> And FWIW I never use h2xs for C code as it is much safer to write > bindings by hand as XS code and even the relatively trivial task > of parsing C declarations is often "wrong".
Same here. I *always* write XS code by hand. If I use h2xs at all, then only to generate all the template files. I also often find the automated API translations (even when they are correct) to be rather un-Perlish. I believe the time spent in making a robust Perl API almost always pays off later in reduced debugging times at the application level. I understand that wrapping a huge API can be a lot of work, but often you can wrap it into a much smaller high-level API for Perl and put more abstraction into the XS code. I know that some people (e.g. Tye McQueen) disagree with this and rather recommend a thin XS layer and then a second abstraction layer in Perl, in which case a working h2xs for C++ would be handy indeed. Cheers, -Jan
