gcc defaults [was Re: building slrn on sparc]

[Dave Love <d.love@dl.ac.uk>]

This important stuff got hidden in a rather non-obvious thread.

"David S. Miller" <davem@redhat.com> writes:

> On Fri, 27 Feb 2004 16:33:14 -0500
> Clint Adams <schizo@debian.org> wrote:
>
>> One difference is that this behavior is the opposite of upstream's
>> (gcc's),

... especially on other, long-established platforms.  It's also
undocumented as far as I can tell -- or, rather, contrary to the
documentation.  To this ex-GCC maintainer it just looks broken, apart
from the practical grief it seems to be causing.

> In fact, this behavior comes from a desire to do the right thing
> when building and bootstrapping gcc.

There seems to be a misunderstanding here of how it should work.

> GCC's bootstrap process expects
> that if the build/host system is sparc64-* that the system compiler,
> as well as the GCC binary produced by the gcc build itself, will both
> produce 64-bit executables by default.  If this is not the case, the
> GCC bootstrap will flat out fail.

Then there appears to be a sparc-gnu-linux-specific bug.  How does it
fail?

There's no such problem on IRIX64, for instance, which I've been
bootstrapping for years (particularly during testing of the original
multilib/autoconf framework), or on Solaris.  I assume likewise on
other 32/64-bit Debian systems.

More to the point, the compiler should be cross-buildable, and
`multilibbing' (compiling abi-specific code) should only be done with
the built compiler.

> In fact, you can ask Ben, that this is where the behavior change came
> from.  After I discussed this very specific issues with Ben over the
> course of several weeks.

Why not discuss it with GCC maintainers?  If you can explain the issue
I may be able to help, though I'm probably out-of-date on the build
process, and I'd prefer not to have to do much experimenting with
bootstraps on the hardware available.