Fix any broken JSDoc syntax. Unfortunately there is no way to document tuples like `{[int, int]}` for a `[width, height]` array. We could possibly fix this and make the code more readable with `{width: number, height: number}` instead of tuples.
Remove single `@fileoverview` from Rectangle.js, since it was the only file with one and was weird to have that one module show up on the docs home page.
Add a `docs` script to `package.json`, run it from `npm test`, so invalid docs will cause tests to fail (eventually we may want to start using eslint-plugin-jsdoc to catch these errors with the linter alone).
In Scratch 2.0, an SVG costume's origin is stored relative to that SVG's
viewbox. In order to use the origin values correctly, then, we need to
compensate for the viewbox offset.
I wasn't so clever with this - I'm sorry. Now that I have managed to
link up my local render & vm projects I am able to fix the issue
properly. Many appologies!
When a new skin is added, previously the rotation center was set to `[0, 0]`. In the case of costumes and backdrops added from libraries, the center should be calculated from the bounding box of the imported skin.
Toward LLK/scratch-gui#18
Scratch uses y=up, whereas WebGL and the Canvas element both use y=down.
To make this work, we set a non-conventional projection matrix when
rendering to the stage. Some of the internal renders (touching color,
stamp, etc.) were using a y-flipped projection matrix, though, which
means that the output was double-flipped and would appear incorrect. For
the touching-color block this just meant a cosmetic issue: the debug
canvas looks upside-down. For pen stamp, though, it's a real issue since
the stamp was appearing upside down on the stage.
This change ensures that the projection matrix for every internal render
follows WebGL / Canvas conventions, and that we only y-flip on the final
render to the stage.
Also clean up the code around our calls to `readPixels`:
- Use `Uint8Array` instead of `Buffer` to hold the pixels
- Use `ImageData.data.set` instead of a `for` loop when copying pixels
to a canvas
- `Drawable` now removes its `Skin` on `dispose`, disconnecting events.
- Creating a new `Drawable` doesn't always create a new `Skin` any more:
now it can share an existing skin.
- Don't throw when asked to update the properties of a `Drawable` that
has already been destroyed. This seems like a VM bug but was causing
overwhelming output in the browser for some projects.
- Add & correct a few comments
- Make `Drawable` IDs work like `Skin` IDs
- Remove some dead code
- Fix costume resolution in `BitmapSkin` before first `setBitmap` call
- Rename `Drawable` & `Skin` management variables in `RenderWebGL` to
clarify their purpose and better distinguish them from each other
- Remove problematic premature optimization in `_drawThese`
- Fix missing return statement in `Skin`'s `setRotationCenter`
- Fix mismatch between getTexture & getUniforms in the `Skin` class
