b423a0f9ef
There's a lot of stuff wrong with the representation of these types: TyFnDef doesn't actually uniquely identify a function, TyFnPtr is used to represent method calls, TyFnDef in the sub-expression of a cast isn't correctly reified, and probably some other stuff I haven't discovered yet. Splitting them seems like the right first step, though.
510 lines
21 KiB
Rust
510 lines
21 KiB
Rust
// Copyright 2012-2013 The Rust Project Developers. See the COPYRIGHT
|
|
// file at the top-level directory of this distribution and at
|
|
// http://rust-lang.org/COPYRIGHT.
|
|
//
|
|
// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
|
|
// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
|
|
// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
|
|
// option. This file may not be copied, modified, or distributed
|
|
// except according to those terms.
|
|
|
|
use middle::def_id::DefId;
|
|
use middle::infer::InferCtxt;
|
|
use middle::ty::outlives::{self, Component};
|
|
use middle::subst::Substs;
|
|
use middle::traits;
|
|
use middle::ty::{self, ToPredicate, Ty, TyCtxt, TypeFoldable};
|
|
use std::iter::once;
|
|
use syntax::ast;
|
|
use syntax::codemap::Span;
|
|
use util::common::ErrorReported;
|
|
|
|
/// Returns the set of obligations needed to make `ty` well-formed.
|
|
/// If `ty` contains unresolved inference variables, this may include
|
|
/// further WF obligations. However, if `ty` IS an unresolved
|
|
/// inference variable, returns `None`, because we are not able to
|
|
/// make any progress at all. This is to prevent "livelock" where we
|
|
/// say "$0 is WF if $0 is WF".
|
|
pub fn obligations<'a,'tcx>(infcx: &InferCtxt<'a, 'tcx>,
|
|
body_id: ast::NodeId,
|
|
ty: Ty<'tcx>,
|
|
span: Span)
|
|
-> Option<Vec<traits::PredicateObligation<'tcx>>>
|
|
{
|
|
let mut wf = WfPredicates { infcx: infcx,
|
|
body_id: body_id,
|
|
span: span,
|
|
out: vec![] };
|
|
if wf.compute(ty) {
|
|
debug!("wf::obligations({:?}, body_id={:?}) = {:?}", ty, body_id, wf.out);
|
|
let result = wf.normalize();
|
|
debug!("wf::obligations({:?}, body_id={:?}) ~~> {:?}", ty, body_id, result);
|
|
Some(result)
|
|
} else {
|
|
None // no progress made, return None
|
|
}
|
|
}
|
|
|
|
/// Returns the obligations that make this trait reference
|
|
/// well-formed. For example, if there is a trait `Set` defined like
|
|
/// `trait Set<K:Eq>`, then the trait reference `Foo: Set<Bar>` is WF
|
|
/// if `Bar: Eq`.
|
|
pub fn trait_obligations<'a,'tcx>(infcx: &InferCtxt<'a, 'tcx>,
|
|
body_id: ast::NodeId,
|
|
trait_ref: &ty::TraitRef<'tcx>,
|
|
span: Span)
|
|
-> Vec<traits::PredicateObligation<'tcx>>
|
|
{
|
|
let mut wf = WfPredicates { infcx: infcx, body_id: body_id, span: span, out: vec![] };
|
|
wf.compute_trait_ref(trait_ref);
|
|
wf.normalize()
|
|
}
|
|
|
|
pub fn predicate_obligations<'a,'tcx>(infcx: &InferCtxt<'a, 'tcx>,
|
|
body_id: ast::NodeId,
|
|
predicate: &ty::Predicate<'tcx>,
|
|
span: Span)
|
|
-> Vec<traits::PredicateObligation<'tcx>>
|
|
{
|
|
let mut wf = WfPredicates { infcx: infcx, body_id: body_id, span: span, out: vec![] };
|
|
|
|
// (*) ok to skip binders, because wf code is prepared for it
|
|
match *predicate {
|
|
ty::Predicate::Trait(ref t) => {
|
|
wf.compute_trait_ref(&t.skip_binder().trait_ref); // (*)
|
|
}
|
|
ty::Predicate::Equate(ref t) => {
|
|
wf.compute(t.skip_binder().0);
|
|
wf.compute(t.skip_binder().1);
|
|
}
|
|
ty::Predicate::RegionOutlives(..) => {
|
|
}
|
|
ty::Predicate::TypeOutlives(ref t) => {
|
|
wf.compute(t.skip_binder().0);
|
|
}
|
|
ty::Predicate::Projection(ref t) => {
|
|
let t = t.skip_binder(); // (*)
|
|
wf.compute_projection(t.projection_ty);
|
|
wf.compute(t.ty);
|
|
}
|
|
ty::Predicate::WellFormed(t) => {
|
|
wf.compute(t);
|
|
}
|
|
ty::Predicate::ObjectSafe(_) => {
|
|
}
|
|
}
|
|
|
|
wf.normalize()
|
|
}
|
|
|
|
/// Implied bounds are region relationships that we deduce
|
|
/// automatically. The idea is that (e.g.) a caller must check that a
|
|
/// function's argument types are well-formed immediately before
|
|
/// calling that fn, and hence the *callee* can assume that its
|
|
/// argument types are well-formed. This may imply certain relationships
|
|
/// between generic parameters. For example:
|
|
///
|
|
/// fn foo<'a,T>(x: &'a T)
|
|
///
|
|
/// can only be called with a `'a` and `T` such that `&'a T` is WF.
