1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298
//! An alternative solver based around the SLG algorithm, which //! implements the well-formed semantics. This algorithm is very //! closed based on the description found in the following paper, //! which I will refer to in the comments as EWFS: //! //! > Efficient Top-Down Computation of Queries Under the Well-formed Semantics //! > (Chen, Swift, and Warren; Journal of Logic Programming '95) //! //! However, to understand that paper, I would recommend first //! starting with the following paper, which I will refer to in the //! comments as NFTD: //! //! > A New Formulation of Tabled resolution With Delay //! > (Swift; EPIA '99) //! //! In addition, I incorporated extensions from the following papers, //! which I will refer to as SA and RR respectively, that //! describes how to do introduce approximation when processing //! subgoals and so forth: //! //! > Terminating Evaluation of Logic Programs with Finite Three-Valued Models //! > Riguzzi and Swift; ACM Transactions on Computational Logic 2013 //! > (Introduces "subgoal abstraction", hence the name SA) //! > //! > Radial Restraint //! > Grosof and Swift; 2013 //! //! Another useful paper that gives a kind of high-level overview of //! concepts at play is the following, which I will refer to as XSB: //! //! > XSB: Extending Prolog with Tabled Logic Programming //! > (Swift and Warren; Theory and Practice of Logic Programming '10) //! //! While this code is adapted from the algorithms described in those //! papers, it is not the same. For one thing, the approaches there //! had to be extended to our context, and in particular to coping //! with hereditary harrop predicates and our version of unification //! (which produces subgoals). I believe those to be largely faithful //! extensions. However, there are some other places where I //! intentionally dieverged from the semantics as described in the //! papers -- e.g. by more aggressively approximating -- which I //! marked them with a comment DIVERGENCE. Those places may want to be //! evaluated in the future. //! //! Glossary of other terms: //! //! - WAM: Warren abstract machine, an efficient way to evaluate Prolog programs. //! See <http://wambook.sourceforge.net/>. //! - HH: Hereditary harrop predicates. What Chalk deals in. //! Popularized by Lambda Prolog. #![feature(conservative_impl_trait)] #![feature(crate_in_paths)] #![feature(crate_visibility_modifier)] #![feature(dyn_trait)] #![feature(in_band_lifetimes)] #![feature(match_default_bindings)] #![feature(macro_vis_matcher)] #![feature(step_trait)] #![feature(universal_impl_trait)] #![feature(underscore_lifetimes)] #[macro_use] extern crate chalk_macros; extern crate stacker; use crate::context::{Context, InferenceContext}; use std::collections::HashSet; use std::cmp::min; use std::usize; pub mod context; pub mod fallible; pub mod forest; pub mod hh; mod derived; mod logic; mod simplify; mod stack; mod strand; mod table; mod tables; index_struct! { pub struct TableIndex { // FIXME: pub b/c Fold value: usize, } } /// The StackIndex identifies the position of a table's goal in the /// stack of goals that are actively being processed. Note that once a /// table is completely evaluated, it may be popped from the stack, /// and hence no longer have a stack index. index_struct! { struct StackIndex { value: usize, } } /// The `DepthFirstNumber` (DFN) is a sequential number assigned to /// each goal when it is first encountered. The naming (taken from /// EWFS) refers to the idea that this number tracks the index of when /// we encounter the goal during a depth-first traversal of the proof /// tree. #[derive(Copy, Clone, Debug, PartialEq, Eq, PartialOrd, Ord)] struct DepthFirstNumber { value: u64, } /// The paper describes these as `A :- D | G`. #[derive(Clone, Debug, PartialEq, Eq, Hash)] pub struct ExClause<C: Context, I: InferenceContext<C>> { /// The substitution which, applied to the goal of our table, /// would yield A. pub subst: I::Substitution, /// Delayed literals: things that we depend on negatively, /// but which have not yet been fully evaluated. pub delayed_literals: Vec<DelayedLiteral<C>>, /// Region constraints we have accumulated. pub constraints: Vec<I::RegionConstraint>, /// Subgoals: literals that must be proven pub subgoals: Vec<Literal<C, I>>, } #[derive(Clone, Debug, PartialEq, Eq, Hash)] struct SimplifiedAnswers<C: Context> { answers: Vec<SimplifiedAnswer<C>>, } #[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)] pub struct SimplifiedAnswer<C: Context> { /// A fully instantiated version of the goal for which the query /// is true (including region constraints). pub subst: C::CanonicalConstrainedSubst, /// If this flag is set, then the answer could be neither proven /// nor disproven. In general, the existence of a non-empty set of /// delayed literals simply means the answer's status is UNKNOWN, /// either because the size of the answer exceeded `max_size` or /// because of a negative loop (e.g., `P :- not { P }`). pub ambiguous: bool, } #[derive(Clone, Debug)] enum DelayedLiteralSets<C: Context> { /// Corresponds to a single, empty