Re: Alpha, S390 on 2.2.5pre1

[Ulrich Weigand <weigand at immd1 dot informatik dot uni-erlangen dot de>]

Ulrich Drepper wrote:
>"Martin Schwidefsky" <schwidefsky@de.ibm.com> writes:
>> Uli Weigand with his gcc 3.1 experiments found another one. make
>> tests failed for elf/filter because libgcc_s.so could not be
>> found. The reason is that the link rule for filtmod1.so misses
>> $(static-libgcc).
>
>This shouldn't happen.  filtmod1 is build using the build-module macro
>which is defined using the build-module-helper macro which in turn
>uses static-libgcc.  There must be something else wrong (in anything).
>I haven't seen problems on x86.

I may be missing something, but elf/Makefile does appear to suggest
that filtmod1 is built specifically *not* using build-module, but
via a special rule (that was missing static-libgcc):

# filtmod1.so has a special rule
$(filter-out $(objpfx)filtmod1.so, $(test-modules)): $(objpfx)%.so: $(objpfx)%.os
        $(build-module)
                                                                                                                $(objpfx)filtmod1.so: $(objpfx)filtmod1.os $(objpfx)filtmod2.so
        $(LINK.o) -shared -o $@ -B$(csu-objpfx) $(LDFLAGS.so) \
                  -L$(subst :, -L,$(rpath-link)) \
                  -Wl,-rpath-link=$(rpath-link) \
                  $< -Wl,-F,$(objpfx)filtmod2.so

The reason for the special rule apparently being the -F option.

(This is from the pre1 tarball.  Has it changed in the meantime?)

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  weigand@informatik.uni-erlangen.de