[neuhausi]

Summary

plot_with_settings currently renders only static R graphics (ggplot, trellis, grob, base). An htmlwidget returned by plot_r() falls through to the "other" type and never renders. This adds a render path for interactive htmlwidget objects (e.g. canvasXpress, plotly), as discussed in pharmaverse/teal#1757.

What changed

• plot_with_settings_srv() gains a "widget" plot_type; htmlwidget plots render through the htmlwidgets shiny binding (shinyWidgetOutput / shinyRenderWidget) instead of renderPlot/plotOutput. The widget's class drives the correct JS output binding.
• Height/width settings and full-screen apply as before; download produces a self-contained HTML file via htmlwidgets::saveWidget.
• htmlwidgets added to Imports; canvasXpress added to Suggests (example + tests only).
• Added an @examplesIf requireNamespace("canvasXpress") example on plot_with_settings, reusing the existing example structure.
• Tests: plot_type detection for htmlwidget vs ggplot via shiny::testServer.

Design notes

• Opt-in, no default change. Static-plot behaviour is untouched; the widget path triggers only when an app author returns an htmlwidget from plot_r. Existing callers (including teal.modules.clinical outputs) are unaffected — the interactive render is something a module opts into, e.g. routing tm_g_km / tm_g_forest_* through canvasXpress for the final render only.
• The render path is generic to htmlwidget, so other widget packages benefit too.

Verification

• shiny::testServer confirms plot_type() resolves to "widget" for a canvasXpress object and the UI emits the widget output binding.
• Verified end-to-end in a running shiny app: a canvasXpress scatter renders inside the plot_with_settings card with the resize/download chrome and interactive behaviour.

Happy to adjust anything — naming, the example, or the download behaviour.

🤖 Generated with Claude Code