R: Completions: Adding parentheses inside of debug() and undebug()

[juliasilge]

In usability testing with @simonpcouch today, we noticed that using autocomplete inside of debug() or undebug() gets you parentheses that keep it from working. For example, you want:

debug(dplyr::filter)

but what you get with autocomplete is:

debug(dplyr::filter())

which then errors during debugging.

We've done related work in #1563 but we do know we are inside of debug()/undebug() in this situation and could maybe not add the parentheses?

[DavisVaughan]

I think @lionel- wants to eventually tackle this in a generic way with his declare() declaration idea. It would allow debug() (and any other function) to specify that the first argument is expected to be a function. We'd process that information and in the completion engine we'd understand that we were inside a function where this was the case and we'd not add () to functions when that declaration was active

That's a while away though

[DavisVaughan]

In the meantime, RStudio does have a set of base functions where it doesn't add () for, which we can copy
https://github.com/rstudio/rstudio/blob/088dfc97d473d4e3fb1e356d7fc3092a6fb35cfc/src/cpp/session/modules/SessionRCompletions.R#L2923-L2929

[DavisVaughan]

This heuristic based on the formals is also interesting (i.e. is the argument name f, fn, fun, or func?)
https://github.com/rstudio/rstudio/blob/088dfc97d473d4e3fb1e356d7fc3092a6fb35cfc/src/cpp/session/modules/SessionRCompletions.R#L2935-L2942

[jennybc]

I have thought about opening an issue about this as well. lookup::lookup() is another context where you often have to fight to not get parentheses. But debug() is the main place I encounter this.

[jennybc]

I also wish the automatic parentheses were not added when completing a function name after ?. AFAICT they are harmless and yet they still offend me. Below, I submit ?toupper() but I would prefer ?toupper.

Screen.Recording.2024-05-10.at.3.09.56.PM.mov

[jennybc]

Update: I now know the issue of adding parentheses after ? has its own issue: #1818

[jaredlander]

I know this isn't exactly the right issue, but it's related. Can we just turn off parenthesis after autocompletion altogether? I have turned off this feature in RStudio for years and never want the parentheses added. It seems like other VS Code language extensions have options to turn off that feature and I would prefer to disable it here too. Heck, I'd even turn off matching closing parentheses/quotes/brackets too. I know not everyone likes that, but it would be nice to have the option.

[lionel-]

@jaredlander I think that makes sense, I've opened an issue at #5377

To disable delimiter auto-closing while typing you can set:

"editor.autoClosingBrackets": "never"

[jaredlander]

I figured out the auto closing thing shortly after posting and even how to do it just for R. Thanks for opening #5377.

[jennybc]

QA Notes

In the R Console:

• Apply the str() or args() function to a function, while accepting a completion for that inner function. Let's use the new.env() function as an example. Start typing args(new.) and allow the completions to pop-up. Accept the completion for new.env. There should be no trailing parenthese. You should complete to args(new.env), which works and reveals the arguments, not to args(new.env()), which just returns NULL.

# GOOD, what you should see
args(new.env)
#> function (hash = TRUE, parent = parent.frame(), size = 29L) 
#> NULL

# BAD, what you should NOT see
args(new.env())
#> NULL

• This feature currently works off a fixed list of special functions: args(), debug(), debugonce(), help(), trace(), str(), so feel free to experiment with any of these.

^{Created on 2025-02-10 with reprex v2.1.1}

[testlabauto]

Verified Fixed

Positron Version(s) : 2025.03.0-11
OS Version          : OSX

Test scenario(s)