Re: [Axiom-developer] Which source distributions should we list?

[C Y]

--- Bill Page <address@hidden> wrote:

> Frankly I am not sure at all how we should manage the
> integration of the SVN-based repositories with Tim's new
> git version of Silver. I was especially disappointed to see
> that in the creation of the new git repository Tim has
> removed all of the prior revision history. :-(

If I'm not mistaken, Tim has already suggested the only workable
approach - isolate smaller changes which can be incrementally merged
into the master branch.  This makes for understandable changes which
allow incremental testing for breakage (and thus making it easier to
locate what broke).  IIRC this is a direct result of Torvald's own
philosophy of kernel development.  It makes some sense, but does imply
overhead (in the form of breaking a large change down into small parts)
when big changes are being considered.  I can think of a couple cases
in the past where large changes had a hard time getting merged.

At the current state of divergence, I don't think it's going to matter
if the old silver is preserved - the deltas are going to be large and
will need to be sorted out by hand anyway.
 
> Anyway as I understand it, right now Tim is manually updating
> both the git repository on axiom-developer and the "daly" branch
> on SVN in parallel. That does not sound like a particularly
> enlightened approach to me. :-( There's got to be a better way,
> right?

I thought git-svn worked with the new url?

The Git philosophy, as I understand it, is people work on a main trunk,
branching off to test individual ideas when are then folded back into
the master branch as they reach a viable point.  What we have, by
contrast, is a lot of large divergences by many people.  I don't think
this is necessarily the end of the world, although I do understand
Tim's frustration with it.  When there are major differences of
opinion, one tends to get major deltas.  But I would much rather see
this work go forward for later resolution than have it not happen at
all.

Perhaps Git could be used as follows - those who are working in similar
directions (Gaby and Waldek with wh-sandbox and build-improvements, for
example) could pick a common master and branch off of that.  Their
master will look rather different from silver, but they can at least
share code between those two trees and fold back into the common master
when one or the other solves a particular problem.  Once the main
points of divergence are resolved between the autoconf system and
silver by the silver maintainer (Tim) the silver master will look more
like the build-improvements and wh-sandbox work.  Thus, the main master
between those two trees can collapse gradually back into silver, and
work can then proceed from that master.

Those are interesting scenarios, but the most important thing is the
work gets done.  We can wade through diffs by hand if we need to - the
key is getting the problems solved.

Cheers,
CY


      
____________________________________________________________________________________Shape
 Yahoo! in your own image.  Join our Network Research Panel today!   
http://surveylink.yahoo.com/gmrs/yahoo_panel_invite.asp?a=7