Stefan Sperling wrote on Sun, Oct 21, 2012 at 18:08:27 +0200: > On Sun, Oct 21, 2012 at 05:47:28PM +0200, Daniel Shahaf wrote: > > Please address my earlier review question, asking why this needs to be > > implemented in the library and can't be done satisfactorily in user > > (C or Python) code. > > Well... let me turn this around on you and ask: > Why should it not be in the library? > You're implying that there is some good, apparently self-evident, > reason that it should not be done in the library. When asking others > to argue against that it helps to explain your own reasoning first. > > My reasoning is that using a callback to issue a warning about > failed verification is adding value to an established notification > mechanism that is already used for similar tasks in other contexts. > So it's great to expand the notification system a bit to support > the use case of reporting bad revisions during a verification run. > > I still think that it should just be the default mode of operation > anyway, with a summary at the end as Julian suggested. But I'm > equally happy with a --keep-going option if that is easier for
Verify could be not a dump but a dag.c tree crawl restricted to node-revs younger than START_REV. It could take a notify_func and call it whenever it encounters a corrupt noderev. That's the blue-sky option, of course --- I'm not signing up for this and I don't expect others to. > Prabhu to implement. I'm not going to have time to write the code > myself so I'm happy to accept whatever approach the patch submitter > believes is best (as long as the approach is within reason, which > I think it is).