Restore the JsRuntimeHost pin lost in the master merge - #12
Merged
CedricGuillemet merged 4 commits intoAug 25, 2026
Merged
Conversation
Merging master into nativeDawn resolved the CMakeLists.txt conflict in favour of master's pin, reverting the branch from cb988ba (JsRuntimeHost master) back to 9271f13 from four weeks earlier. That silently dropped three fixes: BabylonJS#223 keep escaped handles alive when their escapable scope closes BabylonJS#225 QuickJS: report success from the napi_throw family BabylonJS#227 Fix V8 N-API leaking every promise ever created Losing BabylonJS#227 reintroduced the promise leak, and the full validation sweep went from completing to aborting partway through with peak private bytes climbing from 5.4 GB to 12.1 GB. before: ran=632 passed=628, peak 5397 MB merged: crash at 'Sponza Clustered Lighting (node material pbr)', exit 3, peak 12134 MB after: ran=632 passed=627, peak 5797 MB, 0 validation errors The remaining delta is Iridescence NME, a known flake that passes in isolation (0.250% against a 2.5% tolerance); the other four are the pre-existing Draco WASM LinkErrors. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 88569c10-a7ff-4373-9a58-afa9c68b8c09
Collaborator
Author
|
Upstream now carries this pin too: BabylonJS#1848 merged as That makes this PR a straight alignment with |
…e stale reasons Five tests were excluded on every graphics API because bgfx/D3D11 fails them, even though their own recorded reason said Dawn passes. Dropping WebGPU from excludedGraphicsApis keeps bgfx quarantined while restoring the coverage on Dawn: SOGS with SH no texture lookup 0.019% (allowed 5%) Gaussian Splatting Debug Part Test 1.530% (allowed 5%) Gaussian Splatting Debug Part Auto Tracking Test 2.190% (allowed 5%) Gaussian Splatting Part Serialize Test 0.748% (allowed 5%) needDepthPrePass extended material families exact match All five were re-measured in isolation and then in a full in-order sweep, so none of them depends on running alone. Three more carried the same "Dawn passes this test" claim but do not. Re-measured and recorded what they actually do: Gaussian Splatting GLTF 36.513% off (allowed 2.5%) Sprite maps 96.170% off (allowed 2.5%) GUI 4.066% off (allowed 4%) Sweep: ran=637 passed=633 failed=4, zero validation errors. The four failures are the pre-existing Draco WASM LinkErrors. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 88569c10-a7ff-4373-9a58-afa9c68b8c09
The device descriptor requested no optional features at all. Features are opt-in exactly like limits, and requestDevice() hands the JS layer this one pre-created device, so every optional capability the hardware had was permanently invisible to Babylon.js. The most visible casualty was float32-filterable. Without it r32float and rgba32float textures are UnfilterableFloat, so pairing one with a filtering sampler fails validation with "None of the supported sample types (UnfilterableFloat) ... match the expected sample types (Float)", taking the whole render pass and command buffer down with it. Across the OpenPBR anisotropy tests that was 264 validation errors; it is now 0, and 18 of those 23 tests render correctly (several pixel-exact) where all 23 failed before. Dawn's own extensions start at 0x00050000 and are excluded: several change device semantics or need matching toggles. Full sweep: ran=636 passed=632, the 4 failures being the pre-existing Draco WASM LinkError tests. No measurable runtime cost (7m23s vs 7m48s before). Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 88569c10-a7ff-4373-9a58-afa9c68b8c09
Requesting the adapter's standard features unblocked a large batch of tests that were globally excluded. Re-probing all 180 globally-excluded tests with features enabled found 85 now rendering correctly, against 21 on the same probe before that change -- the new passes are almost entirely float-texture work (IBL/CDF prefiltering, OpenPBR, voxel and screen-space shadowing, IES profiles), several of whose exclusion reasons literally read "CDF renderer / IBL prefiltering fails with 'FLOAT' undefined". These stay excluded on bgfx, where they still crash, hang or mis-render, so each entry swaps excludeFromAutomaticTesting for the non-WebGPU API list and keeps its original reason. Two of the 85 are deliberately left quarantined: "SOGS with SH" and "Draco Mesh Compression (fallback)" pass in an excluded-only chunk but their own reasons already document them aborting or drifting after earlier tests in a full run, which the sweep confirmed. Full in-order sweep: ran=719 passed=715, up from ran=636 passed=632. The 4 failures are the same pre-existing Draco WASM ones, unchanged by this commit. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 88569c10-a7ff-4373-9a58-afa9c68b8c09
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The
nativeDawnbranch currently aborts partway through the validation sweep. Mergingmasterin resolved theCMakeLists.txtconflict in favour of master's JsRuntimeHost pin, reverting this branch fromcb988ba(JsRuntimeHost master) to9271f13from four weeks earlier and silently dropping three fixes:napi_throwfamilyLosing BabylonJS#227 reintroduced the promise leak. The sweep went from completing to aborting, with peak private bytes more than doubling.
ran=632 passed=628Sponza Clustered Lighting (node material pbr)ran=632 passed=627, 0 validation errorsThe one-test delta is
Iridescence NME, a known flake that passes in isolation at 0.250% against a 2.5% tolerance. The other four failures are the pre-existing Draco WASMLinkErrors.cb988bais current JsRuntimeHostmaster, so this is a restore rather than a bump. Note thatBabylonJS/BabylonNativemasteris still on9271f13and carries the same promise leak on the V8 path.