Open Quuxplusone opened 15 years ago
Bugzilla Link | PR4259 |
Status | NEW |
Importance | P normal |
Reported by | Török Edwin (edwin+bugs@etorok.eu) |
Reported on | 2009-05-24 04:49:50 -0700 |
Last modified on | 2009-05-25 15:03:42 -0700 |
Version | unspecified |
Hardware | PC Linux |
CC | efriedma@quicinc.com, llvm-bugs@lists.llvm.org, nicholas@mxc.ca |
Fixed by commit(s) | |
Attachments | |
Blocks | |
Blocked by | |
See also |
(In reply to comment #0)
> Using svn r72361 that has bug reported in PR4258, if I let bugpoint run
further
> to simplify instructions in the testcase, it crashes because it creates a
> broken module.
>
> Testcase is same as for PR4258, but this time for bugpoint, instead of opt:
> http://llvm.org/bugs/attachment.cgi?id=3028&action=edit
No, thats the wrong file. The real testcase is too big to upload to bugzilla,
so here it is:
http://edwintorok.googlepages.com/bugpoint-reduced-blocks.bc.gz
Mmm, bugpoint plus debug info is trouble. I would try using the -strip-debug pass to remove debug info before trying to reduce a testcase with bugpoint.
(In reply to comment #2)
> Mmm, bugpoint plus debug info is trouble. I would try using the -strip-debug
> pass to remove debug info before trying to reduce a testcase with bugpoint.
>
Yes I did that, and then bugpoint was able to reduce the testcase.
However I still think bugpoint should be fixed not to crash on debug info.
It may happen that somebody needs to reduce a bug that occurs only in presence
of debug info.
Yes, bugpoint + debug info don't get along right now. The fix (IMO) is to make the verifier allow debug info intrinsics to take null arguments.