At this point, I'm sure everyone saw the power-plant that was just
checked into notes. Karl is going to add some further thoughts.
Everybody take your time to digest that doc. :-)
But to start some conversation (assuming you've *read* the doc first),
here are a couple of questions on the table:
1. In order to do a merge, the client must create two sets of
fulltexts in some temporary area. It then makes context diffs
between the fulltexts, and applies the diffs to your working
copy as local mods. We've talked about just asking for the 2
sets of fulltexts directly; we've also talked about doing fancy
things to make the 2 sets of fulltexts "materialize" via binary
diffs over the wire.
Given how rarely merges occur, is it worth jumping through hoops
to reconstruct the fulltexts? Maybe it's just simpler to have
the client grab the fulltexts directly. What do people think?
2. Last night Karl realized that we could just run
'svn diff -r X:Y URL'
and apply the context diffs right to our wc! In other words,
Philip's fancy new diff-editor is already doing what we want.
Cmpilato points out, however, that a *true* merge should be able
to generate context diffs between any 2 arbitrary urls, which
Philip's system can't yet do. Kfogel counters with the notion
that for 99% of all merges, the two urls are identical anyway.
So, what method to use? And, do we need to stay flexible and
accept 2 different urls?
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Sat Oct 21 14:36:58 2006