Hi stack you can see the new TestKeyValue case:
if the last kv of previous block and the first kv of current block have same postfix and just 1 offset diff, e.g. 100abcdefg and 101abcdefg,
before 9518, the getShortMidpointKey() will fallback to the default right kv, say 101abcdefg.
after 9518, it'll return "101", a shorter faked value, still reasonable, right?
And i found this corner case existing in current hbase test cases as well, so i'd like to let it go into community codebase also.