Re: [PATCH] Disable bprob.exp for cross targets

[Hans-Peter Nilsson <hp at bitrange dot com>]

On Tue, 30 Sep 2003, Eric Christopher wrote:

> On Tue, 2003-09-30 at 08:36, Joern Rennecke wrote:
> > > > Can we safely test if we are building with in-source newlib at that point?
> > >
> > > Should be able to, yes. Otherwise none of the other execute tests would
> > > link either.
> >
> > Huh?  We can't link any target execute tests when we are configuring a
> > cross toolchain - we have to build it first!
>
> By the time we're testing we've already built newlib. If you haven't,
> then it's your fault.

But it's a matter of the bprob support functions in libgcc;
namely gcc/libgcov.c being compiled, right?.  So it seems a
chicken-and-egg problem: we can't configure newlib until gcc is
built, and we need the newlib headers when building libgcc.
The toplevel configure knows of this case:

# If using newlib, add --with-newlib to the extra_host_args so that gcc/configure
# can detect this case.

So it seems a wart in the gcc build process to not be able to
handle this, but instead requiring installed headers to fulfill
the dependency for building the full libgcc.  I guess libgcc
should be moved to toplevel, and depend on maybe-newlib or
something.  (Huh, we're slowly getting the same maze of
dependencies as when building gcc+glibc.)

The --with-newlib option seems misnamed: you're supposed to be
able to build with a previously installed newlib, which I think
is a valid interpretation of what to expect from --with-newlib.
Maybe calling it --with-newlib-source would be better.  BTW, a
bug in libstdc++-v3 configury stops pre-installed builds with
newlib at the moment.  I don't remember the PR.

brgds, H-P