I haven't delved into the depths of Apache::Registry and Maypole to determine where the errors get eaten and not reraised. I believe that the reason are unchecked eval-block expressions, but I can't offer any hard facts for that assumption.

Would you care to write a Devel::UncheckedEvalBlock ala Initial Devel::UncheckedOps, a macro for perl and <rant>CPAN modules failing to check for write errors</rant>? It'd just be the example assuming there was some sort of magic in those hashes to generate some regexp code over a hypothetical stringification of an optree and that changing the stringified optree would change the actual optree...

$optree =~
    s(
        (?<= $STATEMENT{ $CONTAINS{ $OP_ENTERTRY } } )
        (?! $OP_NEXTSTATEMENT
            $STATEMENT{ $CONTAINS{ $GVSV{ '*@' } } } )
    )
    {
        warn $@ if $@;
    }xg
[download]