all 'gccas' tests fail due to recent gccas changes

[lattner]

Bugzilla Link	1167
Resolution	INVALID
Resolved on	Feb 22, 2010 12:50
Version	trunk
OS	MacOS X

Extended Description

The new 'gccas' doesn't output timing info and stats in the right place, causing the nightly tester to report
gccas failing every test.

[llvmbot]

I cannot reproduce this.

gccas is not used by llvm-test at all any more. The stats and pass timings are
generated by opt from TEST.nightly.Makefile. This was recently updated to use
opt instead of gccas. The GCCAS column in the performance reports are generated
by this rule in TEST.nightly.Makefile:

$(PROGRAMS_TO_TEST:%=Output/%.nightly.compile.report.txt):
Output/%.nightly.compile.report.txt: Output/%.llvm.bc $(LOPT)
@​echo > $@
@​echo '$(LOPT) Output/$.linked.bc -f -o /dev/null $(TIMEOPT)
$(EXTRA_LOPT_OPTIONS) >>$@ 2>&1'
@-if ($(LOPT) Output/$.linked.bc -f -o /dev/null $(TIMEOPT)
$(EXTRA_LOPT_OPTIONS) >>$@ 2>&1)
;then
echo "TEST-PASS: compile $(RELDIR)/$" >> $@;
printf "TEST-RESULT-compile: " >> $@;
grep "Total Execution Time" $@.info >> $@;
echo >> $@;
printf "TEST-RESULT-compile: " >> $@;
wc -c $< >> $@;
echo >> $@;
else
echo "TEST-FAIL: compile $(RELDIR)/$" >> $@;
fi

I don't see anything wrong with it.

All my nightly tests passed just fine last night (with the exception of the
usual failures). The only regression was 254.gap which I think you tracked down
to a problem in the code generator.

From looking at the sabre-ppc32 nightly test last night, it appears that your
llvm-gcc is out of date. A number of the CFrontend/C++Frontend tests failed and
if llvm-gcc wasn't up to date it could produce all the failures you saw.

We should really make the nightly tester build its own llvm-gcc.

[llvmbot]

Did your patch:

http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20070129/043898.html

fix this?

[lattner]

no, that fixed 'make bugpoint-gccas'

[lattner]

I will update my llvm-gcc, if it doesn't fix this, I'll reopen. Thanks!