Re: [Gnu-arch-users] Retrieving per-file data from csets/revs

[Tom Lord]

    > From: Samium Gromoff <address@hidden>

    > At 02 Dec 2003 13:56:24 +0100,
    > Robin Farine wrote:

    > > >>>>> Samium Gromoff writes:

    > >     > You have a 10M changeset in a remote archive accessible via
    > >     > an 33.6kbaud modem.

    > >     > And you want to see what was changed in one specific file, without
    > >     > downloading the whole monster-delta.

    > >     > How do you do that?

    > > You write an Arch smart server ;)

    > I think it`s at least one more reason to do so.

    > (Yes, a hint to the future implementors of the smart server)

In the sense that arch browsers count as a kind of "smart server"
that's true.

In general, there's a gazillion different "views" one might construct 
for a given set of arch revisions:  index by identifiers, index by
file histories, index by revision creators, index by the merge graph,
etc.

All of those different views suggest creating different ancillary data
structures to optimize different kinds of access patterns to the raw
arch data.    

The basic dumb-file-system archive format optimizes the access I think
is most commonly needed by working programmers.   It does a decent job
at compression.   It does a decent job of minimizing network traffic
for common usage patterns of active project participants.

For all those other views: start building ancillary tools (whether you
want to call them "smart servers", "browsers", or whatever).

-t