in reply to Re: Re (tilly) 2: passing subroutine arguments directly into a hash
in thread passing subroutine arguments directly into a hash


Manipulating $Carp::CarpLevel would be a bad idea normally simply because $Carp::CarpLevel is an internal interface that is not intended to be used outside of the core. But it gets worse.

$Carp::CarpLevel is simply a horrible hack. I refuse to use it. In fact getting rid of mistakes in Exporter that were caused by misunderstandings of how $Carp::CarpLevel works was important enough for me to decide to rewrite Carp. And I did the rewrite with a much saner alternative, and I was going to proceed and kill $Carp::CarpLevel entirely.

That was a project that I decided to give up on once I realized that the warnings pragma had likewise gotten it totally wrong internally, and I was not up to reversing it and correcting the misunderstandings. (Incidentally virtually everywhere where it was used, I found that it was misunderstood.) So for the indefinite future $Carp::CarpLevel is likely to remain an undocumented internal interface which is deprecated. And if someone else gets irritated with the fact that warnings will sometimes skip a ton of levels that it shouldn't skip, and turn a croak into a confess coming from the guts of the internals, it could easily go away.

Therefore I strongly think that it should not be used. Ever.

Incidentally a case in point for why the internal hackery is such a bad design. In the above, your game with the carp level just broke anyone else trying to play with it. The right way if you wanted to do it is to do:

local $Carp::CarpLevel = $Carp::CarpLevel + 1;
Of course this only works if you are going to confess or cluck at the end. If you are going to carp or croak, you will be so outta luck because it doesn't work anything like you would predict.

Incidentally in Perl 5.8 the right way to achieve the effect that you want will be to export the functions from a module, and in that module do:

$Carp::Ignore{ __PACKAGE__ } = 1;
Which tells the rewritten Carp that the current package is one which it should not start a stack backtrace from. :-(I don't remember if I left this documented or not. I really should wrap that project up a bit better.)-: