DataFrame(x=Int[], y=Int)

[jariji]

Shouldn't this be DimensionMismatch?

julia> DataFrame(x=Int[], y=Int)
0×2 DataFrame
 Row │ x      y        
     │ Int64  DataType 
─────┴─────────────────

Inspired by https://discourse.julialang.org/t/column-types-in-dataframes/108532

[bkamins]

You mean that if all passed vectors have length 0 then also passing non-vector values should error?

Indeed this is what R does.

[jariji]

I guess the result is harder to understand because the constructor uses implicit broadcasting, but if I think about it in terms of normal broadcasting rules there's no real reason for this to error. So maybe it's okay.

[bkamins]

Yes - broadcasting is done here and that is why it was allowed. The issue is that:

DataFrame(x=some_vector, y=scalar)

is a useful pattern for populating the y column with a constant. The question is if some_vector is empty should we special case it and error (we do not; R does).