let indicator allow strings

[pdeffebach]

Fixed #2283

[bkamins]

Apart from the small comments there is a small problem.
You have left out a test:

Symbol(indicator_cols[i]) == indicator

which will always fail is indicator is a string.

However, now as I look at it it actually seems it can be removed if we move line:

select!(joined, Not([df1_ind, df2_ind]))

before line:

unique_indicator = indicator

(if it is not clear for you why then I can fix this after this PR is merged)

[pdeffebach]

Thanks for the review.

Indeed I cannot figure out this error you described. I can't figure out a way to write code that throws an error about comparing a symbol to a string. I think you may have to fix it after the merge.

[bkamins]

OK - I can clean up the code later.

For now can you please just add after line https://github.com/JuliaData/DataFrames.jl/pull/2284/files#diff-3d9293bcc4eac4333160c4d203ae91d7R244 (i.e. as a first statement executed if indicator !== nothing is true)
the following line:

indicator = Symbol(indicator)

it will fix the bug (not in the cleanest way, but will do for now) as then we are sure that in Symbol(indicator_cols[i]) == indicator) we have symbols both on left and right hand sides.

I cannot fix it via web interface unfortunately.

[pdeffebach]

Added! let me know what else to add

[bkamins]

Looks good. Thank you!

[bkamins]

Thank you!