scheme is that something could pull in a module containing another END block using require or eval "use something" and that END block could fiddle with $? before you get to save it. If this is happening you could try calling my_exit instead where

sub my_exit {
  my $value = shift;
  $saved_value = $value;
  exit $value;
}
[download]

See also Changing the return number for "die".

In addition to the solutions already offered, if you don't care whether the END blocks run there is always POSIX::_exit.