Re: [Gnu-arch-users] Re: Avoiding ancestor scan during get with revision library

[Aaron Bentley]

Miles Bader wrote:

> Aaron Bentley <address@hidden> writes:
>
> > Actually, it's looking for CONTINUATION files, not reading patchlogs, 
> > but it's the same principle.
>
>
> Actually, since archives are read-only why not store a file in each
> patch directory gives the patch-number of the last previous
> `interesting' patch (continuation, import) in that version was?
>
> Since e.g. a continuation revision forms a sort of base revision (even
> if it's not called `base-0'), perhaps call this file `BASE'.
>
> Then tla wouldn't have to search at all in so-annotated archives, it
> could just skip directly to the revision named in BASE.  Since it's just
> an optimization, it would be compatible with both old archives and old
> versions of tla too -- if there's no BASE file, just do the current slow
> search.

Yep, that would work.  I've had similar ideas about determining the 
latest revision, e.g. a RECENT file.  (Currently, we list the version 
directory, which is an O(n) operation with the number of revisions.)

Of course, this sort thing increases the program complexity.  Twice as 
many revision-determining codepaths to test.  Nothing's free.

Aaron