Improved performance with large HTML files

[sapphi-red]

Description

I reduced copying AST like #11108.

On this case, the result of yarn run perf:benchmark became about 8.6x faster.

before

[debug] resolve config from 'test.html'
[debug] loaded options `{"useTabs":false,"tabWidth":2,"endOfLine":"lf"}`
[debug] applied config-precedence (cli-override): {"endOfLine":"lf","tabWidth":2,"useTabs":false}
[debug] '--debug-benchmark' option found, measuring formatWithCursor with 'benchmark' module.
[debug] '--debug-benchmark' measurements for formatWithCursor: {
[debug]   "benchmark": "format x 0.80 ops/sec ±3.60% (7 runs sampled)",
[debug]   "hz": 0.8011225145560241,
[debug]   "ms": 1248.2485285714286
[debug] }

after

[debug] resolve config from 'test.html'
[debug] loaded options `{"useTabs":false,"tabWidth":2,"endOfLine":"lf"}`
[debug] applied config-precedence (cli-override): {"endOfLine":"lf","tabWidth":2,"useTabs":false}
[debug] '--debug-benchmark' option found, measuring formatWithCursor with 'benchmark' module.
[debug] '--debug-benchmark' measurements for formatWithCursor: {
[debug]   "benchmark": "format x 6.86 ops/sec ±11.89% (23 runs sampled)",
[debug]   "hz": 6.86268975187975,
[debug]   "ms": 145.71546086956522
[debug] }

refs #4776

Checklist

• I’ve added tests to confirm my change works.
• (If changing the API or CLI) I’ve documented the changes I’ve made (in the docs/ directory).
• (If the change is user-facing) I’ve added my changes to changelog_unreleased/*/XXXX.md file following changelog_unreleased/TEMPLATE.md.
• I’ve read the contributing guidelines.

✨Try the playground for this PR✨

[fisker]

Thanks for working on this.

the result of yarn run perf:benchmark became about 8.6x faster.

👍

[sapphi-red]

@fisker Is there anything I should do for the next step?

[sapphi-red]

I have changed the code to use for loop to recalculate index/next/prev as you suggested, which would be much more readable!
The benchmark result for this code is below.

[debug] resolve config from 'test.html'
[debug] loaded options `{"useTabs":false,"tabWidth":2,"endOfLine":"lf"}`
[debug] applied config-precedence (cli-override): {"endOfLine":"lf","tabWidth":2,"useTabs":false}
[debug] '--debug-benchmark' option found, measuring formatWithCursor with 'benchmark' module.
[debug] '--debug-benchmark' measurements for formatWithCursor: {
[debug]   "benchmark": "format x 5.21 ops/sec ±10.93% (19 runs sampled)",
[debug]   "hz": 5.207000038723637,
[debug]   "ms": 192.04916315789475
[debug] }

[fisker]

I'm not sure about the performance impact, it seems we can use NODES_KEYS as a Set instead of an object. There are lots of for..in loop.

[fisker]

I'm not familiar with this part of code. Why are we using descriptors instead of class getters?

[sapphi-red]

I'm not sure about the performance impact, it seems we can use NODES_KEYS as a Set instead of an object. There are lots of for..in loop.

I tried this now but there was no significant difference (using Set was only about 1ms slower).

[sapphi-red]

I'm not familiar with this part of code. Why are we using descriptors instead of class getters?

I'm not sure.
I thought there is no big reason so I rewrote it to class getters and properties.

commit
008975f	136.69717916666664ms	x 7.32 ops/sec	±13.22%
00099ca	135.36796250000003ms	x 7.39 ops/sec	±11.86%
f6096d2	142.9828695652174ms	x 6.99 ops/sec	±11.72%
eb0b59f	141.92638695652172ms	x 7.05 ops/sec	±11.45%
e9adae2	160.52020476190478ms	x 6.23 ops/sec	±11.29%
381a73c	175.88805ms	x 5.69 ops/sec	±10.47%

[fisker]

LGTM, thank you!