1 | // Copyright 2017 The Rust Project Developers. See the COPYRIGHT |
2 | // file at the top-level directory of this distribution and at |
3 | // http://rust-lang.org/COPYRIGHT. |
4 | // |
5 | // Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or |
6 | // http://www.apache.org/licenses/LICENSE-2.0> or the MIT license |
7 | // <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your |
8 | // option. This file may not be copied, modified, or distributed |
9 | // except according to those terms. |
10 | |
11 | //! A version of the Naive datalog analysis using Datafrog. |
12 | |
13 | use datafrog::{Iteration, Relation, RelationLeaper}; |
14 | use std::time::Instant; |
15 | |
16 | use crate::facts::FactTypes; |
17 | use crate::output::{Context, Output}; |
18 | |
19 | pub(super) fn compute<T: FactTypes>( |
20 | ctx: &Context<'_, T>, |
21 | result: &mut Output<T>, |
22 | ) -> ( |
23 | Relation<(T::Loan, T::Point)>, |
24 | Relation<(T::Origin, T::Origin, T::Point)>, |
25 | ) { |
26 | let timer = Instant::now(); |
27 | |
28 | let (errors, subset_errors) = { |
29 | // Static inputs |
30 | let origin_live_on_entry_rel = &ctx.origin_live_on_entry; |
31 | let cfg_edge = &ctx.cfg_edge; |
32 | let loan_killed_at = &ctx.loan_killed_at; |
33 | let known_placeholder_subset = &ctx.known_placeholder_subset; |
34 | let placeholder_origin = &ctx.placeholder_origin; |
35 | |
36 | // Create a new iteration context, ... |
37 | let mut iteration = Iteration::new(); |
38 | |
39 | // .. some variables, .. |
40 | let subset = iteration.variable::<(T::Origin, T::Origin, T::Point)>("subset" ); |
41 | let origin_contains_loan_on_entry = |
42 | iteration.variable::<(T::Origin, T::Loan, T::Point)>("origin_contains_loan_on_entry" ); |
43 | let loan_live_at = iteration.variable::<((T::Loan, T::Point), ())>("loan_live_at" ); |
44 | |
45 | // `loan_invalidated_at` facts, stored ready for joins |
46 | let loan_invalidated_at = Relation::from_iter( |
47 | ctx.loan_invalidated_at |
48 | .iter() |
49 | .map(|&(loan, point)| ((loan, point), ())), |
50 | ); |
51 | |
52 | // different indices for `subset`. |
53 | let subset_o1p = iteration.variable_indistinct("subset_o1p" ); |
54 | let subset_o2p = iteration.variable_indistinct("subset_o2p" ); |
55 | |
56 | // different index for `origin_contains_loan_on_entry`. |
57 | let origin_contains_loan_on_entry_op = |
58 | iteration.variable_indistinct("origin_contains_loan_on_entry_op" ); |
59 | |
60 | // Unfortunately, we need `origin_live_on_entry` in both variable and relation forms: |
61 | // We need: |
62 | // - `origin_live_on_entry` as a Relation for the leapjoins in rules 3 & 6 |
63 | // - `origin_live_on_entry` as a Variable for the join in rule 7 |
64 | // |
65 | // The leapjoins use `origin_live_on_entry` as `(Origin, Point)` tuples, while the join uses |
66 | // it as a `((O, P), ())` tuple to filter the `((Origin, Point), Loan)` tuples from |
67 | // `origin_contains_loan_on_entry_op`. |
68 | // |
69 | // The regular join in rule 7 could be turned into a `filter_with` leaper but that would |
70 | // result in a leapjoin with no `extend_*` leapers: a leapjoin that is not well-formed. |
71 | // Doing the filtering via an `extend_with` leaper would be extremely inefficient. |
72 | // |
73 | // Until there's an API in datafrog to handle this use-case better, we do a slightly less |
74 | // inefficient thing of copying the whole static input into a Variable to use a regular |
75 | // join, even though the liveness information can be quite heavy (around 1M tuples |
76 | // on `clap`). |
77 | // This is the Naive variant so this is not a big problem, but needs an |
78 | // explanation. |
79 | let origin_live_on_entry_var = |
80 | iteration.variable::<((T::Origin, T::Point), ())>("origin_live_on_entry" ); |
81 | origin_live_on_entry_var.extend( |
82 | origin_live_on_entry_rel |
83 | .iter() |
84 | .map(|&(origin, point)| ((origin, point), ())), |
85 | ); |
86 | |
87 | // output relations: illegal accesses errors, and illegal subset relations errors |
88 | let errors = iteration.variable("errors" ); |
