Stefan Sperling <s...@elego.de>:
> Would reposurgeon in theory be able to use SVN dump files as a destination
> as well as a source?
I have considered this possibility. It would be difficult, and
it would expose me to support issues I don't think I want to deal with.
> Or perhaps the answer is to teach 'svnadmin load' about
> third-party formats, such as git's fast-import (which would then need to
> solve a rather similar data model mapping problem)?
You guys (the Subversion dev group) would be better equipped for the
job. I think there are decent reasons for you to try it. But beware
that there are some boojums among the snarks here.
The basic problem is the nature of the mismatch between Subversion's
ontology and the "gitspace" ontology of fast-import streams. Here
are several of Subversion's ontological premises:
1. Names of containers matter
2. Branches are copies
3. The basic merge operation is cherry-picking
4. User identities are local to an assumed namespace
5. Versions are strictly time-ordered and there is only one clock
In gitspace, every one of those premises is wrong.
I was going to write about this at more length but I have a kung-fu class
to get to. Perhaps I'll start one later tonight.
--
<a href="http://www.catb.org/~esr/";>Eric S. Raymond</a>
