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

[ebotcazou at gcc dot gnu dot org]

------- Additional Comments From ebotcazou at gcc dot gnu dot org  2004-03-01 06:43 -------
Thanks for the new testcase, it will help me to decide what is the most correct
fix for the problem.

Wolfgang, the optimizers are complex pieces of code that implicitly assume that
the code they are fed up with meets minimal requirements of consistency, because
they can't handle every single corner case.  The original testcase triggers an
ICE because of its indeterminate behaviour according to my analysis, that's why
I lowered its severity and priority.  I consider this to be good practice, since
you obviously want to tackle severe problems first.

And I don't think saying "it could show up [in] other places" to increase the
priority of the PR behind the back of the assignee is a better criterion.

I'll not use the same wording the next time I lower a priority.  But, frankly,
when you see

  long long a;

  scanf("%d",a);

  if (!a) {

you sometimes can't help doing so.


-- 


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