changes.md

2026-08-19 (doc screenshot generator)

Added a script to regenerate the screenshots referenced in docs/layout.md directly from the Rust snippets next to them, instead of hand-capturing them by running examples/layout.rs interactively (which silently goes stale whenever a snippet changes).

• examples/gen_doc_screenshots.rs: scans docs/layout.md for fenced ```rust blocks followed by a ![...](file.png) reference, and generates examples/doc_screenshots.rs (a gitignored build artifact, not committed) containing one fn scene_N() -> Scene<Rgb565> per matched snippet plus a main() that renders each headlessly and saves the PNG.
• scripts/gen_doc_screenshots.sh: runs the two examples in sequence (gen_doc_screenshots to codegen, then doc_screenshots --features headless to render). Uses embedded-graphics-simulator's headless feature (SimulatorDisplay + to_rgb_output_image().save_png()), so no SDL/window is needed.
• Cargo.toml: added a [[example]] name = "gen_doc_screenshots" entry; doc_screenshots is deliberately not declared there since an explicit [[example]] entry requires the file to exist at manifest-parse time, which would break every other cargo invocation before the file has been generated once.
• Verified the regenerated PNGs are byte-identical to the previously hand-captured ones, and that editing a single snippet's button label and rerunning only changes that one screenshot.

2026-08-19 (list_view press-then-commit)

src/list_view.rs: input_list now distinguishes PointerDown from PointerUp, same pattern as toggle_group.rs/button.rs.

• ListState gained a private pressed: Option<usize> field.
• PointerDown hit-tests which row the touch landed on (via a new row_index_at() helper shared by both handlers) and stores it in pressed; sets focus and marks dirty, does not change selected.
• PointerUp re-hit-tests the release point and only commits (updating selected, firing OutputAction::Command) if release lands on the same row that was pressed — dragging off the row before lifting cancels it.
• draw_list inverts fill/text for the currently pressed row (even if not yet selected), so the touch is visible before release commits it.

Existing tests (test_list_view, test_list_view_grows_to_fill_parent_width, test_list_view_enter_with_stale_selected_no_panic) needed no changes — test_list_view drives via click_at (synthesized down+up at the same point), so pressed == Some(n) always holds. Verified with cargo test --features std (53 passed).

2026-08-19

src/toggle_group.rs: input_toggle_group now distinguishes PointerDown from PointerUp instead of only reacting to PointerUp, mirroring button.rs's press/release pattern.

• SelectOneOfState gained a private pressed: Option<usize> field.
• PointerDown hit-tests which segment the touch landed on (via a new cell_index_at() helper shared by both handlers) and stores it in pressed; this only sets focus and marks the view dirty, it does not change selected.
• PointerUp re-hit-tests the release point and only commits the selection (updating selected and returning OutputAction::Command) if the release lands on the same segment that was pressed — dragging off the widget before lifting cancels the tap, same as button.rs.
• draw_toggle_group now inverts a segment's fill/text while it is the pressed one (even if not yet selected), so the touch is visible before release commits it.

Existing tests (test_toggle_group, test_toggle_group_empty_no_panic), which drive the widget via click_at (a synthesized down+up at one point), needed no changes and still pass — click_at's down and up share the same point, so pressed == Some(n) always holds. Verified with cargo test --features std (53 passed).

2026-08-16

Parameterized the entire toolkit over the pixel color type instead of hardcoding Rgb565, so it can drive displays in any embedded_graphics::pixelcolor::PixelColor mode (Rgb888, Gray8, BinaryColor, etc.), not just RGB565.

