Re: [Gnu-arch-users] self-contained changesets?

[Tom Lord]

    > From: Joshua Haberman <address@hidden>

    > Arch seems to require that anyone who wants to make their personal
    > changes available to the world have access to a publically available
    > server to host their personal archive.  Is this a safe assumption?

Mostly I think it is -- but that doesn't mean that it isn't worth
providing support for when the assumption is false.   It's also not
quite the case that arch requires that -- although the ways in which
it doesn't require it should probably be made more featureful.

Would you agree that email and netnews provide the other two most
obvious transports?

A simple thing is that you can, of course, pack up changesets to send
via such transports.   Currently, encoding a tar bundle of a changeset
is the way to do that --- certainly (as has been often discussed) a
fancier encoding would be welcome.

A more complex idea is to allow some degree of archive mirroring over
those transports:  monotone has been exploring this design space a bit
-- I think the idea can fit cleanly into arch.


    > Also, for the case where someone wants to make small changes to an
    > arch-versioned tree, the process of creating an archive, branching, and
    > publishing the archive seems like overkill.

Simply packing up a changeset and mailing that works just fine.
Again, a "prettier" format for emailing changesets is desirable.


    > Is it possible to implement a command similar in spirit to "cvs diff?" 
    > What I mean is that such a command would produce a self-contained
    > changeset that doesn't have to be part of an archive, that could then be
    > sent independently (say, as an attachment to email).   Such as changeset
    > would ideally apply as cleanly as possible even if the branch from which
    > it was generated changed in the meantime.

See: 

     % tla mkpatch -H

     % tla revdelta -H

     % tla get-patch -H

     % tla dopatch -H

and, of course there's always:

     % info diff

     % info patch

:-)


    > In other words, I want to be able to generate an independent changeset
    > against a tree that I can apply at any point in the future to that
    > branch or its ancestors and have tla "do the right thing."

-t