Add tab completion to R6 objects

[thomasp85]

This PR adds automatic tab completion of R6 derived objects as a convenience for users and developers. Not super important but definitely a nice-to-have :-)

[wch]

Thanks, this will be useful. Unfortunately, with these modifications, R6 doesn't pass R CMD check.

I think to fix it:

• utils needs to be listed as an import in DESCRIPTION.
• There needs to be a #' @importFrom utils .DollarNames line so that the generic is visible to the package.
• The ::: operator can't be used. (Well, technically, this isn't a show-stopper since it raises a NOTE and not a WARNING or ERROR, but with NOTEs I have to convince CRAN that it's OK, and that's an unpleasant and unsure process.)

[wch]

@kevinushey do you have thoughts about replacing the utils:::findMatches() function with something else, perhaps something that RStudio is using?

[thomasp85]

Sorry for the half-baked PR regarding the imports - I'll fix that. Regarding the note I think there is a strong case for it as it also how it is implemented in utils:

> utils:::.DollarNames.list
function (x, pattern = "") 
{
    findMatches(pattern, names(x))
}
<bytecode: 0x1105207c0>
<environment: namespace:utils>

[wch]

I don't think that the CRAN maintainers will be convinced that, just because a function in utils calls an internal function, that a function from another package should be allowed to call that function. They're generally pretty strict about those things.

[thomasp85]

You're probably more knowledgable on the sentiment of the CRAN maintainers than me, so I'll take your word on it.

[kevinushey]

In RStudio, all we use for R6 classes and objects (as they are just environments under the hood) is:

completions <- ls(object, all.names = TRUE)

Then, for (fuzzy) matching, we use:

.rs.startsWith(tolower(completions), tolower(token))

where .rs.startsWith is implemented as:

.rs.startsWith <- function (strings, string) 
{
    if (!length(string)) 
        string <- ""
    n <- nchar(string)
    (nchar(strings) >= n) & (substring(strings, 1, n) == string)
}

It seems like it would be easy enough to adapt this to the .DollarNames interface, although you might need to do some voodoo to construct the right regular expression.

[thomasp85]

Looking at the internals of findMatches, as well as findFuzzyMatches and findExactMatches which drives it under the hood, there is nothing preventing a copy-paste of the code. This sets a sad precedence for code reuse though... I'll try to hear why it is not exported in the first place and get back, before we have to reinvent the wheel in every package that requires a new tab completion scheme...

[wch]

There is the issue of licensing... R6 is MIT, while R itself is GPL 2 or 3. I don't think it's possible to simply cut and paste the code while respecting the R license.

[kevinushey]

The internals of findExactMatches is literally just a call to grep so just use that.

[thomasp85]

@kevinushey Strange - I remember trying to use grep on refClasses before I knew about findMatches, and it didn't play well with RStudio - half-finished methods would get erased when additional characters were typed... Now it seems fine though..?

I'll revert to using grep and save the trouble of asking Core

[wch]

This no longer is necessary, as tab completion appears to just work now. See https://github.com/wch/R6/issues/43#issuecomment-114224019