[cfe-dev] Clang should natively support fortran

[Joerg Sonnenberger via cfe-dev]

On Wed, Jul 06, 2016 at 12:38:25PM -0400, Ed Maste via cfe-dev wrote:
> On 6 July 2016 at 04:52, Asiri Rathnayake <asiri.rathnayake at gmail.com> wrote:
> > Glad to be of help! :)
> >
> > @Ed: You said:
> >
> > "...heavily modified version of LLVM libunwind in NetBSD, which excludes the
> > HP unwind interface and requires much less stack space."
> >
> > May be we should try and get them (stack usage improvements) into llvm
> > libunwind? We're interested!
> 
> I'm not sure what else was done in NetBSD's version, but I think the
> primary reduction in stack space came from dropping the HP unwind
> interfaces and non-local unwind support, and shrinking the context to
> only what's needed for each platform. I've added Joerg to CC in case
> he has other commentary to add.

Actually, non-local unwind support is pretty much completely unrelated.
Not providing HP unwind by default is a concious decision. It adds a lot
of stupid target dependent code as it requires f.e. a secondary register
numbering. It is also a significant enough chunk of the total code size.
In the context of the primary system unwind library, it doesn't provide
*any* advantage.

Joerg