[Gnu-arch-users] Re: Round II -- new language, arch, furth, etc.

[Stefan Monnier]

> usual concepts of branches and things.  Versions of code are defined
> by what patches they contain, so two versions are closely related if
> they contain similar sets of patches, and the nearest common ancestor
> of two versions can be formed by set intersection (and that ancestor
> may never have existed).  All this requires special properties of the
> patches, so they have to be partly human generated, which is
> presumably an inconvenience.

My understanding is that you don't have to hand-generate patches.
Instead, the "build-revision" step may fail if you specify a set of patches
which cannot be applied one after the other (like cherrypicking).
Providing carefully hand-generated patches can help reduce such failures,
so that darcs can more often automatically build a revision that is the
result of cherrypicking (or as you say "an ancestor that never existed").


        Stefan