Clojurians Log v2

yeah makes sense. babashka + SCI is used like this e.g. on clojure IRC in a bot that can execute arbitrary code examples and is killed on timeout

I think limiting execution time/step and memory consumption of a turing-complete language like SCI is a hard problem. If you want a limited scripting language, you better design it from scratch.

yes, SCI does explicitly not support that, so if you want to limit turing completeness, find something else or write your own

I think turing-complete languages are known for non-deterministic termination. A turing-complete language can't deterministically terminate itself.

you are stating obvious facts about turing completeness. is there a question in there or?

I mean people forget this fact when they try to limit execution steps of SCI. It's probably an unsolvable problem.

So, I think running untrusted SCI scripts in a proper sandboxing environment like babashka/node.js processes is the right solution. You just need to find a sandboxing environment that doesn't consume a lot of RAM.

yep

What I'd really love to have a "metered sci" "You get 10 thousand cycles"

sure, that's been suggested before and I also tried it but it has performance consequences and it'll never be perfect since clojure.core functions aren't interpreted and you can still use unlimited resources with those

hmmm I would think that as long as the turing completeness provided is only by sci's compbination of clojure.core fns, and each of them would by themselves run in some bounded deterministic time, then the overall turing complete action would be time bounded. But yeah I guess you can't accurately measure the underlying cycles

(doall (range))

Hmmm... Yeah, you'd have to detect all those.

you need to control the VM or the host

ask the JVM folks ;)

or Node.js or whatever

wasm maybe

wasm isn't a host

But I don't think I need fine grained cycle knowlege. Just bounded cycles

the infinite seqs are a problem though

even with finite stuff, you can construct stuff that takes longer that you'd wish for

I'm wanting it for a distrubed computation utility token scenario, where you need to just measure work evenly across peers

As long as the cost profile is fair between peers

As long as it is deterministic and gives the same invoice for services rendered on both peers

John, time is the most fundamental currency in the universe. Meter time. Time and attention are the two most fundamental currencies in the universe. Money is an imaginary medium of exchange, but time and attention are real. Even your utility tokens are an imaginary medium of exchange. Money is not real. Your tokens are not real. Time is real. • Enforce a timeout from outside, and charge per an execution that has a hard timeout • Charge per second of execution, and don't enforce a hard timeout. You can charge each second differently for different CPU cores. Run one SCI script at a time in one process. Then, each SCI job gets one dedicated CPU core.

Cloud platforms sell shared/dedicated CPU cores per month. You should do the same perhaps.

Perhaps, sell dedicated/shared CPU cores per month, week, day, second, whatever.

Sell time. Become a time lord.

lol