Make exporting method?

[dkahle]

I think a common case is for users to have pre-defined variables in R that they want to send over to M2. Is there a convenient way to do that? Something like m2_export() or m2_copy() would make sense, but the challenge would be to be careful about what gets mapped to what type/object-wise in the M2 world. It doesn't need to be perfect, just reasoned out and consistent. Some things are obvious: a double gets mapped to a double (that sounds obvious, but we'll have to double check precision to make sure that "double" means the same thing in both places). But others may not be.

Other design details will matter too: can the user provide a list/environment of variable bindings, and they all get copied?; Does it use nonstandard evaluation?; etc. (It may even be reasonable to keep an R-side environment that knows about M2 bindings.)

Thoughts?

[coneill-math]

Porting my comments from #229.

Allow the user to "connect" R and M2 variables that should maintain the same value, and m2r can update these values before/after each call to m2(). If done correctly, it would behave as if the variables share the same memory.

Some thins to note.

• This feature would add significant overhead, as the list of "active" pointers would need to be checked and updated before and after each m2() call.
• This feature would likely require modifying the protocol to allow metadata to be passed back and forth with each m2() call. This should be done anyway, to make the interface more extensible.
• This feature would hinge on the use of synchronous calls, and would likely be impossible if asynchronous calls were added.

[coneill-math]

In short I think this is possible, but we should warn people to use it with caution, as it will be very easy for things to get out of sync no matter how hard we try.