[Gnu-arch-users] Re: towards standards specifications

[Miles Bader]

Jason McCarty <address@hidden> writes:
> > How easy is it to turn a random changeset into an arch commit?  I'm 
> > sure it should be easy, yet I lack the required magic words.
> 
> tla dopatch changeset-dir working-dir ; $EDITOR `tla make-log` ; tla commit

Hmmm, I wonder if this can be done more `naturally', without unpacking
the changeset, repacking it, and then commiting the repacked changeset.

The code for arch_commit basically does:

   arch_prepare_commit_changeset
   ... stuff to put changeset + log in archive ...

Where arch_prepare_commit_changeset does:

   arch_make_changeset
   ... update log (add generated fields) ...
   ... include log in changeset ...
   [... update changelog ...]

arch_make_changeset is what `mkpatch' does.

So one would think it shouldn't be _too_ hard to just add an option to
make commit skip the call to `arch_make_changeset' and use a
pre-prepared changeset instead (adding the log to it &c, just like a
normal commit).

I _think_ (from lightly perusing the code) that in that case commit
wouldn't need a project tree at all, as except for the call to
arch_make_changeset, it only uses it for a few random things that could
possibly be handled otherwise or just skipped.

Of course whether it's worth the effort to do the above, I don't know...
It might make Robert Collins' `pure-merge' stuff a lot more efficient
(I think that can be done with only log-file changes).

Hmmm.

-Miles
-- 
`Life is a boundless sea of bitterness'