• src/gfx.rs: TextStyle and the DrawingContext trait gained a C: PixelColor parameter (TextStyle<'a, C>, DrawingContext<C>); every color-bearing method now takes &C instead of &Rgb565.
• src/lib.rs: ViewStyle, Theme, DrawEvent, LayoutEvent, GuiEvent, and the DrawFn/LayoutFn/InputFn/Callback type aliases all gained <C: PixelColor>. BW_THEME stays a concrete Theme<Rgb565> (the common default); build your own Theme<C> literal for other color types.
• src/view.rs: View → View<C: PixelColor>; src/scene.rs: Scene → Scene<C: PixelColor>, and every dispatch/layout/draw function (click_at, pointer_down_at, pointer_up_at, event_at_focused, pick_at, draw_scene, layout_scene, ...) gained the same parameter.
• src/device.rs: generalized the old FromRgb565 trait into FromColor<Src: PixelColor>, with a blanket identity impl (impl<C> FromColor<C> for C) plus the existing BinaryColor-from-Rgb565 threshold conversion. EmbeddedDrawingContext<'a, T> became EmbeddedDrawingContext<'a, T, C = T::Color>, converting the theme's logical color C into the display's native color T::Color via T::Color: FromColor<C>. This preserves the existing two-tier design (one Theme<Rgb565> can still drive both a full-color display and a 1-bit e-paper display via automatic conversion, as examples/simulator.rs does) while letting the logical color be anything.
• All widget constructors (make_button, make_label, make_panel, make_list_view, make_text_input, make_toggle_button, make_toggle_group, make_tabbed_panel, make_grid_panel) and layout functions became generic over <C: PixelColor>. Their bodies were untouched — they only ever touch color through theme.*.fill/.text, so genericity falls out for free.
• src/grid.rs: the one widget with a hardcoded color (Rgb565::RED for debug gridlines) now uses theme.accented.fill instead, removing its embedded_graphics::pixelcolor dependency entirely.
• src/test.rs: MockDrawingContext/make_mock_theme() were deliberately kept concrete at Rgb565 (not genericized) — every test constructs a scene/theme through one of these, so Rust's local type inference resolves C = Rgb565 throughout each test body for free. As a result almost no test code needed to change; only a handful of test helper functions that had Scene/View/DrawEvent/LayoutEvent/GuiEvent in their own signature (not just their body) needed an explicit <Rgb565> added, since function signatures aren't inferred.
• examples/simulator.rs, examples/custom_view.rs, esptest/src/main.rs: updated call sites accordingly (Theme<Rgb565> on the theme consts, Scene<Rgb565>/View<Rgb565> on a few helper signatures, FromRgb565 → FromColor<Rgb565> bound in run_loop). Behavior is unchanged — verified the simulator's existing dual Rgb565/BinaryColor (e-paper) code path still type-checks against one shared Theme<Rgb565>.

2026-08-15

Replaced the single InputEvent::Tap(Point) event with InputEvent::PointerDown(Point) and InputEvent::PointerUp(Point), so the toolkit can distinguish a press from a release (mouse down/up on desktop, touch down/up on embedded).

• src/scene.rs: added pointer_down_at / pointer_up_at, each hit-testing at the given point and dispatching the corresponding event to the topmost view, exactly like the old click_at. click_at is now a convenience wrapper that calls pointer_down_at then pointer_up_at at the same point (simulates a full click) and is unchanged for existing callers/tests.
• src/button.rs: ButtonState gained a pressed: bool field. PointerDown sets focus + pressed = true (no action yet); PointerUp clears pressed and only fires OutputAction::Command if the button was actually pressed first. The draw function now inverts fill/text color while pressed, so the button visibly depresses and un-depresses. Added a tests module covering press/release/no-op-release/click_at.
• src/text_input.rs, src/list_view.rs, src/toggle_group.rs: their InputEvent::Tap match arms were renamed to InputEvent::PointerUp (selection/focus already only made sense on release).
• src/toggle_button.rs: input_toggle_button previously reacted to any event type; now explicitly filters to InputEvent::PointerUp only, otherwise it would double-toggle (once for down, once for up) since click_at now dispatches both.
• examples/simulator.rs: MouseButtonDown now calls pointer_down_at (previously a no-op); MouseButtonUp calls pointer_up_at instead of click_at.
• esptest/src/main.rs: the touch poll loop had no down/up edge detection (it called click_at every ~100ms poll while a finger stayed down). Added a last_touch_point state so a touch-down fires pointer_down_at once, and release (no touch point reported) fires pointer_up_at once at the last known point.

2026-08-07

Added set_focus_enabled(bool) / is_focus_enabled() to Scene in src/scene.rs. When focus is disabled, set_focused is a no-op: focused stays None, no dirty rects are produced for focus transitions, no focus rings are drawn, and event_at_focused dispatches to nothing. Useful on e-paper touch screens where minimizing dirty rects is critical. Three new tests in focus_disabled_tests verify the flag prevents dirty rect expansion, that the default (enabled) behaviour is unchanged, and that the flag can be toggled at runtime.

2026-08-05 (8)

Fixed layout_list in list_view.rs ignoring h_flex = Grow:

layout_list set the view height based on item count but never touched the width, leaving it at the default 100px regardless of the flex setting. Added a check: when h_flex == Grow, width is set to e.space.w (the space offered by the parent), matching the pattern used by all other layout functions.

Added regression test test_list_view_grows_to_fill_parent_width which places a Grow list view inside a 320px-wide layout_vbox parent and asserts the list width equals 320px. The test failed before the fix and passes after.

