Re: Reporting the Microkernel in the GNU triple?

[Roland McGrath]

> Should we be reporting the microkernel in the GNU triple?

I think we would prefer not to.  Firstly, we don't want to change the
well-established use of `gnu-gnu'.  Second, we would like to consider the
GNU/Hurd ABI per se to be (eventually) something that is independent of
microkernel.  The kernel-os part of the configuration tuple really just
defines a milieu in which the means of compatibility are well defined.  
I think we will be well enough served by using the other means we have
(sonames and symbol version sets and so forth) to distinguish things within
the Hurd family.

> Also, config.guess doesn't report the version number for GNU/Linux -
> anyone know why?  

This is not really the place for that question.  I would guess that noone
really wanted it before.  It might cause problems if it started reporting
e.g. i686-pc-linux2.4.20-gnu, because of things out there that assume that
-linux-gnu or -linux-gnu* will match.  There are other ways of determining
what you want to know from the compilation environment, so there isn't
often a need to key anything off a version number in the tuple.  

> I just spent some time trying to find an elegant solution for a problem
> that wouldn't exist if we didn't have the 0.3 in "i386-unknown-0.3"

That bit did not make sense to me at all.

It has always been GNU convention that os names might have detailed version
numbers on the end, e.g. solaris2.9.  Anything using GNU configure
conventions is broken if it doesn't use wildcard matching on all the os names.
It's usually best to key off other things rather than that, but that
doesn't mean the number shouldn't be there in a canonical tuple.