diagnostics_channel: Channel::binding_data_ dangles after environment cleanup, crashing node:sqlite at exit

まず、提供された node:diagnostics_channel と node:sqlite の再現を実行し、次に Environment::RunCleanup() と SQLite のプロファイルコールバックを通じて Channel::binding_data_ を追跡します。同じ終了シナリオで修正を確認します。これには、明示的な close() を行わずにステップの途中で残されたステートメントも含めます。また、プロセスが SIGSEGV なしで終了することを確認します。

Version

v27.0.0-pre (2dfdb6a4206)

Platform

Darwin 25.6.0 arm64

Subsystem

diagnostics_channel, sqlite

What steps will reproduce the bug?

Channel holds its BindingData as a raw pointer that is never cleared, so any native holder that outlives environment cleanup reads a destroyed object. Here's a replication case using node:sqlite:

const dc = require('node:diagnostics_channel');
const { DatabaseSync } = require('node:sqlite');

dc.subscribe('sqlite.db.query', () => {});

const db = new DatabaseSync(':memory:');
db.exec('CREATE TABLE t(x)');
const ins = db.prepare('INSERT INTO t VALUES (?)');
for (let i = 0; i < 200; i++) ins.run(i);

// Start iterating and abandon it: the statement stays mid-step, so the
// finalize that happens at teardown fires SQLite's profile callback.
const it = db.prepare('SELECT * FROM t').iterate();
it.next();
globalThis.keepAlive = { db, it };
console.log('reached end of script');

How often does it reproduce? Is there a required condition?

Three conditions are required (all included in the example above):

1. A subscriber on sqlite.db.query, so the SQLite profile hook is installed.
2. A statement left mid-step at exit, so that the finalize during teardown has a started statement to profile.
3. No explicit close(), so the statement is finalized by the destructor chain after Environment::RunCleanup() rather than before it.

What is the expected behavior? Why is that the expected behavior?

Exiting a process that used node:sqlite with a sqlite.db.query subscriber
shouldn't crash.

What do you see instead?

SIGSEGV, exit 139, after the script has finished.