New destructor semantics #8861
Assignees
Labels
A-destructors
Area: destructors (Drop, ..)
A-typesystem
Area: The type system
P-medium
Medium priority
Comments
|
Accepted for well-defined |
ghost
assigned 
|
Here is a specific example, from #13246, of unsound code we currently accept due to inadequate rules: This program: struct Foo { a: int }
impl Drop for Foo {
fn drop(&mut self) {
println!("{}", self.a);
}
}
fn main() {
{
let _1 = Foo { a: 1 };
let _2 = Foo { a: 2 };
match Foo { a: 3 } {
_ => {}
}
}
let _4 = Foo { a: 4 };
}should output but instead outputs It appears that if the bare block around the first part of This was originally reported to me in sfackler/rust-postgres#31 as a segfault, but it looks like that's probably just due to heap corruption from a use-after-free caused by this bug. UpdateIR: join: ; preds = %case_body
call void @_ZN3Foo14glue_drop.146917h02f639dd107d06daE(%struct.Foo* %_2)
call void @_ZN3Foo14glue_drop.146917h02f639dd107d06daE(%struct.Foo* %_1)
call void @_ZN3Foo14glue_drop.146917h02f639dd107d06daE(%struct.Foo* %0) ; <-- 3
%7 = getelementptr inbounds %struct.Foo* %_4, i32 0, i32 1
store i8 1, i8* %7
%8 = getelementptr inbounds %struct.Foo* %_4, i32 0, i32 0
store i64 4, i64* %8
call void @_ZN3Foo14glue_drop.146917h02f639dd107d06daE(%struct.Foo* %_4)
ret void |
|
Another problem: use std::cell::RefCell;
fn main() {
let b = {
let a = box RefCell::new(4);
*a.borrow() + 1
};
println!("{}", b);
}As much as I'd love to punt on things for 1.0, this seems like a serious issue we need to fix. |
pcwalton
added a commit
to pcwalton/rust
that referenced
this issue
Sep 16, 2014
to have strictly greater lifetime than the values themselves.
Additionally, remove `#[unsafe_destructor]` in favor of these new
restrictions.
This broke a fair amount of code that had to do with `RefCell`s. We will
probably need to do some work to improve the resulting ergonomics. Most
code that broke looked like:
match foo.bar.borrow().baz { ... }
use_foo(&mut foo);
Change this code to:
{
let bar = foo.bar.borrow();
match bar.baz { ... }
}
use_foo(&mut foo);
This fixes an important memory safety hole relating to destructors,
represented by issue rust-lang#8861.
[breaking-change]
bors
added a commit
that referenced
this issue
Nov 20, 2014
…ctor, r=nikomatsakis (Previously, scopes were solely identified with NodeId's; this refactoring prepares for a future where that does not hold.) Ground work for a proper fix to #8861.
|
Does this rule mean code like this would be invalid? extern crate arena;
fn main() {
struct Foo<'a> { foo: ::std::cell::Cell<Option<&'a Foo<'a>>> }
let foo = arena::TypedArena::new();
let bar = foo.alloc(Foo { foo: ::std::cell::Cell::new(None) });
bar.foo.set(Some(&*bar));
}If so, I'm not really on board with it, as this is a very useful pattern (and is perfectly safe in this case), as this change seems to force the use of |
|
As an addendum: this only seems to be a memory safety issue when two things occur together:
If the struct does not have an explicit destructor, then for each field, either the field is a reference (in which case there is no drop glue, so it can't cause memory unsafety), the field is a non-reference and has a lifetime equal to itself (in which case we can recursively perform the same check), or the field is a non-reference with a lifetime strictly greater than itself (in which case it's fine and we can move on to the next field). This may not apply to |
|
Nominating for -beta milestone. |
|
Moving ahead to the 1.0-beta milestone. |
pcwalton
referenced
this issue
in servo/rust-cssparser
Jan 8, 2015
The tokenizer is now fully incremental, rather than yielding a block/function at a time. The convention is now to take `&mut Parser` as input for parsing, and only consume as necessary.
