major refactoring of conversion related code

[randy3k]

• switch to DataArrays instead of NullableArrays (close DataFrames and DataTables support #174)
• support DataTable
  DataTables doesn't play well with DataFrames Overwrites DataFrames describe function JuliaData/DataTables.jl#33
  Though DataTables support is ready to merge, it was removed from the current PR at 50b1e80
  and will be merged in the future.
• support AxisArrays (close Use NamedArrays.jl to preserve array names #144)
• support Data and time (close Convert Date and DateTime to/from R #163)
• issue subsetting a StrSxp Issues related to subsetting a VectorListSxp, particularly, a StrSxp #86 (not sure what the best practice is, won't fix now)
• sanitize names in conversions (close Sanitize names when copying a data.frame? #112)
• fix Stack Overflow at .../convert-base.jl:56 #176

towards #138

[nalimilan]

support DataTable (with default being DataArrays)

If support for DataTable is added, it should use NullableArrays instead.

[randy3k]

I meant support DataFrame(R"mtcars") and DataTable(R"mtcars"). Of course, the columns of a DataFrame are DataArrays and the columns of a DataTable are NullableArrays.

[nalimilan]

Then I don't understand why you wrote that. :-)

[randy3k]

because I made a stupid typo...

[randy3k]

Will merge soon if there are no other comments.

cc: @simonbyrne @dmbates

[nalimilan]

So you've decided to use AxisArrays rather than NamedArrays? I'm still unsure what's the best approach (e.g. for my package FreqTables). The generality of AxisArray (numeric and categorical axes) is appealing, but the goal of the package does not seem to completely match the R approach to array names. In particular, the printing of names before the data (rather than on each line as in R/NamedArrays) isn't practical for categorical axes.

@mbauman What do you think about this? Are the goals compatible? Could the printing evolve towards what NamedArrays does? i.e. something like this:

4×4 Named Array{Int64,2}
Dim1 ╲ Dim2 │  A   B   C   D
────────────┼───────────────
a           │ 30  20  30  20
b           │ 30  20  30  20
c           │ 20  30  20  30
d           │ 20  30  20  30

[mbauman]

I'd absolutely like to see better printing of AxisArrays. It's still a young project in terms of the number of hours that have been dedicated to it.

I'm not familiar with any homogenous named array in R, so I'm not entirely sure about the use-case, but the goal of AxisArrays is to provide core support for many disparate domains. Then secondary packages can layer on more domain-specific behaviors.

[nalimilan]

R named arrays don't have anything really specific. Names are always strings, indexing by name is supported (but only by putting the index in the position corresponding to the dimension). Names are propagated by some operations like element-wise ops and reductions (which isn't supported by AxisArrays yet IIRC). The way they are printed is similar to the NamedArray example above. I was afraid the way AxisArrays are printed was a design decision, but if you would be OK with something more like NamedArray it would be great.

(Then I'm not a fan of a name, I wouldn't have considered dimension names as axes, but I guess that's a minor point. :-)

[randy3k]

Since the conversion to AxisArrays has to be explicit, i.e. rcopy(AxisArray, foo). It is equally possible to support NamedArrays or AxisArrays or both.