- `simd` - on certain tuple structs, derive the arithmetic operators, which
lower to the target's SIMD instructions, if any; the `simd` feature gate
is necessary to use this attribute.
+- `unsafe_destructor_blind_to_params` - on `Drop::drop` method, asserts that the
+ destructor code (and all potential specializations of that code) will
+ never attempt to read from nor write to any references with lifetimes
+ that come in via generic parameters. This is a constraint we cannot
+ currently express via the type system, and therefore we rely on the
+ programmer to assert that it holds. Adding this to a Drop impl causes
+ the associated destructor to be considered "uninteresting" by the
+ Drop-Check rule, and thus it can help sidestep data ordering
+ constraints that would otherwise be introduced by the Drop-Check
+ rule. Such sidestepping of the constraints, if done incorrectly, can
+ lead to undefined behavior (in the form of reading or writing to data
+ outside of its dynamic extent), and thus this attribute has the word
+ "unsafe" in its name. To use this, the
+ `unsafe_destructor_blind_to_params` feature gate must be enabled.
- `unsafe_no_drop_flag` - on structs, remove the flag that prevents
destructors from being run twice. Destructors might be run multiple times on
the same object with this attribute. To use this, the `unsafe_no_drop_flag` feature
});
let generics = adt.type_scheme(self).generics;
+ // RFC 1238: if the destructor method is tagged with the
+ // attribute `unsafe_destructor_blind_to_params`, then the
+ // compiler is being instructed to *assume* that the
+ // destructor will not access borrowed data via a type
+ // parameter, even if such data is otherwise reachable.
+ if self.has_attr(dtor_method, "unsafe_destructor_blind_to_params") {
+ debug!("typ: {:?} assumed blind and thus is dtorck-safe", adt);
+ return false;
+ }
+
// In `impl<'a> Drop ...`, we automatically assume
// `'a` is meaningful and thus represents a bound
// through which we could reach borrowed data.
return true;
}
+ // RFC 1238: *any* type parameter at all makes this a dtor of
+ // interest (i.e. cannot-assume-parametricity from RFC 1238.)
+ if generics.has_type_params(subst::TypeSpace) {
+ debug!("typ: {:?} has interesting dtor due to type params",
+ adt);
+ return true;
+ }
+
let mut seen_items = Vec::new();
let mut items_to_inspect = vec![impl_did];
while let Some(item_def_id) = items_to_inspect.pop() {
///
/// ----
///
-/// The Drop Check Rule is the following:
+/// The simplified (*) Drop Check Rule is the following:
///
/// Let `v` be some value (either temporary or named) and 'a be some
/// lifetime (scope). If the type of `v` owns data of type `D`, where
///
-/// * (1.) `D` has a lifetime- or type-parametric Drop implementation, and
-/// * (2.) the structure of `D` can reach a reference of type `&'a _`, and
-/// * (3.) either:
-/// * (A.) the Drop impl for `D` instantiates `D` at 'a directly,
-/// i.e. `D<'a>`, or,
-/// * (B.) the Drop impl for `D` has some type parameter with a
-/// trait bound `T` where `T` is a trait that has at least
-/// one method,
+/// * (1.) `D` has a lifetime- or type-parametric Drop implementation,
+/// (where that `Drop` implementation does not opt-out of
+/// this check via the `unsafe_destructor_blind_to_params`
+/// attribute), and
+/// * (2.) the structure of `D` can reach a reference of type `&'a _`,
///
/// then 'a must strictly outlive the scope of v.
///
///
/// This function is meant to by applied to the type for every
/// expression in the program.
+///
+/// ----
+///
+/// (*) The qualifier "simplified" is attached to the above
+/// definition of the Drop Check Rule, because it is a simplification
+/// of the original Drop Check rule, which attempted to prove that
+/// some `Drop` implementations could not possibly access data even if
+/// it was technically reachable, due to parametricity.
+///
+/// However, (1.) parametricity on its own turned out to be a
+/// necessary but insufficient condition, and (2.) future changes to
+/// the language are expected to make it impossible to ensure that a
+/// `Drop` implementation is actually parametric with respect to any
+/// particular type parameter. (In particular, impl specialization is
+/// expected to break the needed parametricity property beyond
+/// repair.)
+///
+/// Therefore we have scaled back Drop-Check to a more conservative
+/// rule that does not attempt to deduce whether a `Drop`
+/// implementation could not possible access data of a given lifetime;
+/// instead Drop-Check now simply assumes that if a destructor has
+/// access (direct or indirect) to a lifetime parameter, then that
+/// lifetime must be forced to outlive that destructor's dynamic
+/// extent. We then provide the `unsafe_destructor_blind_to_params`
+/// attribute as a way for destructor implementations to opt-out of
+/// this conservative assumption (and thus assume the obligation of
+/// ensuring that they do not access data nor invoke methods of
+/// values that have been previously dropped).
+///
pub fn check_safety_of_destructor_if_necessary<'a, 'tcx>(rcx: &mut Rcx<'a, 'tcx>,
typ: ty::Ty<'tcx>,
span: Span,
// borrowed data reachable via `typ` must outlive the parent
// of `scope`. This is handled below.
//
- // However, there is an important special case: by
- // parametricity, any generic type parameters have *no* trait
- // bounds in the Drop impl can not be used in any way (apart
- // from being dropped), and thus we can treat data borrowed
- // via such type parameters remains unreachable.
+ // However, there is an important special case: for any Drop
+ // impl that is tagged as "blind" to their parameters,
+ // we assume that data borrowed via such type parameters
+ // remains unreachable via that Drop impl.
+ //
+ // For example, consider:
+ //
+ // ```rust
+ // #[unsafe_destructor_blind_to_params]
+ // impl<T> Drop for Vec<T> { ... }
+ // ```
//
- // For example, consider `impl<T> Drop for Vec<T> { ... }`,
// which does have to be able to drop instances of `T`, but
// otherwise cannot read data from `T`.
//
// unbounded type parameter `T`, we must resume the recursive
// analysis on `T` (since it would be ignored by
// type_must_outlive).
- //
- // FIXME (pnkfelix): Long term, we could be smart and actually
- // feed which generic parameters can be ignored *into* `fn
- // type_must_outlive` (or some generalization thereof). But
- // for the short term, it probably covers most cases of
- // interest to just special case Drop impls where: (1.) there
- // are no generic lifetime parameters and (2.) *all* generic
- // type parameters are unbounded. If both conditions hold, we
- // simply skip the `type_must_outlive` call entirely (but
- // resume the recursive checking of the type-substructure).
if has_dtor_of_interest(tcx, ty) {
debug!("iterate_over_potentially_unsafe_regions_in_type \
{}ty: {} - is a dtorck type!",
// switch to Accepted; see RFC 320)
("unsafe_no_drop_flag", "1.0.0", None, Active),
+ // Allows using the unsafe_destructor_blind_to_params attribute
+ // (Needs an RFC link)
+ ("unsafe_destructor_blind_to_params", "1.3.0", Some(28498), Active),
+
// Allows the use of custom attributes; RFC 572
("custom_attribute", "1.0.0", None, Active),
("unsafe_no_drop_flag", Whitelisted, Gated("unsafe_no_drop_flag",
"unsafe_no_drop_flag has unstable semantics \
and may be removed in the future")),
+ ("unsafe_destructor_blind_to_params",
+ Normal,
+ Gated("unsafe_destructor_blind_to_params",
+ "unsafe_destructor_blind_to_params has unstable semantics \
+ and may be removed in the future")),
("unwind", Whitelisted, Gated("unwind_attributes", "#[unwind] is experimental")),
// used in resolve