Changing the API to use reference classes?

[nsheff]

Given that the BEDbase() object is a client that connects to a server, it seems to me that this is an appropriate use of a Reference Class, because we don't really need duplicates of this object; we just need 1 client and then whatever functions use it can just use that.

It also makes more sense to me for the user to do something like this:

bb <- BEDbase()


bb$list_beds()

Rather than the current approach of passing the api in to the functions, like bb_list_beds(api).

This has the advantage that I can type bb$ ... and then tab complete and get REPL feedback as to what methods are available.

Maybe this is less idomatic for R, but I find this approach more intuitive and I think reference classes made this approach possible. Would love to hear your thoughts.

So I think the 2 advantages of using reference classes would be:

• improved efficiency, since it wouldn't then require duplicating this client to use it?
• cleaner user interface (in my opinion)

but I'm not super confident in all these things, so it's just an idea for consideration/discussion.

[jwokaty]

I see your points and I headed down this path initially when I was writing bedbaser before it quickly became complicated for a handful of convenience functions I created in the package. I'll give it more thought but I am reluctant to change it close to the release when it's also in review, even though it would mean deprecating things later.