Bug 245817

Pull request: https://github.com/WebKit/WebKit/pull/4822

Committed 255020@main (c0bda7319ffb): <https://commits.webkit.org/255020@main>

Reviewed commits have been landed. Closing PR #4822 and removing active labels.

<rdar://problem/100593236>

Created attachment 462782 [details]
perf test

Using the attached performance test, I've observed the following numbers on my 14" M1 Max MacBookPro.

First, using polyfill for declarative shadow DOM, we observe that:

:Time -> [221, 210.0000000000001, 210, 218, 215, 218.99999999999977, 217.99999999999977, 233, 214, 208, 218, 232, 206, 210, 204.00000000000045, 207.00000000000045, 227, 192, 205, 215] ms
    mean: 214.1 ms
    median: 214.5 ms
    stdev: 9.840303803611327 ms
    min: 192 ms
    max: 233 ms


Using native declarative shadow DOM without streaming (i.e. same as Blink's implementation), we observe that:

:Time -> [163, 174, 198, 158, 157, 157.00000000000023, 158, 180, 157, 156, 158, 168, 159, 162.00000000000045, 158.00000000000045, 158, 169, 158.00000000000045, 160, 163.00000000000045] ms
    mean: 163.5500000000001 ms
    median: 158.50000000000023 ms
    stdev: 10.318481120377072 ms
    min: 156 ms
    max: 198 ms

This mean is 23.6% faster than polyfill.

Finally, using native declarative shadow DOM with streaming (i.e. shadow root is attached at the start tag instead of the end tag), we observe that:

:Time -> [142, 152.00000000000006, 167.99999999999994, 150, 142.0000000000001, 144.9999999999999, 177, 155, 147, 166, 144, 142, 142, 147, 185, 142, 143, 162.99999999999955, 141, 143] ms
    mean: 151.8 ms
    median: 145.99999999999994 ms
    stdev: 13.069047402163593 ms
    min: 141 ms
    max: 185 ms

This mean is 29.1% faster than polyfill and 7% faster than non-streaming version.

Created attachment 462783 [details]
standalone perf test