Re: [patch] Enable plugin tests only when the build and the host systems are the same

[Janis Johnson <janis187 at us dot ibm dot com>]

On Wed, 2009-05-06 at 17:01 -0700, Le-Chun Wu wrote:
> On Wed, May 6, 2009 at 4:12 PM, Joseph S. Myers <joseph@codesourcery.com> wrote:
> > On Wed, 6 May 2009, Le-Chun Wu wrote:
> >
> >> +# The plugin testcases currently only work when
> >> +#   1. the build tree is available,
> >> +#   2. the host supports plugins,
> >> +#   3. the build system and the host system are the same (because we use
> >> +#      HOSTCC to build plugins in the tests while gcc was built with the build
> >> +#      CC).
> >
> > This is confused.  GCC was built with the host compiler, not the build
> > compiler.  It's always a compiler running on the host, built with the host
> > compiler, that is being tested as it generates code for the target.
> >
> 
> Joseph,
> 
> Thanks for the explanation. I think I was indeed (and still am) confused. :-)
> 
> So in a canadian cross compilation, when I do a "make check" in the
> build tree on the build system, I would have thought the GCC being
> tested was built with the build compiler, not the host compiler. Is it
> right? (But even in that case, HOSTCC, which is set to the build
> compiler, is also the one that's used to build GCC, so the new check
> that I added in this patch is not necessary.) I think the only case
> where this patch makes sense is when running "make check" on the host
> system while the build tree was created on the build system. In that
> case, GCC was built with the build compiler and HOSTCC is set to the
> host compiler, and this patch will prevent the plugin tests from being
> executed. But will this case ever happen? (i.e. Is it even allowed to
> run "make check" on the host with a GCC built on the build system?) Or
> maybe I am totally clueless. :-)

I'm also confused, particularly since I only do native builds, and
sincerely hope that those of you with experience testing various kinds
of cross builds can figure this out.

Janis