2026-08-05 (7)

Fixed gap accounting bug in layout_hbox and layout_vbox:

Previously, gap was applied during the child positioning pass but was never deducted from the space offered to Grow children. With N total children and a gap G, the Grow children were collectively given G × (N−1) extra pixels, causing them to overflow the container silently.

Fix: before dividing remaining space among Grow children, subtract gap × (total_children − 1) in both layout_hbox (avail_horizontal_space) and layout_vbox (vert_leftover).

Added two regression tests (test_hbox_grow_child_accounts_for_gap, test_vbox_grow_child_accounts_for_gap) that were written first to confirm the overflow, then pass after the fix.

2026-08-05 (6)

Minor fixes from code review issues #19–#20:

• #19: Size::empty() in geom.rs now returns Size { w: 0, h: 0 } instead of the magic sentinel Size { w: -99, h: -99 }. is_empty() treats any w < 1 || h < 1 as empty, so zero is semantically identical but not misleading.
• #20: TextInputState cursor methods in text_input.rs now navigate by char boundary instead of by byte. cursor_back uses char_indices().next_back() to find the previous char's byte offset; cursor_forward uses chars().next() and len_utf8() to advance past the current char; insert_char advances by key.len_utf8(). This prevents panics on multi-byte char boundaries if non-ASCII content reaches the widget.

2026-08-05 (5)

Performance fixes from code review issues #17–#18:

• #17: Scene::get_children_ids now returns &[ViewId] instead of Vec<ViewId>, eliminating the heap allocation and full clone on every call. All call sites that need to mutate the scene after iterating children (layouts, draw, tabbed_panel, grid) collect to an owned Vec with .to_vec() first; pure read-only callers (pick, dump, filtered getter) iterate the slice directly.
• #18: draw_view now accepts an accumulated offset: Point and skips entire view subtrees whose global bounds do not intersect scene.dirty_rect. Bounds::intersects was added to geom.rs to support the check. When dirty_rect.is_empty() (full-redraw path), culling is disabled and all views are drawn. draw_scene passes Point::zero() as the initial offset.
• Added Bounds::intersects unit test in geom.rs.
• Added two unit tests in scene.rs (gated on headless feature): test_dirty_rect_culls_non_overlapping_view verifies that a view outside the dirty region is not drawn; test_empty_dirty_rect_draws_all_views verifies that an empty dirty_rect disables culling.

2026-08-05 (4)

API design fixes from code review issues #14–#16:

• #14: make_label, make_header_label (label.rs) and make_text_input (text_input.rs) now accept name: &ViewId instead of name: &'static str, matching every other widget constructor. Updated all call sites in grid.rs tests and examples/simulator.rs.
• #15: ListState::new_with and SelectOneOfState::new_with now return Self instead of Box<dyn Any>. Boxing moved to the call site in make_list_view and make_toggle_group. Removed now-unused core::any::Any imports from both files.
• #16: PanelState gains an explicit impl Default (border_visible cannot derive because its default is true). SelectedState gains #[derive(Default)]. Both keep their new() methods as thin delegates to default().

2026-08-05 (3)

Idiomatic Rust cleanups from code review issues #7–#13:

• #7: scene.rs set_focused and event_at_focused — replaced is_some() + unwrap() pairs with if let Some(...).
• #8: scene.rs click_at — changed handlers: &Vec<Callback> to handlers: &[Callback].
• #9: scene.rs get_children_ids_filtered — replaced .map(...).flatten() with .filter_map(...).
• #10: moved &'static str → ViewId conversion from an impl Into in grid.rs to a proper impl From<&'static str> for ViewId in view.rs where it applies unconditionally.
• #11: grid.rs — removed two empty no-op if view.h_flex == Shrink {} / if view.v_flex == Shrink {} branches; removed now-unused Shrink import.
• #12: layouts.rs — removed redundant .clone() calls on Copy types (Insets, Flex, Size) in layout_vbox, layout_hbox, and layout_std_panel.
• #13: layouts.rs layout_vbox fold — removed misleading return inside the closure (returns from the closure, not the function).