This was referenced Jan 26, 2015
pnkfelix
added a commit
to pnkfelix/rust
that referenced
this issue
Jan 27, 2015
This new variant introduces finer-grain code extents, i.e. we now track that a binding lives only for a suffix of a block, and (importantly) will be dropped when it goes out of scope *before* the bindings that occurred earlier in the block. Both of these notions are neatly captured by marking the block (and each suffix) as an enclosing scope of the next suffix beneath it. This is work that is part of the foundation for issue rust-lang#8861. (It actually has been seen in earlier posted pull requests; I have just factored it out into its own PR to ease my own rebasing.) ---- These finer grained scopes do mean that some code is newly rejected by `rustc`; for example: ```rust let mut map : HashMap<u8, &u8> = HashMap::new(); let tmp = Box::new(2); map.insert(43, &*tmp); ``` This will now fail to compile with a message that `*tmp` does not live long enough, because the scope of `tmp` is now strictly smaller than that of `map`, and the use of `&u8` in map's type requires that the borrowed references are all to data that live at least as long as the map. The usual fix for a case like this is to move the binding for `tmp` up above that of `map`; note that you can still leave the initialization in the original spot, like so: ```rust let tmp; let mut map : HashMap<u8, &u8> = HashMap::new(); tmp = box 2; map.insert(43, &*tmp); ``` Similarly, one can encounter an analogous situation with `Vec`: one would need to rewrite: ```rust let mut vec = Vec::new(); let tmp = 'c'; vec.push(&tmp); ``` as: ``` let tmp; let mut vec = Vec::new(); tmp = 'c'; vec.push(&tmp); ``` ---- In some corner cases, it does not suffice to reorder the bindings; in particular, when the types for both bindings need to reflect exactly the *same* code extent, and a parent/child relationship between them does not work. In pnkfelix's experience this has arisen most often when mixing uses of cyclic data structures while also allowing a lifetime parameter `'a` to flow into a type parameter context where the type is *invariant* with respect to the type parameter. An important instance of this is `arena::TypedArena<T>`, which is invariant with respect to `T`. (The reason that variance is relevant is this: *if* `TypedArena` were covariant with respect to its type parameter, then we could assign it the longer lifetime when it is initialized, and then convert it to a subtype (via covariance) with a shorter lifetime when necessary. But `TypedArena` is invariant with respect to its type parameter, and thus if `S` is a subtype of `T` (in particular, if `S` has a lifetime parameter that is shorter than that of `T`), then a `TypedArena<S>` is unrelated to `TypedArena<T>`.) Concretely, consider code like this: ```rust struct Node<'a> { sibling: Option<&'a Node<'a>> } struct Context<'a> { // because of this field, `Context<'a>` is invariant with respect to `'a`. arena: &'a TypedArena<Node<'a>>, ... } fn new_ctxt<'a>(arena: &'a TypedArena<Node<'a>>) -> Context<'a> { ... } fn use_ctxt<'a>(fcx: &'a Context<'a>) { ... } let arena = TypedArena::new(); let ctxt = new_ctxt(&arena); use_ctxt(&ctxt); ``` In these situations, if you try to introduce two bindings via two distinct `let` statements, each is (with this commit) assigned a distinct extent, and the region inference system cannot find a single region to assign to the lifetime `'a` that works for both of the bindings. So you get an error that `ctxt` does not live long enough; but moving its binding up above that of `arena` just shifts the error so now the compiler complains that `arena` does not live long enough. SO: What to do? The easiest fix in this case is to ensure that the two bindings *do* get assigned the same static extent, by stuffing both bindings into the same let statement, like so: ```rust let (arena, ctxt): (TypedArena, Context); arena = TypedArena::new(); ctxt = new_ctxt(&arena); use_ctxt(&ctxt); ``` Due to the new code rejections outlined above, this is a ... [breaking-change]
bors
added a commit
that referenced
this issue
Jan 27, 2015
Add `CodeExtent::Remainder` variant; pre-req for new scoping/drop rules. This new enum variant introduces finer-grain code extents, i.e. we now track that a binding lives only for a suffix of a block, and (importantly) will be dropped when it goes out of scope *before* the bindings that occurred earlier in the block. Both of these notions are neatly captured by marking the block (and each suffix) as an enclosing scope of the next suffix beneath it. This is work that is part of the foundation for issue #8861. (It actually has been seen in earlier posted pull requests, in particular #21022; I have just factored it out into its own PR to ease my own near-future rebasing, and also get people used to the new rules.) ---- These finer grained scopes do mean that some code is newly rejected by `rustc`; for example: ```rust let mut map : HashMap<u8, &u8> = HashMap::new(); let tmp = Box::new(2); map.insert(43, &*tmp); ``` This will now fail to compile with a message that `*tmp` does not live long enough, because the scope of `tmp` is now strictly smaller than that of `map`, and the use of `&u8` in map's type requires that the borrowed references are all to data that live at least as long as the map. The usual fix for a case like this is to move the binding for `tmp` up above that of `map`; note that you can still leave the initialization in the original spot, like so: ```rust let tmp; let mut map : HashMap<u8, &u8> = HashMap::new(); tmp = box 2; map.insert(43, &*tmp); ``` Similarly, one can encounter an analogous situation with `Vec`: one would need to rewrite: ```rust let mut vec = Vec::new(); let tmp = 'c'; vec.push(&tmp); ``` as: ```rust let tmp; let mut vec = Vec::new(); tmp = 'c'; vec.push(&tmp); ``` ---- In some corner cases, it does not suffice to reorder the bindings; in particular, when the types for both bindings need to reflect exactly the *same* code extent, and a parent/child relationship between them does not work. In pnkfelix's experience this has arisen most often when mixing uses of cyclic data structures while also allowing a lifetime parameter `'a` to flow into a type parameter context where the type is *invariant* with respect to the type parameter. An important instance of this is `arena::TypedArena<T>`, which is invariant with respect to `T`. (The reason that variance is relevant is this: *if* `TypedArena` were covariant with respect to its type parameter, then we could assign it the longer lifetime when it is initialized, and then convert it to a subtype (via covariance) with a shorter lifetime when necessary. But `TypedArena` is invariant with respect to its type parameter, and thus if `S` is a subtype of `T` (in particular, if `S` has a lifetime parameter that is shorter than that of `T`), then a `TypedArena<S>` is unrelated to `TypedArena<T>`.) Concretely, consider code like this: ```rust struct Node<'a> { sibling: Option<&'a Node<'a>> } struct Context<'a> { // because of this field, `Context<'a>` is invariant with respect to `'a`. arena: &'a TypedArena<Node<'a>>, ... } fn new_ctxt<'a>(arena: &'a TypedArena<Node<'a>>) -> Context<'a> { ... } fn use_ctxt<'a>(fcx: &'a Context<'a>) { ... } let arena = TypedArena::new(); let ctxt = new_ctxt(&arena); use_ctxt(&ctxt); ``` In these situations, if you try to introduce two bindings via two distinct `let` statements, each is (with this commit) assigned a distinct extent, and the region inference system cannot find a single region to assign to the lifetime `'a` that works for both of the bindings. So you get an error that `ctxt` does not live long enough; but moving its binding up above that of `arena` just shifts the error so now the compiler complains that `arena` does not live long enough. * SO: What to do? The easiest fix in this case is to ensure that the two bindings *do* get assigned the same static extent, by stuffing both bindings into the same let statement, like so: ```rust let (arena, ctxt): (TypedArena, Context); arena = TypedArena::new(); ctxt = new_ctxt(&arena); use_ctxt(&ctxt); ``` ---- Due to the new code restrictions outlined above, this is a ... [breaking-change]
|
for people who come to this issue url due to its many references on github, the code base, and IRC: I suspect the most relevant thing for you to look at right now is the Sound Generic Drop RFC PR: rust-lang/rfcs#769 |
bors
added a commit
that referenced
this issue
Feb 11, 2015
This is a resurrection and heavy revision/expansion of a PR that pcwalton did to resolve #8861. The most relevant, user-visible semantic change is this: #[unsafe_destructor] is gone. Instead, if a type expression for some value has a destructor, then any lifetimes referenced within that type expression must strictly outlive the scope of the value. See discussion on rust-lang/rfcs#769
|
The description for #21972 accidentally closed this ticket, but there is still work to be done; reopening. Also, assigning to myself since I pretty much own this topic at this point. |
|
reclassifying P-high, not 1.0 since everything for 1.0 is pretty much done (hopefully `unsafe_destructor will indeed be gone). |
|
closing; the remaining (hypothetical or far-future) work items are tracked on #22321 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Tracking issue for RFC #769, Sound Generic Drop.
Original description follows:
We agreed in a meeting to replace the unsafe destructor code with the rule that a value with a destructor must only contain values of lifetime strictly greater than the value to be destructed. The idea is to prevent values from referencing one another. Permitting borrowed values in destructors enables a number of RAII causes and helps to eliminate the need for once fns. This also relies on the change to prevent
&pointers from being placed within managed boxes (which I think has already been made).I have to figure out precisely how to formalize this rule still. =)
Nominating for backwards compat.
Update from @pnkfelix : Much of this is implemented, and so pnkfelix is listing out specific subtasks, issues, and/or PRs, to try to track what is done and what is left to do:
CodeExtent::Remaindervariant; pre-req for new scoping/drop rules. #21657Dropfrom adding new bounds on a type param; Issue Prohibit specialized drops #8142<iterator-expr>only live for the for-loop itself. (i.e. turn for-loops into a terminating scope). Addresses the bulk Issue new scoping rules for safe dtors can yield spurious semi-colon or trailing unit expr #21114. Resolved by makefor PAT in ITER_EXPR { ... }a terminating-scope for ITER_EXPR. #21984#[unsafe_destructor]attribute entirely; Issue Once confident in dropck: remove#[unsafe_destructor]attribute #22196Far future / No longer deemed necessary
Vec<T>,Box<T>,Rc<T>, etc, covariant with respect toT. Discussed on issue new scoping rules for safe dtors may benefit from variance on type params #21198 (may not be a strict requirement for landing new destructor semantics).(See also #22321)
The text was updated successfully, but these errors were encountered: