Re: dwarfcmp performance

[Roland McGrath <roland at redhat dot com>]

> On second thought, no it probably wasn't.  I probably picked canon dwarf 
> branch for that.  That I'm on dwarf-hacking now is still true, so what 
> I'm looking at is actually small-scale optimization.

Ok.  That is encouraging news.  That is, it's good to know that our
expectations beforehand about what was going on were not so wildly wrong.
It's very good to know that these week-long cases did all complete
eventually, since we had no explanation for infinite iteration had there
been any of that.  It's also good to know that the caching approach does
indeed produce vaguely reasonable times.  That means it is probably viable
to pursue the more intelligent "contingent caching" approach I discussed
before to solve the problem of false positives.  Otherwise we would need an
entirely new and different approach to the reference comparison problem.
It may very well still be that a different approach would be far better,
but we haven't yet imagined one at all.


Thanks,
Roland