set. None, /// Some (non-zero) number of non-empty sets. Some(HashSet<DelayedLiteralSet<C>>), } /// A set of delayed literals. The vector in this struct must /// be sorted, ensuring that we don't have to worry about permutations. /// /// (One might expect delayed literals to always be ground, since /// non-ground negative literals result in flounded /// executions. However, due to the approximations introduced via RR /// to ensure termination, it *is* in fact possible for delayed goals /// to contain free variables. For example, what could happen is that /// we get back an approximated answer with `Goal::CannotProve` as a /// delayed literal, which in turn forces its subgoal to be delayed, /// and so forth. Therefore, we store canonicalized goals.) #[derive(Clone, Debug, Default)] struct DelayedLiteralSet<C: Context> { delayed_literals: Vec<DelayedLiteral<C>>, } #[derive(Clone, Debug)] pub enum DelayedLiteral<C: Context> { /// Something which can never be proven nor disproven. Inserted /// when truncation triggers; doesn't arise normally. CannotProve(()), /// We are blocked on a negative literal `~G`, where `G` is the /// goal of the given table. Because negative goals must always be /// ground, we don't need any other information. Negative(TableIndex), /// We are blocked on a positive literal `Li`; we found a /// **conditional** answer (the `CanonicalConstrainedSubst`) within the /// given table, but we have to come back later and see whether /// that answer turns out to be true. Positive(TableIndex, C::CanonicalConstrainedSubst), } /// Either `A` or `~A`, where `A` is a `Env |- Goal`. #[derive(Clone, Debug)] pub enum Literal<C: Context, I: InferenceContext<C>> { // FIXME: pub b/c fold Positive(I::GoalInEnvironment), Negative(I::GoalInEnvironment), } /// The `Minimums` structure is used to track the dependencies between /// some item E on the evaluation stack. In particular, it tracks /// cases where the success of E depends (or may depend) on items /// deeper in the stack than E (i.e., with lower DFNs). /// /// `positive` tracks the lowest index on the stack to which we had a /// POSITIVE dependency (e.g. `foo(X) :- bar(X)`) -- meaning that in /// order for E to succeed, the dependency must succeed. It is /// initialized with the index of the predicate on the stack. So /// imagine we have a stack like this: /// /// // 0 foo(X) <-- bottom of stack /// // 1 bar(X) /// // 2 baz(X) <-- top of stack /// /// In this case, `positive` would be initially 0, 1, and 2 for `foo`, /// `bar`, and `baz` respectively. This reflects the fact that the /// answers for `foo(X)` depend on the answers for `foo(X)`. =) /// /// Now imagine that we had a clause `baz(X) :- foo(X)`, inducing a /// cycle. In this case, we would update `positive` for `baz(X)` to be /// 0, reflecting the fact that its answers depend on the answers for /// `foo(X)`. Similarly, the minimum for `bar` would (eventually) be /// updated, since it too transitively depends on `foo`. `foo` is /// unaffected. /// /// `negative` tracks the lowest index on the stack to which we had a /// NEGATIVE dependency (e.g., `foo(X) :- not { bar(X) }`) -- meaning /// that for E to succeed, the dependency must fail. This is initially /// `usize::MAX`, reflecting the fact that the answers for `foo(X)` do /// not depend on `not(foo(X))`. When negative cycles are encountered, /// however, this value must be updated. #[derive(Copy, Clone, Debug)] struct Minimums { positive: DepthFirstNumber, negative: DepthFirstNumber, } impl<C: Context> DelayedLiteralSets<C> { fn is_empty(&self) -> bool { match *self { DelayedLiteralSets::None => true, DelayedLiteralSets::Some(_) => false, } } } impl<C: Context> DelayedLiteralSet<C> { fn is_empty(&self) -> bool { self.delayed_literals.is_empty() } fn is_subset(&self, other: &DelayedLiteralSet<C>) -> bool { self.delayed_literals .iter() .all(|elem| other.delayed_literals.binary_search(elem).is_ok()) } } impl Minimums { const MAX: Minimums = Minimums { positive: DepthFirstNumber::MAX, negative: DepthFirstNumber::MAX, }; /// Update our fields to be the minimum of our current value /// and the values from other. fn take_minimums(&mut self, other: &Minimums) { self.positive = min(self.positive, other.positive); self.negative = min(self.negative, other.negative); } fn minimum_of_pos_and_neg(&self) -> DepthFirstNumber { min(self.positive, self.negative) } } impl DepthFirstNumber { const MIN: DepthFirstNumber = DepthFirstNumber { value: 0 }; const MAX: DepthFirstNumber = DepthFirstNumber { value: ::std::u64::MAX, }; fn next(&mut self) -> DepthFirstNumber { let value = self.value; assert!(value < ::std::u64::MAX); self.value += 1; DepthFirstNumber { value } } } /// Because we recurse so deeply, we rely on stacker to /// avoid overflowing the stack. fn maybe_grow_stack<F, R>(op: F) -> R where F: FnOnce() -> R, { // These numbers are somewhat randomly chosen to make tests work // well enough on my system. In particular, because we only test // for growing the stack in `new_clause`, a red zone of 32K was // insufficient to prevent stack overflow. - nikomatsakis stacker::maybe_grow(256 * 1024, 2 * 1024 * 1024, op) }