I finally got around to investigate the brokenness in execution
tests for mmix-knuth-mmixware. Thankfully it was not a "real"
breakage, but only due to gcc now emitting warnings while
building the status-wrapper, and the dejagnu machinery
subsequently barfing in a kind of cached-lossage way. Looking
around, I found drow's message at
<URL:http://gcc.gnu.org/ml/gcc/2004-08/msg00961.html>. That
would work (it's in dejagnu CVS), but as mec points out, we do
need something locally in gcc for this. We need it not only
until dejagnu-1.4.5 is released, but as the comment points out
in the patch, for whenever we "feel like breaking" the testglue
code again by emitting new warnings.
A few .exp's had more-or-less inadequate copies of the
testglue-building code, so I saw it as a inseparable change ;-)
to move them all to one place. This place is not gcc-defs.exp
(where it is loaded from for the benefit of all .exp:s loading
that file) but in a separate file. This is for the benefit of
other testsuites that want to refer to it without catching the
other definitions, notably libstdc++-v3. I did not change the
testsuite for libjava or other lib* testsuites that may need
changing, because I can't test that kind of change: it'd need a
simulator cross target that runs libjava and needs a status
wrapper. Quite spuriously I choose to update treelang.exp
anyway (though treelang doesn't build as cross), because the
change is identical to the other files in gcc/testsuite/lib.
This is in contrast to libjava, which needs to change more (yet
trivial) stuff e.g. wrap_compile_flags to wrap_flags.
Tested cross to mmix-mmixware-sim (needing wrapper) and cross to
cris-axis-linux-gnu (no wrapper) and native (no wrapper ;-)
i686-pc-linux-gnu (except the libstdc++-v3 change, which I forgot there).
Ok to commit?