On Mar 21, 2007, at 11:43 AM, John Sundman wrote:
My brain showed early signs of exploding the first time I read
this, but eventually my rational thought mechanism kicked in and it
started to make sense.
summary: Perforce has the ability to track multiple sets of changes
made to a single source tree. Subversion doesn't. Subversion makes it
easy to duplicate a local tree -- you just copy the root directory --
and it makes it easy to create a new branch -- 'svn copy' takes about
5 seconds. So the natural working style with Subversion takes
advantage of these abilities rather than trying to cram as much work
as possible into each local source tree.
We've tried to implement changesets on top of Subversion, but with
somewhat limited success and continued issues. My proposal is that we
shift to a working style that is more adapted to Subversion.
Forgive me for coming late to the party, but does Subversion know
enough to not send stuff it doesn't need to send? Or does every
checkin take forever?
I'm not proposing that *everything* gets checked in. I'm proposing
that all *changes* get checked in, rather that checking in a subset.
If you think about it, checking all changes in is really the safest,
most conservative approach, because you are checking in exactly (*)
what you have tested locally against.
(*) not quite -- as you know, you can still forget to 'svn add' a new
file.