2026-08-05 (2)

Fixed correctness bugs from code review issues #3–#6:

• #3 (already fixed): layouts.rs hbox flex space correctly uses available_space.w — carried over from the prior "fix hbox layout bug" commit. Updated the test_hbox_fixed_width assertion to match correct values (child2 width 120px, child3 x=160).
• #4: scene.rs move_view_to_parent now also calls self.parents.insert(child, parent) so get_parent_for_view returns the correct parent after a move. Added assertion to parent_child test.
• #5: scene.rs remove_parent_and_children is now recursive: replaces remove_view(&kid) with remove_parent_and_children(&kid) so grandchildren are fully cleaned up from all three maps. Added remove_parent_and_children_cleans_grandchildren regression test.
• #6: geom.rs Point::scaled and Bounds::scaled now use scale.min(i32::MAX as u32) as i32 instead of scale as i32 to avoid silent wrapping on large scale values.

2026-08-05

Fixed crash/panic risks from code review issues #1 and #2:

• toggle_group.rs input_toggle_group / draw_toggle_group: added is_empty() early-return guard before bounds.size.w / items.len() division to prevent divide-by-zero on an empty items list.
• list_view.rs TextAction::Enter path: replaced unchecked items[state.selected] with items.get(state.selected) so a stale selected index (e.g. after external truncation) cannot panic.
• Added regression tests: test_toggle_group_empty_no_panic and test_list_view_enter_with_stale_selected_no_panic.

2026-08-04 (2)

Fixed remaining crash paths in list_view.rs when the list is empty or has zero-height cells:

• input_list / draw_list: guard against cell_height <= 0 (can happen when the view has no height yet) in addition to the empty-items guard added previously.
• TextAction::Enter: wrapped in !state.items.is_empty() check to prevent an out-of-bounds index on an empty vec.

2026-08-04

Added inline comments to test_cliprect_nested explaining how nested parent offsets accumulate into scene coordinates and what each assertion proves (src/lib.rs).

Added test_cliprect_toggle_group_in_panel — verifies that clicking a toggle group inside a panel marks only the toggle group's global bounds dirty, not the panel or full scene (src/lib.rs). Also adds make_panel and make_toggle_group imports to the test module.

Fixed divide-by-zero panic in list_view.rs when the list has no items: input_list and draw_list now return early if state.items.is_empty().

2026-08-03

Fixed cargo test failures caused by optional crates (test-log, embedded_graphics_simulator, env_logger) being used without feature gates:

• src/layouts.rs, src/lib.rs, src/tabbed_panel.rs, src/scene.rs: Added #[cfg(any(feature = "std", feature = "headless"))] to test modules that depend on test_log or cross-reference helpers from other feature-gated test modules.
• Cargo.toml: Added [[example]] entry for custom_view with required-features = ["std"] so it is not compiled during a plain cargo test.

cargo test (no features) now runs 10 tests; cargo test --features std runs all 25.

2026-07-27 13:30

Added examples/orientation.rs — demonstrates display orientation using an inline RotatedDisplay<D> wrapper.

embedded-graphics-transform 0.1.0 targets embedded-graphics-core 0.3 (embedded-graphics 0.7.x) and is incompatible with this project's 0.8.x (core 0.4). Instead, RotatedDisplay is implemented directly in the example using the same coordinate transforms the crate uses:

• Rotate0: identity
• Rotate90: (phys_w-1-y, x)
• Rotate180: (phys_w-1-x, phys_h-1-y)
• Rotate270: (y, phys_h-1-x)

The wrapper implements DrawTarget and Dimensions so that draw_content receives a canvas whose bounding_box() reflects the logical (post-rotation) size, making the scene automatically adapt between landscape (0°/180°) and portrait (90°/270°).

Press R to cycle through all four rotations.

cargo run --example orientation --features std

2026-07-27 12:35

Made EmbeddedDrawingContext generic over display color type (src/device.rs).

Added pub trait FromRgb565: PixelColor with implementations for Rgb565 (identity) and BinaryColor (black→Off, anything else→On). All drawing structs and functions (EmbeddedDrawingContext, ScaledDisplay, draw_ttf_glyphs) now use T: DrawTarget with T::Color: FromRgb565 instead of the hard-coded T: DrawTarget<Color = Rgb565>. Colors from the UI layer (always Rgb565) are converted at the draw boundary via T::Color::from_rgb565(...).

To use a BinaryColor display:

let mut display: SimulatorDisplay<BinaryColor> = SimulatorDisplay::new(size);
let mut ctx = EmbeddedDrawingContext::new(&mut display);

2026-07-27 12:10

Fixed TTF text centering in ctx.text() (src/device.rs). The text() method is used by draw_centered_text, which buttons call with bounds.center() as the position.

