Unable to convert Julia Array{Tuple,1} to R object with @rput

[gtollefson]

When I try to convert an Array{Tuple,1} using @rput I get the following error:

ERROR: MethodError: no method matching sexp(::Tuple{Int64,Int64})

Is this a bug or intended functionality. How can I convert this julia object into R?

[randy3k]

There is no tuple correspondence defined now. You will need to convert each tuple into an array first.

[simonbyrne]

we could convert tuples to an R list?

[Non-Contradiction]

There is one problem in converting tuples to plain R lists because plain R lists get converted to Array{Any} in Julia, so converting back and forth will change the type of Julia variable. The way to handle this in JuliaCall is to convert tuples into R list with S3 class JuliaTuple. Before loading the R library JuliaCall:

julia> using RCall

julia> RObject((2,3))
ERROR: MethodError: no method matching sexpclass(::Tuple{Int64,Int64})
Closest candidates are:
  sexpclass(::Nothing) at /Users/lichangcheng/.julia/packages/RCall/RPlFw/src/convert/default.jl:218
  sexpclass(::Missing) at /Users/lichangcheng/.julia/packages/RCall/RPlFw/src/convert/default.jl:219
  sexpclass(::Symbol) at /Users/lichangcheng/.julia/packages/RCall/RPlFw/src/convert/default.jl:222
  ...
Stacktrace:
 [1] sexp(::Tuple{Int64,Int64}) at /Users/lichangcheng/.julia/packages/RCall/RPlFw/src/convert/default.jl:214
 [2] RObject(::Tuple{Int64,Int64}) at /Users/lichangcheng/.julia/packages/RCall/RPlFw/src/convert/default.jl:209
 [3] top-level scope at none:0

Loading JuliaCall:

julia> R"library(JuliaCall)"
RObject{StrSxp}
[1] "JuliaCall" "stats"     "graphics"  "grDevices" "utils"     "datasets" 
[7] "methods"   "base"     


julia> R"julia_setup()"
┌ Warning: RCall.jl: Julia version 1.0.2 at location /Applications/Julia-1.0.app/Contents/Resources/julia/bin will be used.
│ Loading setup script for JuliaCall...
└ @ RCall ~/.julia/packages/RCall/RPlFw/src/io.jl:110
┌ Warning: RCall.jl: Finish loading setup script for JuliaCall.
└ @ RCall ~/.julia/packages/RCall/RPlFw/src/io.jl:110
RObject{EnvSxp}
<environment: 0x7fc87d6f72e0>

After loading JuliaCall:

julia> RObject((2,3))
RObject{VecSxp}
[[1]]
[1] 2

[[2]]
[1] 3

attr(,"class")
[1] "JuliaTuple"

And Julia objects with types like Array{Tuple, 1} are also handled like this:

julia> RObject([(1,2), (3,4)])
RObject{EnvSxp}
Julia Object of type Array{Tuple{Int64,Int64},1}.
Tuple{Int64,Int64}[(1, 2), (3, 4)]

Basic operations on the converted object are also possible in R:

julia> a = [(1,2), (3,4)]
2-element Array{Tuple{Int64,Int64},1}:
 (1, 2)
 (3, 4)

julia> @rput a
2-element Array{Tuple{Int64,Int64},1}:
 (1, 2)
 (3, 4)

julia> R"a[1]"
RObject{VecSxp}
[[1]]
[1] 1

[[2]]
[1] 2

attr(,"class")
[1] "JuliaTuple"


julia> R"a[1][1]"
RObject{VecSxp}
[[1]]
[1] 1



julia> R"a[1][[1]]"
RObject{IntSxp}
[1] 1

[simonbyrne]

Lots of things don't round trip, so that isn't exactly a requirement.

I do like the idea of using S3 classes, but would need some testing.

[Non-Contradiction]

Yes, I think it is only a desirable feature, especially considering tuples are quite common in Julia, it is helpful to have an R class converting to tuples in Julia and to have tuples in Julia converting to it.

[randy3k]

FYI, reticulate converts R list to python tuple and vice versa.

Do we want to follow their standard https://rstudio.github.io/reticulate/#type-conversions

[gtollefson]

Until this is developed @Non-Contradiction, I will follow @randy3k's suggestion. Because I'd like to convert a large set of tuples to R objects, I'll split the tuples into two vectors, convert to R objects, and convert to tuples within R instead. Thanks for looking into this!