patch rereview requested for PRs 6860, 10467 and 11741

[Richard Kenner]

    | If you hadn't used the word "silly", I'd agree.  Showing that all a patch
    | does is turn something that used to ICE into something else is an
    | exceptionally convincing proof that the patch can't break a bootstrap or
    | introduce regressions.

    However, that sole argument is not sufficiently convincing, int that
    it may turn an ICE into bad code generation.  I'd rather an ICE than a
    silent bad code generation.

Of course.  But read carefully what I said, which is that it's convincing
proof *only* that it can't break things or cause regressions.  That's
important (and can be the *most* important thing during some stages of
development), but does not remove the requirement to show in other ways that
the patch is correct (e.g., your example of replacing an ICE with incorrect
code).

So we are agreeing.