Re: more about the external target headers/libraries issue

[Dan Kegel <dank at kegel dot com>]

William A. Gatliff wrote:
> I think it's just a problem of tuning the gcc build process so that it
> stops before it needs those headers, but _after_ enough of the
> compiler is built that it can compile headerless code.  Then it's up
> to newlib and glibc to make sure _they_ can build using their own
> headers (which I think we're at already).
>
> The logic in my head goes something like this.
>
> As I see it, the gcc-3.x libgcc2 has merely gotten polluted with some
> header-requiring constructs, where the 2.95.x libgcc2 was relatively
> header-free.  The 3.x "bootstrap" build target, all-gcc, won't finish
> until it has built libgcc2.  It can't build libgcc2 in a bootstrap,
> because it has no header files.  But I can't build headers for it
> because the gcc bootstrap build process refuses to give me a compiler
> until it has a libgcc2.
>
> The thing is, I don't _need_ libgcc2 to build libraries.  At all.
>
> What I'm looking at is somehow removing the libgcc2 dependency from
> the all-gcc target, so that the build process will install xgcc and I
> can get on with building those header files that libgcc2 needs.
That would be the best of all worlds.  Go for it!

Even when you're done, though, and gcc no longer depends on
glibc, I still think gnu ought to maintain that
build-an-entire-working-toolchain script
(greatly simplified by the change you proposed)
because that would better document the remaining
loose ends (like having to change the PATH in the middle
of the script, etc.).
- Dan

--
Dan Kegel
Linux User #78045
http://www.kegel.com