src: avoid deferred gc/cleanup for Buffer.from

[jasnell]

/cc @addaleax @Daninet

Previously, the code path would allocated a tracked ArrayBuffer that defers cleanup and deallocation of the underlying data with a native SetImmediate. Avoid the unnecessary deferral by just allocating a BackingStore directly and writing into it.

This should be considered an alternative to #38336

Fixes: #38300
Refs: #38336

[XadillaX]

After testing:

const str = 'x'.repeat(10000);

console.log('start', process.memoryUsage());

console.time('1');
  for(let i = 0; i < 1e6; i++) {
    Buffer.from(str); // memory leak happens here
    if (i % 1e5 === 0) {
      // await wait(100);
      gc();
      console.log('step', i, 'rss', process.memoryUsage().rss);
    }
  }
console.timeEnd('1');

  gc();
  console.log('end', process.memoryUsage());

The new PR #38337 runs about 3.9s in my computer. And 39f42d1a2d (before using BackingStore) runs about 3.7s. I think this performance loss is acceptable.

[mcollina]

lgtm

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/37635/

[jasnell]

@nodejs/build ... is there a known issue with the node-test-commit-arm-fanned cluster? It is failing consistently on multiple PRs

[richardlau]

@nodejs/build ... is there a known issue with the node-test-commit-arm-fanned cluster? It is failing consistently on multiple PRs

I cleaned up leftover node process on three of the pi 2's this morning. AFAIK things should be okay now.

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/37643/

[jasnell]

@richardlau ... no joy it seems. See https://ci.nodejs.org/job/node-test-binary-arm-12+/10459/RUN_SUBSET=1,label=pi2-docker/console for #38331, for instance.

[targos]

Landed in 720fdf2