How to deal with tinyints?

[aersam]

Tinyint's are a bit annoying. In spark there is tinyint, and so is in MS SQL Server. But spark's Data Type is signed, while t-sql's is not. DuckDB has UTinyInt and TINYINT which is great.

Spark turn's SQL Server's into SMALLINT, since it's own data type does not cover values > 127. So should sqlglot do this as well?

print(
    sqlglot.parse_one(
        "SELECT cast(255 as tinyint)",
        dialect="tsql",
    ).sql("spark")
) # now , smallint or tinyint? Currently is tinyint

[georgesittas]

So the issue is that Spark's TINYINT is a 1-byte signed integer whereas T-SQL's TINYINT is unsigned? If that's the case then we could change the parsing of T-SQL's TINYINT so that it produces a data type node with UTINYINT as its values, and then generate TINYINT back when targeting T-SQL for UTINYINT. Then in Spark we could make UTINYINT generate into SMALLINT or something.

[aersam]

Probably makes most sense. I thought there are other RDBMS which choose TINYINT to be unsigned, but looks like an MSFT-thing.

[georgesittas]

Check out #3486

[aersam]

Looks good, thanks a lot! Maybe a test for transpilation from/to duckdb which supports both tinyint and utinyint?