Re: [Gnu-arch-users] situations where cached revisions are not so good

[Robert Collins]

Firstly,
   paraquoting Tom, from awhile back on IRC:

      there are a lot of tweaks that can be done [to archive 
      retrieval efficency], and (most) of them can be placed 
      in a smart server. I hope folk play with this ontop of
      the arch core.

Secondly,
  There are a number of things that can be done:
* Alternative deltas (in addition to the standard patchset):
  arbitrary deltas ('summary changesets are a special case of these').
Examples include:
 - summary changesets.
 - /dev/null->revision   (aka what 'import' generates)
 - cross-branch changesets (i.e. a changeset from mainline-555 ->
branch-32)
* least cost retrieval logic in the client:
  - build a SPF spanning tree out from the needed revision, stopping as
soon as all unterminated paths are more costly. (which will generate a
path through the archive to one of:
  - something in the revision library
  - a pristine tree the client knows about
  - the tree version of the project tree
  - a /dev/null->revision

  the exact termination point will depend on the SPF found result).

A smart server is largely orthogonal to the above, but:
  - can generate arbitrary changesets (so the clients existing nodes in
the revision graph would be a very useful hint to a smart server)
  - doesn't exist today ;).

Rob
-- 
GPG key available at: <http://members.aardvark.net.au/lifeless/keys.txt>.

signature.asc
Description: This is a digitally signed message part