1991689 - 1167.55 - 5.17% perf_reftest_singletons svg-text-textLength-1.html / perf_reftest_singletons thai-reflow.html + 10 more (Linux, OSX, Windows) regression on Tue September 23 2025

Status: RESOLVED FIXED

(Core :: Layout: Text and Fonts, defect, P3)

Description

[Treeherder Bug Filer]

Perfherder has detected a talos performance regression from push ff7bf296a64e46a6e33cc3cd0ebe683d8ae8556f. As author of one of the patches included in that push, we need your help to address this regression.

Please acknowledge, and begin investigating this alert within 3 business days, or the patch(es) may be backed out in accordance with our regression policy. Our guide to handling regression bugs has information about how you can proceed with this investigation.

If you have any questions or need any help with the investigation, please reach out to [e-mail]. Alternatively, you can find help on Slack by joining #perf-help, and on Matrix you can find help by joining #perftest.

Regressions:

Ratio	Test	Platform	Options	Absolute values (old vs new)
1168%	perf_reftest_singletons svg-text-textLength-1.html	macosx1470-64-shippable	e10s fission stylo webrender	1,428.31 -> 18,104.47
1036%	perf_reftest_singletons svg-text-textLength-1.html	linux1804-64-shippable-qr	e10s fission stylo webrender	1,315.87 -> 14,953.32
513%	perf_reftest_singletons svg-text-textLength-1.html	windows11-64-24h2-shippable	e10s fission stylo webrender	1,423.46 -> 8,724.13
21%	perf_reftest_singletons resize-lengthy-textarea.html	macosx1470-64-shippable	e10s fission stylo webrender	134.58 -> 162.93
17%	perf_reftest_singletons bidi-resolution-1.html	linux1804-64-shippable-qr	e10s fission stylo webrender	209.28 -> 244.37
14%	perf_reftest_singletons bidi-resolution-1.html	windows11-64-24h2-shippable	e10s fission stylo webrender	68.94 -> 78.52
13%	perf_reftest_singletons resize-lengthy-textarea.html	linux1804-64-shippable-qr	e10s fission stylo webrender	156.01 -> 175.67
11%	perf_reftest_singletons resize-lengthy-textarea.html	windows11-64-24h2-shippable	e10s fission stylo webrender	103.02 -> 114.68
11%	perf_reftest_singletons insert-subtree-not-nth-edge-has-pseudo.html	macosx1470-64-shippable	e10s fission stylo webrender	14.75 -> 16.31
6%	perf_reftest_singletons thai-reflow.html	macosx1470-64-shippable	e10s fission stylo webrender	75.76 -> 80.67
5%	perf_reftest_singletons thai-reflow.html	linux1804-64-shippable-qr	e10s fission stylo webrender	73.47 -> 77.29
5%	perf_reftest_singletons thai-reflow.html	windows11-64-24h2-shippable	e10s fission stylo webrender	30.69 -> 32.27

Improvements:

Ratio	Test	Platform	Options	Absolute values (old vs new)
12%	perf_reftest_singletons remove-children-positional-2.html	windows11-64-24h2-shippable	e10s fission stylo webrender	130.13 -> 114.08

Details of the alert can be found in the alert summary, including links to graphs and comparisons for each of the affected tests.

If you need the profiling jobs you can trigger them yourself from treeherder job view or ask [e-mail] to do that for you.

You can run all of these tests on try with ./mach try perf --alert 46941

The following documentation link provides more information about this command.

Comment 1

[BugBot [:suhaib / :marco/ :calixte]]

Set release status flags based on info from the regressing bug 1988596

Comment 2

[Ting-Yu Lin [:TYLin] (PDT, UTC-7)]

This is a nightly-only regression, and is expected per bug 1988596 comment 15.

Here is a performance comparison of current Nightly vs Nightly + bug 1987997.

With Bug 1987997 lands, it will bring the tests performance on par with Nightly builds with text-autospace disabled.

