Re: [Gnu-arch-users] question about greedy library and history

[Aaron Bentley]

David Allouche wrote:
> On Thu, Apr 08, 2004 at 12:20:09AM -0400, Aaron Bentley wrote:
>
> > David Allouche wrote:
> >
> > > On Mon, Apr 05, 2004 at 11:16:34AM -0400, Aaron Bentley wrote:
> > >
> > >
> > > > David Allouche wrote:
> > >
> > > Do not get me wrong, the old code _did_ allow hardlinks to cross
> > > tag and archive boundaries when building a revlib, didn't it?
> >
> > As far as I can figure it, the old code *did* cross archive boundaries, 
> > but only if the revlib contained the immediate ancestor.  I guess that 
> > way you don't need access to the ancestor archive.  That subtlety was 
> > lost on me earlier, but of course, it makes a big difference.
>
>
> Okay, that's good to know.
>
> What is the new rule? If the revlib is non-sparse, under which
> circumstances will the the parent revision be built, and under which
> circumstances will the cachedrev be used instead (breaking hardlinks)?

In addition to hardlinking in all the cases the current code does, the 
backbuilder will also hard link if
- there's no closer pristine
- there's no import or cached revision 50 steps closer

> > implementing aba isn't hard, and it'd probably be shorter in a real 
> > programming language.
>
> According to what I was able to extract from Rob, that's already
> essentially done, and more. But he feels like he needs a bigger critical
> mass of actual features before making a public release. Or maybe he was
> just pulling my leg.

Features, smeatures.  Release early, release often.  Make me obsolete.

Aaron