feat(snowflake)!: Annotate type for snowflake REGR_SLOPE function

[fivetran-kwoodbeck]

Annotate type for snowflake REGR_SLOPE function.

Documentation:
https://docs.snowflake.com/en/sql-reference/functions/regr_slope

[geooo109]

Did you also check if this function exists in other sql dialects ? (from the docs online)

[fivetran-kwoodbeck]

Did you also check if this function exists in other sql dialects ? (from the docs online)

I assume that by "exists in other sql dialects", you are asking whether it's present elsewhere in SQLGlot - no, it's not. The function itself is supported by some other databases.

[geooo109]

Did you also check if this function exists in other sql dialects ? (from the docs online)

I assume that by "exists in other sql dialects", you are asking whether it's present elsewhere in SQLGlot - no, it's not. The function itself is supported by some other databases.

Yeap, I was referring to other databases, not SQLGlot codebase. So, when we introduce a new function expression, the goal is to have a common representation (1 AST node) for all the dialects + add parsing tests (validate_identity or validate_all) for all the dialects that support the targeted function. For example, you can imagine that a function with the same name can exist in two different dialects and support different number of arguments, in this case our AST node must handle both of these cases.

I did a quick look up and found that spark, databricks, presto, trino, oracle, postgres, and duckdb (you can re-check it also by yourself I maybe missed something) support this function and for all of these your representation is sufficient. So, we can include a validate_identity test like self.validate_identity("REGR_SLOPE(y, x)") in each of their test files (test_{dialect_name}.py), or use the test_dialect.py file and do something like this PR (gather all the tests in one file): https://github.com/tobymao/sqlglot/pull/5576/files

[geooo109]

This function returns: If any of the input expressions is of type DECFLOAT, the returned type is DECFLOAT. Otherwise, the returned type is FLOAT.

Currently, SQLGlot doesn't support DECFLOAT , should we add a new type here or use a mapping? @VaggelisD

[VaggelisD]

We should add support for DECFLOAT, it's different than FLOAT and has variable scale so we probably can't make it a mapping.

[geooo109]

yeap agree thanks.

@fivetran-kwoodbeck so let's add a new datatype called DECFLOAT, then we have to make a custom annotation function that first annotates the args of RegrSlope, and based on the types we will output the correct type.

[fivetran-kwoodbeck]

Did you also check if this function exists in other sql dialects ? (from the docs online)

I assume that by "exists in other sql dialects", you are asking whether it's present elsewhere in SQLGlot - no, it's not. The function itself is supported by some other databases.

Yeap, I was referring to other databases, not SQLGlot codebase. So, when we introduce a new function expression, the goal is to have a common representation (1 AST node) for all the dialects + add parsing tests (validate_identity or validate_all) for all the dialects that support the targeted function. For example, you can imagine that a function with the same name can exist in two different dialects and support different number of arguments, in this case our AST node must handle both of these cases.

I did a quick look up and found that spark, databricks, presto, trino, oracle, postgres, and duckdb (you can re-check it also by yourself I maybe missed something) support this function and for all of these your representation is sufficient. So, we can include a validate_identity test like self.validate_identity("REGR_SLOPE(y, x)") in each of their test files (test_{dialect_name}.py), or use the test_dialect.py file and do something like this PR (gather all the tests in one file): https://github.com/tobymao/sqlglot/pull/5576/files

Sounds good. I double checked and also found a few extra, they were all added to the test_dialect.py same as the example you shared.