Easiest way to interrupt VM when specific predicate is the goal (syscall)?

[leonardoalt]

Hi all, here's a potentially odd question: what would be the easiest way for me to interrupt the VM's main loop when a specific predicate becomes the goal? What I want to do is basically implement the equivalent of syscalls for some imaginary platform.

When such a predicate is "called", the execution is interrupted and goes back to the host running the VM. The host solves the predicate and gives some substitutions back to the VM to resume execution from exactly that point with the answers given by the host.

I found that machine::mod::try_execute and machine::system_calls::fast_calls are good places to start, but maybe there's even a mechanism already for that kind of stuff internally so wanted to ask before I try to reinvent anything.

Thanks!

[bakaq]

Not sure if applicable, but that sounds a lot like freeze/2 from library(freeze) (see also library(when) and library(cont)).

If I understand this in another way (running code "outside of Scryer Prolog"), this also sounds a lot like library(ffi).

[leonardoalt]

Thanks for the quick reply!
It does go in a similar direction, but my case is different because I want to manage a stateful host that makes calls to the VM. It's similar to how a RISCV interpreter would handle an ecall.

[bakaq]

I'm not sure at what level you would want this, but it's definitely possible to do this with just Prolog and the libraries I cited (and things like assertz/1 and bb_put/2 for state, although doing state completely declaratively is a more powerful alternative). Do you need this to be handled by the Scryer Prolog interpreter at a low level? Why?

[leonardoalt]

That's a good point. I definitely would prefer if what I want to do can be done 100% in Prolog. I think I do agree with you, but I haven't managed to sketch out the solution in my head. Say I have a host memory DB (to simplify, the host could be a lot more complex), and every time the predicate db_read(DB, K, V). is called, we need to read from the DB (and db_write(DB1, K, V, DB2) analogously).

• FFI wouldn't be great for this
• We can assume the DB is too large to be passed in the query.
• Sockets could work
• Sockets for first reads, internal Prolog assert/bb for reads/writes, final socket call for new writes

Unless I'm missing something, these feel like the easiest ways to implement it. A closer integration on the Rust library level would make all this communication a lot simpler, at the expense of modifying the Rust code.

I'll try to implement the last option above, see if I can make it work. Thanks again!

[triska]

One solution that may be worth considering is meta-interpretation: Prolog makes it easy to write a program called a meta-interpreter, which takes an existing Prolog program and interprets it as Prolog would, and this gives you the opportunity to interpret goals also differently than normal Prolog would do it.

[jjtolton]

Hi all, here's a potentially odd question: what would be the easiest way for me to interrupt the VM's main loop when a specific predicate becomes the goal? What I want to do is basically implement the equivalent of syscalls for some imaginary platform.

Can you be a little more specific about what you mean by "interrupt the VM's main loop"? This is an area I'm also researching for embedded C# video game stuff. As @triska mentioned, a meta interpreter is probably a key technique, and there are a few lower lever techniques such as get_single_char/1 that can pause execution. But choosing the right lower level construct is important here.