Re: [PATCH] [WIP] Port feature/native-comp to Windows.

[Nicolas Bértolo]

> I think that's the misunderstanding: libgccjit *is* GCC.  No external
> compiler is required.

I understand that. But we need some extra files, like libgcc.a and
some extra binaries: as, ld, and collect2.
If understood the libgccjit sources correctly it will try to find
them in a directory like: /usr/lib/gcc/x86_64-w64-mingw32/10.0.0

The contents of that directory are what I meant when said "GCC installation".

Nicolas.

El dom., 10 may. 2020 a las 16:50, Andrea Corallo (<address@hidden>) escribió:
>
> Nicolas Bértolo <address@hidden> writes:
>
> >> But up front, we can simply use the apriori knowledge about the way the 
> >> setjmp
> >> macro is expanded in the version used for building Emacs.
> >
> > I do not understand this. Do you mean that we should just add the necessary
> > expansions to the emit_setjmp() function?
> >
> >> But that's a problem on any platform, not just on Windows, isn't it?
> >> It would mean that when you upgrade GCC, you need to rebuild Emacs in
> >> order to be able to continue natively-compiling Lisp files, even if
> >> your Emacs version stays the same.  Is this reasonable?  What will
> >> GNU/Linux users do that install binary distributions of Emacs -- will
> >> they have to install a very particular version of GCC to be able to
> >> compile to native code, and never upgrade it, except together with
> >> upgrading Emacs?  Sounds very inconvenient to me.
> >
> > Exactly. I suppose libgccjit has a stable ABI, so it should not be a 
> > problem to
> > update libgccjit. The user will need to ensure that the version of libgccjit
> > loaded by Emacs has its corresponding GCC installation present.
>
> I think that's the misunderstanding: libgccjit *is* GCC.  No external
> compiler is required.
>
> --
> address@hidden