Improve data and pdata macros about dataarrays.jl HOT 7 CLOSED

There's also this:

julia> @data [NA]
ERROR: no method convert(Type{None},ASCIIString)
 in setindex! at array.jl:412
 in parsedata at /home/milan/.julia/DataArrays/src/literals.jl:13
 in literaldata at /home/milan/.julia/DataArrays/src/literals.jl:44

But:

julia> @data [NA, 1]
2-element DataArray{Int64,1}:
  NA
 1  

from dataarrays.jl.

This is an issue with every constructor literal we've had: NA doesn't give you enough type information to do anything sensible. We should fix it to produce DataArray{NAtype}, but that's still not worth having at the end of the day.

from dataarrays.jl.

A DataArray{NAtype} vector isn't very useful by itself, but sometimes you need it e.g. to represent a missing variable of the desired length, and when concatenated with another vector it will be promoted to a meaningful type.

from dataarrays.jl.

Ok. We should see if there's a sane way to fix this.

Another bug: @data [pi] sees pi as a symbol.

from dataarrays.jl.

I just pushed a big fix for this with bb2d5ef.

In this fix, I throw an error in response to @data [NA] to indicate a failure of type inference. We can hardcode this edge case, but I'm not sure that would be helpful since the rest of DataArrays is built around the assumption that a DataArray{NAtype} never occurs in valid code.

from dataarrays.jl.

Unfortunately my fix still leaves lots of stupid edge cases of its own:

julia> @data [sin(4.0), cos(7.4)]
2x2 DataArray{Any,2}:
 sin  4.0
 cos  7.4

from dataarrays.jl.

Closed by f10d15f

from dataarrays.jl.