Interactive stat_bin with showSelected (#158 PoC)

[ANAMASGARD]

Summary

This PR starts work on #158: making stat_bin work with showSelected so histograms update when the user changes selection.

What in Commit 1

Added tests/testthat/test-renderer3-stat-bin-showSelected.R.

The test checks that a faceted histogram with showSelected="facet":

• compiles without the current StatBin guard warning
• renders in the browser
• recalculates counts when facet selection changes
• keeps bin positions stable

Why

Issue #158 asks whether stats should move out of compile-time processing. This PR defines the expected behavior for a small stat_bin + showSelected proof-of-concept before implementation.

How the fix will work (Commit 2 )

1. R exports pre-stat data, showSelected columns, and panel x_breaks.
2. JS filters by selection, recomputes bin counts, then draws bars.

Selection changes counts, not the bin grid.

Local run shows expected failure on master

[tdhock]

this is interesting but lowest priority, let’s work on the other easier PRs first

[codecov]

Codecov Report

❌ Patch coverage is 92.19858% with 11 lines in your changes missing coverage. Please review.
✅ Project coverage is 78.51%. Comparing base (8f009ed) to head (3310c34).

Files with missing lines	Patch %	Lines
R/plot-build.r	83.67%	8 Missing ⚠️
R/z_animint.R	86.66%	2 Missing ⚠️
R/z_knitr.R	66.66%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #343      +/-   ##
==========================================
+ Coverage   73.07%   78.51%   +5.43%     
==========================================
  Files         165      166       +1     
  Lines        8933     9071     +138     
  Branches        0      186     +186     
==========================================
+ Hits         6528     7122     +594     
+ Misses       2405     1949     -456     

Flag	Coverage Δ
javascript	97.36% <100.00%> (+16.29%)	⬆️
r	69.22% <83.58%> (+0.02%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[Copilot]

Pull request overview

This PR adds a proof-of-concept for browser-side stat_bin recalculation when showSelected changes, with stable bin boundaries.

Changes:

• Exports pre-stat data and bin breaks from R.
• Adds JavaScript histogram binning.
• Integrates the new path into rendering and HTML/knitr output.
• Updates tests, coverage handling, NEWS, and package metadata.

Reviewed changes

Copilot reviewed 13 out of 13 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Changes and final review comments
v8-to-istanbul.js	Merges coverage from multiple snapshots.
tests/testthat/test-renderer3-stat-bin.R	Updates histogram rendering expectations.
tests/testthat/test-renderer3-stat-bin-showSelected.R	Adds interactive histogram tests.
tests/testthat/helper-HTML.R	Captures JavaScript coverage snapshots.
R/z_knitr.R	Loads the new JavaScript dependency.
R/z_animint.R	Routes interactive StatBin layers; moderate issue: unbuilt plot scales can leave mapped non-position aesthetics incorrectly serialized.
R/plot-build.r	Critical issue: effective default closed handling rejects direct stat_bin() calls. Moderate issues: explicit breaks are ignored, and facet panel assignments can be overwritten.
R/geom-.r	Moderate issue: coord_flip renames raw x before JavaScript binning, producing zero-height bins.
NEWS.md	Documents the proof of concept.
inst/htmljs/stat-bin.js	Moderate issues: weights are ignored, and generated rows discard styling and interaction metadata.
inst/htmljs/index.html	Loads stat-bin.js.
inst/htmljs/animint.js	Invokes browser-side binning.
DESCRIPTION	Updates package metadata and version.

Suppressed comments (10)

R/geom-.r:341

• Because this path is selected solely by StatBin, it rewrites a caller-supplied geom to rect. A valid call such as stat_bin(geom="line", position="identity", binwidth=1, showSelected=...) is therefore silently rendered as bars instead of honoring the documented geom override. Restrict the JS path to GeomBar/histograms or reject other geoms.

      g$geom <- "rect"

R/geom-.r:320

• Bypassing the non-identity-stat guard here also permits combining this PoC with clickSelects. The browser aggregation emits no clickSelects field, while the existing click handler calls update_selector(..., d.clickSelects), so clicking a bar selects undefined instead of a value. Reject clickSelects for this path or define an unambiguous bin-level value.

    js_stat_bin <- identical(l$js_stat, "bin")
    if(!(js_stat_bin && stat.type == "StatBin")){
      checkForNonIdentityAndSS(stat.type, has.show, is.show, l,
                              g$classed, names(g.data), names(g$aes))

