Re: [Gnu-arch-users] Re: [GNU-arch-dev] Re: how to fix a bad log message?

[Aaron Bentley]

John S. Yates, Jr. wrote:
> I understand the model.  I understand the purity/simplicity/rigor arguments.
> But I also recognize that in the real world things are not nearly as clean
> as you all want to make it.

Don't misunderstand me.  I don't think that Arch's "mistakes are 
forever" model is a good thing.  The decision to tightly couple branches 
and revision ancestry was a bad choice, in my opinion, and the results 
of this decision are responsible for much of the ugliness in the Arch UI.

Other models are possible that would make it safe to delete revisions, 
by ensuring that every revision has a unique name.  Martin Pool's 
working on an RCS like that, and I have high hopes for it.

> If you want arch to be widely adopted then it needs to accommodate real use
> cases from users of increasingly lesser technical competence.

Agreed.  I don't dispute the need for deleting revisions.  I just don't 
see what to do.  It would be a bad idea for the tool to encourage 
dangerous behavior.  Especially since archive consistency problems may 
not be apparent for a long time after the damage is done.

One thing that's safe: when you delete a revision, tag its ancestor into 
a new branch, and continue development from there.  (never commit to the 
old branch again)

In some cases, you actually need to delete a revision, e.g. if you 
commit a file containing nuclear launch codes.  In other cases, you just 
want to undo the effect of a revision, and Arch does provide ways of 
doing that.

> Even the most limited undo scheme could actually become a distinctive arch
> capability, unmatched in most (all?) competing RCSs.

I agree.  Some form of undo-commit would be very useful.  But the Arch 
model doesn't provide for it.  Rather than breaking the model to force 
it, I think it makes more sense to change the model to accomodate it.

Aaron

--
Aaron Bentley
Director of Technology
Panometrics, Inc.