Re: PATCH to split up cpp_read_main_file

[Geoff Keating <geoffk at geoffk dot org>]

> Date: Tue, 05 Aug 2003 21:06:24 -0700
> From: Per Bothner <per@bothner.com>

> Geoff Keating wrote:
> > You did check that intermodule analysis still works (GCC bootstraps
> > with --enable-intermodule) after your patch, right?
> 
> I did not, since I didn't even know of the existence of the
> --enable-intermodule configure flag.  When I asked you about
> testing so it didn't break your changes, you just mentioned
> compiling a trivial pair of files, and if you mentioned the
> --enable-intermodule flag then I forgot about it.
> 
> Generally, I don't think you can expect people to routinely test
> with special configure flags.  Features that are only enabled
> by special configure flags aren't fundamentally different from
> features that are in separate development branches - except that
> hopefully it will be easier to keep them in synch.  The most you
> can reasonbly expect is that if someone breaks a non-default
> feature is that they try to fix it, but (just as with target-specific
> breakage) it's up to you to test it.

If you change target-specific code, or code that is exercised by only
some targets, you're supposed to test (at least one of) those targets:
<http://gcc.gnu.org/contribute.html#testing>.

Likewise, since this code is part of the small amount of code in GCC
that's specifically concerned with intermodule analysis, I think that
a patch to it should be tested using the only testsuite we currently
have for intermodule analysis, which is bootstrapping GCC with
--enable-intermodule.

Yes, I know that we should have a real testsuite.  Apple's had an
employee working on it full-time for several weeks.  Dejagnu seems to
be resisting the change with some vigour.  I have some testcases ready
and waiting for a test framework; however, those testcases don't
cover anything more than a bootstrap with --enable-intermodule would,
except for error handling.

-- 
- Geoffrey Keating <geoffk@geoffk.org>