Comment by btrask | Hacker News Reader

Comment by btrask | Hacker News Reader

Parent Full thread

It's not just getpid. It's everything that interacts with processes. fork(2), wait(2), kill(1/2)... Any external management tools (supervisors, debuggers, etc)... Imagine trying to administer such a program, where even tools like top(1) couldn't track it over time.

Here's one that I think you'll find compelling: threads. What do you do if you have multiple background threads running when one thread wants to exit the process? You could just exit the thread, but I think that would defeat the point.

Imagine you're using a language, and you find this feature in the docs. Then it lists all the caveats and things you have to give up. Would you still use it, or would you make do without it?

Maybe I haven't convinced you but I hope you can agree it's a major tradeoff. I think that explains why a general purpose language hasn't tried to make it a standard feature.

You have at least convinced me that anyone using it would need to know very well what they are doing and what is going to happen behind the scenes. And that it's better suited to the higher-level languages (node, ruby family) rather than those that give you more fine-grained control over processes.

You've also convinced me that it's better as syntactic sugar for splitting into true separate processes behind the scenes, rather than a first-class language construct. This would solve many of the issues you raised. But I do still think there's value in such a concept.

Perhaps it could be a good fit for serverless functions. You write it as one function, but it happens as two, and the developer doesn't have to worry about the glue code.

In terms of jumping to the right place in the code, several languages have the concept of continuations, it doesn't seem like a huge leap to something like this.

> Perhaps it could be a good fit for serverless functions.

I was thinking of that too. :)

If you're interested in the technical mechanism, it's possible to load a core dump and resume execution. I found https://web.archive.org/web/20091026234235/http://geocities.... which explains the (surprisingly complex) process of resuming from a core dump.

Also see https://criu.org/ and https://checkpointing.org/

Cool, very interesting! That just convinces me even more that it's better to leave the OS out of it and let each language's runtime figure out its own way of serializing its state and restoring it for a second invocation.

Reply on news.ycombinator.com