[Bug optimization/14235] [3.4/3.5 Regression] ICE in verify_local_live_at start (flow.c:546)

["bangerth at dealii dot org" <gcc-bugzilla at gcc dot gnu dot org>]

------- Additional Comments From bangerth at dealii dot org  2004-03-01 15:20 -------
In response to Eric's comment #14: I think I may have been misunderstood. 
I certainly didn't want to tread on anyone's toes with my comment, and 
I also didn't change any priorities (but that comment wasn't obviously 
meant for me anyway). I think what I wanted to say is that when we 
reduce long testcases to small ones, we often end up with totally 
nonsensical code that might well be undefined. If I can, I frequently 
remove initialization code if it preserves the ICE, and I also usually 
try to reduce most variables to integers, so that is where the dumbness 
comes from. At the same time, the original testcase may well have been 
totally meaningful and well-defined. My statement was just that looking 
at a reduced testcase doesn't really give a good picture of whether we 
have an ICE-on-well-defined or an ICE-on-undefined. This PR is actually 
a good example, because the reporter came up with a perfectly valid testcase 
after your comment, so the "dumbness" was introduced upon testcase reduction. 
 
That being said: I'm sorry if I offended someone, I clearly am not :-) 
 
W. 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14235