There was a bug with how we were removing a NestedAnimation from export. The export process for an artboard does this:

- Collect all the object in an artboard which should export.
  - This is done by calling ```component.exportsWith(artboard)``` on each component.
  - The ones which return true are included in the export list.
- Assign runtime ids (indices) to each object in that list and build a remapping table to make sure edit time ids are written as their matching runtime id (index).
- Export each object in the list by calling ```writeRuntime``` in the exact order of the runtime ids as the runtime ids are indices into this export list.

We were earlying out of exporting by returning early in ```writeRuntime```. This causes a problem because the export list now is missing an item which consumes an id at that export index. If an object does not wish to be exported, it should return ```false``` in its ```exportsWith``` method.

Diffs=
cc15ffa4f6 Fix NestedAnimation skipping export too late (#8573)

A long time ago, I only sent out dirt.nslicer when anything about the vector n-slicer changed. However, this caused clipping paths inside of a vector n-slicer to not update in time. I was worried to miss a spot, so in addition to sending dirt.nslicer, I also sent out dirt.path.. recursively...

Then it turns out that if there are texts inside of the n-sliced node, they will also listen to dirt.path, and that caused some performance issues. So now we are back to wanting to send dirt.nslicer, and have places that listen to it calculate things.

The way I audited this... I first did a search for all ComponentDirt.path in the Dart codebase:
![image](https://github.com/user-attachments/assets/281df80c-3320-46f2-9ff7-c3ee6e40065a)

Then for all of them, I see whether the check occurred in a update() function. If it does, then I added an dirt.nslicer condition too. This was after the first commit. I'm hoping that doing it this way means that the checks are complete, meaning I'm not missing any dirt.nslicer checks.

I then commented each out, and see whether something will not update in time. So e.g. if the update is on a layout node, I'd check whether a hug layout node responds to an nsliced node's width/height. (in this case, I don't need the nslicer check, so i removed it). This was the second commit. I'm hoping that doing it this way means all the dirt.nslicer checks are necessary.

Then the idea is...if we are doing sth complete and necessary, it's the right amount.

Diffs=
1d23ae5782 editor: be more mindful of sending dirt recursively when n-slicing (#8576)

Diffs=
01e014f82d Add an android_gms job to browserstack (#8583)

ANGLE advertises support for EXT_clip_cull_distance on this device but doesn't implement it correctly. Just never use this extension on ANGLE. If we're getting ANGLE, then the device most likely supports Vulkan and we will switch over to there shortly.

Diffs=
8030dc05eb Fix clip planes on S22/Xclipse/ANGLE (#8582)

Diffs=
869816556f Fix invalid memory access (#8546)

Renamed the define to RENDERING_MODE_MSAA as we talked about yesterday.

Diffs=
e10bb4aafd Rename USING_DEPTH_STENCIL to RENDERING_MODE_MSAA (#8584)

Remove code that made ANGLE behave differently on lots_of_images due to slowness. Now that GMs run in parallel and we have the GMREGISTER_SLOW macro, the longer test time is invisible.

Use the same number of subdivisions for mandoline on every rendering mode. It had too many subdivisions in rasterOrdering anyway and didn't render well. Since we aren't going to change it, just run tests that we are committed to keep rendering correctly.

Diffs=
a2b0cb230a Normalize GM tests (#8586)

added a new data converter by request from Duo.
It defines an input range and maps it to an output range.
It includes some extra features like clamping the value, applying a modulo in order to make it loop, and reversing it.
And it also takes an interpolator such that the mapping doesn't need to be lineal.
@

https://github.com/user-attachments/assets/d18b1269-8b5e-4aad-9504-9f5f3dcdaad9

Diffs=
88543fa792 Nnnn range data converter (#8585)

* Bundle 32-bit binaries in the test apk as well. Not all devices on browserstack are 64-bit.

* Save the gm pngs to different directories based on android os version. Android introduced a lot of changes to external storage at v11.

* Work around an invalid gl error being generatd by Moto G7 Play.

* Migrate the android_gms_browserstack runner to ubuntu 22.04. This is a long-running job that spends most its time waiting on browserstack and the network. It's not worth tying up a cirrus runner.

* Begin syncing android_gms_browserstack results with s3.

Diffs=
36e73f6b4b Fixup android_gms_browserstack (#8599)

This PR addresses a runtime limitation where we previously needed to trigger a pointer event to confirm if a listener existed at a specific location. With this update, users in game engines can choose whether pointer events in Rive should block raycasts from hitting items behind a Rive graphic. This change allows for hit testing without triggering the event itself.

I believe this is already available in Flutter, so this PR extends it to other runtimes as well

Diffs=
17474d3e2c feat: expose hit test to runtimes (#8598)

* added `-H` flag that uses a new histogram compare mode for checking diffs of images on `diff.py` and `image_diff.py`
* added `NO_VENV` environment variable check for disabling `pyenv` for local development in `diff.py`

Diffs=
807f9b20ac image diff hiistogram mode (#8601)

More experimenting on top of https://github.com/rive-app/rive/pull/8556

This breaks some APIs that still need to be fixed up elsewhere.

The basic concept here is that when you request a system font you can tell it whether you're requesting to use it a Harfbuzz shaped system font or a CoreText (System) shaped system font.

We prioritize Harfbuzz first so that we have predictable performance and results across edit and runtime. To do this the fallback process now comes with an index. Iteration will keep happening until no font is returned (or all glyphs are found).

If the registered fallbacks look like this:
```
RiveFont.fallbackFonts = [
  UIFont(name: "PingFangSC-Semibold", size: 12)!,
  UIFont(name: "Hiragino Sans", size: 12)!
]
```

The fallback process will return:
```
iter 0: Ping Fang Harfbuzz Shaped
iter 1: Hiragino San Harfbuzz Shaped
iter 2: Ping Fang Coretext Shaped
iter 3: Hiragino Sans Coretext Shaped
```

We also use Coretext last as usually shaping with Coretext causes Apple's shaper to do its own fallbacks (it calls them cascades like css). So iter 2 (in the example above) will be the final iteration as all glyphs get filled via Apple's own fallbacks.

![CleanShot 2024-11-14 at 21 16 55@2x](https://github.com/user-attachments/assets/c12e158b-5b8e-41d4-8d9b-8184a316a079)

Diffs=
3eefba5039 CoreText fallback shaper ex (#8568)

Co-authored-by: David Skuza <[email protected]>
Co-authored-by: Luigi Rosso <[email protected]>

currently it's 1000, which I think is too low (i was being conservative). This bumps it up to 10k

Diffs=
7be5f73b5a editor: update the repeat limit for n-slicing (#8604)