Thanks Jeff for this very useful explanation. I guess locking is not
needed as long as the system is well understood by everyone (which was not
the case for us, sorry).
Sylvain
On Thu, 22 Jul 2010, Ralph Castain wrote:
On Jul 22, 2010, at 8:01 AM, Jeff Squyres wrote:
On Jul 22, 2010, at 9:54 AM, Ralph Castain wrote:
We used to have the branches "locked" to all but the gatekeeper to prevent this
kind of mistake. Did this change? Or did you forget to lock the 1.5 branch?
We changed it for v1.5 -- they're *not* locked because Ralph (back when Ralph was the RM
from LANL with me) and I anticipated letting developers commit their "larger"
features over to the release branch themselves and spare some GK pain.
Ah yes - I remember now. However, I thought we were going to lock it, and then
selectively unlock it when a particular update was approved. Guess I
misunderstood at the time.
Anyway, I would suggest doing it that way.
E.g., an ORTE refresh may take some pain and iteration to get right. So let
Ralph do it himself rather than do it via George/GK proxy.
Similar for hwloc issues. Similar for ROMIO refresh. ...etc.
Instead, however, we might want to change the permissions to lock it down, and
then selectively unlock it. Hindsight is 20-20. :-)
--
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/
_______________________________________________
devel mailing list
de...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel
_______________________________________________
devel mailing list
de...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel
