Correct regression in type-inference caused by failing to reconfirm that
the object trait matches the required trait during trait selection. The
existing code was checking that the object trait WOULD match (in a
probe), but never executing the match outside of a probe.
This corrects various regressions observed in the wild, including
issue #26952. Fixes#26952.
r? @eddyb
cc @frankmcsherry
the object trait matches the required trait during trait selection. The
existing code was checking that the object trait WOULD match (in a
probe), but never executing the match outside of a probe.
This corrects various regressions observed in the wild, including
issue #26952. Fixes#26952.
TyClosure variant; thread this through wherever closure substitutions
are expected, which leads to a net simplification. Simplify trans
treatment of closures in particular.
Transition to the new object lifetime defaults, replacing the old defaults completely.
r? @pnkfelix
This is a [breaking-change] as specified by [RFC 1156][1156] (though all cases that would break should have been receiving warnings starting in Rust 1.2). Types like `&'a Box<Trait>` (or `&'a Rc<Trait>`, etc) will change from being interpreted as `&'a Box<Trait+'a>` to `&'a Box<Trait+'static>`. To restore the old behavior, write the `+'a` explicitly. For example, the function:
```rust
trait Trait { }
fn foo(x: &Box<Trait>) { ... }
```
would be rewritten as:
```rust
trait Trait { }
fn foo(x: &'a Box<Trait+'a>) { ... }
```
if one wanted to preserve the current typing.
[1156]: https://github.com/rust-lang/rfcs/blob/master/text/1156-adjust-default-object-bounds.md
This PR modernizes some names in the type checker. The only remaining snake_case name in ty.rs is `ctxt` which should be resolved by @eddyb's pending refactor. We can bike shed over the names, it would just be nice to bring the type checker inline with modern Rust.
r? @eddyb
cc @nikomatsakis
region-bound is expected to change in Rust 1.3, but don't use it for
anything in this commit. Note that this is not a "significant" part of
the type (it's not part of the formal model) so we have to normalize
this away or trans starts to get confused because two equal types wind
up with distinct LLVM types.
This patch implements the next chunk of flattening out the type checking context. In a series of patches I moved around the necessary state and logic in order to delete the `Typer` and `ClosureTyper` traits. My next goal is to clean the interfaces and start to move the normalization code behind them.
r? @nrc I hope my PR is coherent, doing this too late at night ;)
This commit finalizes the work of the past commits by fully moving the fulfillment context into
the InferCtxt, cleaning up related context interfaces, removing the Typer and ClosureTyper
traits and cleaning up related intefaces
Update all uses of FulfillmentContext to be ones obtained via
an InferCtxt. This is another step of flattening the type
checking context into a single piece of state.
This first patch starts by moving around pieces of state related to
type checking. The goal is to slowly unify the type checking state
into a single typing context. This initial patch moves the
ParameterEnvironment into the InferCtxt and moves shared tables
from Inherited and ty::ctxt into their own struct Tables. This
is the foundational work to refactoring the type checker to
enable future evolution of the language and tooling.
When we successfully resolve a trait reference with no type/lifetime parameters, like `i32: Foo` or `Box<u32>: Sized`, this is in fact globally true. This patch adds a simple global to the tcx to cache such cases. The main advantage of this is really about caching things like `Box<Vec<Foo>>: Sized`. It also points to the need to revamp our caching infrastructure -- the current caches make selection cost cheaper, but we still wind up paying a high cost in the confirmation process, and in particular unrolling out dependent obligations. Moreover, we should probably do caching more uniformly and with a key that takes the where-clauses into account. But that's for later.
For me, this shows up as a reasonably nice win (20%) on Servo's script crate (when built in dev mode). This is not as big as my initial measurements suggested, I think because I was building my rustc with more debugging enabled at the time. I've not yet done follow-up profiling and so forth to see where the new hot spots are. Bootstrap times seem to be largely unaffected.
cc @pcwalton
This is technically a [breaking-change] in that functions with unsatisfiable where-clauses may now yield errors where before they may have been accepted. Even before, these functions could never have been *called* by actual code. In the future, such functions will probably become illegal altogether, but in this commit they are still accepted, so long as they do not rely on the unsatisfiable where-clauses. As before, the functions still cannot be called in any case.
again, do it once and then just remember the expanded form. At the same
time, filter globally nameable predicates out of the environment, since
they can cause cache errors (and they are not necessary in any case).
that are known to have been satisfied *somewhere*. This means that if
one fn finds that `SomeType: Foo`, then every other fn can just consider
that to hold.
Unfortunately, there are some complications:
1. If `SomeType: Foo` includes dependent conditions, those conditions
may trigger an error. This error will be repored in the first fn
where `SomeType: Foo` is evaluated, but not in the other fns, which
can lead to uneven error reporting (which is sometimes confusing).
2. This kind of caching can be unsound in the presence of
unsatisfiable where clauses. For example, suppose that the first fn
has a where-clause like `i32: Bar<u32>`, which in fact does not
hold. This will "fool" trait resolution into thinking that `i32:
Bar<u32>` holds. This is ok currently, because it means that the
first fn can never be calle (since its where clauses cannot be
satisfied), but if the first fn's successful resolution is cached, it
can allow other fns to compile that should not. This problem is fixed
in the next commit.
There is no subtyping relationship between the types (or their non-freshened
variants), so they can not be merged.
Fixes#22645Fixes#24352Fixes#23825
Should fix#25235 (no test in issue).
Should fix#19976 (test is outdated).
There is no subtyping relationship between the types (or their non-freshened
variants), so they can not be merged.
Fixes#22645Fixes#24352Fixes#23825
Should fix#25235 (no test in issue).
