The names of our UDFs are sometimes Get<Something>UDF and sometimes just <Something>UDF, we should find a standard way to do that.
The UDFs sometimes test for null values being passed to them, and sometimes not. It seems there's some confusion about how to actually do this properly.
The UDFs sometimes use the Generic UDF Helper class to check arguments on initialize, and sometimes they duplicate that logic. We should clean this up.
| Subject | Repo | Branch | Lines +/- | |
|---|---|---|---|---|
| Standardized UDF naming | analytics/refinery/source | master | +1 K -881 |
Null treatment might be different per UDFs as nulls in hive depend on the columns they operate on .
Changes need to be backwards compatible so if we rename classes we need to have a stub with the old name.
We have about 20 UDFs:
https://github.com/wikimedia/analytics-refinery-source/tree/master/refinery-hive/src/main/java/org/wikimedia/analytics/refinery/hive
I'm quite happy with the current refactor, pending successful smoke testing within Hive (which I'm doing now):
https://gerrit.wikimedia.org/r/#/c/327237/