Introduce UnpackedKind
This adds an `UnpackedKind` type as a typesafe counterpart to `Kind`. This should make future changes to kinds (such as const generics!) more resilient, as the type-checker will be able to catch more potential issues.
r? @eddyb
cc @yodaldevoid
Fix exponential projection complexity on nested types
This implements solution 1 from https://github.com/rust-lang/rust/issues/38528#issuecomment-366263076.
The code quality is currently extremely poor, but we can improve them during review.
Blocking issues:
- we probably don't want a quadratic deduplication for obligations.
- is there an alternative to deduplication?
Based on #48315.
Needs changelog. Noticable improvement on compile time is expected.
Fix#38528Close#39684Close#43757
This adds an `UnpackedKind` type as a typesafe counterpart to `Kind`. This should make future changes to kinds (such as const generics!) more resilient, as the type-checker should catch more potential issues.
A new section is added to both both struct and trait doc pages.
On struct/enum pages, a new 'Auto Trait Implementations' section displays any
synthetic implementations for auto traits. Currently, this is only done
for Send and Sync.
On trait pages, a new 'Auto Implementors' section displays all types
which automatically implement the trait. Effectively, this is a list of
all public types in the standard library.
Synthesized impls for a particular auto trait ('synthetic impls') take
into account generic bounds. For example, a type 'struct Foo<T>(T)' will
have 'impl<T> Send for Foo<T> where T: Send' generated for it.
Manual implementations of auto traits are also taken into account. If we have
the following types:
'struct Foo<T>(T)'
'struct Wrapper<T>(Foo<T>)'
'unsafe impl<T> Send for Wrapper<T>' // pretend that Wrapper<T> makes
this sound somehow
Then Wrapper will have the following impl generated:
'impl<T> Send for Wrapper<T>'
reflecting the fact that 'T: Send' need not hold for 'Wrapper<T>: Send'
to hold
Lifetimes, HRTBS, and projections (e.g. '<T as Iterator>::Item') are
taken into account by synthetic impls
However, if a type can *never* implement a particular auto trait
(e.g. 'struct MyStruct<T>(*const T)'), then a negative impl will be
generated (in this case, 'impl<T> !Send for MyStruct<T>')
All of this means that a user should be able to copy-paste a synthetic
impl into their code, without any observable changes in behavior
(assuming the rest of the program remains unchanged).
remove intercrate ambiguity hints
The scheme was causing overflows during coherence checking (e.g. #47139). This is sort of a temporary fix; the proper fix I think involves reworking trait selection in deeper ways.
cc @sgrif -- this *should* fix diesel
cc @qnighy -- I'd like to discuss you with alternative techniques for achieving the same end. =) Actually, it might be good to put some energy into refactoring traits first.
r? @eddyb
No longer parse it.
Remove AutoTrait variant from AST and HIR.
Remove backwards compatibility lint.
Remove coherence checks, they make no sense for the new syntax.
Remove from rustdoc.
DefaultImpl is a highly confusing name for what we now call auto impls,
as in `impl Send for ..`. The name auto impl is not formally decided
but for sanity anything is better than `DefaultImpl` which refers
neither to `default impl` nor to `impl Default`.
update FIXME(#6298) to point to open issue 15020
update FIXME(#6268) to point to RFC 811
update FIXME(#10520) to point to RFC 1751
remove FIXME for emscripten issue 4563 and include target in `test_estimate_scaling_factor`
remove FIXME(#18207) since node_id isn't used for `ref` pattern analysis
remove FIXME(#6308) since DST was implemented in #12938
remove FIXME(#2658) since it was decided to not reorganize module
remove FIXME(#20590) since it was decided to stay conservative with projection types
remove FIXME(#20297) since it was decided that solving the issue is unnecessary
remove FIXME(#27086) since closures do correspond to structs now
remove FIXME(#13846) and enable `function_sections` for windows
remove mention of #22079 in FIXME(#22079) since this is a general FIXME
remove FIXME(#5074) since the restriction on borrow were lifted
Migrate a slew of metadata methods to queries
This PR intends to make more progress on #41417, knocking off some low-hanging fruit.
Closes#44190
cc #44137
This commit moves the calculation of the `LanguageItems` structure into a
query rather than being calculated before the `TyCtxt` exists, with the eventual
end goal of removing some `CrateStore` methods.
rustc: Remove `specialization_cache` in favor of a query
This commit removes the `specialization_cache` field of `TyCtxt` by moving it to
a dedicated query, which it turned out was already quite easily structured to do
so!
cc #44137
Remove the trait selection impl in method::probe
This removes the hacky trait selection reimplementation in `method::probe`, which occasionally comes and causes problems.
There are 2 issues I've found with this approach:
1. The older implementation sometimes had a "guess" type from an impl, which allowed subtyping to work. This is why I needed to make a change in `libtest`: there's an `impl<A> Clone for fn(A)` and we're calling `<for<'a> fn(&'a T) as Clone>::clone`. The older implementation would do a subtyping between the impl type and the trait type, so it would do the check for `<fn(A) as Clone>::clone`, and confirmation would continue with the subtyping. The newer implementation directly passes `<for<'a> fn(&'a T) as Clone>::clone` to selection, which fails. I'm not sure how big of a problem that would be in reality, especially after #43690 would remove the `Clone` problem, but I still want a crater run to avoid breaking the world.
2. The older implementation "looked into" impls to display error messages. I'm not sure that's an advantage - it looked exactly 1 level deep.
r? @eddyb
