Re: refs/notes/amlog problems, was Re: [PATCH v3 01/20] linear-assignment: a function to solve least-cost assignment problems

> Johannes Schindelin <Johannes.Schindelin@gmx.de> writes:
> 
> > To summarize, there are two commits recorded for that Message-Id, the
> > later one not mapped back, and neither is the correct commit that made it
> > into `master`.
> >
> > It would be nice to figure out what went wrong there, and how to fix it
> > for the future (and also to fix up the existing mis-mappings in `amlog`).
> 
> I think what happened is that I used to have post-rewrite, but
> because it did not solve the real issue of multiple commits existing
> for the same message ID (either because of amending, or because of
> running "am" multiple times while looking for the best base to
> contruct a topic branch for the series that contains it) *and* the
> one that will eventually used in the final history may not be the
> last one (e.g. I may "am" twice to see if an older base I use in my
> second attempt is a better one than the base I originally used, and
> the patches may even apply cleanly to the older history, but may
> turn out to need semantic adjustment, at which point I would discard
> that second attempt and use the old commit from the first attempt
> that built on a newer base), I stopped using it.
> 
> The mid-to-commit, for it to be relialble, needs to keep mapping for
> all the commits created from a single message, instead of being the
> last-one-survives mapping.  I just didn't have that much interest
> back when I decided it was not worth and dropped the post-rewrite, I
> think.