PATCH: Fix libstdc++ testsuite on systems without ranlib

[Rainer Orth]

Paolo Bonzini <bonzini@gnu.org> writes:

[I had never seen this mail, and been ill for 3 weeks, thus the late
response.]

> On 02/03/2010 07:27 PM, Rainer Orth wrote:
>> By adding some logging to libstdc++-v3/testsuite/lib/libstdc++.exp
>> (v3-build_support), I found that this happens because RANLIB isn't set
>> in the environment (the Makefile has .NOEXPORT:).
>
> It should still be in the environment if the testsuite is run from the
> toplevel (via $(NORMAL_TARGET_EXPORTS)).  Can you confirm that, or see why
> not?  Were you invoking the testsuite manually from the libstdc++ build
> directory?

I'm pretty sure I saw both problems at different times: during
bootstrap, RANLIB=: was probably passed in via the environment and broke
the build as described.  When I saw this, I went into
mips-sgi-irix5.3/libstdc++-v3/testsuite and ran make check there (with a
debugging version of libstdc++.exp to investigate, at which point I was
bitten by the fact that the Makefile didn't pass RANLIB in the
environment.

> The toplevel changes are okay now, but please change : to true also in the
> test for STRIP.  (And commit to both gcc and src).

Done for gcc.  Wrt. src, I see there are many other uncommited changes,
some even requireing additional files.  I'd like to leave this alone for
the moment.

> The libstdc++-v3 changes are safe, so I'm fine with putting them in now but
> first I'd like to understand my doubt above.

I've installed them as well, including the quoting Ralf suggested.

Thanks.
        Rainer

-- 
-----------------------------------------------------------------------------
Rainer Orth, Center for Biotechnology, Bielefeld University