Test failures in IE 11

[mramato]

After #2859, there are still 5 test failures in IE 11 that we should investigate

Renderer/CubeMap draws with a cube map with premultiplied alpha
Renderer/Texture renders with premultiplied alpha
Scene/Pick drill picks multiple objects
Scene/Pick does not drill pick when show is false
Scene/Primitive can disable picking when asynchronous

[pjcozzi]

With Cesium 1.17, there's two need test failures in IE 11:

• Scene/WebMapServiceImageryProvider pickFeatures works with THREDDS XML format
• Scene/WebMapServiceImageryProvider pickFeatures works with msGMLOutput format

[kring]

@RacingTadpole can you take a look at this?

[RacingTadpole]

Sure. For my future ref, I've just tried the tests in a Win8 IE10 environment, and get the error:
Failed: promise rejected: TypeError: Unable to get property 'value' of undefined or null reference. The offending code is in https://github.com/AnalyticalGraphicsInc/cesium/blob/master/Specs/Scene/WebMapServiceImageryProviderSpec.js#L559 and https://github.com/AnalyticalGraphicsInc/cesium/blob/master/Source/Scene/GetFeatureInfoFormat.js#L230 .
I'll take a look soon.

[RacingTadpole]

Oops, sorry. Yes, I only fixed the two new failures, THREDDS XML and msGMLOutput.

[hpinkos]

New failures in 1.18:
Scene/DeviceOrientationCameraController rotates for alpha
Scene/DeviceOrientationCameraController rotates for beta
Scene/DeviceOrientationCameraController rotates for gamma

@bagnell?

[lilleyse]

The errors I see in 1.19:

These tests work when run individually:
Scene/Pick drill picks multiple objects
Scene/Pick does not drill pick when show is false

These tests are really close to the expected results, seems like precision / rounding issues.
Renderer/CubeMap draws with a cube map with premultiplied alpha
Renderer/Texture renders with premultiplied alpha

Renderer/CubeMap fails to generate mipmaps (width) is failing because IE does not support non-power-of-two cube maps (source)

[hpinkos]

New failure in 1.24
Scene/GroundPrimitive creating a synchronous primitive throws if initializeTerrainHeights wasn't called

[lilleyse]

Failures in 1.25

From before:

Renderer/CubeMap draws with a cube map with premultiplied alpha
Renderer/Texture renders with premultiplied alpha
Scene/GroundPrimitive creating a synchronous primitive throws if initializeTerrainHeights wasn't called

New ones:
DataSources/KmlDataSource when a LineString is clamped to ground and tesselated, entity has a corridor geometry and ColorProperty (Object.assign not supported in IE11)

Widgets/Viewer/Viewer does not crash when tracking an object with a position property whose value is undefined.

These two only pass when run individually:
Scene/GeometryRendering WallGeometry pick
Scene/GeometryRendering Extruded PolygonGeometry pick

These all seem to be related:
DataSources/RectangleGeometryUpdater color material sets onTerrain to true
DataSources/RectangleGeometryUpdater fill is true sets onTerrain to true
DataSources/RectangleGeometryUpdater dynamic updater on terrain
DataSources/RectangleGeometryUpdater Checks that an entity without height and extrudedHeight and with a color material is on terrain
DataSources/PolygonGeometryUpdater color material sets onTerrain to true
DataSources/PolygonGeometryUpdater fill is true sets onTerrain to true
DataSources/PolygonGeometryUpdater dynamic updater on terrain
DataSources/PolygonGeometryUpdater Checks that a polygon without per position heights is on terrain
DataSources/PolygonGeometryUpdater Checks that an entity without height and extrudedHeight and with a color material is on terrain
DataSources/GeometryVisualizer StaticGroundGeometryColorBatch updates color attribute after rebuilding primitive
DataSources/EllipseGeometryUpdater color material sets onTerrain to true
DataSources/EllipseGeometryUpdater fill is true sets onTerrain to true
DataSources/EllipseGeometryUpdater dynamic updater on terrain
DataSources/EllipseGeometryUpdater Checks that an entity without height and extrudedHeight and with a color material is on terrain
DataSources/CorridorGeometryUpdater color material sets onTerrain to true
DataSources/CorridorGeometryUpdater fill is true sets onTerrain to true
DataSources/CorridorGeometryUpdater dynamic updater on terrain
DataSources/CorridorGeometryUpdater Checks that an entity without height and extrudedHeight and with a color material is on terrain

[lilleyse]

Now the only failures that happen consistently are:

Renderer/CubeMap draws with a cube map with premultiplied alpha
Renderer/Texture renders with premultiplied alpha

Other test fails when run with all, but pass when run individually.

