Re: g++ 3.0

[Matt Zimmerman <mdz@debian.org>]

On Tue, Jul 10, 2001 at 02:37:09PM +0200, Filip Van Raemdonck wrote:

> On Sun, Jul 08, 2001 at 10:45:31PM -0400, Matt Zimmerman wrote:
> > If the problem occurs only on certain architectures, you may want to try it
> > yourself on as many as possible before switching entirely to g++-3.0.
> 
> What is gained by this? Either he uses g++-3.0 for building these himself or
> he will still have to deal with all the problems.
> If he builds them with g++-3.0 he may just as well Build-Depend: on it and
> be done with it. In fact, if he doesn't Build-Depend: on g++-3.0 but does
> build manually with it, it might cause major trouble (NMU's, or just someone
> trying to build himself from sources without knowing he needs 3.0)

I stated in the following paragraph what could be gained from such a test.
Obviously, if the package were meant to be built with a different compiler on
certain architectures, this would be enforced in the debian/rules file, and
there would be no difference between manual and automated builds.

> > On some
> > Debian architectures, 3.0 is already the default compiler (see the source for
> > gcc-defaults, I think).  If those are the same architectures where your package
> > has problems, then there is no problem for Debian.
> 
> I think if those were the architectures, he wouldn't know of any problems...

That is not necessarily true.  Most likely, not all users of the program are
Debian users.  There is no telling which compiler versions are in use when
these problems surface.

> > As long as it doesn't link against any other C++ libraries, it should work,
> > yes.  Whether or not this is a good idea, I'm not sure.  There are issues with
> > g++-3.0 other than ABI compatibility (such as debugging).  It seems to be
> > available on all platforms, though, so it might be worth a try if it turns out
> > to be necessary.  It might be wise to restrict the change to the problematic
> > architectures, though.
> 
> If the package builds fine with g++-3.0, and must be built with it even on
> some arches, I don't see why one would want to restrict this usage. It will
> become the default compiler eventually anyway.

Read what I wrote.  g++-3.0 is a big step, and when it becomes the default
compiler, there will be a transition plan.  "Just build random packages with
3.0" is not a transition plan.  Who's to say that building with g++-3.0 on i386
will not introduce new bugs?

-- 
 - mdz