Re: [patch] port libffi to x86/msvc

[Timothy Wall <twalljava at dev dot java dot net>]

On Jan 13, 2010, at 11:17 PM, Dan Witte wrote:

> > I haven't looked at the patch yet, but ffi_call_SYSV and
> > ffi_call_STDCALL are not identical, since stdcall must pop its own
> > args from the stack.  I think I originally wrote it to share code
> > between two functions, with only a little bit on the ends for  
> > stdcall,
> >
> > but someone added some debug/stack metadata that required the two
> > functions be distinct, thus the duplicated code.
>
> Hmm. Both functions save 'esp' on entry, and restore it on exit, so  
> whether the caller or callee pops the args doesn't matter, and both  
> functions can be identical. Is this correct?
>
> (Looking at history, in an older libffi release, they were different  
> and the SYSV variant adjusted 'esp' appropriately after the foreign  
> call. But this was dropped when 'esp' was saved & restored.)
>
> I have no idea what the eh_frame stuff is doing, so if the two  
> functions need to be distinct for that reason, I can produce a new  
> patch. Presumably the functions can still be merged for MSVC, just  
> not for gcc. Should I do that?

I may be thinking of the closure code.  At any rate, if the code is  
truly identical, and the stdcall tests still pass (and they actually  
cover what you're changing), then there's no reason not to remove the  
duplicate code.