[Svnmerge] [PATCH] Cache svn output

[Rich Williams]

On 9/19/06, Giovanni Bajo <rasky at develer.com> wrote:
>
> the patch is very interesting, and I have been pondering something similar for
> quite of some time. I believe the correct way of doing this is by caching
> specific bits of information extracted from the SVN output, instead of being
> just a "dumb" mapping cmdline -> output. For instance, any "svn log" issued by
> svnmerge is different in its arguments (different ranges) but could still be
> cached properly (per-rev) and reused (modulo revprop edits, for which a
> rebuild-cache command could be devised).

I did consider the over-lapping case briefly, but didn't really find much call
for it in the kind of usage we're seeing. Looking at the command output
I currently have cached after running avail/merge between a number of
branches - only 3 of the 96 cached commands have the potential for
overlap, and those commands are typical the 'one per run' ones (e.g.
log --verbose --quiet -rREV:REV) the big saving here is on all the smaller,
shorter commands (e.g. 'log --incremental -v -r REV:REV' and 'svn propget -r
REV') which get repeated between runs, because the same revisions are
queried in many different cases.

> Thanks for proposing it. I should have some time to work on svnmerge next
> month, and I will see if I can come up with a more comprehensive patch. Feel
> free to ping me again :)

Thanks for considering it.

There are a few other things we'd like to add ...

- somewhere to store some defaults (we always use -b, but if you forget,
you can sometimes get confusing results).

- a way to limit the considered revisions to a single author (i.e. show me
which of my changes are on trunk, but not on the release branch)

- a way of suppressing the revisions which are just setting up branches
(i.e. Initialized merge tracking via svnmerge).

Do any/all of these things sound like things you'd consider adding?

Have fun,

Rich