-1 for backwards compatibility issues described.


Corey, I'm really sorry for the churn. I thought I ran both forward and backward compatibility modes last time (-old 1.6.1 -new 1.6.2 as well as -old 1.6.2 -new 1.6.1), but I must have just eyeballed the output of the 1.6.1 -> 1.6.2 report for problems with forward compatibility.

I ran things again this time (as a formality) and the 1.6.2 -> 1.6.1 check turned up 2 issues.

1) minicluster.ServerType added enum members

Specifically TRACER and MONITOR. This changes the public API because ServerType is in it, and a client built against 1.6.2 could refer to these enum values and then get a NoSuchFieldError if they try to go back to 1.6.1. This only shows up as a low severity "other" issue in the 1.6.1 -> 1.6.2 check, which is probably why I didn't see it.

2) core.client.mapreduce.AbstractInputFormat.getConfiguration(JobContext) changed from package-private to public

This causes the method to show up as a new part of the public API. This issue only shows up in the 1.6.2 -> 1.6.1 check below.

Here are the specific report outputs for others to look:

* 1.6.0 -> 1.6.2 (added things are fine, because the change might be from 1.6.0 -> 1.6.1)
* 1.6.1 -> 1.6.2 (nothing should be added, but it's easier to just pay attention to the next one)
* 1.6.2 -> 1.6.1 (under a semver patch increment, this should be just as strong an assertion as the reverse)

On Fri, Jan 23, 2015 at 8:02 PM, Corey Nolet <cjnolet@apache.org> wrote:

    Please consider the following candidate for Apache Accumulo 1.6.2

    Branch: 1.6.2-rc2
    SHA1: 34987b4c8b4d896bbf2d26be8e70f70976614c0f

    (Append ".sha1", ".md5" or ".asc" to download the signature/hash for a given artifact.)

    Signing keys available at: https://www.apache.org/dist/accumulo/KEYS

    Testing: All unit, integration and functional tests are passing. 

    The vote will be extended as a result of the weekend and will be open until Tuesday, January 28th 12:00AM UTC (1/27 8:00PM ET, 1/27 5:00PM PT)