Struct rustc_trait_selection::solve::EvalCtxt

pub struct EvalCtxt<'a, 'tcx> {
    infcx: &'a InferCtxt<'tcx>,
    pub(super) var_values: CanonicalVarValues<'tcx>,
    predefined_opaques_in_body: PredefinedOpaques<'tcx>,
    pub(super) max_input_universe: UniverseIndex,
    pub(super) search_graph: &'a mut SearchGraph<'tcx>,
    pub(super) nested_goals: NestedGoals<'tcx>,
    tainted: Result<(), NoSolution>,
}

Fields

infcx: &'a InferCtxt<'tcx>

The inference context that backs (mostly) inference and placeholder terms instantiated while solving goals.

NOTE: The InferCtxt that backs the EvalCtxt is intentionally private, because the InferCtxt is much more general than EvalCtxt. Methods such as take_registered_region_obligations can mess up query responses, using At::normalize is totally wrong, calling evaluate_root_goal can cause coinductive unsoundness, etc.

Methods that are generally of use for trait solving are intentionally re-declared through the EvalCtxt below, often with cleaner signatures since we don’t care about things like ObligationCauses and Spans here. If some InferCtxt method is missing, please first think defensively about the method’s compatibility with this solver, or if an existing one does the job already.

var_values: CanonicalVarValues<'tcx>

predefined_opaques_in_body: PredefinedOpaques<'tcx>

max_input_universe: UniverseIndex

The highest universe index nameable by the caller.

When we enter a new binder inside of the query we create new universes which the caller cannot name. We have to be careful with variables from these new universes when creating the query response.

Both because these new universes can prevent us from reaching a fixpoint if we have a coinductive cycle and because that’s the only way we can return new placeholders to the caller.

search_graph: &'a mut SearchGraph<'tcx>

nested_goals: NestedGoals<'tcx>

tainted: Result<(), NoSolution>

Implementations

impl<'tcx> EvalCtxt<'_, 'tcx>

pub(super) fn assemble_and_evaluate_candidates<G: GoalKind<'tcx>>( &mut self, goal: Goal<'tcx, G> ) -> Vec<Candidate<'tcx>>

fn assemble_candidates_after_normalizing_self_ty<G: GoalKind<'tcx>>( &mut self, goal: Goal<'tcx, G>, candidates: &mut Vec<Candidate<'tcx>> )

If the self type of a goal is a projection, computing the relevant candidates is difficult.

To deal with this, we first try to normalize the self type and add the candidates for the normalized self type to the list of candidates in case that succeeds. We also have to consider candidates with the projection as a self type as well

fn assemble_impl_candidates<G: GoalKind<'tcx>>( &mut self, goal: Goal<'tcx, G>, candidates: &mut Vec<Candidate<'tcx>> )

fn assemble_builtin_impl_candidates<G: GoalKind<'tcx>>( &mut self, goal: Goal<'tcx, G>, candidates: &mut Vec<Candidate<'tcx>> )

fn assemble_param_env_candidates<G: GoalKind<'tcx>>( &mut self, goal: Goal<'tcx, G>, candidates: &mut Vec<Candidate<'tcx>> )

fn assemble_alias_bound_candidates<G: GoalKind<'tcx>>( &mut self, goal: Goal<'tcx, G>, candidates: &mut Vec<Candidate<'tcx>> )

pub(super) fn validate_alias_bound_self_from_param_env<G: GoalKind<'tcx>>( &mut self, goal: Goal<'tcx, G> ) -> QueryResult<'tcx>

Check that we are allowed to use an alias bound originating from the self type of this goal. This means something different depending on the self type’s alias kind.

• Projection: Given a goal with a self type such as <Ty as Trait>::Assoc, we require that the bound Ty: Trait can be proven using either a nested alias bound candidate, or a param-env candidate.

• Opaque: The param-env must be in Reveal::UserFacing mode. Otherwise, the goal should be proven by using the hidden type instead.

fn assemble_alias_bound_candidates_for_builtin_impl_default_items<G: GoalKind<'tcx>>( &mut self, goal: Goal<'tcx, G>, candidates: &mut Vec<Candidate<'tcx>> )

Assemble a subset of builtin impl candidates for a class of candidates called “defaulted” built-in traits.