|
|
/// For `&'a T` to be WF, `T: 'a` must hold. So we can assume `T: 'a`.
|
|
#[derive(Debug)]
|
|
pub enum ImpliedBound<'tcx> {
|
|
RegionSubRegion(ty::Region, ty::Region),
|
|
RegionSubParam(ty::Region, ty::ParamTy),
|
|
RegionSubProjection(ty::Region, ty::ProjectionTy<'tcx>),
|
|
}
|
|
|
|
/// Compute the implied bounds that a callee/impl can assume based on
|
|
/// the fact that caller/projector has ensured that `ty` is WF. See
|
|
/// the `ImpliedBound` type for more details.
|
|
pub fn implied_bounds<'a,'tcx>(
|
|
infcx: &'a InferCtxt<'a,'tcx>,
|
|
body_id: ast::NodeId,
|
|
ty: Ty<'tcx>,
|
|
span: Span)
|
|
-> Vec<ImpliedBound<'tcx>>
|
|
{
|
|
// Sometimes when we ask what it takes for T: WF, we get back that
|
|
// U: WF is required; in that case, we push U onto this stack and
|
|
// process it next. Currently (at least) these resulting
|
|
// predicates are always guaranteed to be a subset of the original
|
|
// type, so we need not fear non-termination.
|
|
let mut wf_types = vec![ty];
|
|
|
|
let mut implied_bounds = vec![];
|
|
|
|
while let Some(ty) = wf_types.pop() {
|
|
// Compute the obligations for `ty` to be well-formed. If `ty` is
|
|
// an unresolved inference variable, just substituted an empty set
|
|
// -- because the return type here is going to be things we *add*
|
|
// to the environment, it's always ok for this set to be smaller
|
|
// than the ultimate set. (Note: normally there won't be
|
|
// unresolved inference variables here anyway, but there might be
|
|
// during typeck under some circumstances.)
|
|
let obligations = obligations(infcx, body_id, ty, span).unwrap_or(vec![]);
|
|
|
|
// From the full set of obligations, just filter down to the
|
|
// region relationships.
|
|
implied_bounds.extend(
|
|
obligations
|
|
.into_iter()
|
|
.flat_map(|obligation| {
|
|
assert!(!obligation.has_escaping_regions());
|
|
match obligation.predicate {
|
|
ty::Predicate::Trait(..) |
|
|
ty::Predicate::Equate(..) |
|
|
ty::Predicate::Projection(..) |
|
|
ty::Predicate::ObjectSafe(..) =>
|
|
vec![],
|
|
|
|
ty::Predicate::WellFormed(subty) => {
|
|
wf_types.push(subty);
|
|
vec![]
|
|
}
|
|
|
|
ty::Predicate::RegionOutlives(ref data) =>
|
|
match infcx.tcx.no_late_bound_regions(data) {
|
|
None =>
|
|
vec![],
|
|
Some(ty::OutlivesPredicate(r_a, r_b)) =>
|
|
vec![ImpliedBound::RegionSubRegion(r_b, r_a)],
|
|
},
|
|
|
|
ty::Predicate::TypeOutlives(ref data) =>
|
|
match infcx.tcx.no_late_bound_regions(data) {
|
|
None => vec![],
|
|
Some(ty::OutlivesPredicate(ty_a, r_b)) => {
|
|
let components = outlives::components(infcx, ty_a);
|
|
implied_bounds_from_components(r_b, components)
|
|
}
|
|
},
|
|
}}));
|
|
}
|
|
|
|
implied_bounds
|
|
}
|
|
|
|
/// When we have an implied bound that `T: 'a`, we can further break
|
|
/// this down to determine what relationships would have to hold for
|
|
/// `T: 'a` to hold. We get to assume that the caller has validated
|
|
/// those relationships.
