Module solve 

The next-generation trait solver, currently still WIP.

As a user of rust, you can use -Znext-solver to enable the new trait solver.

As a developer of rustc, you shouldn’t be using the new trait solver without asking the trait-system-refactor-initiative, but it can be enabled with InferCtxtBuilder::with_next_trait_solver. This will ensure that trait solving using that inference context will be routed to the new trait solver.

For a high-level overview of how this solver works, check out the relevant section of the rustc-dev-guide.

Re-exports

use derive_where::derive_where; 🔒

use rustc_type_ir::lang_items::SolverTraitLangItem; 🔒

use rustc_type_ir::transmute::Condition; 🔒

use rustc_type_ir as ty; 🔒

use rustc_type_ir::Const; 🔒

use rustc_type_ir::Interner; 🔒

use rustc_type_ir::Region; 🔒

use rustc_type_ir::TraitClause; 🔒

use rustc_type_ir::TypeVisitableExt; 🔒

use rustc_type_ir::Upcast; 🔒

use thin_vec::ThinVec; 🔒

use thin_vec::thin_vec; 🔒

use tracing::instrument; 🔒

use tracing::instrument; 🔒

pub use self::eval_ctxt::EvalCtxt;

pub use self::eval_ctxt::GenerateProofTree;

pub use self::eval_ctxt::SolverDelegateEvalExt;

pub use self::eval_ctxt::evaluate_root_goal_for_proof_tree_raw_provider;

pub use self::eval_ctxt::fast_path;

use crate::delegate::SolverDelegate; 🔒

use crate::solve::assembly::Candidate; 🔒

use rustc_type_ir::inherent::*; 🔒

Modules

assembly 🔒

Code shared by trait and projection goals for candidate assembly.

effect_goals 🔒

Dealing with host effect goals, i.e. enforcing the constness in T: const Trait or T: [const] Trait.

eval_ctxt 🔒

inspect

normalizes_to 🔒

project_goals 🔒

search_graph 🔒

trait_goals 🔒

Dealing with trait goals, i.e. T: Trait<'a, U>.

Structs

AccessedOpaques

ExternalConstraintsData

Additional constraints returned on success.

Goal

A goal is a statement, i.e. predicate, we want to prove given some assumptions, i.e. param_env.

GoalEvaluation

The result of evaluating a goal.

GoalStalledOn

The conditions that must change for a goal to warrant

MaybeInfo

NestedNormalizationGoals

NoSolution

QueryInput

RerunNonErased

A bit like NoSolution, but for functions that normally cannot fail unless they accessed opaues. (See TypingMode::ErasedNotCoherence). Getting OpaquesAccessed doesn’t mean there truly is no solution. It just means that we want to bail out of the current query as fast as possible, possibly by returning NoSolution if that’s fastest. This is okay because when you get OpaquesAccessed we’re guaranteed that we’re going to retry this query in the original typing mode to get the correct answer.

Response

Enums

AdtDestructorKind

Indicates that a impl Drop for Adt is const or not.

AliasBoundKind

BuiltinImplSource

CandidatePreferenceMode

Which trait candidates should be preferred over other candidates? By default, prefer where bounds over alias bounds. For marker traits, prefer alias bounds over where bounds.

CandidateSource

Possible ways the given goal can be proven.

Certainty

ComputeGoalFastPathOutcome

For some goals we can trivially answer some questions without going through canonicalization. There are three options:

ExternalRegionConstraints

FetchEligibleAssocItemResponse

GoalSource

Why a specific goal has to be proven.

GoalStalledOnOpaques

HasChanged

Whether evaluating this goal ended up changing the inference state.

IsIdentity

The variant order here matters for the PartialOrd/Ord impls.

MaybeCause

Why we failed to evaluate a goal.

NoSolutionOrRerunNonErased

OpaqueTypesJank

Supporting not-yet-defined opaque types in HIR typeck is somewhat challenging. Ideally we’d normalize them to a new inference variable and just defer type inference which relies on the opaque until we’ve constrained the hidden type.

ParamEnvSource

RerunCondition

Information about how we accessed opaque types This is what the trait solver does when each states is encountered:

RerunReason

Mainly for debugging, to keep track of the source of the rerunning in TypingMode::ErasedNotCoherence.

SizedTraitKind

Which sizedness trait - Sized, MetaSized? PointeeSized is omitted as it is removed during lowering.

SmallCopySet

A small set of up to 3 Copy elements, used as an optimization in RerunCondition. The entire set can be Copyed because of this requirement.

StalledOnCoroutines

SucceededInErased

TyOrConstInferVar

Helper for InferCtxt::ty_or_const_infer_var_changed (see comment on that), used for traits::fulfill’s list of stalled_on inference variables and for merging ambiguity errors caused by the same inference variable during error reporting.

VisibleForLeakCheck

Whether the given region constraint should be considered/ignored for leak check. In most part of the compiler, this should be Yes, except for applying constraints from the nested goals in next-solver. Unreachable is used in places in which leak check isn’t done, e.g. borrowck.

Constants

FIXPOINT_STEP_LIMIT 🔒

How many fixpoint iterations we should attempt inside of the solver before bailing with overflow.

Traits

RerunResultExt

Functions

flatten_answer_tree

Flatten the Condition tree into a conjunction of predicates.

has_no_inference_or_external_constraints 🔒

has_only_region_constraints 🔒

Type Aliases

CanonicalInputData

CanonicalResponse

QueryResult

The result of evaluating a canonical query.

QueryResultOrRerunNonErased