For example, we always know that T: Pointee is implemented, but we do not always know what <T as Pointee>::Metadata actually is! See the comment in EvalCtxt::validate_alias_bound_self_from_param_env for more detail.

fn assemble_object_bound_candidates<G: GoalKind<'tcx>>( &mut self, goal: Goal<'tcx, G>, candidates: &mut Vec<Candidate<'tcx>> )

fn assemble_coherence_unknowable_candidates<G: GoalKind<'tcx>>( &mut self, goal: Goal<'tcx, G>, candidates: &mut Vec<Candidate<'tcx>> )

pub(super) fn merge_candidates( &mut self, candidates: Vec<Candidate<'tcx>> ) -> QueryResult<'tcx>

If there are multiple ways to prove a trait or projection goal, we have to somehow try to merge the candidates into one. If that fails, we return ambiguity.

impl<'tcx> EvalCtxt<'_, 'tcx>

pub(super) fn canonicalize_goal( &self, goal: Goal<'tcx, Predicate<'tcx>> ) -> (Vec<GenericArg<'tcx>>, CanonicalInput<'tcx>)

Canonicalizes the goal remembering the original values for each bound variable.

pub(in solve) fn evaluate_added_goals_and_make_canonical_response( &mut self, certainty: Certainty ) -> QueryResult<'tcx>

To return the constraints of a canonical query to the caller, we canonicalize:

• var_values: a map from bound variables in the canonical goal to the values inferred while solving the instantiated goal.
• external_constraints: additional constraints which aren’t expressible using simple unification of inference variables.

pub(in solve) fn make_ambiguous_response_no_constraints( &self, maybe_cause: MaybeCause ) -> CanonicalResponse<'tcx>

Constructs a totally unconstrained, ambiguous response to a goal.

Take care when using this, since often it’s useful to respond with ambiguity but return constrained variables to guide inference.

fn compute_external_query_constraints( &self ) -> Result<ExternalConstraints<'tcx>, NoSolution>

pub(super) fn instantiate_and_apply_query_response( &mut self, param_env: ParamEnv<'tcx>, original_values: Vec<GenericArg<'tcx>>, response: CanonicalResponse<'tcx> ) -> Result<(Certainty, Vec<Goal<'tcx, Predicate<'tcx>>>), NoSolution>

After calling a canonical query, we apply the constraints returned by the query using this function.

This happens in three steps:

• we instantiate the bound variables of the query response
• we unify the var_values of the response with the original_values
• we apply the external_constraints returned by the query

fn compute_query_response_substitution( &self, original_values: &[GenericArg<'tcx>], response: &CanonicalResponse<'tcx> ) -> CanonicalVarValues<'tcx>

This returns the substitutions to instantiate the bound variables of the canonical response. This depends on the original_values for the bound variables.

fn unify_query_var_values( &self, param_env: ParamEnv<'tcx>, original_values: &[GenericArg<'tcx>], var_values: CanonicalVarValues<'tcx> ) -> Result<Vec<Goal<'tcx, Predicate<'tcx>>>, NoSolution>

fn register_region_constraints( &mut self, region_constraints: &QueryRegionConstraints<'tcx> )

fn register_opaque_types( &mut self, param_env: ParamEnv<'tcx>, opaque_types: &[(OpaqueTypeKey<'tcx>, Ty<'tcx>)] ) -> Result<(), NoSolution>

impl<'a, 'tcx> EvalCtxt<'a, 'tcx>

pub(super) fn solver_mode(&self) -> SolverMode

fn evaluate_canonical_goal( tcx: TyCtxt<'tcx>, search_graph: &'a mut SearchGraph<'tcx>, canonical_input: CanonicalInput<'tcx> ) -> QueryResult<'tcx>

The entry point of the solver.

This function deals with (coinductive) cycles, overflow, and caching and then calls EvalCtxt::compute_goal which contains the actual logic of the solver.

Instead of calling this function directly, use either EvalCtxt::evaluate_goal if you’re inside of the solver or InferCtxtEvalExt::evaluate_root_goal if you’re outside of it.

fn evaluate_goal( &mut self, is_normalizes_to_hack: IsNormalizesToHack, goal: Goal<'tcx, Predicate<'tcx>> ) -> Result<(bool, Certainty, Vec<Goal<'tcx, Predicate<'tcx>>>), NoSolution>

Recursively evaluates goal, returning whether any inference vars have been constrained and the certainty of the result.

