[rameshvarun]

Fixes #826

Problem

When an application run through the tsx CLI handles SIGINT or SIGTERM with process.once(), tsx can exit before asynchronous cleanup finishes.

tsx installs a hidden listener to relay signals to its parent process and emulate Node's default exit when no application listeners remain. That listener could run after an application's one-time listener. Node removes a one-time listener before invoking it, so tsx then saw no remaining application listeners and called process.exit().

An application could start closing a database connection, then exit before its cleanup completes. The same handler completes when run with node or node --import tsx.

Changes

tsx now runs its hidden relay listener before normal application listeners. It observes a one-time listener before Node removes it, allowing the handler to complete asynchronous cleanup and control the exit code.

The regression sends SIGINT to an asynchronous process.once() handler and confirms that cleanup finishes before exit. The test fails on the pre-fix implementation with exit code 130 and passes with exit code 42 after cleanup completes.

This covers ordinary process.once() registration. A handler added later with process.prependOnceListener() can still intentionally take precedence over the relay.

[privatenumber]

This issue is now resolved in v4.23.4.

If you're able to, your sponsorship would be very much appreciated.