try_divide function does not accept floating point numbers only integers

[canimus]

Function try_divide does not accept floating point numbers:

spark.sql("select try_divide(10.0, 2.2)").show()

ERROR:

AnalysisException: Execution error: Function 'try_divide' user-defined coercion failed with: Execution error: Unsupported Data Type: Spark `try_divide` function expects Int32, Int64 o Interval(YearMonth) / Int32, got Decimal128(3, 1), Decimal128(2, 1). No function matches the given name and argument types 'try_divide(Decimal128(3, 1), Decimal128(2, 1))'. You might need to add explicit type casts.
        Candidate functions:
        try_divide(UserDefined)

[shehabgamin]

Thanks for reporting, should only be a few lines of code to address.

We can add support for this in the next release, which is planned for Tuesday.

[linhr]

I tried a quick fix in Codex, but then I realized that this may be a much bigger task to match Spark behavior properly. For example, for this particular case, the inputs are of the decimal type, and the result should be decimal. A naive fix would cast the input to double which is wrong.

We need to audit the logic for spark_{plus,minus,multiply,divide,modulo} and SparkTry{Add,Subtract,Mult,Div,Mod}. There are various TODOs in the code and it seems the types and type coercion logic could be more rigorous to properly support all input type combinations to fully match Spark behavior.

It would also be interesting to see if Comet has put efforts in this topic and whether we can get some inspiration.