fn compute_goal( &mut self, goal: Goal<'tcx, Predicate<'tcx>> ) -> QueryResult<'tcx>

pub(super) fn try_evaluate_added_goals( &mut self ) -> Result<Certainty, NoSolution>

impl<'tcx> EvalCtxt<'_, 'tcx>

pub(super) fn probe<T>( &mut self, f: impl FnOnce(&mut EvalCtxt<'_, 'tcx>) -> T ) -> T

pub(super) fn tcx(&self) -> TyCtxt<'tcx>

pub(super) fn next_ty_infer(&self) -> Ty<'tcx>

pub(super) fn next_region_infer(&self) -> Region<'tcx>

pub(super) fn next_const_infer(&self, ty: Ty<'tcx>) -> Const<'tcx>

pub(super) fn next_term_infer_of_kind(&self, kind: Term<'tcx>) -> Term<'tcx>

Returns a ty infer or a const infer depending on whether kind is a Ty or Const. If kind is an integer inference variable this will still return a ty infer var.

pub(super) fn term_is_fully_unconstrained( &self, goal: Goal<'tcx, ProjectionPredicate<'tcx>> ) -> bool

Is the projection predicate is of the form exists<T> <Ty as Trait>::Assoc = T.

This is the case if the term is an inference variable in the innermost universe and does not occur in any other part of the predicate.

pub(super) fn eq<T: ToTrace<'tcx>>( &mut self, param_env: ParamEnv<'tcx>, lhs: T, rhs: T ) -> Result<(), NoSolution>

pub(super) fn sub<T: ToTrace<'tcx>>( &mut self, param_env: ParamEnv<'tcx>, sub: T, sup: T ) -> Result<(), NoSolution>

pub(super) fn eq_and_get_goals<T: ToTrace<'tcx>>( &self, param_env: ParamEnv<'tcx>, lhs: T, rhs: T ) -> Result<Vec<Goal<'tcx, Predicate<'tcx>>>, NoSolution>

Equates two values returning the nested goals without adding them to the nested goals of the EvalCtxt.

If possible, try using eq instead which automatically handles nested goals correctly.

pub(super) fn instantiate_binder_with_infer<T: TypeFoldable<TyCtxt<'tcx>> + Copy>( &self, value: Binder<'tcx, T> ) -> T

pub(super) fn instantiate_binder_with_placeholders<T: TypeFoldable<TyCtxt<'tcx>> + Copy>( &self, value: Binder<'tcx, T> ) -> T

pub(super) fn resolve_vars_if_possible<T>(&self, value: T) -> Twhere T: TypeFoldable<TyCtxt<'tcx>>,

pub(super) fn fresh_substs_for_item(&self, def_id: DefId) -> SubstsRef<'tcx>

pub(super) fn translate_substs( &self, param_env: ParamEnv<'tcx>, source_impl: DefId, source_substs: SubstsRef<'tcx>, target_node: Node ) -> SubstsRef<'tcx>

pub(super) fn register_ty_outlives(&self, ty: Ty<'tcx>, lt: Region<'tcx>)

pub(super) fn register_region_outlives(&self, a: Region<'tcx>, b: Region<'tcx>)

pub(super) fn well_formed_goals( &self, param_env: ParamEnv<'tcx>, arg: GenericArg<'tcx> ) -> Option<impl Iterator<Item = Goal<'tcx, Predicate<'tcx>>>>

Computes the list of goals required for arg to be well-formed

pub(super) fn is_transmutable( &self, src_and_dst: Types<'tcx>, scope: Ty<'tcx>, assume: Assume ) -> Result<Certainty, NoSolution>

pub(super) fn can_define_opaque_ty(&mut self, def_id: LocalDefId) -> bool

pub(super) fn register_opaque_ty( &mut self, a: OpaqueTypeKey<'tcx>, b: Ty<'tcx>, param_env: ParamEnv<'tcx> ) -> Result<(), NoSolution>

pub(super) fn unify_existing_opaque_tys( &mut self, param_env: ParamEnv<'tcx>, key: OpaqueTypeKey<'tcx>, ty: Ty<'tcx> ) -> Vec<CanonicalResponse<'tcx>>

pub(super) fn try_const_eval_resolve( &self, param_env: ParamEnv<'tcx>, unevaluated: UnevaluatedConst<'tcx>, ty: Ty<'tcx> ) -> Option<Const<'tcx>>

impl<'tcx> EvalCtxt<'_, 'tcx>

pub(super) fn normalize_opaque_type( &mut self, goal: Goal<'tcx, ProjectionPredicate<'tcx>> ) -> QueryResult<'tcx>

impl<'tcx> EvalCtxt<'_, 'tcx>

pub(super) fn compute_projection_goal( &mut self, goal: Goal<'tcx, ProjectionPredicate<'tcx>> ) -> QueryResult<'tcx>