Should fix#19976 (test is outdated).
There were still some mentions of `~[T]` and `~T`, mostly in comments and debugging statements. I tried to do my best to preserve meaning, but I might have gotten some wrong-- I'm happy to fix anything :)
The error message was misleading, so I adjusted it, and I also added the long diagnostics for this error (resolves one point in #24407).
I was unsure about how to phrase the error message. Is “generic parameter binding” the correct term for this?
Puts implementations in bins hashed by the fast-reject key, and
only looks up the relevant impls, reducing O(n^2)-ishness
Before: 688.92user 5.08system 8:56.70elapsed 129%CPU (0avgtext+0avgdata 1208164maxresident)k, LLVM 379.142s
After: 637.78user 5.11system 8:17.48elapsed 129%CPU (0avgtext+0avgdata 1201448maxresident)k LLVM 375.552s
Performance increase is +7%-ish
The former stopped making sense when we started interning substs and made
TraitRef a 2-word copy type, and I'm moving the latter into an arena as
they live as long as the type context.
Closes#17841.
The majority of the work should be done, e.g. trait and inherent impls, different forms of UFCS syntax, defaults, and cross-crate usage. It's probably enough to replace the constants in `f32`, `i8`, and so on, or close to good enough.
There is still some significant functionality missing from this commit:
- ~~Associated consts can't be used in match patterns at all. This is simply because I haven't updated the relevant bits in the parser or `resolve`, but it's *probably* not hard to get working.~~
- Since you can't select an impl for trait-associated consts until partway through type-checking, there are some problems with code that assumes that you can check constants earlier. Associated consts that are not in inherent impls cause ICEs if you try to use them in array sizes or match ranges. For similar reasons, `check_static_recursion` doesn't check them properly, so the stack goes ka-blooey if you use an associated constant that's recursively defined. That's a bit trickier to solve; I'm not entirely sure what the best approach is yet.
- Dealing with consts associated with type parameters will raise some new issues (e.g. if you have a `T: Int` type parameter and want to use `<T>::ZERO`). See rust-lang/rfcs#865.
- ~~Unused associated consts don't seem to trigger the `dead_code` lint when they should. Probably easy to fix.~~
Also, this is the first time I've been spelunking in rustc to such a large extent, so I've probably done some silly things in a couple of places.
Rather than storing the relations between free-regions in a global
table, introduce a `FreeRegionMap` data structure. regionck computes the
`FreeRegionMap` for each fn and stores the result into the tcx so that
borrowck can use it (this could perhaps be refactored to have borrowck
recompute the map, but it's a bid tedious to recompute due to the
interaction of closures and free fns). The main reason to do this is
because of #22779 -- using a global table was incorrect because when
validating impl method signatures, we want to use the free region
relationships from the *trait*, not the impl.
Fixes#22779.
skolemize_late_bound_regions essentially copies the entire type (most of the times it shouldn't, but it does), and match_impl runs millions of times.
Times compiling rustc, tested with
$ make -j4 rustc-stage1
$ ( time RUSTFLAGS=-Z time-passes make -j4 rustc-stage2 ) # need LLVM time for calibration
Before:
real 21m44.960s
user 29m38.812s
sys 0m14.944s
After:
real 19m31.445s
user 26m47.260s
sys 0m14.952s
Making this is a 10% performance improvement.
LLVM passes took 867 seconds before, 862 seconds after.
table, introduce a `FreeRegionMap` data structure. regionck computes the
`FreeRegionMap` for each fn and stores the result into the tcx so that
borrowck can use it (this could perhaps be refactored to have borrowck
recompute the map, but it's a bid tedious to recompute due to the
interaction of closures and free fns). The main reason to do this is
because of #22779 -- using a global table was incorrect because when
validating impl method signatures, we want to use the free region
relationships from the *trait*, not the impl.
Fixes#22779.
`Trait`, prefer the object. Also give a nice error for attempts to
manually `impl Trait for Trait`, since they will be ineffectual.
Fixes#24015.
Fixes#24051.
Fixes#24037.
Fixes#23853.
Fixes#21942.
cc #21756.
The primary purpose of this PR is to add blanket impls for the `Fn` traits of the following (simplified) form:
impl<F:Fn> Fn for &F
impl<F:FnMut> FnMut for &mut F
However, this wound up requiring two changes:
1. A slight hack so that `x()` where `x: &mut F` is translated to `FnMut::call_mut(&mut *x, ())` vs `FnMut::call_mut(&mut x, ())`. This is achieved by just autoderef'ing one time when calling something whose type is `&F` or `&mut F`.
2. Making the infinite recursion test in trait matching a bit more tailored. This involves adding a notion of "matching" types that looks to see if types are potentially unifiable (it's an approximation).
The PR also includes various small refactorings to the inference code that are aimed at moving the unification and other code into a library (I've got that particular change in a branch, these changes just lead the way there by removing unnecessary dependencies between the compiler and the more general unification code).
Note that per rust-lang/rfcs#1023, adding impls like these would be a breaking change in the future.
cc @japaric
cc @alexcrichton
cc @aturon
Fixes#23015.
local only if matches `FUNDAMENTAL(LocalType)`, where `FUNDAMENTAL`
includes `&T` and types marked as fundamental (which includes `Box`).
Also apply these tests to negative reasoning.
trait matching more tailored. We now detect recursion where the
obligations "match" -- meaning basically that they are the same for some
substitution of any unbound type variables.
When testing whether a default method predicates are satisfiable,
combine normalization with this check so that we also skip the
default method if normalization fails. Fixes#23485.
r? @nrc (I tried to address your nit from before as well)
