Changes in version 0.99.3                        

  - Selection reporting. Lasso selections are always reported (Shiny
    input$<plotId>_selected plus a sp-selection DOM event for custom
    hosts); cross-plot selection mirroring is scoped to the syncPlots
    group.
  - In-plot toolbar (toolbarPosition = "left" / "top" / "none"): pan,
    lasso, zoom-to-selection, reset, screenshot.
  - zoomOnSelection auto-frames the lasso selection.
  - Encode size / opacity by data via sizeBy / opacityBy (a numeric
    column or vector).
  - Frosted-glass legend with legendOpacity / legendBlur, round colour
    dots and per-category counts.
  - Smoother panning - the d3 axis redraw is coalesced to one frame per
    view event.

                       Changes in version 0.99.2                        

  - Bundled example dataset. data(reglScatterExample) loads a small
    (3,400-cell) synthetic single-cell-style UMAP data.frame with a
    categorical cell-type column and a continuous marker gradient, so
    users can plot something immediately without downloading data or
    installing a Suggests package. Used by the examples, vignette and
    tests.
  - Seurat support. reglScatterplot() now dispatches on Seurat objects:
    coordinates come from Embeddings() (reduction matched
    case-insensitively, default "UMAP" with umap/tsne/pca fallback), and
    colorBy / groupBy resolve against meta.data columns or features via
    FetchData(). Works with both Seurat v4 (slot) and v5 (layer)
    objects. SeuratObject added to Suggests.
  - Plain matrix input. reglScatterplot(m) accepts a numeric coordinate
    matrix (e.g. a UMAP embedding or prcomp()$x), using the first two
    columns by default; x / y may select other columns by index or name.
  - Monocle3. A cell_data_set is a SingleCellExperiment subclass, so it
    flows through the existing SCE dispatch unchanged - now covered by a
    test and documented.
  - AnnData support. reglScatterplot() now dispatches on in-memory
    AnnData objects (from anndataR or the anndata CRAN package):
    coordinates from an obsm embedding (default "UMAP" -> "X_umap",
    matched case-insensitively), colorBy / groupBy resolved against obs
    columns or var_names features, with assay selecting the layer (X / a
    layer / "raw"). AnnData read via zellkonverter::readH5AD() continues
    to use the SingleCellExperiment path. anndataR added to Suggests.
  - filterBy now shows interactive distribution sliders outside Shiny.
    Each numeric filterBy column gets a mini histogram of its
    distribution with a dual-handle range brush; dragging the handles
    selects a range, dims the out-of-range bars and filters the points
    live. Previously the slider UI only existed when a Shiny app
    supplied it (the data was loaded but nothing showed).
  - Per-plot filter/legend state. Legend category toggling and filterBy
    ranges are now scoped to each plot instead of a page-global
    registry, so independent plots on one page (e.g. several widgets in
    a Jupyter notebook, which all share window) no longer filter each
    other. Shiny still broadcasts filters across a dashboard via its
    message handlers.
  - Legend dragging no longer stretches the box. A
    bottom-/right-anchored legend kept its far-edge anchor when dragged,
    pinning all four edges so it ballooned to full height. Dragging now
    re-anchors to top/left cleanly. Legend height also trimmed to fit
    its contents.
  - title is now drawn on screen. Previously the title argument only
    appeared in PNG/SVG/PDF exports; it now renders as a centred caption
    at the top of the plot (coloured with axisColor).
  - Client-side plot sync without Shiny. Passing the same syncPlots
    group to several plots now links their pan/zoom in plain HTML / R
    Markdown / the Viewer, not just in Shiny. Plots are now keyed by
    their logical plotId (previously the internal DOM id won, so a
    syncPlots group never matched and sync silently did nothing outside
    Shiny). Verified with a headless-browser test (js/test-sync.mjs).
  - Legend theming. The legend header background now follows legendBg
    (it was a fixed light strip that looked wrong on dark themes), and
    the header title text follows legendText instead of a CSS variable
    that RStudio's dark Qt theme could override to a near-invisible
    colour (the "white legend title" bug). The legend also has a hard
    height cap so it can't span the whole plot when the container height
    is indefinite (e.g. a knitted R Markdown).
  - Download button honours an explicit enableDownload = TRUE. The
    export button was hidden inside any iframe (RStudio Viewer,
    knitted-HTML preview, Jupyter); an explicit opt-in now always shows
    it.
  - Crisp rendering in the RStudio Viewer (pixelRatio). The widget now
    renders the WebGL backing store at max(devicePixelRatio, 2) by
    default instead of relying on window.devicePixelRatio, which the
    RStudio Viewer (an embedded browser) reports as 1 even on HiDPI
    screens - the cause of the previously soft / low-quality look in the
    Viewer. A new pixelRatio argument overrides this (e.g. pixelRatio
    = 1 to save GPU memory, 3 for export-quality sharpness); very large
    data (n > 500000) keeps the true ratio to bound the pixel count.
  - No CDN at runtime - the widget now works offline. All browser
    dependencies (regl-scatterplot, d3, pickr, html2canvas, jspdf) are
    bundled into inst/htmlwidgets/reglScatterplot.js with esbuild,
    instead of being fetched from esm.sh / cdnjs via dynamic import()
    and <script> injection. Standalone HTML reports, the RStudio Viewer
    and firewalled / air-gapped environments now render without a
    network connection. Build sources and instructions live in js/
    (excluded from the package build).

