What is the status of "CREATE FUNCTION"?

[qrilka]

I was doing some basic calculations using datafusion-cli and found that I repeat a couple of function invocations there. So I thought that it could be cool to have that in a function so I tried and saw this:

> create function foo();
Invalid or Unsupported Configuration: Function factory has not been configured``

Searching the documentation of Datafusion gave me no good results so I wonder what does this error mean (maybe it could be improved?) and if it's even planned to support UDFs defined from CLI and not in custom Rust code (which would need some extra work to plug it into datafusion-cli)

[Omega359]

The core framework to support it exists however it is otherwise unimplemented to the best of my knowledge. See

• https://github.com/apache/datafusion/blob/main/datafusion/core/tests/user_defined/user_defined_scalar_functions.rs
• https://github.com/apache/datafusion/blob/main/datafusion-examples/examples/function_factory.rs

[qrilka]

Thanks @Omega359 , but all of that seem to assume some custom Rust code, maybe datafusion-cli could report some more meaningful error like "custom functions are not supported in datafusion-cli"?

[alamb]

Did anyone make an issue / ticket to improve the error message? I agree that the current message is pretty obtuse

[qrilka]

@alamb I don't know about anything like that, the original question was about finding out the current situation/plans around UDFs i.e. about finding out whether such ticket is a valid one

[alamb]

Got it it -- thank you. If the error message is clear about the status then I agree nothing else is needed.

I was imagining we could help future users avoid confusion with a clearer error message

[qrilka]

I wouldn't say that the message is clear - if it were then this thread won't be needed. And I still don't understand the state of UDFs in datafusion-cli - I see that UDFs could be added by adding custom code but from a point of view of datafusion-cli user that doesn't look like an easy option: as I see it ATM to add a couple of simple function I'd need to create a custom fork of datafusion-cli with a small portion of extra Rust code.

[alamb]

I see that UDFs could be added by adding custom code but from a point of view of datafusion-cli user that doesn't look like an easy option: as I see it ATM to add a couple of simple function I'd need to create a custom fork of datafusion-cli with a small portion of extra Rust code.

Yes I agree -- user defined functions are not supported by datafusion-cli (or anything in the core crate either)