Comment 3

[Emilio Cobos Álvarez [:emilio]]

Seems like we're going to do no-autospace by default which will fix this.

Comment 4

[Jonathan Kew [:jfkthame]]

By far the biggest regressions here are the SVG text ones. It looks like the SVGTextFrame code has some pretty bad performance characteristics, which get magnified by the added work of text-autospace processing. The good news is that I think we can mitigate a lot of this with some judicious caching, so that we don't constantly re-create PropertyProvider instances and re-measure the same ranges of text over and over.

I've got some WIP patches here, waiting on tryserver to see how things go...

Comment 5

[Jonathan Kew [:jfkthame]]

Attachment: Bug 1991689 - Cache advances of text ranges measured by TextRenderedRun::GetClipEdges in the root SVGTextFrame to accelerate ReflowSVG. r=#layout-jp-market-reviewers

Comment 6

[Jonathan Kew [:jfkthame]]

Attachment: Bug 1991689 - Cache a PropertyProvider in SVGTextFrame so that we don't have to create a fresh one for every call to the same textframe. r=#layout-jp-market-reviewers

Comment 7

[Jonathan Kew [:jfkthame]]

With these patches, we end up with substantially better results on the svg-text-textLength-1 test than before text-autospace was landed:

https://perf.compare/subtests-compare-results?baseRev=30b09d0a0bc99a7fc15db39635d26b4242ea3c55&baseRepo=try&newRev=c2d4a071d54eb8d5e047e92faea2df0d8c6c2886&newRepo=try&framework=1&baseParentSignature=4763178&newParentSignature=4763178

(The try run seems to show a few other improvements as well, though I'm unsure if these are real given that the patches here should only affect SVG. The recently-landed patches in bug 1991943 may have helped some of those other tests a bit, though.)

Comment 8

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/4bc276362491
https://hg.mozilla.org/integration/autoland/rev/e38bccd1f4c1
Cache advances of text ranges measured by TextRenderedRun::GetClipEdges in the root SVGTextFrame to accelerate ReflowSVG. r=firefox-svg-reviewers,longsonr
https://github.com/mozilla-firefox/firefox/commit/aa124118714b
https://hg.mozilla.org/integration/autoland/rev/66e00f899737
Cache a PropertyProvider in SVGTextFrame so that we don't have to create a fresh one for every call to the same textframe. r=firefox-svg-reviewers,longsonr

Comment 9

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/5416d09b794a
https://hg.mozilla.org/integration/autoland/rev/05cf3e73d4f6
Revert "Bug 1991689 - Cache a PropertyProvider in SVGTextFrame so that we don't have to create a fresh one for every call to the same textframe. r=firefox-svg-reviewers,longsonr" for causing multiple failures.

Comment 10

[Iulian Moraru]

Revert for causing multiple failures.

• Push with failures

• Failure log for bc failures

• Failure log for wpt failures

• Failure log for wpt failures 2

• Backout link

Comment 11

[Jonathan Kew [:jfkthame]]

Ah, I see -- we need to be more careful to avoid the risk of using a stale PropertyProvider, as things like style changes might have invalidated it. I think I have this fixed locally, but will wait for more Try results before updating and re-landing.

Comment 12

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/cc8d67d595ce
https://hg.mozilla.org/integration/autoland/rev/d39196d22690
Cache advances of text ranges measured by TextRenderedRun::GetClipEdges in the root SVGTextFrame to accelerate ReflowSVG. r=firefox-svg-reviewers,longsonr
https://github.com/mozilla-firefox/firefox/commit/56772201997e
https://hg.mozilla.org/integration/autoland/rev/c4911116c376
Cache a PropertyProvider in SVGTextFrame so that we don't have to create a fresh one for every call to the same textframe. r=firefox-svg-reviewers,longsonr

Comment 13

[Serban Stanca [:SerbanS]]

https://hg.mozilla.org/mozilla-central/rev/d39196d22690
https://hg.mozilla.org/mozilla-central/rev/c4911116c376