On Mon, 16 Jul 2012 22:31:26 +0200 Laszlo Ersek <ler...@redhat.com> wrote:
> On 07/16/12 19:12, Luiz Capitulino wrote: > > On Fri, 13 Jul 2012 21:11:15 +0200 > > Paolo Bonzini <pbonz...@redhat.com> wrote: > > > > >> I think doing it for error_set was just for symmetry and to avoid > >> introducing excessive complexity. > > > > We already check if the error is set in several places, and I don't think > > it will add much complexity. I still think that an assert() is better. > > If that means that the generated traversal code takes responsibility to > call any visitor callback with a fresh error receptacle, IOW I can go > ahead and just use error_set() in OptsVisitor and any firing assert will > be blamed on the generator: fine :) If that means it's finding bugs then that's great. On the other hand, if it shows only false positives and we end up having to re-work the code just to avoid that, then I'd agree on not having an assert().
