--watch --on-success should send SIGTERM before sending SIGKILL

[JakeIsMeh]

Clear and concise description of the problem

when using tsdown --watch --on-success "node ..." to run a backend service (such as a web server with database connections), tsdown forcibly kills the child process immediately upon a file change to spawn the new process, resulting in graceful shutdown handlers not running.

Suggested solution

change the default restart behavior to send a graceful termination signal first (like SIGTERM or SIGINT), with a fallback to SIGKILL after a timeout if the process hangs.

Alternative

currently a workaround would be skipping --on-success and using concurrently and node --watch or tsx watch, but that requires yet another devDependency which isnt the end of the world but kind of defeats tsdown's elegance of being a reliable all-in-one devtool.

Additional context

tsx watch sends SIGTERM first before sending SIGKILL after 5 seconds (src)

nodemon sends SIGUSR2 first before sending SIGKILL I think

Validations

• Follow our Code of Conduct
• Read the Contributing Guide.
• Check that there isn't already an issue that request the same feature to avoid creating a duplicate.

[JakeIsMeh]

Investigated further, seems this is a tree-kill issue. By default, it delegates to Node's process.kill (which issues a SIGTERM by default if no signal is specified) on Darwin (macOS) and Linux, but directly calls taskkill /f (force kill!) on Windows for some reason. (taskkill has no notion of unix signals). tsx just ignores trees and only kills the top most child and hopes that it trickles down to the rest.