I've not tried, but the compiler should choke at it. It seems that you don't need it, because the factorial function caches values up to 33, while for larger numbers uses the log-gamma to approximate the factorial you're looking for. Given that it's an approximation, why don't you re-implement gammln in straight Perl and try to get rid of the C function in a first step?

Anyway, if you want some BigDouble you should look for some ad-hoc lib (like GMP cited above) - if it were as simple as you're trying, those libs probably wouldn't exist :) Moreover, are you really convinced you need to compute such big factorials? (This is becoming OT, so if you want write about it in some math forum or, if you want my opinion, write me to flavio (at) polettix (dot) it).