[Bryan Chen (bryanchen-d)]

Main CI 491287 failed on macOS ARM64 Run unit tests (Browser, Webkit) with Animations › creates consecutive pixelated rainbow confetti animations, where allParticlesUseBalancedTiming was false. The same commit passed in the previous main CI.

Cause

triggerConfettiAnimation computes each keyframe offset as time / duration, using the full-precision random duration (for example 1490.6732…). The test multiplies the offset step by effect.getTiming().duration and requires the result to be <= 40 + 1e-10. WebKit reports that duration rounded to whole microseconds (1490.673). Whenever the rounding goes up, each 40 ms step reads as up to about 1e-5 ms longer, which fails the 1e-10 tolerance. The duration is random, so the test fails about half the time on WebKit. Chromium and Electron report the exact value.

I confirmed this in local WebKit with a temporary probe over 40 animations. Every failing case had d with exactly 3 decimals and maxGap − 40 between 5e-6 and 1.3e-5, while easing, keyframe counts and duration bounds were all fine.

Change (test only)

The tolerance goes from 1e-10 to 1e-3 ms, which covers microsecond rounding, and a comment says why. That's still 4 orders of magnitude tighter than anything visible, so the test still checks that the steps are 40 ms. The product is unchanged.

Validation (local, --run animations.test.ts)

• Before, WebKit: 5 of 10 runs failed.
• After:
  • WebKit: 0 of 20 runs failed.
  • Chromium: 0 of 20 runs failed.
  • Electron: 9 passing.
• eslint is clean.

Authored with Copilot

[Bryan Chen (bryanchen-d)]

Lee Murray (@mrleemurray) this is a test-only follow-up to #345182. The confetti timing check failed on main CI's WebKit job (491287). WebKit reports getTiming().duration rounded to whole microseconds, but the keyframe offsets come from the unrounded random duration, so a 40 ms step can read as about 1e-5 ms over the test's 1e-10 tolerance. That happens about half the time. This PR loosens the tolerance to 1e-3 ms. The local probe and before/after runs are in the description.

Co-authored with Copilot

[Copilot]

🟢 Approval recommended

No active review issues were identified for the test-only adjustment accommodating WebKit duration rounding.

No open issues 🎉

What changed in this PR

This test-only change addresses intermittent WebKit failures in the confetti animation timing test by allowing for microsecond rounding of reported durations. Production animation behavior is unchanged.

Changes:

• Increases the timing tolerance from 1e-10 to 1e-3 ms while retaining the existing timing checks.
• Adds a comment explaining why WebKit needs the tolerance.

File	Description
src/​vs/​base/​test/​browser/​ui/​animations/​animations.test.ts	Adjusts the confetti timing assertion tolerance and documents WebKit’s duration rounding.

_{Review effort: Balanced}

[Lee Murray (mrleemurray)]

Bryan Chen (@bryanchen-d) gotcha, thanks for updating this!

[Copilot]

🟢 Approval recommended

No active review issues were identified for the test-only adjustment accommodating WebKit duration rounding.

No open issues 🎉

_{Review effort: Balanced}

[Bryan Chen (bryanchen-d)]

Closing this as a duplicate of #345214, which merged the same 1e-3 tolerance change to animations.test.ts. Thanks Lee Murray (@mrleemurray) for reviewing.

Co-authored with Copilot