hnrklssn wrote:
> > >
> >
> >
> > +1, it's why I stopped recommending this approach in my reviews.
>
> While I agree for the most part, there are quite a few diagnostics that are
> absurdly verbose (particularly ones with "did you mean to FLOOP it?"). I
> think partial matches are good/fine, but they need enough of the diag to not
> be ambiguous.
I don't necessarily oppose the use of truncated checks where it makes sense, I
oppose the implicit default on the matcher's part. An
`expected-note-re{{.*FLOOP.*}}` would be a reasonable explicit opt-in. A new
mode like `expected-note-part{{FLOOP}}` would be even better, making the old
mode behave like `expected-note-re{{^FLOOP$}}`. As an (admittedly extreme)
example of footgunnery with this implicit behaviour, we had bug downstream
where there were two versions of an error, something like this: `cannot use FOO
in context asdf` vs `cannot use FOO in context asdf; did you mean to use BAR?`.
The `; did you mean to use BAR?` version was accidentally emitted in some cases
where it didn't make sense to do so, but the tests still passed because the
diagnostic *did* contain `cannot use FOO in context asdf`.
https://github.com/llvm/llvm-project/pull/108658
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