89 | let subset_errors = iteration.variable::<(T::Origin, T::Origin, T::Point)>("subset_errors" ); |
90 | |
91 | // load initial facts: |
92 | |
93 | // Rule 1: the initial subsets are the non-transitive `subset_base` static input. |
94 | // |
95 | // subset(Origin1, Origin2, Point) :- |
96 | // subset_base(Origin1, Origin2, Point). |
97 | subset.extend(ctx.subset_base.iter()); |
98 | |
99 | // Rule 4: the issuing origins are the ones initially containing loans. |
100 | // |
101 | // origin_contains_loan_on_entry(Origin, Loan, Point) :- |
102 | // loan_issued_at(Origin, Loan, Point). |
103 | origin_contains_loan_on_entry.extend(ctx.loan_issued_at.iter()); |
104 | |
105 | // .. and then start iterating rules! |
106 | while iteration.changed() { |
107 | // Cleanup step: remove symmetries |
108 | // - remove origins which are `subset`s of themselves |
109 | // |
110 | // FIXME: investigate whether is there a better way to do that without complicating |
111 | // the rules too much, because it would also require temporary variables and |
112 | // impact performance. Until then, the big reduction in tuples improves performance |
113 | // a lot, even if we're potentially adding a small number of tuples |
114 | // per round just to remove them in the next round. |
115 | subset |
116 | .recent |
117 | .borrow_mut() |
118 | .elements |
119 | .retain(|&(origin1, origin2, _)| origin1 != origin2); |
120 | |
121 | // Remap fields to re-index by keys, to prepare the data needed by the rules below. |
122 | subset_o1p.from_map(&subset, |&(origin1, origin2, point)| { |
123 | ((origin1, point), origin2) |
124 | }); |
125 | subset_o2p.from_map(&subset, |&(origin1, origin2, point)| { |
126 | ((origin2, point), origin1) |
127 | }); |
128 | |
129 | origin_contains_loan_on_entry_op |
130 | .from_map(&origin_contains_loan_on_entry, |&(origin, loan, point)| { |
131 | ((origin, point), loan) |
132 | }); |
133 | |
134 | // Rule 1: done above, as part of the static input facts setup. |
135 | |
136 | // Rule 2: compute the subset transitive closure, at a given point. |
137 | // |
138 | // subset(Origin1, Origin3, Point) :- |
139 | // subset(Origin1, Origin2, Point), |
140 | // subset(Origin2, Origin3, Point). |
141 | subset.from_join( |
142 | &subset_o2p, |
143 | &subset_o1p, |
144 | |&(_origin2, point), &origin1, &origin3| (origin1, origin3, point), |
145 | ); |
146 | |
147 | // Rule 3: propagate subsets along the CFG, according to liveness. |
148 | // |
149 | // subset(Origin1, Origin2, Point2) :- |
150 | // subset(Origin1, Origin2, Point1), |
151 | // cfg_edge(Point1, Point2), |
152 | // origin_live_on_entry(Origin1, Point2), |
153 | // origin_live_on_entry(Origin2, Point2). |
154 | subset.from_leapjoin( |
155 | &subset, |
156 | ( |
157 | cfg_edge.extend_with(|&(_origin1, _origin2, point1)| point1), |
158 | origin_live_on_entry_rel.extend_with(|&(origin1, _origin2, _point1)| origin1), |
159 | origin_live_on_entry_rel.extend_with(|&(_origin1, origin2, _point1)| origin2), |
160 | ), |
161 | |&(origin1, origin2, _point1), &point2| (origin1, origin2, point2), |
162 | ); |
163 | |
164 | // Rule 4: done above as part of the static input facts setup. |
165 | |
166 | // Rule 5: propagate loans within origins, at a given point, according to subsets. |
167 | // |
168 | // origin_contains_loan_on_entry(Origin2, Loan, Point) :- |
169 | // origin_contains_loan_on_entry(Origin1, Loan, Point), |
170 | // subset(Origin1, Origin2, Point). |
171 | origin_contains_loan_on_entry.from_join( |
172 | &origin_contains_loan_on_entry_op, |
173 | &subset_o1p, |
174 | |&(_origin1, point), &loan, &origin2| (origin2, loan, point), |
175 | ); |
176 | |
177 | // Rule 6: propagate loans along the CFG, according to liveness. |
178 | // |
179 | // origin_contains_loan_on_entry(Origin, Loan, Point2) :- |
180 | // origin_contains_loan_on_entry(Origin, Loan, Point1), |
181 | // !loan_killed_at(Loan, Point1), |
182 | // cfg_edge(Point1, Point2), |
183 | // origin_live_on_entry(Origin, Point2). |
184 | origin_contains_loan_on_entry.from_leapjoin( |
185 | &origin_contains_loan_on_entry, |
186 | ( |
187 | loan_killed_at.filter_anti(|&(_origin, loan, point1)| (loan, point1)), |
