rustc_infer/infer/outlives/
env.rs

1use rustc_data_structures::fx::{FxHashSet, FxIndexSet};
2use rustc_data_structures::transitive_relation::TransitiveRelationBuilder;
3use rustc_middle::{bug, ty};
4use tracing::debug;
5
6use super::explicit_outlives_bounds;
7use crate::infer::GenericKind;
8use crate::infer::free_regions::FreeRegionMap;
9use crate::traits::query::OutlivesBound;
10
11/// The `OutlivesEnvironment` collects information about what outlives
12/// what in a given type-checking setting. For example, if we have a
13/// where-clause like `where T: 'a` in scope, then the
14/// `OutlivesEnvironment` would record that (in its
15/// `region_bound_pairs` field). Similarly, it contains methods for
16/// processing and adding implied bounds into the outlives
17/// environment.
18///
19/// Other code at present does not typically take a
20/// `&OutlivesEnvironment`, but rather takes some of its fields (e.g.,
21/// `process_registered_region_obligations` wants the
22/// region-bound-pairs). There is no mistaking it: the current setup
23/// of tracking region information is quite scattered! The
24/// `OutlivesEnvironment`, for example, needs to sometimes be combined
25/// with the `middle::RegionRelations`, to yield a full picture of how
26/// (lexical) lifetimes interact. However, I'm reluctant to do more
27/// refactoring here, since the setup with NLL is quite different.
28/// For example, NLL has no need of `RegionRelations`, and is solely
29/// interested in the `OutlivesEnvironment`. -nmatsakis
30#[derive(Clone)]
31pub struct OutlivesEnvironment<'tcx> {
32    pub param_env: ty::ParamEnv<'tcx>,
33    free_region_map: FreeRegionMap<'tcx>,
34    /// FIXME: Your first reaction may be that this is a bit strange. `RegionBoundPairs`
35    /// does not contain lifetimes, which are instead in the `FreeRegionMap`, and other
36    /// known type outlives are stored in the `known_type_outlives` set. So why do we
37    /// have these at all? It turns out that removing these and using `known_type_outlives`
38    /// everywhere is just enough of a perf regression to matter. This can/should be
39    /// optimized in the future, though.
40    region_bound_pairs: RegionBoundPairs<'tcx>,
41    known_type_outlives: Vec<ty::PolyTypeOutlivesPredicate<'tcx>>,
42    /// Assumptions that come from the well-formedness of coroutines that we prove
43    /// auto trait bounds for during the type checking of this body.
44    higher_ranked_assumptions: FxHashSet<ty::ArgOutlivesPredicate<'tcx>>,
45}
46
47/// "Region-bound pairs" tracks outlives relations that are known to
48/// be true, either because of explicit where-clauses like `T: 'a` or
49/// because of implied bounds.
50pub type RegionBoundPairs<'tcx> = FxIndexSet<ty::OutlivesPredicate<'tcx, GenericKind<'tcx>>>;
51
52impl<'tcx> OutlivesEnvironment<'tcx> {
53    /// Create a new `OutlivesEnvironment` from normalized outlives bounds.
54    pub fn from_normalized_bounds(
55        param_env: ty::ParamEnv<'tcx>,
56        known_type_outlives: Vec<ty::PolyTypeOutlivesPredicate<'tcx>>,
57        extra_bounds: impl IntoIterator<Item = OutlivesBound<'tcx>>,
58        higher_ranked_assumptions: FxHashSet<ty::ArgOutlivesPredicate<'tcx>>,
59    ) -> Self {
60        let mut region_relation = TransitiveRelationBuilder::default();
61        let mut region_bound_pairs = RegionBoundPairs::default();
62
63        // Record relationships such as `T:'x` that don't go into the
64        // free-region-map but which we use here.
65        for outlives_bound in explicit_outlives_bounds(param_env).chain(extra_bounds) {
66            debug!("add_outlives_bounds: outlives_bound={:?}", outlives_bound);
67            match outlives_bound {
68                OutlivesBound::RegionSubParam(r_a, param_b) => {
69                    region_bound_pairs
70                        .insert(ty::OutlivesPredicate(GenericKind::Param(param_b), r_a));
71                }
72                OutlivesBound::RegionSubAlias(r_a, alias_b) => {
73                    region_bound_pairs
74                        .insert(ty::OutlivesPredicate(GenericKind::Alias(alias_b), r_a));
75                }
76                OutlivesBound::RegionSubRegion(r_a, r_b) => match (r_a.kind(), r_b.kind()) {
77                    (
78                        ty::ReStatic | ty::ReEarlyParam(_) | ty::ReLateParam(_),
79                        ty::ReStatic | ty::ReEarlyParam(_) | ty::ReLateParam(_),
80                    ) => region_relation.add(r_a, r_b),
81                    (ty::ReError(_), _) | (_, ty::ReError(_)) => {}
82                    // FIXME(#109628): We shouldn't have existential variables in implied bounds.
83                    // Panic here once the linked issue is resolved!
84                    (ty::ReVar(_), _) | (_, ty::ReVar(_)) => {}
85                    _ => bug!("add_outlives_bounds: unexpected regions: ({r_a:?}, {r_b:?})"),
86                },
87            }
88        }
89
90        OutlivesEnvironment {
91            param_env,
92            known_type_outlives,
93            free_region_map: FreeRegionMap { relation: region_relation.freeze() },
94            region_bound_pairs,
95            higher_ranked_assumptions,
96        }
97    }
98
99    pub fn free_region_map(&self) -> &FreeRegionMap<'tcx> {
100        &self.free_region_map
101    }
102
103    pub fn region_bound_pairs(&self) -> &RegionBoundPairs<'tcx> {
104        &self.region_bound_pairs
105    }
106
107    pub fn known_type_outlives(&self) -> &[ty::PolyTypeOutlivesPredicate<'tcx>] {
108        &self.known_type_outlives
109    }
110
111    pub fn higher_ranked_assumptions(&self) -> &FxHashSet<ty::ArgOutlivesPredicate<'tcx>> {
112        &self.higher_ranked_assumptions
113    }
114}
rustc_infer/infer/outlives/env.rs

rustc_infer/infer/outlives/
env.rs