[bagnell]

Test failures in 1.26

Scene/LabelCollection can render after adding a label
Scene/GroundPrimitive renders with distance display condition per instance attribute

These tests pass when run individually:
Scene/GeometryRendering PolygonGeometry pick
Scene/GeometryRendering Extruded PolygonGeometry pick

[tfili]

New test failures in 1.28

DataSources/CorridorGeometryUpdater A time-varying color causes ground geometry to be dynamic
DataSources/EllipseGeometryUpdater A time-varying color causes ground geometry to be dynamic
DataSources/PolygonGeometryUpdater A time-varying color causes ground geometry to be dynamic
DataSources/RectangleGeometryUpdater A time-varying color causes ground geometry to be dynamic
Scene/LabelCollection renders label with distanceDisplayCondition

@bagnell I'm assuming the first 4 are because of lack of ground primitives in IE and the other one caused by the same issue as Scene/LabelCollection can render after adding a label?

[bagnell]

I'm assuming the first 4 are because of lack of ground primitives in IE and the other one caused by the same issue as Scene/LabelCollection can render after adding a label

Yes, but don't we check if ground primitives are supported and fall back to regular primitives?

[tfili]

@bagnell That's exactly whats happening. It's falling back to a regular primitive. The test will get different results depending whether a Primitive or GroundPrimitive is used. Behavior is correct, the test is just broken for browsers without GroundPrimitive support.

[hpinkos]

New test failure: Renderer/Context get the fragment depth extension

There's also a bunch of tests that fail when you try to run all of the Scene tests at the same time. Looks like something is going weird with the context? RuntimeError: WebGL Error: CONTEXT_LOST_WEBGL lost: undefined(17408);
Those tests pass individually, however.

[lilleyse]

Errors I'm seeing now:

Core/loadCRN returns a promise that resolves to an compressed texture when the request loads - @bagnell can you check that crn textures are working in IE
Renderer/CubeMap fails to generate mipmaps (width)
Scene/GeometryRendering PolygonGeometry pick
Scene/GeometryRendering Extruded PolygonGeometry pick

And then many CONTEXT_LOST_WEBGL errors like above - all in PointPrimitiveCollection

[hpinkos]

New test failures for 1.36:

• Scene/Expression constructs regex
• Scene/Cesium3DTileset verify batched points statistics
• Scene/Cesium3DTileset debugColorizeTiles for pnts with batch table
• 7 Test failures in Scene/PointCloud3DTileContent
• 15 test failures in Scene/ShadowMap

[pjcozzi]

In the upcoming 1.41 release on Windows, IE 11, I just get:

• Scene/Model renders textured box with embedded binary CRN texture
• Scene/Model raises animation start, update, and stop events when removeOnStop is true

[hpinkos]

I'm seeing a ton of test failures for 1.43

• DataSources/CzmlDataSource checks validation document
• Renderer/freezeRenderState works as expected
• Scene/ClassificationPrimitive renders with invert classification and an opaque color
• Scene/GroundPrimitive renders with invert classification and an opaque color
• Scene/PointCloud3DTileContent rebuilds shader style when expression changes
• Scene/Vector3DTilePoints renders a point
• 28 Geometry3DTileContent render tests
• 12 Vector3DTileGeometry render tests
• 15 test failures in Scene/ShadowMap

A lot of the failures look like this:
Expected to render [0,0,255,255], but actually rendered [[object Uint8Array]].
It seems like those should be pretty easy to fix

[shunter]

The validation document failure is strange. I see it too both in IE11 and Edge. Has this failed in previous releases? The failure is a tiny rounding error in two cartesians converted from spherical coordinates, but essentially it's doing the same operation twice and getting different results.

[hpinkos]

@shunter yeah that's really weird. it looks like it started failing in the 1.38 release

[hpinkos]

Actually, the same test fails in firefox too. I'm going to open a separate issue for that one

[lilleyse]

23 failures now. Some of them actually don't look too bad, makes me optimistic that we may eventually hit zero. This was run pre-Draco updates (https://github.com/AnalyticalGraphicsInc/cesium/pull/6405/files)

