-
Type:
New Feature
-
Status: Open
-
Priority:
Major
-
Resolution: Unresolved
-
Affects Version/s: 1.4.0
-
Fix Version/s: None
-
Component/s: Table SQL / API
-
Labels:None
Umbrella issue for adding scalar functions such as mathematical-function LOG, date-functions DATEADD, string-functions LPAD, etc.
How to contribute a built-in scalar function
Thank you very much for contributing a built-in function. In order to make sure your contributions are in a good direction, it is recommended to read the following instructions.
- Investigate the behavior of the function that you are going to contribute in major DBMSs. This is very important since we have to understand the exact semantics and type inference of the function.
- It is recommended to add function for both SQL and Table API (Java, Scala, Python).
- For every scalar function, add corresponding docs which should include a SQL, a Java/Scala/Python version in ./docs/dev/table/functions/systemFunctions.md. Make sure your description of the function is accurate. Please do not simply copy documentation from other projects, especially if the projects are not Apache licensed.
- Take overflow, handling of NULL, other data types and exceptions into consideration.
There are different generations of function stacks in the Flink code base. The newest generation has been added in FLINK-20522 and is the recommended way of implementing a function similar to a regular user-defined function.
Adding a built-in function requires changes in only 2 classes: BuiltInFunctionDefinitions for the definition and the class that contains the actual runtime logic.
Tests should be added in org.apache.flink.table.planner.functions.
See the reference implementation of the IFNULL function here:
https://github.com/apache/flink/commit/df46f8948279d1ed89d7113919b4c4f720cf4199