Use an R module

[tshort]

This is a feature request to make Rif work more like @stevengj's PyCall. With PyCall, one can do things like:

using PyCall
@pyimport math
math.sin(math.pi / 4) - sin(pi / 4)  # returns 0.0

It would be great if Rif could make an R module and fill it with R function calls, so that Rif.call isn't needed.

Steven's handling of keyword arguments is also nice.

[lgautier]

Yes, this is looking nice. I followed part this on the mailing-list (just too busy with other things to be more active at the moment).

I just had a quick glance and PyCall is using macros to define wrappers (functions) on the fly (the Python-R rpy2 interface is doing something similar, except that it is possible to define "callable" classes in Python so no need for macros).

R has a number of little surprises up its sleeve: "." (dot) is a valid character in a symbol name, as well as in a parameter's name but hopefully I'll be able to port most of the workarounds I implemented in rpy2.

[diegozea]

Hi Igautier!! Do you have some hints for implement this? I was thinking on ask to R for the functions and signatures in the first import, and generate a .jl file with the wrap library.

[lgautier]

The way I'd like to do it is pretty much the way I am doing it with rpy2:

• define a namespace in the host language (here Julia) for each R package
• bind the R objects to symbols in that namespace, using a naming conversion (.->_ majoritarily).

Generating a .jl file can reduce the computer's work at import time, but will require a number of checks at import time (like checking the version of the R package). From experience with rpy2, this is not worth the trouble.

[stevengj]

PyCall doesn't really need macros either, the macros are just to clean up the syntax slightly. Instead of @pyimport foo as bar, you can equivalently do const bar = pywrap(pyimport("foo")).

[lgautier]

I see. A nice side of macros.

What it could look like in Rif is:

stats = rimport("stats")

# the original R name is var.test - a similar conversion to what is done in rpy2's `importr()` takes place 
stats.var_test(x, y)

[lgautier]

How do I get Julia REPL console to autocomplete after a "." (dot) ? I am guessing that I should define a method, but it is not clear to me which one.

A pointer at the exact place in PyCall implementating what let's it work with it would be fine.

[stevengj]

In PyCall, the pywrap function actually defines an anonymous module. So, the lookup of the fields after the . is static, not dynamic.

[lgautier]

I see. Pretty much like what is in rpy2 then.

I am almost done, I think, proven that someone can shed light on what is happening with the following bit inpywrap:

    eval(m, Expr(:toplevel, consts..., :(pymember(s) = getindex($(o), s)),
                 Expr(:export, exports...)))

[lgautier]

This is in with the latest commits, although it is in a rough form (and Rif needs a bit of work in several places).

The following is already working:

using Rif

rstats = Rif.rwrap("stats")