• Renderer/freezeRenderState works as expected
• Scene/Expression constructs regex
• Core/CircleGeometry computing rectangle property
• Scene/GoogleEarthEnterpriseImageryProvider resolves readyPromise
• Scene/GoogleEarthEnterpriseImageryProvider resolves readyPromise with Resource
• Scene/GoogleEarthEnterpriseImageryProvider returns false for hasAlphaChannel
• Scene/GoogleEarthEnterpriseImageryProvider can provide a root tile
• Widgets/CesiumWidget/CesiumWidget resizing triggers a render in requestRender mode
• Scene/BillboardCollection picks a billboard using scaleByDistance
• Scene/ClassificationPrimitive renders with invert classification and an opaque color
• Scene/GroundPrimitive renders with invert classification and an opaque color
• Scene/Model renders textured box with embedded binary CRN texture
• Scene/Model raises animation start, update, and stop events when removeOnStop is true
• Scene/Model throws runtime error when loading a gltf with uint32 indices if OES_element_index_uint is disabled
• Scene/Model loads a glTF with KHR_draco_mesh_compression extension
• Scene/Model loads a glTF with KHR_draco_mesh_compression extension with integer attributes
• Scene/Model error decoding a draco compressed glTF causes model loading to fail
• Scene/Model loads a draco compressed glTF and dequantizes in the shader
• Scene/Model loads a draco compressed glTF and dequantizes in the shader, skipping generic attributes
• Scene/PointCloud3DTileContent rebuilds shader style when expression changes
• Scene/PointCloud3DTileContent Rebuilds shaders when clipping planes are enabled, change between union and intersection, or change count
• Scene/Vector3DTilePoints renders a point
• Renderer/CubeMap fails to generate mipmaps (width)

[hpinkos]

From the 1.50 release:

• A bunch of dynamic DataSources/GeometryVisualizer tests Polygon Intervals and Availability broken in IE #7103
• Scene/ClassificationPrimitive update throws when an incompatible Appearance is set (likely related to above)
• Scene/ClassificationPrimitive renders with invert classification and an opaque color
• Scene/GroundPrimitive renders with invert classification and an opaque color
• Scene/Model renders textured box with embedded binary CRN texture
• Scene/Pick drillPick can drill pick batched Primitives with show attribute
• Scene/Pick drillPick can drill pick batched Primitives without show attribute
• Scene/Pick pickFromRay picks the globe
• Scene/Vector3DTileGeometry renders with inverted classification
• Scene/Vector3DTilePoints renders a point
• Scene/Vector3DTilePolygons renders with inverted classification

[lilleyse]

From the 1.51 release:

• Renderer/freezeRenderState works as expected
• Scene/Expression constructs regex
• Widgets/CesiumWidget/CesiumWidget resizing triggers a render in requestRender mode
• Scene/BillboardCollection picks a billboard using scaleByDistance
• Scene/BillboardCollection picks a billboard using translucencyByDistance
• Scene/ClassificationPrimitive renders with invert classification and an opaque color
• Scene/ClassificationPrimitive update throws when an incompatible Appearance is set
• Scene/Model renders textured box with embedded binary CRN texture
• Scene/Model raises animation start, update, and stop events when removeOnStop is true
• Scene/Model loads a glTF 2.0 with doubleSided set to true
• Scene/Vector3DTileGeometry renders with inverted classification
• Scene/Vector3DTilePolygons renders with inverted classification
• Scene/Vector3DTilePoints renders a point
• Widgets/Viewer/Viewer does not crash when tracking an object with a position property whose value is undefined.

These tests pass when run individually:

• Scene/BillboardCollection picks a billboard using scaleByDistance
• Scene/BillboardCollection picks a billboard using translucencyByDistance
• Scene/Model raises animation start, update, and stop events when removeOnStop is true
• Scene/Model loads a glTF 2.0 with doubleSided set to true

[likangning93]

I'm seeing Scene/GroundPrimitive renders with invert classification and an opaque color in 1.51 as well, but that might be associated with Scene/ClassificationPrimitive renders with invert classification and an opaque color.

[mramato]

Just a reminder that we ow run IE11 tests as part of CI, so all non-webgl should be passing but aren't (15 failures): https://ci.appveyor.com/project/AnalyticalGraphics/cesium/

I'm not sure how useful this GitHub issue is anymore, unless we made a concerted effort to clean up WebGL IE11 failures.

[ggetz]

New ones as of 1.54:

• Core/EllipsoidRhumbLine intersection with longitude handles N-S lines with different longitude
• Renderer/freezeRenderState works as expected
• A host of Renderer/BuiltinFunctions relating to oct encoding/IBL
• Renderer/CubeMap sub copies images to a cube map
• Renderer/CubeMap sub copies array buffers to a cube map
• Scene/BillboardCollection is picked
• Scene/Cesium3DTileset most detailed height queries tests
• Scene/Model renders textured box with embedded binary CRN texture
• Scene/Pick drillPickFromRay tests
• Scene/PolylineCollection renders with model matrix

[ggetz]

Closing due to dropped support for IE11