                       Changes in version 0.99.0                        

  - First Bioconductor submission.
  - Package restructured into one file per responsibility (utils.R,
    colors.R, scatterplot.R, sce.R, shiny.R).
  - Functions renamed to Bioconductor camelCase style: my_scatterplot()
    → reglScatterplot(), my_scatterplotOutput() /
    renderMy_scatterplot() → reglScatterplotOutput() /
    renderReglScatterplot(), enableMyScatterplotSync() →
    enableReglScatterplotSync(), updateMyScatterplotSize() →
    updateReglPointSize(), to_base64() → toBase64(). Argument naming has
    been harmonised to camelCase (group_var → groupBy, filter_vars →
    filterBy, size → pointSize, gene_names → pointLabels, center_zero →
    centerZero, filtered_indices → filteredIndices, selected_indices →
    selectedIndices, legend_title → legendTitle).
  - library() calls inside package code removed; explicit importFrom
    declarations added throughout.
  - Input validation added: coordinates, palettes, and filter arguments
    now fail fast with informative messages.
  - JavaScript widget bugs fixed:
      - empty ResizeObserver body now drives the public resize() so
        standalone HTML, RStudio Viewer and Jupyter outputs reflow
        correctly,
      - stale factory width / height references replaced with live
        container.getBoundingClientRect() measurements,
      - the previously undeclared prevNumPoints is now properly scoped,
      - CDN imports for d3 and regl-scatterplot are wrapped in try /
        catch and display a user-visible error when offline.
  - sizingPolicy() configured for the RStudio Viewer and Jupyter so
    rendering is correct outside Shiny.
  - Tests added under tests/testthat/ covering utilities, colour
    resolution and widget construction.
  - Vignette reglScatterplot.Rmd added, including a SingleCellExperiment
    walk-through.
  - DESCRIPTION updated for Bioconductor (biocViews, R >= 4.4.0,
    URL/BugReports, VignetteBuilder, Suggests for testing).
  - Auto-fit on resize: when the user has not yet panned/zoomed,
    container resize events re-run the data-aspect-preserving fit,
    eliminating the "bottom rows clipped" problem in Hyprland tiles,
    RStudio Viewer at non-default browser zoom, and flex/grid layouts
    that settle late.
  - Single-cell support: reglScatterplot(sce, x = "UMAP", colorBy = ...)
    pulls coordinates from reducedDim() and resolves colorBy against
    colData columns or feature rownames automatically. SpatialExperiment
    inputs use spatialCoords() when x = "spatial".
  - Z-channel quantisation (Uint16) cuts the standalone-HTML payload by
    another ~25% on top of the X/Y quantisation already in place. At 5M
    points the file drops from ~78 MB to ~39 MB.
  - New rangePadding argument (default 0.4) controls how much whitespace
    is added on each side of the data range. This is the recommended
    dial when clusters appear clipped at the edges of very small
    viewports - the previous fixed margin was not enough at narrow
    window sizes.
  - Known viewport behaviour: the widget renders inside its htmlwidgets
    container, which honours an explicit height = N pixel value
    verbatim. If the host window is shorter than N (e.g. a small browser
    tab, a tile in a tiling window manager), the bottom of the widget is
    clipped by the host viewport - not by reglScatterplot. Pass height =
    "100%" (or leave the argument out) to let the widget fill whatever
    vertical space is available.