One nice idea in Camp is to have a separate 'primary' format that you consult
for reading a repo and a 'secondary' format that you consult for writing to it.

Right now, Darcs only has the equivalent of the primary format.

I think it would be useful to introduce the notion of a secondary or local
format, which would contain things you absolutely do not have to care about
unless you actually want to write to the repository.

The use case I have in mind is for the --no-working-tree option.  Darcs apply
should not attempt to touch the working directory if a repo was initialised as such.

I think the right way to introduce the local format is when there are local
format options available to write "localformat" into _darcs/format (which
indicates that this repo requires support for the "localformat" mechanism). 
This could cause compatible darcsen to check _darcs/localformat (or whatever).

Petr was working on this over the sprint, and has devised a new format
mechanism, a very simple one that allows us to accomodate things like 'optional'
features (the pristine tarball) and also no-working-tree.

If I remember correctly, it involves splitting things into key-value pairs (fmt
borrowed by some RFC):

pristine: hashed
semantics: darcs-2
working-dir: enabled
optional: pristine-tarball

I'm adding Ganesh to this because darcs rebase will need to have a
format for rebasing repos.

I'm also unassigning Petr, who I think is busy making Darcs fast :-)

And I'm assigning back to kowey again. By now it's not realistic to
design a new format before the 2.5 release. Is there a way to complete
issue1277 without completing this one?

Bumping to 2.6.

Looks like it's not a prerequisite to issue1277 (although it would be
nice).  I'm afraid this will have to fall to the collective.  Maybe
something for the 2010-10 sprint agenda.

Three case studies:

- packs (optimize --http?)
- no working directory
- rebase

See also issue1099 (format mechanism should also extend to patch bundles)