perf: improve encoding

[anonrig]

Before:

[bench:run] │ Tests               │ Samples │          Result │ Tolerance │ Difference with slowest │
[bench:run] |─────────────────────|─────────|─────────────────|───────────|─────────────────────────|
[bench:run] │ undici - fetch      │       1 │ 1312.32 req/sec │  ± 0.00 % │                       - │
[bench:run] |─────────────────────|─────────|─────────────────|───────────|─────────────────────────|
[bench:run] │ http - no keepalive │       1 │ 2640.12 req/sec │  ± 0.00 % │              + 101.18 % │
[bench:run] |─────────────────────|─────────|─────────────────|───────────|─────────────────────────|
[bench:run] │ http - keepalive    │       1 │ 3609.65 req/sec │  ± 0.00 % │              + 175.06 % │
[bench:run] |─────────────────────|─────────|─────────────────|───────────|─────────────────────────|
[bench:run] │ undici - request    │       1 │ 4525.88 req/sec │  ± 0.00 % │              + 244.87 % │
[bench:run] |─────────────────────|─────────|─────────────────|───────────|─────────────────────────|
[bench:run] │ undici - pipeline   │       1 │ 5737.63 req/sec │  ± 0.00 % │              + 337.21 % │
[bench:run] |─────────────────────|─────────|─────────────────|───────────|─────────────────────────|
[bench:run] │ undici - dispatch   │       1 │ 5812.50 req/sec │  ± 0.00 % │              + 342.92 % │
[bench:run] |─────────────────────|─────────|─────────────────|───────────|─────────────────────────|
[bench:run] │ undici - stream     │       1 │ 7318.62 req/sec │  ± 0.00 % │              + 457.68 % │

After:

[bench:run] │ Tests               │ Samples │          Result │ Tolerance │ Difference with slowest │
[bench:run] |─────────────────────|─────────|─────────────────|───────────|─────────────────────────|
[bench:run] │ undici - fetch      │       1 │ 1326.74 req/sec │  ± 0.00 % │                       - │
[bench:run] |─────────────────────|─────────|─────────────────|───────────|─────────────────────────|
[bench:run] │ http - no keepalive │       1 │ 2635.04 req/sec │  ± 0.00 % │               + 98.61 % │
[bench:run] |─────────────────────|─────────|─────────────────|───────────|─────────────────────────|
[bench:run] │ http - keepalive    │       1 │ 3757.30 req/sec │  ± 0.00 % │              + 183.20 % │
[bench:run] |─────────────────────|─────────|─────────────────|───────────|─────────────────────────|
[bench:run] │ undici - request    │       1 │ 5283.23 req/sec │  ± 0.00 % │              + 298.21 % │
[bench:run] |─────────────────────|─────────|─────────────────|───────────|─────────────────────────|
[bench:run] │ undici - dispatch   │       1 │ 5651.02 req/sec │  ± 0.00 % │              + 325.93 % │
[bench:run] |─────────────────────|─────────|─────────────────|───────────|─────────────────────────|
[bench:run] │ undici - pipeline   │       1 │ 6096.91 req/sec │  ± 0.00 % │              + 359.54 % │
[bench:run] |─────────────────────|─────────|─────────────────|───────────|─────────────────────────|
[bench:run] │ undici - stream     │       1 │ 7483.29 req/sec │  ± 0.00 % │              + 464.04 % │

[KhafraDev]

just the smallest change, lgtm otherwise

outdated

[ronag]

@mcollina isn't there security implication to this? i.e. what happens if a partial symbol stays buffered in the encoder between requests?

[mcollina]

is there still a perf improvement?

[anonrig]

Yes I updated the description of the pull request.

[mcollina]

it does not seem to add much.

[anonrig]

Mainly two reasons:

1. The benchmark does not include headers
2. https://www.youtube.com/watch?v=jrDikOz8YSU

[ronag]

I don't think this is worth it.

[KhafraDev]

TextEncoder's constructor only sets a single symbol¹, and TextDecoder seems to call into C++ land once². With those benchmarks, I don't think there's really any performance gain considering creating a new TextEncoder shouldn't be slow.

Footnotes

1. https://github.com/nodejs/node/blob/8cae0a117ecb0dd2c9f9bbb05fdd53f114906990/lib/internal/encoding.js#L327 ↩

2. https://github.com/nodejs/node/blob/8cae0a117ecb0dd2c9f9bbb05fdd53f114906990/lib/internal/encoding.js#L401 ↩

[anonrig]

I'm working on adding a fast path to TextDecoder for UTF8. I'll close this pull request.

[anonrig]

With the addition of the fast path utf8 pull request, here's the new result for the benchmark:

[bench:run] │ Tests               │ Samples │          Result │ Tolerance │ Difference with slowest │
[bench:run] |─────────────────────|─────────|─────────────────|───────────|─────────────────────────|
[bench:run] │ undici - fetch      │       1 │ 1668.14 req/sec │  ± 0.00 % │                       - │
[bench:run] |─────────────────────|─────────|─────────────────|───────────|─────────────────────────|
[bench:run] │ http - no keepalive │       1 │ 2820.24 req/sec │  ± 0.00 % │               + 69.07 % │
[bench:run] |─────────────────────|─────────|─────────────────|───────────|─────────────────────────|
[bench:run] │ http - keepalive    │       1 │ 4053.76 req/sec │  ± 0.00 % │              + 143.01 % │
[bench:run] |─────────────────────|─────────|─────────────────|───────────|─────────────────────────|
[bench:run] │ undici - request    │       1 │ 4929.85 req/sec │  ± 0.00 % │              + 195.53 % │
[bench:run] |─────────────────────|─────────|─────────────────|───────────|─────────────────────────|
[bench:run] │ undici - pipeline   │       1 │ 5840.36 req/sec │  ± 0.00 % │              + 250.11 % │
[bench:run] |─────────────────────|─────────|─────────────────|───────────|─────────────────────────|
[bench:run] │ undici - stream     │       1 │ 7391.04 req/sec │  ± 0.00 % │              + 343.07 % │
[bench:run] |─────────────────────|─────────|─────────────────|───────────|─────────────────────────|
[bench:run] │ undici - dispatch   │       1 │ 8433.93 req/sec │  ± 0.00 % │              + 405.59 % │

[mcollina]

I like the improvement on all the methods.