Autogenerated suffixes can clash

[knuesel]

I was bit by this issue in a case that looked like this:

julia> df = DataFrame(x=1:3);

julia> select(df, :x => (x->2x), :x => (x->3x))
ERROR: ArgumentError: duplicate output column name: :x_function

I guess this works as documented:

the generated name is created by concatenating source column name and function name by default (see examples below).

But it would be nice if name auto-generation was smarter, to guarantee unique names.

Another case that is maybe worse as it overwrites data:

julia> f(x) = 2x;

julia> df = DataFrame(x=1:3, x_f=0)
3×2 DataFrame
 Row │ x      x_f   
     │ Int64  Int64 
─────┼──────────────
   1 │     1      0
   2 │     2      0
   3 │     3      0

julia> transform!(df, :x => f)
3×2 DataFrame
 Row │ x      x_f   
     │ Int64  Int64 
─────┼──────────────
   1 │     1      2
   2 │     2      4
   3 │     3      6

Finally, a case that is a bit contrived:

julia> g(x) = 3x;

julia> f_g(x) = 1;

julia> df = DataFrame(x=1:3, x_f=0);

julia> transform(df, :x => f_g, :x_f => g)
ERROR: ArgumentError: duplicate output column name: :x_f_g

[bkamins]

All this works as expected. Still we may re-consider adding makeunique kwarg to these functions that is why I keep it open.

The general idea is that it is safer in production code to throw error than silently modify generated (or passed by the user explicitly) column name.