[nigrosimone]

For #1336. After #1337 a string can still write the pool once for nothing: fewer characters than free bytes, but more UTF-8 bytes. This is costly for two-byte strings: up to 3 bytes per unit, and V8 writes them to UTF-8 much slower than ASCII.

Now a two-byte string tries the first write only if Length() * 3 fits in the free pool, otherwise it is measured first and alloc gives the pool or malloc. One-byte strings do not change.

Trade-off, app.publish('t', message) with no subscribers, Windows, Node 26, min of 11 rounds:

• two-byte that does not fit, 50 K chars of €: from 316 to 168 us
• two-byte that fits after the measure, € + 60 KB ascii: from 69 to 79 us (+14%)
• ascii and latin1 from 20 B to 1 MB: same

The 3x check for all strings, as in #1336, costs also +40% to ascii between 43 and 128 KB, so I kept one-byte strings out.

Test: tests/strings.js has a two-byte string past the pool and one that fits after the measure.

[github-actions]

Benchmark: 850b7d1 against f4fc521

Scenario	Base req/s	Head req/s	Head / base	Rounds	Noise	Busy	CPU per request
http/hello-world	114720	114214	0.995x (-0.5%)	0.99 to 1.00	±2.5%	100% / 100%	8.71 / 8.75 us (1.005x)
http/headers	98104	97502	0.993x (-0.7%)	0.99 to 0.99	±1.6%	100% / 100%	10.18 / 10.25 us (1.007x)
http/json-post	85126	84857	0.996x (-0.4%)	1.00 to 1.00	±1.0%	100% / 100%	11.74 / 11.77 us (1.004x)
http/cached	129020	128512	0.996x (-0.4%)	0.99 to 1.00	±0.5%	100% / 100%	7.75 / 7.78 us (1.004x)
ws/echo-20b	117499	118931	1.010x (+1.0%)	1.01 to 1.02	±1.7%	100% / 100%	8.50 / 8.40 us (0.990x)
ws/echo-4kb	102356	102790	1.005x (+0.5%)	1.00 to 1.01	±1.2%	100% / 100%	9.76 / 9.72 us (0.995x)

4 rounds of 3s per scenario, each round loading base, head and a second process of each one, from a copy of its build, after the other, in an order that changes every round. "Head / base" is the median of the per-round ratios of req/s and "Rounds" their range. "Noise" is how far base/base and head/head, the same code on both sides, got from 1 in this same run: that is what the machine did, so a row is marked only when the median moved further than that, at least 2%, and every round moved the same way: 👀 slower, 🏆 faster. "Busy" is the cpu time of the server process over the round and "CPU per request" that time divided by the requests it answered, with the head/base ratio judged against its own same-code band and bold when it cleared it. When busy is well under 100% the load generator set the pace, the req/s are its and the cpu per request is the column to read. Only the ratios are comparable across runs, the absolute figures depend on the runner.

Node v26.10.0, AMD EPYC 7763 64-Core Processor, 4 cores (server on cpu 0, load on 2,3), http_load_test and load_test with 100 connections.

[uNetworkingAB]

One byte Latin-1 can cause 2 byte Utf-8

[nigrosimone]

Yes, latin1 can take 2 bytes per char, so a one-byte string can still fill the pool for nothing, as on master. I left it on purpose: V8 writes one-byte strings to UTF-8 fast, so that waste is small, while a 2x check would also measure first every ascii string between 64 and 128 KB, the +40% I mesured for the 3x check on that range. If you prefer it safe for latin1 too, I can use 2 for one-byte strings.