[SystemZ] [release_34] clang regressions

Hi Richard,

The 2 llvm regressions I reported are now fixed (thanks again!) but now I see 7 clang regressions:

FAIL: Clang :: Analysis/misc-ps.c (257 of 16072)
******************** TEST ‘Clang :: Analysis/misc-ps.c’ FAILED ********************
Script:

İsmail Dönmez <ismail@donmez.ws> writes:

The 2 llvm regressions I reported are now fixed (thanks again!) but now I
see 7 clang regressions:

Yeah, sorry, I was seeing these too, but because I don't know clang
it took me a while to debug which part of it was being miscompiled.
I've just committed a fix for four of them as r196267 (trunk) and
r196268 (branch). That leaves the Index failures:

     Clang :: Index/TestClassDecl.m
     Clang :: Index/TestClassForwardDecl.m
     Clang :: Index/c-index-getCursor-test.m

which I'm looking at now.

FWIW, my normal testing procedure involves bootstrapping clang in
Release and Release+Asserts mode and then checking that the stage 2
and stage 3 objects are identical. I do the other testing
(projects/testsuite, etc.) on an unbootstrapped compiler and run
them first, since failures in those testsuites are easier to track
down than bootstrap failures.

That seems to have missed three separate bugs though, so from now on I'll
also run the testsuite on each bootstrap stage.

Thanks,
Richard

Hi,

Richard Sandiford <rsandifo@linux.vnet.ibm.com> writes:

İsmail Dönmez <ismail@donmez.ws> writes:

The 2 llvm regressions I reported are now fixed (thanks again!) but now I
see 7 clang regressions:

Yeah, sorry, I was seeing these too, but because I don't know clang
it took me a while to debug which part of it was being miscompiled.
I've just committed a fix for four of them as r196267 (trunk) and
r196268 (branch). That leaves the Index failures:

     Clang :: Index/TestClassDecl.m
     Clang :: Index/TestClassForwardDecl.m
     Clang :: Index/c-index-getCursor-test.m

which I'm looking at now.

I've just committed a fix for those as r196370 (trunk) and r196371 (branch).
The results should be clean now.

Thanks,
Richard

Confirmed as fixed, thanks!