[WIP] Initial draft of engine book #326
Conversation
book/src/engine/logic.md
Outdated
|
|
||
| ## Overview | ||
|
|
||
| `chalk-engine` solves a `Goal` using a depth-first search. When asked to solve a |
There was a problem hiding this comment.
This isn't strictly true. We don't use depth-first search. If we did, we would always fully explore one strand before starting on another. The docs I wrote before claimed we use a breadth-first search but this isn't really true either. I'm actually not sure if our current search algorithm can be cleanly characterized. It's one of the things I've been wondering about -- we might want to think about how to specify the behavior more rigorously. The kind of "disconnected" way that our current solver works makes it a bit less obvious to me how to do that.
The minikanren algorithm is quite interesting in that (by the nature of its design) winds up spending a lot more time traversing and working with answers that have "gotten farther" in the process than it does with those that haven't. Or at least so it was explained to me. To be honest, I have to kind of review how it works, and maybe it'd be good for us to spend some time trying to characterize and work through how our own solver would behave.
Anyway for now let's just not say depth-first search, since to me that suggests something very specific (prolog-style solving) which is not the case. Maybe something like "using a hybrid search strategy with elements of depth- and breadth-first search"?
There was a problem hiding this comment.
Yeah, true, it's not depth- or breadth-first. It would be good to characterize exactly what we do, but I think that will be more useful when we have benches and can check how different search schemes perform.
| On the other hand, `Context::Goal` represents an opaque goal generated | ||
| externally. As such, it may contain any extra information or may be interned. | ||
| When solving a logic predicate, Chalk will lazily convert `Context::Goal`s | ||
| into `HHGoal`s. |
There was a problem hiding this comment.
I intend to add some chapters to the type section describing the sorts of goals that appear in our logic etc in a more "semantic way", so maybe we should link over there eventually I guess.
jackh726
force-pushed
the
engine_book
branch
from
e5663d7 to
de87a52
Compare
Works for me |
No description provided.