fn normalize_anon_const( &mut self, goal: Goal<'tcx, ProjectionPredicate<'tcx>> ) -> QueryResult<'tcx>

impl<'tcx> EvalCtxt<'_, 'tcx>

fn disqualify_auto_trait_candidate_due_to_possible_impl( &mut self, goal: Goal<'tcx, TraitPredicate<'tcx>> ) -> Option<QueryResult<'tcx>>

fn probe_and_evaluate_goal_for_constituent_tys( &mut self, goal: Goal<'tcx, TraitPredicate<'tcx>>, constituent_tys: impl Fn(&EvalCtxt<'_, 'tcx>, Ty<'tcx>) -> Result<Vec<Ty<'tcx>>, NoSolution> ) -> QueryResult<'tcx>

Convenience function for traits that are structural, i.e. that only have nested subgoals that only change the self type. Unlike other evaluate-like helpers, this does a probe, so it doesn’t need to be wrapped in one.

pub(super) fn compute_trait_goal( &mut self, goal: Goal<'tcx, TraitPredicate<'tcx>> ) -> QueryResult<'tcx>

impl<'a, 'tcx> EvalCtxt<'a, 'tcx>

fn compute_type_outlives_goal( &mut self, goal: Goal<'tcx, TypeOutlivesPredicate<'tcx>> ) -> QueryResult<'tcx>

fn compute_region_outlives_goal( &mut self, goal: Goal<'tcx, RegionOutlivesPredicate<'tcx>> ) -> QueryResult<'tcx>

fn compute_coerce_goal( &mut self, goal: Goal<'tcx, CoercePredicate<'tcx>> ) -> QueryResult<'tcx>

fn compute_subtype_goal( &mut self, goal: Goal<'tcx, SubtypePredicate<'tcx>> ) -> QueryResult<'tcx>

fn compute_closure_kind_goal( &mut self, goal: Goal<'tcx, (DefId, SubstsRef<'tcx>, ClosureKind)> ) -> QueryResult<'tcx>

fn compute_object_safe_goal(&mut self, trait_def_id: DefId) -> QueryResult<'tcx>

fn compute_well_formed_goal( &mut self, goal: Goal<'tcx, GenericArg<'tcx>> ) -> QueryResult<'tcx>

fn compute_alias_relate_goal( &mut self, goal: Goal<'tcx, (Term<'tcx>, Term<'tcx>, AliasRelationDirection)> ) -> QueryResult<'tcx>

fn compute_const_arg_has_type_goal( &mut self, goal: Goal<'tcx, (Const<'tcx>, Ty<'tcx>)> ) -> QueryResult<'tcx>

impl<'tcx> EvalCtxt<'_, 'tcx>

fn set_normalizes_to_hack_goal( &mut self, goal: Goal<'tcx, ProjectionPredicate<'tcx>> )

fn add_goal(&mut self, goal: Goal<'tcx, Predicate<'tcx>>)

fn add_goals( &mut self, goals: impl IntoIterator<Item = Goal<'tcx, Predicate<'tcx>>> )

fn try_merge_responses( &mut self, responses: &[CanonicalResponse<'tcx>] ) -> Option<CanonicalResponse<'tcx>>

Try to merge multiple possible ways to prove a goal, if that is not possible returns None.

In this case we tend to flounder and return ambiguity by calling [EvalCtxt::flounder].

fn flounder( &mut self, responses: &[CanonicalResponse<'tcx>] ) -> QueryResult<'tcx>

If we fail to merge responses we flounder and return overflow or ambiguity.

Trait Implementations

impl<'tcx> OverflowHandler<'tcx> for EvalCtxt<'_, 'tcx>

fn search_graph(&mut self) -> &mut SearchGraph<'tcx>

fn repeat_while_none<T>( &mut self, on_overflow: impl FnOnce(&mut Self) -> Result<T, NoSolution>, loop_body: impl FnMut(&mut Self) -> Option<Result<T, NoSolution>> ) -> Result<T, NoSolution>

fn with_incremented_depth<T>( &mut self, on_overflow: impl FnOnce(&mut Self) -> T, body: impl FnOnce(&mut Self) -> T ) -> T

Layout

Note: Most layout information is completely unstable and may even differ between compilations. The only exception is types with certain repr(...) attributes. Please see the Rust Reference’s “Type Layout” chapter for details on type layout guarantees.

Size:96 bytes