FAI

Henning Glawe glaweh at physik.fu-berlin.de
Wed Jul 27 08:28:42 CEST 2005


On Tue, Jul 26, 2005 at 11:18:37PM +0200, Henning Sprang wrote:
> 1) "developers should be able to share code" was one of the requirements
> for the new revision control system. If all have only read access, and
> it's not possible with subversion, to restrict write access per
> developer per branch, then migration will only help with moving
> directories and files around, which, in my opinion, is not enough to
> solve the problems we face if we want to work on FAI with more people.
> And sharing code between each other by sending .tar.gz files by mail
> (until then?) is much more than only slightly inefficient.

it would be the same situation as now... well, it would be slightly better,
because subversion allows moving files (well, cvsps, on which cvs2darcs,
which I am currently using, is based, tries to determine moved files by
some heuristics).
very important:
every developer must be able to create branches on his own because of the
weak cherrypicking capabilities of svn: each feature should be developed in
its own branch, afterwards we tell thomas to 'pull in the "foo" branch to
trunk' (which has to happen quite often). if we don't have that possibility, 
we will run into major merging problems (well, basically each of the
developers would have his own fork of fai, with nearly no possibility to
track the trunk...).

> I will see if svk helps with 1), or ask Henning G., how would, for
> example, darcs help us to share the fai-rpm code until it eventually
> gets into the FAI subversion tree. How do I share code with someone else
> in that decentralised system? I need some place on the Internet to put
> my changes, right? Does that mean I have to upload my darcs repo to an
> http server after every change?

if you have darcs installed on the webserver:
 - if mail is working there, you can use 'darcs send' to send your changes
   to the repository
 - if you have ssh access, you can use 'darcs push' also through an ssh
   connection.

both, of cause, also work without a webserver ;). if you want to share code,
you can give all developers ssh access, which will work in both directions...

darcs is also lot easier when it comes to merging, because it tracks 
changesets and their merging history, so you don't merge the same changeset
twice (history-aware merging is one of the major weaknesses in svn).

> This is important, because, if the code is never shared, the
> requirement that everything that should go in FAI must be tested can
> never be met, so, by defintion, the code will never go into FAI if
> there's no means of sharing it.

y're right; that's why i publish my fai stuff on a webserver...

-- 
c u
henning



More information about the linux-fai-devel mailing list