|
|
fn implied_bounds_from_components<'tcx>(sub_region: ty::Region,
|
|
sup_components: Vec<Component<'tcx>>)
|
|
-> Vec<ImpliedBound<'tcx>>
|
|
{
|
|
sup_components
|
|
.into_iter()
|
|
.flat_map(|component| {
|
|
match component {
|
|
Component::Region(r) =>
|
|
vec!(ImpliedBound::RegionSubRegion(sub_region, r)),
|
|
Component::Param(p) =>
|
|
vec!(ImpliedBound::RegionSubParam(sub_region, p)),
|
|
Component::Projection(p) =>
|
|
vec!(ImpliedBound::RegionSubProjection(sub_region, p)),
|
|
Component::EscapingProjection(_) =>
|
|
// If the projection has escaping regions, don't
|
|
// try to infer any implied bounds even for its
|
|
// free components. This is conservative, because
|
|
// the caller will still have to prove that those
|
|
// free components outlive `sub_region`. But the
|
|
// idea is that the WAY that the caller proves
|
|
// that may change in the future and we want to
|
|
// give ourselves room to get smarter here.
|
|
vec!(),
|
|
Component::UnresolvedInferenceVariable(..) =>
|
|
vec!(),
|
|
}
|
|
})
|
|
.collect()
|
|
}
|
|
|
|
struct WfPredicates<'a,'tcx:'a> {
|
|
infcx: &'a InferCtxt<'a, 'tcx>,
|
|
body_id: ast::NodeId,
|
|
span: Span,
|
|
out: Vec<traits::PredicateObligation<'tcx>>,
|
|
}
|
|
|
|
impl<'a,'tcx> WfPredicates<'a,'tcx> {
|
|
fn cause(&mut self, code: traits::ObligationCauseCode<'tcx>) -> traits::ObligationCause<'tcx> {
|
|
traits::ObligationCause::new(self.span, self.body_id, code)
|
|
}
|
|
|
|
fn normalize(&mut self) -> Vec<traits::PredicateObligation<'tcx>> {
|
|
let cause = self.cause(traits::MiscObligation);
|
|
let infcx = &mut self.infcx;
|
|
self.out.iter()
|
|
.inspect(|pred| assert!(!pred.has_escaping_regions()))
|
|
.flat_map(|pred| {
|
|
let mut selcx = traits::SelectionContext::new(infcx);
|
|
let pred = traits::normalize(&mut selcx, cause.clone(), pred);
|
|
once(pred.value).chain(pred.obligations)
|
|
})
|
|
.collect()
|
|
}
|
|
|
|
/// Pushes the obligations required for `trait_ref` to be WF into
|
|
/// `self.out`.
|
|
fn compute_trait_ref(&mut self, trait_ref: &ty::TraitRef<'tcx>) {
|
|
let obligations = self.nominal_obligations(trait_ref.def_id, trait_ref.substs);
|
|
self.out.extend(obligations);
|
|
|
|
let cause = self.cause(traits::MiscObligation);
|
|
self.out.extend(
|
|
trait_ref.substs.types
|
|
.as_slice()
|
|
.iter()
|
|
.filter(|ty| !ty.has_escaping_regions())
|
|
.map(|ty| traits::Obligation::new(cause.clone(),
|
|
ty::Predicate::WellFormed(ty))));
|
|
}
|
|
|
|
/// Pushes the obligations required for `trait_ref::Item` to be WF
|
|
/// into `self.out`.
|
|
fn compute_projection(&mut self, data: ty::ProjectionTy<'tcx>) {
|
|
// A projection is well-formed if (a) the trait ref itself is
|
|
// WF WF and (b) the trait-ref holds. (It may also be
|
|
// normalizable and be WF that way.)
|
|
|
|
self.compute_trait_ref(&data.trait_ref);
|
|
|
|
if !data.has_escaping_regions() {
|
|
let predicate = data.trait_ref.to_predicate();
|
|
let cause = self.cause(traits::ProjectionWf(data));
|
|
self.out.push(traits::Obligation::new(cause, predicate));
|
|
}
|
|
}
|
|
|
|
/// Push new obligations into `out`. Returns true if it was able
|
|
/// to generate all the predicates needed to validate that `ty0`
|
|
/// is WF. Returns false if `ty0` is an unresolved type variable,
|
|
/// in which case we are not able to simplify at all.
