better disallowmissing error message

[bkamins]

Improve the error message in case column conversion fails, e.g.:

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

julia> disallowmissing(df)
ERROR: ArgumentError: Missing value found in column number 2 

julia> disallowmissing!(df)
ERROR: ArgumentError: Missing value found in column number 2

[bkamins]

doctests failure is unrelated. I fix it in #2967.

[jtrakk]

Thanks.

I don't use column numbers much so a name would be more familiar to me.

Maybe reporting a row too would be useful? I'm not sure.

[bkamins]

Fixed. Is it better now?

[jtrakk]

Cheers.

Closes #2965

[nalimilan]

Instead of this, we could just do error || continue in the if row !== nothing branch above.

[bkamins]

We could not do continue because we still need to execute line 2088, so instead we would need to write inside if row !== nothing:

if error
    col = _names(df)[i]
    throw(ArgumentError("Missing value found in column :$col in row $row"))
else
    y = x
end

Do you think it would be simpler to understand by the potential readers of the code?
I used this explicit logic because when I read the old condition:

if !error && Missing <: eltype(x) && any(ismissing, x)

I had to spend some time to make sure it was OK (although I think I have written it some time ago).

Therefore I would vote for leaving things as they are as it is I think easier to understand if some user wants to inspect the code.

[nalimilan]

OK. But I'd use try... catch when !error too.

[bkamins]

I have changed disallowmissing and disallowmissing! the way that I think you wanted.

[nalimilan]

OK. But I'd use try... catch when !error too.

[bkamins]

Thank you!