hadoop-pig-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Olga Natkovich" <ol...@yahoo-inc.com>
Subject keeping Load-Store branch and trunk in synch
Date Thu, 17 Dec 2009 00:57:31 GMT


As many of you now we are actively working on Load-Store redesign for
Pig. This work is documented in
http://wiki.apache.org/pig/LoadStoreRedesignProposal and is happening on
a separate branch with intent to merge is back into the trunk early next
year. We need to figure out a way to keep the L-S redesign branch as
close as possible to the trunk. Here is the proposal. Please, let us
know what you think.


We need to periodically merge from trunk to the branch. Most code will
merge as is. However, the changes in the area of Hadoop integration or
Input/OutputFormat handling will be tricky. For those (hopefully few)
patches, the proposal is to make the owner of the patch to create a L-S
redesign compatible patch and get it applied directly to the branch. The
second part is to make sure that the patch from the trunk is not applied
at the time of the next merge. Currently, merges are done starting from
the point of the last merge. The information of the last SVN number is
recorded in the commit message of the previous merge and can be found
from the svn log. The following modification to the process is proposed
to accommodate the patches that need different versions for trunk and
L-S branch:

(0) Lock the trunk and the branch for the duration of the merge and
patch application.
(1)     Synch the L-S branch to the point of the patch on the trunk
(2)     Apply the patch to both the trunk and the branch
(3)     Record the SVN number that resulted after the patch has been
applied to the trunk is. (This basically skips the just committed patch
from the next merge.)




  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message