Re: [testsuite, build] Convert boehm-gc testsuite to DejaGnu (PR boehm-gc/11412)

[Rainer Orth <ro at CeBiTec dot Uni-Bielefeld dot DE>]

Hi Iain,

> One issue on powerpc-darwin9****
> (which I would expect to manifest on *-*-darwin*)
[...]
> [[[ **** this is failing because of ranlib output:
>
> Executing on host: ../libtool --silent --tag=CC --mode=link  /GCC/
> gcc-4-6-trunk-build/gcc/xgcc -B/GCC/gcc-4-6-trunk-build/gcc/
> staticrootslib.lo /GCC/gcc-4-6-tr
> unk-build/powerpc-apple-darwin9/./boehm-gc/libgcjgc.la  -O2  -version-
> info 1:2:0 -no-undefined -rpath /nowhere -Wc,-shared-libgcc /GCC/
> gcc-4-6-trunk-build/power
> pc-apple-darwin9/./boehm-gc/libgcjgc_convenience.la -I/GCC/gcc-4-6-
> trunk-build/powerpc-apple-darwin9/./boehm-gc/include -I/GCC/gcc-live-
> trunk/boehm-gc/testsuite
> /../include  -Wc,-shared-libgcc -lpthread -lm   -m32 -o 
> libstaticrootslib.la    (timeout = 300)
> ranlib: file: .libs/libstaticrootslib.a(backgraph.o) has no symbols
> <snip>
> ranlib: file: .libs/libstaticrootslib.a(win32_threads.o) has no symbols
> output is:
> ranlib: file: .libs/libstaticrootslib.a(backgraph.o) has no symbols
> <snip>
> ranlib: file: .libs/libstaticrootslib.a(win32_threads.o) has no symbols
>
> I would imagine that a similar situation must have been handled somewhere
> else in the test-suite - but not sure where... ]]]

the only other place in the GCC testsuite where ranlib is called is in
libstdc++-v3/testsuite/lib/libstdc++.exp, but there it's called via
remote_exec and the output ignored.

While one might consider handling this in upstream libtool when --silent
is passed, the other solution would be to defined a boehm-gc-prune proc
that handles this.  I'll give it a try, but that may take until the next
weekend.

>> Ok for mainline (either now, or, more likely, after 4.6 branches?
>
> my 0.02 euro...
> since AFAIK these tests are
> (a) currently unreported by the regression testers, and ...
> (b) only applied for the native lib unless one does it by hand ...
> - ISTM that we would only be improving test coverage by applying this for
> 4.6.

While I agree with that reasoning (and I've already found a boehm-gc bug
due to the new testsuite), my caveat was that this may add testsuite
noise so short before the 4.6.0 release that there might not be enough
time to fix that.  But I could certainly be persuaded to apply the patch
now :-)

Thanks.
	Rainer

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