R/plot-build.r:363

• These breaks are computed from the final built scales, which are retrained after StatBin on the generated xmin/xmax values when facet shrink is enabled. That range can be wider than the original pre-stat x range, causing the JS grid to gain or shift bins (often an extra empty edge bin) instead of matching the bins used by the R stat. Compute the breaks from the pre-stat panel x dimensions before the statistic is applied.

js_stat_bin_x_breaks <- function(built, binwidth) {
  layout <- built$panel$layout
  breaks_list <- list()
  for(panel_i in layout$PANEL) {
    scales <- panel_scales(built$panel, panel_i)
    bins <- bin_breaks_width(scales$x$dimension(), binwidth)
    breaks_list[[as.character(panel_i)]] <- bins$breaks

R/plot-build.r:315

• The validator rejects an explicit boundary, but StatBin$setup_params also accepts the deprecated origin alias and converts it to a boundary. With origin=..., the initial R stat uses the requested alignment while js_stat_bin_x_breaks() always computes the default alignment, so bars are assigned to a different grid. Reject origin here as well or carry the normalized boundary into the JS parameters.

  if(!is.null(params$boundary)){
    stop("JS stat_bin does not support boundary=.", call.=FALSE)
  }

R/z_animint.R:394

• CoordFlip is still allowed into this path, but Geom$export_animint() swaps serialized x/y column names for flipped coordinates while compute_stat_bin() always reads row.x and emits x bounds. Consequently a flipped interactive histogram has no usable x values for counting and renders zero-height or incorrectly oriented bars. Reject coord_flip() here or carry the coordinate orientation through the JavaScript binning and rendering path.

        L$stat_params$x_breaks <- js_stat_bin_x_breaks(
          ggplot.info$built, L$stat_params$binwidth)

R/z_animint.R:391

• get_pre_stat_layer_data() has already assigned each row to its facet PANEL, but this overwrites that assignment by copying every row into every panel. A faceted histogram whose showSelected variable is not the facet variable (for example facet_grid(facet~.) + showSelected="year") will therefore count all facets in every panel instead of preserving the facet split. Retain the original panel assignment, or make any replication conditional on an explicitly defined facet-selection behavior.

        panel_ids <- ggplot.info$built$panel$layout$PANEL
        df <- js_stat_bin_all_panels(df, panel_ids)

inst/htmljs/stat-bin.js:53

• For multiple selection, data can contain rows from several selected values, but d.group is computed by add_group() after showSelected* is added as an aesthetic; discrete showSelected columns therefore become part of the default group whenever no explicit group is mapped. The same visual group is consequently binned separately per selected value, producing overlapping per-value bars instead of counts over the combined selection. Use a stable visual grouping that excludes showSelected unless grouping by it was explicitly requested.

        groups = d3.nest().key(function(d) {
                return d.group;
        }).sortKeys(d3.ascending).entries(data);

inst/htmljs/stat-bin.js:65

• ymax is always set to the raw count, regardless of the layer's calculated y mapping. For example, geom_histogram(aes(y=..density..), showSelected=...) is rendered as counts rather than density (and the other stat_bin computed variables are likewise ignored). Compute the requested stat value or reject non-count y mappings before selecting this JS implementation.

                                ymin: 0,
                                ymax: counts[bin_i],
                                fill: fill_val,

inst/htmljs/stat-bin.js:29

• When CoordFlip is used, Geom$export_animint renames the pre-stat x column to y before this function runs. rows[row_i].x is then undefined, so every value is skipped and the histogram counts are zero; the generated bounds are not flipped either. Preserve the source x for binning and explicitly swap the generated coordinates, or reject this coordinate system.

                if (x_val === null || isNaN(x_val)) {
                        continue;
                }
                bin_i = bin_index_for_stat_bin(Number(x_val), breaks);

tests/testthat/helper-HTML.R:126

• takePreciseCoverage() only reads the active session; it does not stop it. Calling startPreciseCoverage() again can fail with Precise coverage has already been started, and that warning will escape animint2HTML()'s expect_no_warning() checks in the JS_coverage suite. Stop the session before restarting it.

    remDr$Profiler$startPreciseCoverage(callCount = TRUE, detailed = TRUE)

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.