amoeba_grid

Releases

1. ▸▾0.3.021 Jul 2026

   New — `AmoebaGridConfig.bodyDragActivation` (BodyDragActivation): controls how a drag on a card *body* begins.

   • New — AmoebaGridConfig.bodyDragActivation (BodyDragActivation): controls how a drag on a card body begins.
     • immediate (default, unchanged behavior): a pan anywhere on the body moves the card — even through hit-opaque content.
     • longPress: a body move starts only after a ~350ms hold (then movement), feeding the same drag-engine path pans use. The card stops competing in the pan arena, so taps, drag-scrolls, and other gestures inside card content win instantly. On activation the card raises its lift cue as the pick-up signal. Edge/corner handles keep immediate drag in both modes — they are chrome, not content.

   Open source →
2. ▸▾0.2.021 Jul 2026

   New — `AmoebaShell`: shape-aware header/body card scaffold. The header pins inside the highest solid span that can fit it (never a notch, never a too-

   • New — AmoebaShell: shape-aware header/body card scaffold. The header pins inside the highest solid span that can fit it (never a notch, never a too-narrow arm); the body receives the full remaining shape below it, so flow widgets keep re-flowing through notches. Exists because the obvious compositions both fail — see its dartdoc.

   • Shape-following padding: CardOutline.trace(extraInset:) erodes the outline uniformly (convex radii shrink, concave grow); AmoebaCardGeometry.erodedPath(inset) exposes it. AmoebaPadding now clips its child to the eroded silhouette by default (clipToShape: false to opt out) — padding hugs the edge the way rectangle padding hugs a rectangle, enforced no matter what the child paints. Content can no longer leak onto the page background through empty bounding-box regions.

   • AmoebaListView flow correctness:

     • rows adopt the narrower span when they touch (or come within a geometry-derived clearance of) a horizontal notch/step edge, instead of rendering flush against it;
     • rows probe the outline's corner arcs and walk inward to clear them;
     • rows probe the SAME eroded surface AmoebaPadding clips to (contentClip), so the clip can never shear glyphs the layout thought were safe;
     • slots wholly outside the shape collapse instead of borrowing a distant span;
     • flush sides: a zero inset on any side (any combination) makes content run straight to that edge past the corner arcs — the rounded outline trims the corner pixels, like an edge-to-edge list in a rounded container. Non-zero insets keep the arc-probe behavior.

   • Fix — drag-morph chamfering: lerpOutline resampled outlines into a coarse polyline, flattening every corner arc to a chamfer mid-morph. Samples now scale with perimeter and reconnect via midpoint quadratics — corners stay round while dragging.

   • Developer experience:

     • AmoebaGridDiagnostics.showPaddingOverlay (+ paddingOverlayInset) paints every card's padding band translucent red in debug builds;
     • an AmoebaListView nested inside AmoebaContentArea (windowed, notch-free scope) warns in debug that rows will never re-flow;
     • a box/geometry width mismatch (plain Padding between the card and a flow widget) warns about broken span alignment.

   • Fix — pressing a card grabbed the WRONG card after a resize/re-layout. RawGestureDetector constructs its recognizer once and only re-runs the initializer on rebuild, so the GridMetrics closed over in the recognizer's constructor (the pointer-down hit test) froze at first-build values while paint used fresh metrics. After the cell extent re-resolved — a window resize, or a drag/rearrange that changed the occupied bounds — hit-testing and painting used different cell sizes and a press grabbed a card offset from the pointer (worse the further it was scrolled). The hit test now reads the controller's live metrics. Regression test: test/scroll_resize_hittest_test.dart.

   • Fix — side-handle inward L-gesture. Pulling a side handle IN and then perpendicular used to drop the perpendicular drag entirely (it fell through to a plain strip retract), so "pull in, then down" only ever carved a single row. The inward gesture now mirrors the outward one and removes the full notch block, deepening it along the perpendicular axis.

   • Fix — submissive resolution reworked to the runner model. When an aggressor overlaps a card it now: cedes the overlap (trimmed); keeps the LARGER side when bisected; and when fully smothered becomes a runner that retreats OPPOSITE the aggressor's advance at the smother frame (not first contact) and takes full space at its original size, shrinks to fit a partial gap, or drops to a 1x1 (an open adjacent cell, or into the aggressor, which cedes that cell). A runner never displaces another card and is never killed. This replaces the previous chained "retreater-becomes-aggressor" behaviour, which could ping-pong two cards until one was flung across the grid, and could drop the smaller half of a bisect or vanish a cornered card. Covered by test/submissive_ac_test.dart and test/resize_gesture_test.dart.