Should column names be symbols?

[hayd]

DataFrames.jl allows only symbol column names, whilst it looks like TimeSeries column names must be UTF8String.

JuliaData/DataFrames.jl#509

[quinnj]

+1, I think it would be good to mirror DataFrames here. I think they've put a lot of thought into this and it will be better to be consistent in the long run (even though we don't have an explicit dependency).

[milktrader]

I haven't evaluated the discussion yet as to why they switched, but I'm open to the idea. What are the main benefits of doing this?

[quinnj]

I think one of the main ideas was preparing for the planned feature of overloading the dot operator so you can access columns like R. df = DataFrame(...); df.col1.

[hayd]

There is also the issue with only being valid identifiers (presumably for the overloading reason). JuliaData/DataFrames.jl#520 & JuliaData/DataFrames.jl#617

[garborg]

I think parsing formula expressions is another reason: JuliaData/DataFrames.jl#618. I'm now sure what else there might be.

[milktrader]

JuliaData/DataFrames.jl#618 (comment)

[milktrader]

This feature doesn't allow spaces in column names, so that would be a loss. There would need to be some new constraint to enforce that there are no spaces in the colnames field. There are actually quite a bit of column names in time series data that does include a space.

[milktrader]

Not sure the pros outweigh the cons. Some developments in Base JuliaLang/julia#9945 (comment)

Closing for now.