Interactive, reproducible figures inside teal

A teal.widgets PR adds a render path for interactive htmlwidgets, so plot_with_settings can return a chart you can hover, select on, and download as self-contained HTML — opt-in, and static plots stay exactly as they were.

If you build Shiny apps with teal, you know plot_with_settings — the module that wraps a plot with resize, full-screen, and download controls. Until now it rendered only static R graphics: ggplot, lattice, grob, base. Return anything else from plot_r and it quietly fell through and never drew.

That leaves a gap. The analyst looking at a volcano plot, a Kaplan–Meier curve, or a forest plot often wants to hover a point, select a group, and take the exact figure with them — not screenshot a PNG. In a submission-grade workflow they also want that figure to be reproducible: the chart and the code that made it, together.

A small change with a big surface

A PR on teal.widgets adds a render path for interactive htmlwidgets. If plot_r() returns an htmlwidget — canvasXpress, plotly, anything with a Shiny binding — plot_with_settings renders it through the widget’s own binding, keeping the familiar card chrome. Height, width, and full-screen work as before, and download produces a self-contained HTML file you can email or archive.

It is strictly opt-in. Every existing static-plot caller behaves exactly as it did; the interactive path triggers only when a module author chooses to return a widget.

library(shiny)
library(canvasXpress)
library(teal.widgets)

pg <- as.data.frame(na.omit(palmerpenguins::penguins))

ui <- bslib::page_fluid(plot_with_settings_ui("p"))

server <- function(input, output, session) {
  plot_with_settings_srv(
    "p",
    plot_r = reactive(canvasXpress(
      data = pg, graphType = "Scatter2D",
      xAxis = list("flipper_length_mm"), yAxis = list("body_mass_g"),
      colorBy = "species"
    ))
  )
}
shinyApp(ui, server)

The chart that lands in the card is a real canvasXpress widget — the same one you can render on its own here. Hover a point, click to select, use the toolbar to export:

library(canvasXpress)

pg <- as.data.frame(na.omit(palmerpenguins::penguins))

canvasXpress(
  data      = pg,
  graphType = "Scatter2D",
  xAxis     = list("flipper_length_mm"),
  yAxis     = list("body_mass_g"),
  colorBy   = "species",
  title     = "Palmer penguins — flipper length vs body mass"
)

Inside plot_with_settings, that interactivity sits in the usual card, with the resize and download controls you already use:

Three canvasXpress plots inside plot_with_settings — select an element, open the customizer, download as HTML.

Describe the chart instead of configuring it

Because the render path is generic, you inherit whatever the widget can do. canvasXpress, for instance, can build a chart from a plain-English prompt — no graphType, no axes:

plot_with_settings_srv(
  "p",
  plot_r = reactive(canvasXpress(
    data   = pg,
    prompt = "Create a scatter plot of bill_depth_mm vs body_mass_g and color by species"
  ))
)

Try it live

There’s a running demo — three canvasXpress plots inside plot_with_settings cards:

canvasxpress.shinyapps.io/teal-canvasxpress-demo

or install the PR build and run it yourself:

pak::pak("pharmaverse/teal.widgets#371")

Why it matters

Interactive dashboards usually mean leaving the tools you trust for a BI platform, and losing reproducibility in the trade. This keeps both: the module model and reproducible, code-defined figures of teal, plus the interactivity and portability of an htmlwidget. The figure is still code — it just moves now.

The discussion that started it is teal#1757; the PR is teal.widgets#371. Feedback from teal users is welcome, especially on how you’d want widget support wired into the module API.