• Horizontal: The TTF branch was ignoring style.halign and starting the cursor at position.x directly, so all text drew from the center point rightward. Now Align::Center offsets the cursor left by (total_w + first_xmin) / 2 so visible glyphs are centered on the position.
• Vertical: The TTF branch passed position.y directly as the baseline, placing text below center. Now Align::Center uses font.horizontal_line_metrics to compute baseline_y = position.y + (ascent + descent) / 2, which aligns the visual midpoint of the text with the given position (falls back to size * 0.25 offset if metrics are unavailable).

2026-07-27 18:00

Added --scale=N command-line option to the simulator example. The scale defaults to 2 if not specified.

cargo run --example simulator --features std -- --scale=1
cargo run --example simulator --features std -- --scale=3

2026-07-27 17:45

Fixed compile error in examples/simulator.rs: fontdue::Font::from_bytes expects impl Deref<Target=[u8]>; passing &Vec<u8> resolves to Vec<u8>, not [u8]. Fixed by passing bytes.as_slice().

2026-07-27 17:30

Added TTF font option to the simulator example and documented font usage in README.

• examples/simulator.rs: Added get_ttf_font() helper (gated behind #[cfg(feature = "ttf")]) that loads a system font at runtime (Geneva on macOS, DejaVu Sans on Linux) using OnceLock<Option<fontdue::Font>>. Added a TTF button to the font-size toolbar (also #[cfg(feature = "ttf")]) and a "font-ttf" command handler in handle_events. Run with cargo run --example simulator --features std,ttf.
• README.md: Replaced the Themes section with updated field names (standard, panel, selected, accented). Added a Fonts subsection explaining FontKind::Bitmap vs FontKind::TrueType, with std (OnceLock) and no_std (MaybeUninit) usage examples.

2026-07-27 17:00

Added optional TrueType font support via fontdue, gated behind the ttf Cargo feature.

• Cargo.toml: Added fontdue = { version = "0.9", optional = true, default-features = false } and ttf = ["fontdue"] feature. The default-features = false enables fontdue's no_std + alloc mode, matching the rest of the crate.
• src/font.rs (new): Defines FontKind enum with Bitmap(MonoFont<'static>) and (when ttf feature enabled) TrueType { font: &'static fontdue::Font, size: f32 } variants. Implements Copy, Clone, and Debug. Provides metric helpers: char_width(), char_height(), baseline(), str_width().
• src/lib.rs: Re-exports FontKind. Theme.font and Theme.bold_font changed from MonoFont<'static> to FontKind. Added #[derive(Clone, Copy)] to Theme. Updated BW_THEME constant.
• src/gfx.rs: TextStyle.font changed from &'a MonoFont<'static> to FontKind (by value). draw_centered_text updated accordingly.
• src/util.rs: calc_size and calc_bounds updated to take FontKind and use metric helpers.
• src/device.rs: fill_text dispatches on FontKind. TTF rendering uses draw_ttf_glyphs (alpha threshold >127). Fixed pre-existing bug where text centering metrics were hardcoded to FONT_6X10 regardless of the active font.
• All widget files updated to pass FontKind by value and use .char_width()/.char_height()/.str_width() instead of .character_size.*.
• examples/simulator.rs and examples/custom_view.rs: Updated Theme construction to wrap font constants with FontKind::Bitmap(...).

Developer usage: iris-ui = { features = ["ttf"] }, then construct fontdue::Font in a static and pass FontKind::TrueType { font, size } in the theme.

2026-07-27 16:10

Added integer scale factor support to Scene and EmbeddedDrawingContext.

• src/geom.rs: Added Bounds::scaled(scale) and Point::scaled(scale) helpers.
• src/scene.rs: Added scale: u32 field (default 1), Scene::new_with_scale(bounds, scale) constructor, and scene.scale() getter. All layout, picking, and dirty tracking remain in logical coordinates.
• src/device.rs: Added scale field to EmbeddedDrawingContext and new_with_scale() constructor. Geometric primitives (rect, line) multiply coordinates by scale. Text uses a ScaledDisplay wrapper that turns each logical pixel into a scale×scale block, achieving true pixel-doubling of bitmap fonts.
• examples/simulator.rs: Wired scale into display size, drawing context, clip, and mouse input (divides physical coords by scale before hit-testing). Enabled scale=2 in make_scene().

Default scale is 1, so all existing call sites (including the ESP target) are unchanged.

2026-07-27 15:30

• Upgraded embedded-graphics-simulator from 0.7.0 to 0.8.0, which pulls in sdl2 0.38.0.
• Fixes a panic ("trying to construct an enum from an invalid value 0x207") caused by newer SDL2 system library (2.28+) emitting event types that sdl2 0.37.0 did not recognize.