Why does @byrows! contain the bang (!) at the end of its name when it is not modifying the input argument?

[hurak]

Why does @byrows! contain the bang (!) at the end of its name when it is not modifying the input argument?

An example

First, define some DataFrame df

julia> df = DataFrame(x = 1:3, y = [2, 1, 2])
3×2 DataFrame
│ Row │ x     │ y     │
│     │ Int64 │ Int64 │
├─────┼───────┼───────┤
│ 1   │ 1     │ 2     │
│ 2   │ 2     │ 1     │
│ 3   │ 3     │ 2     │

Now do some modifications using @byrows!

@byrow! df if :x > :y; :x = :y *3 end
3×2 DataFrame
│ Row │ x     │ y     │
│     │ Int64 │ Int64 │
├─────┼───────┼───────┤
│ 1   │ 1     │ 2     │
│ 2   │ 3     │ 1     │
│ 3   │ 6     │ 2     │

and yet the original DataFrame df remains intact:

julia> df
3×2 DataFrame
│ Row │ x     │ y     │
│     │ Int64 │ Int64 │
├─────┼───────┼───────┤
│ 1   │ 1     │ 2     │
│ 2   │ 2     │ 1     │
│ 3   │ 3     │ 2     │

[henriquebecker91]

Great question, I stumbled in this behavior again today. I think it is a bug, because the documentation uses as the first example of @byrow!:

@byrow! df if :A > :B; :A = :B * :C end

That makes no sense without the input argument mutation. Then, in the sequence, in the context of the special syntax for new columns, the documentation says:

Note that the returned AbstractDataFrame includes these new columns, but the original df is not affected.

As this was an exception of the behavior just for the case in which new columns are added.

[pdeffebach]

I think this is a bug as well, @bkamins do you agree?.

@henriquebecker91 or @hurak Would either mind opening a PR to deprecate this behavior?

That is, people will still be able to use @byrow! but they will get a warning and be told to use @byrow instead.

[henriquebecker91]

Done: #154

[bkamins]

Yes - @byrow is better I think.