so I've been trying to think of the best way to solve the implicit coersion of sql. in python, we usually want things to be pretty explicit (explicit coercion), but in SQL, its generally accepted that you can use values as long as they can be cast to the datatype. ex: in python

date_df = daft.from_pydict({"date_str":["2020-01-01","2020-01-02","2020-01-03"]})
date_df = date_df.with_column("date", col("date_str").str.to_date("%Y-%m-%d"))

# we probably want the user to use an explicit date object, or a casted string
date_df.filter(col("date") == datetime.date(2020, 1, 1)).explain()
date_df.filter(col("date") == daft.lit("2020-01-01").str.to_date("%Y-%m-%d")).explain()

but in sql, implicit coercions are almost always allowed.

where date = '2020-01-01'; -- implicit
where date = cast('2020-01-01' as date); --explicit

Was hoping others might have some ideas on how we can solve this.

Yes, the planning config is applied when constructing the plan/optimizations etc The execution config is passed into each task when stuff is running

I think we do allow some implicit coercions elsewhere, such as i64 == f64. So should we just change the logic to allow

str == date

comparisons as well?

I think this is reasonable. Are there any examples of implicit conversions we want to do in SQL but would want to disallow in Dataframe?

I opened up a separate issue to disable all implicit conversions https://github.com/Eventual-Inc/Daft/issues/3338