Fix: generate_series function support string type

[getChan]

Which issue does this PR close?

Closes #11922.

Rationale for this change

generate_series function does not accept string arguments.
However, the as-is test provide string arguments.

I think it would be better if the type signature of generate_series was changed.
https://github.com/apache/datafusion/blob/main/datafusion/functions-nested/src/range.rs#L129

What changes are included in this PR?

Only sqllogictest

Are these changes tested?

Yes.

Are there any user-facing changes?

No.

[jayzhan211]

@getChan I think what we need is to support string type but not modifying the test to other already supported type

D create table date_table(start date, stop date, step interval);
D insert into date_table values (DATE '1992-01-01', DATE '1993-01-02', INTERVAL '1' MONTH);
D select generate_series(start, '1993-03-01', INTERVAL '1 year') from date_table;
┌────────────────────────────────────────────────────────────────────┐
│ generate_series("start", '1993-03-01', CAST('1 year' AS INTERVAL)) │
│                            timestamp[]                             │
├────────────────────────────────────────────────────────────────────┤
│ [1992-01-01 00:00:00, 1993-01-01 00:00:00]                         │
└────────────────────────────────────────────────────────────────────┘

We should able to return the same result as Duckdb's

[getChan]

In that case, I guess we should extend the above issue to "Add string type support to generate_series function".

[jayzhan211]

I expect you get the same answer as if you explicitly cast the second argument to date:

The goal is make the query return the same result as the one you explicitly cast the argument.

In that case, I guess we should extend the above issue to "Add string type support to generate_series function".

Yes, I think what we need is to support string type, or cast the string to date. Not sure what is the exact solution. But I expect we should get the correct result for that query

[alamb]

Yes, I think what we need is to support string type, or cast the string to date. Not sure what is the exact solution. But I expect we should get the correct result for that query

The type coercsion system is supposed to handle adding this kind of cast.

I wonder if we can could remove the Any(3) signature a new signature to generate_series :

datafusion/datafusion/functions-nested/src/range.rs

Lines 125 to 129 in e24a5dd

	TypeSignature::Exact(vec![Int64]),
	TypeSignature::Exact(vec![Int64, Int64]),
	TypeSignature::Exact(vec![Int64, Int64, Int64]),
	TypeSignature::Exact(vec![Date32, Date32, Interval(MonthDayNano)]),
	TypeSignature::Any(3),

I think that would then ensure DataFusion actually casts the arguments to the supported types

[alamb]

Marking as draft as I think this PR is no longer waiting on feedback. Please mark it as ready for review when it is ready for another look

[getChan]

I wonder if we can could remove the Any(3) signature a new signature to generate_series :

removed Any(3) signature. and fix some type coerce code for range, generate_series

[jayzhan211]

👍

[jayzhan211]

Thanks @getChan

[jayzhan211]

I will file an ticket for this issue

[alamb]

Thank you 🙏