*Update:*
There is a blocker bug [1] found for 1.16 with TPCDS performance runs and
Gautam is working on it currently. Once that is fixed and there are no
other blocker bugs I will prepare RC0. Since the branch is already created
for 1.16.0 on apache side [2], it will be helpful if everyone can do some
initial testing. This will help to find any potential issues before RC
candidate is created and avoid delays with release.

[1]: https://issues.apache.org/jira/browse/DRILL-7182
[2]: https://github.com/apache/drill/commits/1.16.0

Thanks,
Sorabh

On Mon, Apr 15, 2019 at 7:42 PM Ted Dunning <[email protected]> wrote:

>
>
> On Mon, Apr 15, 2019 at 12:35 PM Sorabh Hamirwasia <
> [email protected]> wrote:
>
>> ...
>>
>> @Ted Dunning <[email protected]> - I am not sure if I understood
>> correctly but I think the reason master is locked and two active branches
>> are not chosen is to reduce the overhead of cherry-picking commits from
>> master to release branch. And also if master is not locked then the
>> scenario which Arina mentioned can still happen if a required commit for
>> 1.16 is between 2 commits intended for 1.17 only.
>>
>> For now *** Please treat master branch as locked and don't merge any
>> commit until further notice ***
>>
>
>
> Yes. I get it. I understand the lock motivation. That means that master is
> effectively a 1.16-RC branch. Somebody who needs to commit changes to 1.17
> will (should) create a 1.17 branch from which they will eventually
> cherry-pick commits back to master. At the very least, they will have a
> private copy of master that is effectively a separate branch that they will
> have to rebase as changes go on to master to get the release in shape.
>
> This means that there *will* be at least two branches. Probably more than
> two since there is no formal place to put the 1.17 changes that are pending.
>
> As such, I would suggest that making this situation explicit and public
> would be easier for people. It would help people to either create a 1.16
> branch, committing fixes there for RC problems and cherry-picking to or
> from master as needed OR create a 1.17 branch and use master as the 1.16
> branch (which is a bit confusing because it changes peoples normal
> procedures). Neither strategy requires that master be locked. The former
> leaves master as the place all new stuff goes. The latter follows the "all
> development on a branch" orthodoxy.
>
>
>
>
>
>

Reply via email to