Re: struct by value copies

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

On Sep 2, 2008, at 2:22 PM, Thomas Heller wrote:

> Timothy Wall schrieb:
> > I'm wondering if the responsibility of making a copy of struct by
> > value arguments lies with the caller of ffi_call or with libffi.
> >
> > I've been looking through different targets to see if this is done
> > anywhere, and don't see copies made anywhere, so I would guess
> > responsibility lies with the caller of ffi_call.
> >
> > Anyone know for sure?
>
> Are you asking in the context of your win64 port?  I do not know
> the answer but I suspect that ffi_call makes the copy itself.

Yes, this is the context.

> In my Python win64 libffi implementation there is still the bug that  
> these
> copies are not made (for structures that do not fit into a register)
> although I have never seen this bug on other platforms.

I'll check on X86_WIN32 when I clean this stuff up.

> I did not yet have the time to debug or fix this problem.

I'll post the patches when I'm done.  I've got a solution which does  
the right thing, assuming that it is the responsibility of ffi_call to  
allocate the copies.

BTW, I started from your port, thank you very much :)

T.