Re: The Gnus repository is switching to Git as of 2010-04-19

[Stephen J. Turnbull]

Stefan Monnier writes:

 > I know that's how it works.  But I don't know what [git's
 > automatic-detection-implementation of copy/rename] means in terms of
 > the behavior when merging changes, including addition/removal of files.
 > IIUC it understands movement of files, but does it understand movement
 > of directories (e.g. if Gnus adds a new file lisp/gnus-new.el, will it
 > appear in Emacs as lisp/gnus/gnus-new.el or lisp/gnu-new.el?; similarly
 > if Emacs adds a new file lisp/new.el will it appear in Gnus as
 > lisp/new.el?).

Bazaar can't handle that and won't anytime soon as a standard feature
(no support for nested trees, how to do it remains controversial among
the core Bazaar developers).  git supports that with submodules
(speaking theoretically, I haven't tried this at home), but the
subtrees represented by git submodules must have identical structure.

 > I don't personally care whether the moves&removal&adds are done on the
 > Bzr or Git side.  AFAIK it's as easy to do it on either side.

Not.  See above.  You can take your chances with one of the Bazaar
plugins (bzr-externals, scmproj), but given the level of demand for
this feature, the fact that none of lifeless, jam, or aaron have
jumped on it suggests that the chances are unhappy that you'll bet on
the right horse, and in any case the documentation will undoubtedly
suck because the semantics will be buried deep in the Python code.

OTOH, git has a standard, built-in feature, with well-defined-and-
documented, simple semantics.

 > It should be just as easy for a branch as for the trunk.

Maybe.  I worry about ghost revisions appearing when you do a git->bzr
sync, and that's where they are most likely since git users branch
with abandon then abandon the branches, while it's no fun to try to
work with Bazaar that way.  Note that ghost revisions causes nasty
bugs in bzr as recently as a few weeks ago.