|
|
fn compute(&mut self, ty0: Ty<'tcx>) -> bool {
|
|
let mut subtys = ty0.walk();
|
|
while let Some(ty) = subtys.next() {
|
|
match ty.sty {
|
|
ty::TyBool |
|
|
ty::TyChar |
|
|
ty::TyInt(..) |
|
|
ty::TyUint(..) |
|
|
ty::TyFloat(..) |
|
|
ty::TyError |
|
|
ty::TyStr |
|
|
ty::TyParam(_) => {
|
|
// WfScalar, WfParameter, etc
|
|
}
|
|
|
|
ty::TySlice(subty) |
|
|
ty::TyArray(subty, _) => {
|
|
if !subty.has_escaping_regions() {
|
|
let cause = self.cause(traits::SliceOrArrayElem);
|
|
match traits::trait_ref_for_builtin_bound(self.infcx.tcx,
|
|
ty::BoundSized,
|
|
subty) {
|
|
Ok(trait_ref) => {
|
|
self.out.push(
|
|
traits::Obligation::new(cause,
|
|
trait_ref.to_predicate()));
|
|
}
|
|
Err(ErrorReported) => { }
|
|
}
|
|
}
|
|
}
|
|
|
|
ty::TyBox(_) |
|
|
ty::TyTuple(_) |
|
|
ty::TyRawPtr(_) => {
|
|
// simple cases that are WF if their type args are WF
|
|
}
|
|
|
|
ty::TyProjection(data) => {
|
|
subtys.skip_current_subtree(); // subtree handled by compute_projection
|
|
self.compute_projection(data);
|
|
}
|
|
|
|
ty::TyEnum(def, substs) |
|
|
ty::TyStruct(def, substs) => {
|
|
// WfNominalType
|
|
let obligations = self.nominal_obligations(def.did, substs);
|
|
self.out.extend(obligations);
|
|
}
|
|
|
|
ty::TyRef(r, mt) => {
|
|
// WfReference
|
|
if !r.has_escaping_regions() && !mt.ty.has_escaping_regions() {
|
|
let cause = self.cause(traits::ReferenceOutlivesReferent(ty));
|
|
self.out.push(
|
|
traits::Obligation::new(
|
|
cause,
|
|
ty::Predicate::TypeOutlives(
|
|
ty::Binder(
|
|
ty::OutlivesPredicate(mt.ty, *r)))));
|
|
}
|
|
}
|
|
|
|
ty::TyClosure(..) => {
|
|
// the types in a closure are always the types of
|
|
// local variables (or possibly references to local
|
|
// variables), we'll walk those.
|
|
//
|
|
// (Though, local variables are probably not
|
|
// needed, as they are separately checked w/r/t
|
|
// WFedness.)
|
|
}
|
|
|
|
ty::TyFnDef(..) | ty::TyFnPtr(_) => {
|
|
// let the loop iterate into the argument/return
|
|
// types appearing in the fn signature
|
|
}
|
|
|
|
ty::TyTrait(ref data) => {
|
|
// WfObject
|
|
//
|
|
// Here, we defer WF checking due to higher-ranked
|
|
// regions. This is perhaps not ideal.
|
|
self.from_object_ty(ty, data);
|
|
|
|
// FIXME(#27579) RFC also considers adding trait
|
|
// obligations that don't refer to Self and
|
|
// checking those
|
|
|
|
let cause = self.cause(traits::MiscObligation);
|
|
self.out.push(
|
|
traits::Obligation::new(
|
|
cause,
|
|
ty::Predicate::ObjectSafe(data.principal_def_id())));
|
|
}
|
|
|
|
// Inference variables are the complicated case, since we don't
|
|
// know what type they are. We do two things:
|
|
//
|
|
// 1. Check if they have been resolved, and if so proceed with
|
|
// THAT type.
|
|
// 2. If not, check whether this is the type that we
|
|
// started with (ty0). In that case, we've made no
|
|
// progress at all, so return false. Otherwise,
|
|
// we've at least simplified things (i.e., we went
|
|
// from `Vec<$0>: WF` to `$0: WF`, so we can
|
|
// register a pending obligation and keep
|
|
// moving. (Goal is that an "inductive hypothesis"
|
|
// is satisfied to ensure termination.)
|
|
ty::TyInfer(_) => {
|
|
let ty = self.infcx.shallow_resolve(ty);
|
|
if let ty::TyInfer(_) = ty.sty { // not yet resolved...
|
|
if ty == ty0 { // ...this is the type we started from! no progress.
|
|
return false;
|
|
}
|
|
|
|
let cause = self.cause(traits::MiscObligation);
|
|
self.out.push( // ...not the type we started from, so we made progress.
|
|
traits::Obligation::new(cause, ty::Predicate::WellFormed(ty)));
|
|
} else {
|
|
// Yes, resolved, proceed with the
|
|
// result. Should never return false because
|
|
// `ty` is not a TyInfer.
