Column completion in the 'i' of a data.table command

[matsburg]

The regex that's used to detect whether the user is typing inside the brackets of a df only matches once a comma has been typed i.e. 'df[,' matches and completions are shown, but 'df[' doesn't and no completions are shown.

For users of the data.table package, completions before the first comma would be really helpful since column names can be used directly for filtering there. This is known as the 'i' in the data.table DT[i, j, by] syntax.

I suggest either changing the regex so that completions are provided after the '[' is typed, or exposing this as an option. I'm happy to create a PR for this if you would like. Thanks for the great plug-in.

[gaalcaras]

Thanks for the feedback. I'm not a regular data.table user so it's good to hear from that point of view.

I'm happy to create a PR for this if you would like.

Please do! This feature shouldn't be on by default since it only works with a specific package − unless I'm mistaken and you can do something useful with column names before the comma in base R? So our first step should be to expose a g:ncmr_dt option.

Let's start with that, and then we can maybe think about doing fancier stuff, like enabling this behavior automatically when the data.table package is loaded. But we can discuss all of that in the PR comments :)

[matsburg]

I've implemented this locally now, but had another question:

Currently, when something like iris[, is typed, a completion is triggered by the ',\s' completion pattern, and the only matches at this point are the columns in iris, hence these are shown.

However, if iris is a data.table, then it would also be valid to type something like iris[, median(Sepal.Width)]. But a new completion is not triggered when e.g. iris[, me has been typed and no completions are shown. If I remove the ',\s' completion pattern then it works as expected, but this seems like too big of a change. Can you think of another way of doing it?

BTW, I think it should be fairly easy to check whether e.g. iris is a data.table and then change the behaviour based on that, but I haven't tried yet.