188 | cfg_edge.extend_with(|&(_origin, _loan, point1)| point1), |
189 | origin_live_on_entry_rel.extend_with(|&(origin, _loan, _point1)| origin), |
190 | ), |
191 | |&(origin, loan, _point1), &point2| (origin, loan, point2), |
192 | ); |
193 | |
194 | // Rule 7: compute whether a loan is live at a given point, i.e. whether it is |
195 | // contained in a live origin at this point. |
196 | // |
197 | // loan_live_at(Loan, Point) :- |
198 | // origin_contains_loan_on_entry(Origin, Loan, Point), |
199 | // origin_live_on_entry(Origin, Point). |
200 | loan_live_at.from_join( |
201 | &origin_contains_loan_on_entry_op, |
202 | &origin_live_on_entry_var, |
203 | |&(_origin, point), &loan, _| ((loan, point), ()), |
204 | ); |
205 | |
206 | // Rule 8: compute illegal access errors, i.e. an invalidation of a live loan. |
207 | // |
208 | // Here again, this join acts as a pure filter and could be a more efficient leapjoin. |
209 | // However, similarly to the `origin_live_on_entry` example described above, the |
210 | // leapjoin with a single `filter_with` leaper would currently not be well-formed. |
211 | // We don't explictly need to materialize `loan_live_at` either, and that doesn't |
212 | // change the well-formedness situation, so we still materialize it (since that also |
213 | // helps in testing). |
214 | // |
215 | // errors(Loan, Point) :- |
216 | // loan_invalidated_at(Loan, Point), |
217 | // loan_live_at(Loan, Point). |
218 | errors.from_join( |
219 | &loan_live_at, |
220 | &loan_invalidated_at, |
221 | |&(loan, point), _, _| (loan, point), |
222 | ); |
223 | |
224 | // Rule 9: compute illegal subset relations errors, i.e. the undeclared subsets |
225 | // between two placeholder origins. |
226 | // Here as well, WF-ness prevents this join from being a filter-only leapjoin. It |
227 | // doesn't matter much, as `placeholder_origin` is single-value relation. |
228 | // |
229 | // subset_error(Origin1, Origin2, Point) :- |
230 | // subset(Origin1, Origin2, Point), |
231 | // placeholder_origin(Origin1), |
232 | // placeholder_origin(Origin2), |
233 | // !known_placeholder_subset(Origin1, Origin2). |
234 | subset_errors.from_leapjoin( |
235 | &subset, |
236 | ( |
237 | placeholder_origin.extend_with(|&(origin1, _origin2, _point)| origin1), |
238 | placeholder_origin.extend_with(|&(_origin1, origin2, _point)| origin2), |
239 | known_placeholder_subset |
240 | .filter_anti(|&(origin1, origin2, _point)| (origin1, origin2)), |
241 | // remove symmetries: |
242 | datafrog::ValueFilter::from(|&(origin1, origin2, _point), _| { |
243 | origin1 != origin2 |
244 | }), |
245 | ), |
246 | |&(origin1, origin2, point), _| (origin1, origin2, point), |
247 | ); |
248 | } |
249 | |
250 | // Handle verbose output data |
251 | if result.dump_enabled { |
252 | let subset = subset.complete(); |
253 | assert!( |
254 | subset |
255 | .iter() |
256 | .filter(|&(origin1, origin2, _)| origin1 == origin2) |
257 | .count() |
258 | == 0, |
259 | "unwanted subset symmetries" |
260 | ); |
261 | for &(origin1, origin2, location) in subset.iter() { |
262 | result |
263 | .subset |
264 | .entry(location) |
265 | .or_default() |
266 | .entry(origin1) |
267 | .or_default() |
268 | .insert(origin2); |
269 | } |
270 | |
271 | let origin_contains_loan_on_entry = origin_contains_loan_on_entry.complete(); |
272 | for &(origin, loan, location) in origin_contains_loan_on_entry.iter() { |
273 | result |
274 | .origin_contains_loan_at |
275 | .entry(location) |
276 | .or_default() |
277 | .entry(origin) |
278 | .or_default() |
279 | .insert(loan); |
280 | } |
281 | |
282 | let loan_live_at = loan_live_at.complete(); |
283 | for &((loan, location), _) in loan_live_at.iter() { |
284 | result.loan_live_at.entry(location).or_default().push(loan); |
285 | } |
286 | } |
287 | |
288 | (errors.complete(), subset_errors.complete()) |
289 | }; |
290 | |
291 | info!( |
292 | "analysis done: {} `errors` tuples, {} `subset_errors` tuples, {:?}" , |
293 | errors.len(), |
294 | subset_errors.len(), |
295 | timer.elapsed() |
296 | ); |
297 | |
298 | (errors, subset_errors) |
299 | } |
300 | |