|
|
assert!(self.compute(ty));
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// if we made it through that loop above, we made progress!
|
|
return true;
|
|
}
|
|
|
|
fn nominal_obligations(&mut self,
|
|
def_id: DefId,
|
|
substs: &Substs<'tcx>)
|
|
-> Vec<traits::PredicateObligation<'tcx>>
|
|
{
|
|
let predicates =
|
|
self.infcx.tcx.lookup_predicates(def_id)
|
|
.instantiate(self.infcx.tcx, substs);
|
|
let cause = self.cause(traits::ItemObligation(def_id));
|
|
predicates.predicates
|
|
.into_iter()
|
|
.map(|pred| traits::Obligation::new(cause.clone(), pred))
|
|
.filter(|pred| !pred.has_escaping_regions())
|
|
.collect()
|
|
}
|
|
|
|
fn from_object_ty(&mut self, ty: Ty<'tcx>, data: &ty::TraitTy<'tcx>) {
|
|
// Imagine a type like this:
|
|
//
|
|
// trait Foo { }
|
|
// trait Bar<'c> : 'c { }
|
|
//
|
|
// &'b (Foo+'c+Bar<'d>)
|
|
// ^
|
|
//
|
|
// In this case, the following relationships must hold:
|
|
//
|
|
// 'b <= 'c
|
|
// 'd <= 'c
|
|
//
|
|
// The first conditions is due to the normal region pointer
|
|
// rules, which say that a reference cannot outlive its
|
|
// referent.
|
|
//
|
|
// The final condition may be a bit surprising. In particular,
|
|
// you may expect that it would have been `'c <= 'd`, since
|
|
// usually lifetimes of outer things are conservative
|
|
// approximations for inner things. However, it works somewhat
|
|
// differently with trait objects: here the idea is that if the
|
|
// user specifies a region bound (`'c`, in this case) it is the
|
|
// "master bound" that *implies* that bounds from other traits are
|
|
// all met. (Remember that *all bounds* in a type like
|
|
// `Foo+Bar+Zed` must be met, not just one, hence if we write
|
|
// `Foo<'x>+Bar<'y>`, we know that the type outlives *both* 'x and
|
|
// 'y.)
|
|
//
|
|
// Note: in fact we only permit builtin traits, not `Bar<'d>`, I
|
|
// am looking forward to the future here.
|
|
|
|
if !data.has_escaping_regions() {
|
|
let implicit_bounds =
|
|
object_region_bounds(self.infcx.tcx,
|
|
&data.principal,
|
|
data.bounds.builtin_bounds);
|
|
|
|
let explicit_bound = data.bounds.region_bound;
|
|
|
|
for implicit_bound in implicit_bounds {
|
|
let cause = self.cause(traits::ReferenceOutlivesReferent(ty));
|
|
let outlives = ty::Binder(ty::OutlivesPredicate(explicit_bound, implicit_bound));
|
|
self.out.push(traits::Obligation::new(cause, outlives.to_predicate()));
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
/// Given an object type like `SomeTrait+Send`, computes the lifetime
|
|
/// bounds that must hold on the elided self type. These are derived
|
|
/// from the declarations of `SomeTrait`, `Send`, and friends -- if
|
|
/// they declare `trait SomeTrait : 'static`, for example, then
|
|
/// `'static` would appear in the list. The hard work is done by
|
|
/// `ty::required_region_bounds`, see that for more information.
|
|
pub fn object_region_bounds<'tcx>(
|
|
tcx: &TyCtxt<'tcx>,
|
|
principal: &ty::PolyTraitRef<'tcx>,
|
|
others: ty::BuiltinBounds)
|
|
-> Vec<ty::Region>
|
|
{
|
|
// Since we don't actually *know* the self type for an object,
|
|
// this "open(err)" serves as a kind of dummy standin -- basically
|
|
// a skolemized type.
|
|
let open_ty = tcx.mk_infer(ty::FreshTy(0));
|
|
|
|
// Note that we preserve the overall binding levels here.
|
|
assert!(!open_ty.has_escaping_regions());
|
|
let substs = tcx.mk_substs(principal.0.substs.with_self_ty(open_ty));
|
|
let trait_refs = vec!(ty::Binder(ty::TraitRef::new(principal.0.def_id, substs)));
|
|
|
|
let mut predicates = others.to_predicates(tcx, open_ty);
|
|
predicates.extend(trait_refs.iter().map(|t| t.to_predicate()));
|
|
|
|
tcx.required_region_bounds(open_ty, predicates)
|
|
}
|