This seems like a good thing to do. It does seem as though there are things in the gccgo export data that don't have directly correspondents in the cmd/compile format (notably init function handling).

Speaking of init functions, why does gccgo gather them up and present them all to the main package, rather than taking the regular and uniform approach of gc, which is (a) make each package initialize its direct dependencies and (b) make init functions idempotent?

This may not be the best place for gccgo design discussions, but it's simply because the gccgo approach is more efficient for large programs.

I'm willing to work on this if someone can show me what needs to be done (I don't have much experience with C/C++).