Neil Booth wrote:
>
> > + #undef CPP_SPEC
> > + #define CPP_SPEC "%{!traditional: -D__STDC__} \
> > + %{mdisable-fpregs:-D__NO_FP__} \
> > + %{F*} %{ansi} %{fno-asm} %{header-mapfile} \
> > + -D__APPLE_CC__=" STRINGIFY(APPLE_CC) " \
> > + %{static:-D__STATIC__}%{!static:-D__DYNAMIC__}\
> > + %{MD:-MD %M} %{MMD:-MMD %M}"
>
> I'm quite sceptical about the MD and MMD stuff. Also, the traditional
> bit can go. We have two different preprocessors now; that solves that
> issue.
Turns out that almost all of this should evaporate; we have a local
change where %M is a dependency file name used when building fat
binaries, which is quite different from the standard meaning of
it as the multilib dir!
> > + #define LINK_COMMAND_SPEC \
> > + "%{!fsyntax-only: \
> > + %{!c:%{!M:%{!MM:%{!E:%{!precomp:%{!S: \
>
> I've not looked at what this is trying to achieve in detail; !M* might
> be more correct in place of the !M && !MM.
The standard LINK_COMMAND_SPEC in gcc.c uses the same sequence,
except -precomp is used to generate precompiled headers and thus
should not be part of this patch; if I whack that, it looks like
I can sponge off the standard spec and just define LINK_SPEC.
Thanks for the comments!
Stan