This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Testing C++ changes
On Wed, 2003-11-19 at 05:17, Jason Merrill wrote:
> On 19 Nov 2003 03:47:22 +0100, Gabriel Dos Reis <gdr@integrable-solutions.net> wrote:
>
> > Gerald Pfeifer <gerald@pfeifer.com> writes:
> >
> > | On Sun, 16 Nov 2003, Andreas Tobler wrote:
> > | > Personally I'd welcome if a core cp change happens that it should be a
> > | > must to bootstrap a complete gcc (except ada, g77 ?)
> > | > Libgcj contains a large part of c++. It's a good way to see if your
> > | > change is safe or if we uncover another problem.
> > |
> > | Given that this would have caught that bootstrap failure I'm tempted to
> > | agree.
> > |
> > | Jason, Mark, Gaby, what do you think?
> >
> > This seems reasonable to me -- even though I'm a bit worried about the
> > impact on build + testing time. But, it is consistent with what we
> > require for C changes.
>
> I don't think a full bootstrap is necessary. Note that libgcj has a
> 'clean-nat' target which just removes the C++ objects so you don't have to
> rebuild all of the Java objects to test a C++ change.
That indeed tests that the modified C++ compiler can build the JNI
code.
Is the proposal just to do that, or also to do "make check" on the Java
library?
I think we really ought to be doing everything we can to prevent
bootstrap breakage.
So, I think we should require that you rebuild the Java library when you
change the C++ front end -- if you are using a platform where GCJ builds
at all.
I'm not sure about running the Java library tests; do the Java people
think that's necessary?
--
Mark Mitchell <mark@codesourcery.com>
CodeSourcery, LLC