Rtems 4.8.0 / Mingw / Coldfire Gcc Bug

[Ralf Corsepius]

On Fri, 2008-02-29 at 08:14 -0600, Joel Sherrill wrote:
> Ralf Corsepius wrote:
> > On Fri, 2008-02-29 at 19:54 +1100, Chris Johns wrote:
> >   
> >> Paul Whitfield wrote:
> >>     
> >>> I just attempted to build rtems 4.8.0 multi lib with the Mingw tools,
> >>>
> >>> However, it stops with a gcc internal compiler error ala
> >>>
> >>> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34439
> >>>
> >>> Any suggestions?
> >>>
> >>>       
> >> It is suppose to be fixed in the latest gcc.
> >>     
> > AFAICT, this bug is present in all GCC releases from GCC's 4.2.x release
> > series. I am not aware about any fix to it (c.f. the BZ above).
> >
> >   
> >>  Ralf has confirmed this how-ever
> >> these tool sets are not available to download.
> >>     
> > This bug seems to be fixed in GCC's 4.3.x release series, which is
> > supposed to be released in "not to distant future".
> >
> > At least, I can't reproduce the breakdown with toolchains having been
> > built from development versions/release candidates from this series.
> >
> > BTW: I pushed rtems-4.9-GCC Linux rpms having been built from GCC-4.3.0
> > release candidate tarballs, which also should be suitable to build
> > rtems-4.8.
> >
> >   
> The tools may be OK but from the 4.8 branch itself will
> need a number of minor changes made to compile and link
> cleanly with 4.3.x.  I need to review my build logs before
> committing but I edited about ~75 files yesterday to
> address 4.3.x issues.  That ignores the miscellaneous
> limits.h changes you and I both have made over the past
> few weeks.
> 
> If the community wants to move to gcc 4.3.x for the 4.8
> branch then I have no problem with it.  It is definitely better
> for the Coldfire and long term maintenance.   I would
> be happy to cut a 4.8.1 when the gcc 4.3.0 RPMs are available.
> For this to happen, I only see a few issues worth addressing:
> 
> + Fix the locale C++ issue.  I see Ralf committed the fix
>     to the issue from a couple of months ago so this will
>     have to be debugged further. :(
> 
> + Backport the minor changes 4.3.x requires with limits.h,
>     etc. and linkcmds.  This is minor and I will go ahead and
>     start committing this type of change to the trunk and 4.8
>     in anticipation.
> 
> + It would also be highly desirable for the release tarball
>     to be bootstrap'ed on Fedora and work on MS-Windows.
>     I really don't know off hand what patches are missing.
> 
> That isn't much too aim for in a quick 4.8.1, is it? :-D
Full d'accord. I am not recommending to switch the rtems-4.8 toolchains
to gcc-4.3.0. 

Gradually backporting fixes to issues (such as my limits.h patches) we
discover during testing gcc-4.3.0 (gcc-4.3.0 has not even been released
yet!) definitely is desirable.

Ralf