[ 
https://issues.apache.org/jira/browse/HIVE-144?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12654878#action_12654878
 ] 

Ashish Thusoo commented on HIVE-144:
------------------------------------

The overhead for that approach is too high. We had gone that route originally 
and were maintaining three different code drops for 0.17 (which we use in 
production at facebook), 0.18 which was a port requested by some external folks 
and 0.19 which was the hadoop version at the time hive was open sourced, and 
making these bug fixes and checking it into 3 different places was 3x the 
amount of work for the developers. Add to that, these branches would invariably 
diverge in the short term because the sequence of the checkins would not be the 
same, it became a big pain to port ones changes from one branch to the other 
because of the numerous merge conflict. So the solution for us, specially 
during this active phase of development was to have a single branch and compile 
it and run it with the different versions of hadoop.

For Hbase this probably works because they are perhaps a tad more mature at 
this point or at least have figured out the dependencies with hadoop to a large 
extent. It seems that they were also having a separate versioning scheme 
(different from hadoop) intially - I can hazard a guess that perhaps it was for 
the same reasons.

Anyway, given the minor tweaks that were needed in the code to port it to 0.18 
and 0.19, we saw this as the better of two evils.

Regarding the unit testing template code - we have really taken the pains to 
ensure that the actual run time code is in QTestUtil.java and the velocity 
templates are just used to generate the front end junit test cases for 
different queries.

> Ql java source copied to build/gen-java
> ---------------------------------------
>
>                 Key: HIVE-144
>                 URL: https://issues.apache.org/jira/browse/HIVE-144
>             Project: Hadoop Hive
>          Issue Type: Bug
>          Components: Build Infrastructure
>    Affects Versions: 0.19.0
>            Reporter: Johan Oskarsson
>            Priority: Trivial
>             Fix For: 0.19.0
>
>
> Since HIVE-98, the build process copies the contents of the ql/src/java 
> directory to build/gen-java where the code generated from Hive.g is located. 
> This makes it a bit more difficult to work on the project in Eclipse, since 
> one has to add the build/gen-java folder to the project and not ql/src/java. 
> This means we'll edit the ql java files in the wrong location. Adding both of 
> these directories will of course result in duplicate entries.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

Reply via email to