Comments on: Painless NP-complete problems: an embedded DSL for SMT solving

By: Chris Yuen

Chris Yuen — Thu, 01 Sep 2011 13:30:13 +0000

Thanks Dons! This is one of those rare Haskell articles that don’t into algebraic Mars-speak which I highly suspect only serves bruise the writer’s ego.

By: dons00

dons00 — Wed, 27 Apr 2011 23:43:52 +0000

Haskell is relatively accessible compared to programming in an SMT solver. Especially for Haskell programmers.

By: Dan Kaminsky

Dan Kaminsky — Wed, 13 Apr 2011 13:04:16 +0000

You do realize the cognitive load of Haskell itself is hilariously out of control, right?

By: Will

Will — Wed, 19 Jan 2011 12:18:24 +0000

To my knowledge SMT solvers are unable to reason (co-)inductively and so can’t construct proofs over lists/trees or any user-defined recursive data-type. This means that you can only solve imperative model-checking style problems over primitive/enumerated data-types.

That being said I think this is a great piece of work and will be very helpful for this style of problem, which is very prevalent in real-life programs.

Note: My hypothesis is that this is a fundamental incompatibility with the DPLL(T) algorithm, in that the theory T would have to do perform the inductive step in its propositional truth check, which is much too complex (not to mention incomplete) for a DPLL(T) theory, and would need to have DPLL assignment steps in between for a complex proof. I’m not sure since I don’t work within SMT solving but I’d love some clarification on this.

By: dons00

dons00 — Tue, 18 Jan 2011 00:00:44 +0000

Sam Martin, current idea is to introduce sharing explicitly in the backend AST, and recover sharing via an “observable sharing” approach. There’s some examples out there (accelerate EDSL, Gill’s paper from 09).

By: dons00

dons00 — Mon, 17 Jan 2011 22:14:32 +0000

Yep, solrize. I’ll be modifying the system to compile to SMT-LIB (so you can choose which backend you want).

By: solrize

solrize — Mon, 17 Jan 2011 21:24:43 +0000

Ah, I see you addressed this somewhat ;)

By: solrize

solrize — Mon, 17 Jan 2011 21:22:20 +0000

Cool! Is there any chance of or interest in bindings to an open source solver (OpenSMT?) instead of Yices? Is Yices really better? I’ve never used anything like this but have been interested for a while.

By: Simon Meier

Simon Meier — Mon, 17 Jan 2011 18:11:38 +0000

Hi Dons, that sounds great. Thanks for sharing it. I’m very interested in progress of your binding and I hope that I’ll get the chance to apply it sometime in the future.

By: Mikhail Glushenkov

Mikhail Glushenkov — Mon, 17 Jan 2011 14:11:08 +0000

> e.g. lazy lists instead of iteratees

Maybe “iterators”?