[html5] Add `desynchronized` WebGL context attribute by UnknownHacker1 · Pull Request #27163 · emscripten-core/emscripten

UnknownHacker1 · 2026-06-23T14:30:16Z

Adds a desynchronized field to EmscriptenWebGLContextAttributes. When set, emscripten_webgl_create_context() passes both desynchronized and its legacy name lowLatency to canvas.getContext(), and emscripten_webgl_get_context_attributes() reads the value back so it round-trips. Defaults to false via the existing memset in emscripten_webgl_init_context_attributes(). Regenerated struct_info (wasm32 + wasm64) and documented the new field.

Tested manually in Chrome: a context created with desynchronized=true reads back 1; default is 0.

On the test: happy to add one, but since whether a browser honors desynchronized is platform-dependent, asserting the readback value could be flaky in CI. Would a test that hooks getContext to check the requested attribute be preferred? Open to guidance.

Fixes #8406.

Add a `desynchronized` field to EmscriptenWebGLContextAttributes. When set, emscripten_webgl_create_context() passes both `desynchronized` and its legacy name `lowLatency` to canvas.getContext(), and emscripten_webgl_get_context_attributes() reads the value back. It defaults to false via the existing memset in emscripten_webgl_init_context_attributes. Regenerated struct_info and documented the new field.

sbc100

You can use is_chrome() in test/test_browser.py to have the test assert different things on chrome to other browsers.

sbc100 · 2026-06-23T16:10:50Z

      'failIfMajorPerformanceCaveat': !!HEAP8[attributes + {{{ C_STRUCTS.EmscriptenWebGLContextAttributes.failIfMajorPerformanceCaveat }}}],
+      'desynchronized': !!HEAP8[attributes + {{{ C_STRUCTS.EmscriptenWebGLContextAttributes.desynchronized }}}],
+      // 'lowLatency' is the legacy name for 'desynchronized'.
+      'lowLatency': !!HEAP8[attributes + {{{ C_STRUCTS.EmscriptenWebGLContextAttributes.desynchronized }}}],


What browser versions require/use the legacy name? Perhaps you could add a comment here linking to caniuse data and then if an #if MIN_CHROME_VERSION < XXX || MIN_FIREFOX_VERSION < YYY around this line.

This will allow is us to one day completely delete this line once we drop support for those older browsers.

Add an assertion to the html5_webgl browser test that the desynchronized attribute round-trips on Chrome, where it is supported, and reads back false on browsers that do not, selected via is_chrome(). lowLatency was Chrome's name for desynchronized in versions 73-74, so only emit it when targeting Chrome < 75.

UnknownHacker1 · 2026-06-23T16:59:46Z

Thanks for your review @sbc100

Done both:

Added a desynchronized round-trip check to the html5_webgl test. It asserts the value comes back true on Chrome and false elsewhere, via is_chrome().
Gated the lowLatency line behind #if MIN_CHROME_VERSION < 75 (it was Chrome's 73-74 name for desynchronized) with a caniuse link, so it's dropped by default and can be deleted once we stop supporting Chrome < 75.

sbc100 · 2026-06-23T17:54:11Z

+      // 'lowLatency' was Chrome's experimental name for 'desynchronized', renamed in Chrome 75.
+      // See https://caniuse.com/?search=desynchronized. Drop once we no longer support Chrome < 75.
+      'lowLatency': !!HEAP8[attributes + {{{ C_STRUCTS.EmscriptenWebGLContextAttributes.desynchronized }}}],
+#endif


Ah, since OLDEST_SUPPORTED_CHROME is currently 85 (see feature_matrix.py) we can just delete this !

sbc100

lgtm with that block removed

sbc100 · 2026-06-23T17:55:35Z

Thanks for picking up that old bug and fixing it! Much appreciated.

UnknownHacker1 · 2026-06-23T18:01:14Z

@sbc100

Removed, thanks! And thanks for the quick reviews.

sbc100 · 2026-06-23T18:04:09Z


+  .. c:member:: bool desynchronized
+
+    If ``true``, requests a "desynchronized" WebGL context, which can lower latency by bypassing the browser's normal compositing/double-buffering of the canvas. This maps to the ``desynchronized`` attribute on the JavaScript ``getContext()`` call (and its legacy name ``lowLatency``). Whether the request is honored depends on the browser and platform; use ``emscripten_webgl_get_context_attributes()`` to read back the value the context was actually created with. Default value is ``false``.


maybe remove the reference to the legacy name here?

UnknownHacker1 · 2026-06-23T18:11:25Z

@sbc100 Done. I think you need to re-enable auto-merge

UnknownHacker1 · 2026-06-23T19:30:45Z

@sbc100 The Codesize check is just the expected webgl sizes shifting from the added glue. Should I rebaseline, or do we just merge through?

UnknownHacker1 · 2026-06-23T19:31:43Z

the error message says that "-- This failure is only a warning and can be ignored"

sbc100 · 2026-06-23T19:47:29Z

Yes, please rebasline.

Do to this you need to make sure you have the very latest emsdk version with emsdk install tot. Then run tools/maint/rebaseline_tests.py.

