Re: Change set based shallow clone

From: Marco Costalba <mcostalba@gmail.com>
Date: Sep 9, 2006, 13:00 UTC
Message-ID: <e5bfff550609090600k4e4b1ae0s1cbee5e3fa01cba@mail.gmail.com>
In-Reply-To: <20060909103157.23388.qmail@science.horizon.com>

On 9 Sep 2006 06:31:57 -0400, linux@horizon.com <linux@horizon.com> wrote:

Show 27 quoted lines

> > If the out of order revisions are a small amount of the total then
> > could be possible to have something like
> >
> > git rev-list --topo-order --with-appended-fixups HEAD
> >
> > Where, while git-rev-list is working _whithout sorting the whole tree
> > first_, when finds an out of order revision stores it in a fixup-list
> > buffer and *at the end* of normal git-rev-lsit the buffer is flushed
> > to receiver, so that the drawing logic does not change and the out of
> > order revisions arrive at the end, already packed, sorted and prepared
> > by git-rev-list.
>
> I don't think I understand your proposal.  The problem arises when
> git-rev-list finds a commit that it should have listed before something
> that it has already output.
>
> Just for example:
>
> Commit D: Ancestor B
> Commit B: Ancestor A
> Commit C: Ancestor B
>
> Commit C is the problem, because if git-rev-list has already output B,
> there's no way to back up and insert it in the right place.
>
> How is waiting to output the already-behind-schedule commit C going
> to help anything?

It helps in a number of ways from receiver point of view.

- Receiver doesn't need to keep a sorted list of loaded revisions.

- Receiver doesn't need to stop parsing input and redraw the graph in
the middle of the loading.

- Receiver doesn't need to  check for possible out of order commits
when loading data, but can be assured data that arrives is
--topo-order consistent (although my not be complete)

- Split the in order parsing code (performance critical and common
case) from fixup-code (run at the end and on a small set of commits).

- Much faster implementation because the sorting is done only in
git-rev-list and only once.

Following your example my suggestion was something like this:

Instead of:

git-rev-list HEAD

Commit D: Ancestor B Commit B: Ancestor A Commit C: Ancestor B Commit A: Ancestor E Commit E: Ancestor F Commit F: Ancestor G .... Commit V: Ancestor Z

git-rev-list --topo-order ----with-appended-fixups HEAD

sends something like:

Commit D: Ancestor B Commit B: Ancestor A Commit A: Ancestor E Commit E: Ancestor F Commit F: Ancestor G .... Commit V: Ancestor Z (* end of correct sorted commits and start of out of order commits*) Commit C: Ancestor B ...... Commit N: Ancestor M

So that receiver drawing code designed with working with --topo-order kind output could as usual draws the graph until Commit V: Ancestor Z (but without waiting for git-rev-list latency!!) and the new fixup code could *at the end* loop across *already sorted* fixup set:

Commit C: Ancestor B ...... Commit N: Ancestor M

and update the graph in one go. Of course some flag/syntax should be introduced to trigger the end of sorted code and beginning of fixups in git-rev-list output stream.

       Marco