8 Gander: R package for code completion

If you’re an RStudio user and don’t want to leave it for VS Code, Gander is the closest thing to Copilot’s inline completion you’ll get. Posit’s own announcement is worth a skim too: Introducing Gander.

Gander’s trick is that it can see inside your R session. Where Copilot is guessing your variable names from text in open scripts, Gander can look at the actual objects in your environment — so it already knows what columns are in dat without you telling it.

If you haven’t already, follow the Gander set-up instructions in Setup before continuing.

Once it’s working, try this: load the benthic data, select a blank line, hit your Gander shortcut, and type a prompt like:

Make a ggplot of pres.topa against CB_cover, coloured by logged

Because Gander can see dat directly, it should get the variable names and their types right on the first try — a nice contrast with the tab-completion experience from the previous modules, where you often had to spell things out yourself.

Try a few more prompts, selecting existing code first and asking Gander to modify it:

Add a stat_smooth to this plot

Convert this ggplot to use theme_classic()

Because Gander can see your R objects, it uses more tokens per request than plain code completion. The Gander documentation has options for controlling how much of your environment gets sent, which is worth reading if you’re watching your API costs.

Challenge

Select the dat %>% group_by(logged) %>% summarize(...) code from the previous module and ask Gander to add a third summary column for secchi. Compare how confidently it picks the right column compared to when you did the same task with comment-driven completion.