…nCanvas

sbc100 · 2026-06-23T22:30:33Z

I you like I can push the codesize updates to this branch? (assuming I have access)

UnknownHacker1 · 2026-06-23T22:35:46Z

@sbc100

Uppdate on two CI things:

Browser tests: fixed the chrome-2gb / wasm64-4gb failures. Those are the OffscreenCanvas variants, and desynchronized isn't honored on OffscreenCanvas, so my round-trip check was asserting the wrong value there. It now only runs on a normal canvas.

Codesize: this one looks like it's failing on the target branch, not this PR. The run checks out main and reports 10 expectation files out of date from LLVM drift (the cxx_* tests, -46 bytes each), ending with "run the Rebaseline Tests github action on the target branch (main)." My change's own sizes all matched. Could you run that on main, or merge through?

UnknownHacker1 · 2026-06-23T22:36:49Z

I you like I can push the codesize updates to this branch? (assuming I have access)

Sorry, didn't see your reply. Yes, that'd be great.

sbc100 · 2026-06-23T22:57:51Z

Oh wait.. the expectation need updating on main. I will take care of that.

sbc100 · 2026-06-23T23:16:16Z

#27170 is now landed.

Can you rebase and update the codesize expectations again? (you might also need to ./emcc --clear-cache)

…ized-canvas # Conflicts: # test/codesize/test_codesize_hello_dylink_all.json

UnknownHacker1 · 2026-06-24T13:26:46Z

@sbc100 The dylink_all codesize baseline keeps conflicting as main drifts. Since you've approved, could you just merge it (or push the codesize update from your side)? I can rebase again but it'll likely drift before it lands.

UnknownHacker1 · 2026-06-24T15:12:06Z

@sbc100 codesize check failed again

sbc100 reviewed Jun 23, 2026

View reviewed changes

sbc100 approved these changes Jun 23, 2026

View reviewed changes

sbc100 changed the title ~~Add desynchronized WebGL context attribute (#8406)~~ [html5] Add desynchronized WebGL context attribute Jun 23, 2026

Remove lowLatency: oldest supported Chrome (85) uses desynchronized

7feb9b3

sbc100 enabled auto-merge (squash) June 23, 2026 18:03

sbc100 reviewed Jun 23, 2026

View reviewed changes

Drop lowLatency mention from docs

1722b34

auto-merge was automatically disabled June 23, 2026 18:08
Head branch was pushed to by a user without write access

sbc100 enabled auto-merge (squash) June 23, 2026 18:23

sbc100 disabled auto-merge June 23, 2026 18:23

sbc100 enabled auto-merge (squash) June 23, 2026 18:23

Rebaseline webgl codesize expectations for the desynchronized attribute

f108d80

auto-merge was automatically disabled June 23, 2026 20:34
Head branch was pushed to by a user without write access

sbc100 enabled auto-merge (squash) June 23, 2026 21:05

Only check desynchronized round-trip on a normal canvas, not Offscree…

31f7331

…nCanvas

auto-merge was automatically disabled June 23, 2026 21:38
Head branch was pushed to by a user without write access

Rebaseline hello_dylink_all codesize for the desynchronized attribute

6d5f06d

UnknownHacker1 added 2 commits June 24, 2026 02:22

Merge remote-tracking branch 'upstream/main' into fix/8406-desynchron…

9a0e8f8

…ized-canvas # Conflicts: # test/codesize/test_codesize_hello_dylink_all.json

Rebaseline dylink codesize for the desynchronized attribute

a551bfe


		.. c:member:: bool desynchronized

		If ``true``, requests a "desynchronized" WebGL context, which can lower latency by bypassing the browser's normal compositing/double-buffering of the canvas. This maps to the ``desynchronized`` attribute on the JavaScript ``getContext()`` call (and its legacy name ``lowLatency``). Whether the request is honored depends on the browser and platform; use ``emscripten_webgl_get_context_attributes()`` to read back the value the context was actually created with. Default value is ``false``.

Conversation

UnknownHacker1 commented Jun 23, 2026 • edited by sbc100 Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

sbc100 left a comment

Choose a reason for hiding this comment

Uh oh!

sbc100 Jun 23, 2026

Choose a reason for hiding this comment

Uh oh!

UnknownHacker1 commented Jun 23, 2026

Uh oh!

sbc100 Jun 23, 2026

Choose a reason for hiding this comment

Uh oh!

sbc100 left a comment

Choose a reason for hiding this comment

Uh oh!

sbc100 commented Jun 23, 2026

Uh oh!

UnknownHacker1 commented Jun 23, 2026

Uh oh!

sbc100 Jun 23, 2026

Choose a reason for hiding this comment

Uh oh!

UnknownHacker1 commented Jun 23, 2026

Uh oh!

UnknownHacker1 commented Jun 23, 2026

Uh oh!

UnknownHacker1 commented Jun 23, 2026

Uh oh!

sbc100 commented Jun 23, 2026

Uh oh!

sbc100 commented Jun 23, 2026

Uh oh!

UnknownHacker1 commented Jun 23, 2026

Uh oh!

UnknownHacker1 commented Jun 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

sbc100 commented Jun 23, 2026

Uh oh!

sbc100 commented Jun 23, 2026

Uh oh!

UnknownHacker1 commented Jun 24, 2026

Uh oh!

UnknownHacker1 commented Jun 24, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

UnknownHacker1 commented Jun 23, 2026 •

edited by sbc100

Loading

UnknownHacker1 commented Jun 23, 2026 •

edited

Loading