src: improve StreamBase write throughput

[addaleax]

Improve performance by transferring information about write status
to JS through an AliasedBuffer, rather than object properties
set from C++.

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• commit message follows commit guidelines

[nodejs-github-bot]

@addaleax build started: https://ci.nodejs.org/blue/organizations/jenkins/node-test-pull-request-lite-pipeline/detail/node-test-pull-request-lite-pipeline/1354/pipeline

[addaleax]

CI: https://ci.nodejs.org/job/node-test-pull-request/18099/
Benchmark CI: https://ci.nodejs.org/view/Node.js%20benchmark/job/benchmark-node-micro-benchmarks/252/
CITGM: https://ci.nodejs.org/view/Node.js-citgm/job/citgm-smoker/1596/

[addaleax]

Benchmarks looked good but had some isolated negative instances – I’d like to make sure these aren’t real performance regressions…

Benchmark CI again (net-c2s.js only, more iterations): https://ci.nodejs.org/view/Node.js%20benchmark/job/benchmark-node-micro-benchmarks/253/
Resume CI: https://ci.nodejs.org/job/node-test-pull-request/18125/

[addaleax]

Think I figured out why the benchmarks had that weird drop for ASCII writes. (Edit: No, my mistake.)

New benchmark CI: https://ci.nodejs.org/view/Node.js%20benchmark/job/benchmark-node-micro-benchmarks/254/
(New CI: https://ci.nodejs.org/job/node-test-pull-request/18133/)
Resume CI: https://ci.nodejs.org/job/node-test-pull-request/18134/

[addaleax]

So … the most recent full benchmark (https://ci.nodejs.org/view/Node.js%20benchmark/job/benchmark-node-micro-benchmarks/254/) shows a significant -8 % regression for one of the large-buffer ASCII variants of a benchmark. Running only that benchmark with a few more iterations (https://ci.nodejs.org/view/Node.js%20benchmark/job/benchmark-node-micro-benchmarks/255/) shows a significant 5 % increase. (I did not reproduce either locally.)

I don’t quite know how to make sense of that – I don’t understand how this patch would lead to a regression, and the high significance levels in the different benchmarks would make my first guess that it’s related to how we benchmark things, so I’d feel comfortable going ahead with it (especially in like of the more common significant improvements), but I wanted to point this out in case anybody felt differently.

[addaleax]

Resume CI again: https://ci.nodejs.org/job/node-test-pull-request/18182/

[targos]

Landed in f01518e

[MylesBorins]

Is this something we would want to backport to LTS? I'm assuming it would likely have to be the entire StreamBase refactor in one go.

[addaleax]

Is this something we would want to backport to LTS?

I think it might be better not to.

I'm assuming it would likely have to be the entire StreamBase refactor in one go.

Not really, no. It’s like all other changes – some commits make sense to backport, others have a nonzero risk of breakage (although I try to run CITGM on all of them).