src: avoid making JSTransferable wrapper object weak

[legendecas]

JSTransferable wrapper object is a short-lived wrapper in the scope of
the serialization or the deserialization. Make the JSTransferable
wrapper object pointer as a strongly-referenced detached BaseObjectPtr
so that a JSTransferable wrapper object and its target object will never
be garbage-collected during a ser-des process, and the wrapper object
will be immediately destroyed when the process is completed.

Fixes: #49852
Fixes: #49844

[anonrig]

For education purposes: What is the purpose of initializing a scope in here?

[legendecas]

A local handle is created with target_.Get(env()->isolate()) in the next line. A handle scope is needed in place to collect the local handle when it goes out of lexical scope.

[H4ad]

The crash still occurs even with this commit:

[00:00:08|%   7| 0/1 files |  4/60 runs | 1/2 configs]: perf_hooks/histogram.js node:internal/worker/io:410
  const message = receiveMessageOnPort_(port?.[kHandle] ?? port);
                  ^

Error: Unable to deserialize cloned data.
    at receiveMessageOnPort (node:internal/worker/io:410:19)
    at structuredClone (node:internal/structured_clone:24:10)
    at main (/home/h4ad/Projects/opensource/node-copy-4/benchmark/perf_hooks/histogram.js:30:22)
    at /home/h4ad/Projects/opensource/node-copy-4/benchmark/common.js:54:9
    at process.processTicksAndRejections (node:internal/process/task_queues:77:11)

Node.js v21.0.0-pre

[legendecas]

Stress test: https://ci.nodejs.org/job/node-stress-single-test/459/

I can reproduce with pummel/test-structuredclone-jstransferable locally on the main branch in 10 runs, and no reproduction with the patch.

[legendecas]

@H4ad the stress test passed. Would you mind sharing more information about your local test? Thank you.

[H4ad]

@legendecas Try with this code:

'use strict';

const assert = require('assert');
const common = require('../common.js');

const { createHistogram } = require('perf_hooks');

const bench = common.createBenchmark(main, {
  n: [1e5],
  operation: ['creation', 'clone'],
});

let _histogram;

function main({ n, operation }) {
  switch (operation) {
    case 'creation': {
      bench.start();
      for (let i = 0; i < n; i++)
        _histogram = createHistogram();
      bench.end(n);

      // Avoid V8 deadcode (elimination)
      assert.ok(_histogram);
      break;
    }
    case 'clone': {
      bench.start();
      for (let i = 0; i < n; i++)
        _histogram = structuredClone(createHistogram());
      bench.end(n);

      // Avoid V8 deadcode (elimination)
      assert.ok(_histogram);
      break;
    }
    default:
      throw new Error(`Unsupported operation ${operation}`);
  }
}

Put it inside benchmark/perf_hooks and then run with this new patch.

[legendecas]

Note that benchmark/compare.js runs the benchmark with the new node binary and the old node binary. So it would report errors for node binary without this patch. Would you mind verifying if the error happened with the new node binary? Because from what I can tell from your posted output result, it seems like you are running with benchmark/compare.js. Please correct me if I'm getting wrong.

[H4ad]

Oh, you are right, I forgot to compile the old version with your patch, sorry about that.

In this case, I can confirm this PR fixes my bug! Thanks for your work!

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/54581/

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/54640/

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/54642/

[mcollina]

lgtm

[nodejs-github-bot]

Landed in 78a1570