[darcs-users] Re: Re: Annotate doesn't recognize anything

[David Roundy]

On Fri, Aug 10, 2007 at 03:37:29PM +0200, Tommy Pettersson wrote:
> On Fri, Aug 10, 2007 at 10:47:38AM +0200, Alexander Staubo wrote:
> > On 8/9/07, Matthew Palmer <mpalmer at hezmatt.org> wrote:
> > > On Thu, Aug 09, 2007 at 12:40:42PM +0200, Alexander Staubo wrote:
> > > > I'm surprised that Darcs does not fall back to the checkpoint to get
> > > > the patches it needs, though.
> > >
> > > You mean retrieve the extra patches from the parent repo?  --lazy does that,
> > > I think (not available in 1.0.9, only in the 1.1.0 pre-releases).
> > 
> > No, I meant look in the patch file in _darcs/checkpoints/..., which
> > looks to me like it contains the entire repo as a single patch. It
> > would be slow to process, but it's better than just failing with the
> > wrong error message, and annotate is horrendously slow already, as
> > previously demonstrated. :)
> 
> Annotate on a file shows for each line of the file which patch
> the line comes from. This information is collected by examining
> the contents of all patches touching the file. The checkpoint
> patch is just a snapshot, so unfortunately it does not provide
> information about which patches originally added the lines. The
> best darcs could do (besides fetching the missing patches) is to
> say all "old" lines comes from the checkpoint patch. But I don't
> think darcs' traditional partial repo format knows _which_
> checkpoint is its "base". Actually I don't even think it can
> tell it _is_ a partial repo with any certainty. So implementing
> this feature could turn out to be tricky. And now the new lazy
> hashed repo format is on its way.

Right, this would be tricky simply because we don't actually *record* that
we are a partial repo, we only know because reading a patch file fails.

> I don't want to say this would be a bad feature as is, but
> perhaps a more general form could be interesting. What if
> Annotate took the --from-patch and --from-tag flags? It could
> hopefully make Annotate faster when only the latest changes are
> of interest. All lines from before the break could be
> un-annotated or marked with the tag in --from-tag.

Another possibility would be an annotate --partial option, which would
potentially speed up annotate even on full repos, but might be easier to
code than the full --from-tag and --from-patch tags.

> I'm not sure how big a change this would be, because I think the
> code used by Annotate works by producing a _complete_ history of
> all changes to a file, and then showing the "surface", but maybe
> the evaluation is lazy in a way that only the needed patches are
> examined. I fear just cutting out the selected patches and feed
> that to the Annotate code, instead of all patches, would result
> in all "old" lines being missing.

I think it wouldn't be too hard to do.  If nothing else, we could probably
just make the checkpoint patch itself be a "pseudopatch" (probably giving
it an artificial name to clarify that it *isn't* a real patch) in the input
to annotate's internals and reuse the existing code.
-- 
David Roundy
Department of Physics
Oregon State University