On 2014-09-22 15:56, John Regehr wrote:
Thanks for the test case. We know what causes bugs of this form-- a pass reports that it made progress but it actually didn't. This breaks a C-Reduce invariant. If this happens often, we can put a check for this into the main C-Reduce, but until this I hadn't seen it happen for quite a while. John
Maybe it's better to add a check on this and save the temp file upon the failure? If somehow the blank transformation happened very early, then it would be quite bad.
- Yang