[PATCH 1 of 4 lazy-manifest v2] manifest.c: new extension code to lazily parse manifests

[Martin von Zweigbergk]

If updates are in bulk without queries in between, it should be possible to
just append the updates to the list. Then, on query, sort with a stable
sort, pick the last entry for each file and drop the rest. Makes sense?

If queries are interleaved, we would have a problem anyway and would have
to do something smarter.

On Tue, Jan 13, 2015, 18:24 Augie Fackler <raf at durin42.com> wrote:

>
> On Jan 13, 2015, at 9:22 PM, Martin von Zweigbergk <martinvonz at google.com>
> wrote:
>
> >
> >
> > On Tue, Jan 13, 2015, 18:11 Augie Fackler <raf at durin42.com> wrote:
> >
> >
> > On Jan 13, 2015, at 7:42 PM, Martin von Zweigbergk <
> martinvonz at google.com> wrote:
> >
> > > As I reported on IRC, the 'setitem' method seems slow. On the Mozilla
> repo, running
> > >
> > >   hg co -C 1813b && hg mv -q intl i18n && time hg ci -qm 'move intl'
> > >
> > > takes 6.3s with current hg and 1m21s with these patches applied. It
> may very be related to your TODO in setitem.
> >
> > Yeah, I think this is the point where we should drop v2, and I’ll work
> on a bulk-add function (shouldn’t take long) to mitigate that problem.
> >
> >
> >
> > I think I once suggested making it transparent to the user by instead
> storing modifications separately and then compacting them on any query
> operation. That should effectively be the same (performance-wise) as an
> explicit bulk update operation, but without the need to update any callers.
>
> But if someone inserts a new entry, then does it again, the binary search
> won’t work, which makes things frustrating. If you’ve got an idea for that,
> I’d love to hear it though (I haven’t come up with anything better.)
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://selenic.com/pipermail/mercurial-devel/attachments/20150114/a9481875/attachment.html>