dragonbones_flutter 0.1.0

Draw DragonBones skeletal animations on a Flutter Canvas. Framework-agnostic: `update(dt)` + `render(Canvas)`, plus a ready-made widget. No game engine dependency, so it drops into any Flutter project [...]

Changelog

0.1.0

First release. Draws DragonBones skeletal animations on a Flutter Canvas through two methods, update(dt) and render(canvas) — with no dependency on any game engine, so the same code drops into a plain Flutter app, an engine component, or a CustomPainter.

• Sprites (drawImageRect) and deformable meshes (drawVertices + ImageShader), including skinned meshes with bone weights and animated FFD.
• Nested child armatures flattened into the draw list, with the parent slot's transform composed onto them.
• Slot colour transforms applied through ColorFilter.matrix, including per-channel offsets; the identity case is free.
• DragonBonesWidget with DragonBonesFit, and a computeBounds() extension for placement and hitboxes.

Built on dragonbones 0.1.0, whose geometry is diffed against the official DragonBones runtime across 51 assets — 757,701 numeric comparisons, 0 mismatches. This package's own tests replay every one of those fixtures through the real renderer to prove no asset is silently skipped.

Known limitations in this release

Inherited from the runtime (the renderer cannot draw what is not evaluated):

• No animation events (EventObject) — nothing can be driven from the animation.
• Animated IK is ignored (static IK works).
• SlotZIndex and SlotAlpha / BoneAlpha timelines are not evaluated, so animated draw order and per-slot fade will be wrong.
• Path constraints and Surface bones are not evaluated.

Specific to this package:

• Alpha, Erase, Invert, Layer and Subtract blend modes have no exact Flutter counterpart and use the closest ui.BlendMode. The other nine map cleanly. No fixture uses the approximated five — unverified.
• Rotated atlas entries have a code path that no fixture exercises — unverified.
• Appearance is not checked by CI. Runners have no GPU; tests rasterise and count painted pixels, which catches "wired wrong and drew nothing", not "drew it wrong".
• No benchmarks. Allocation behaviour differs from upstream because pooling was dropped.