When we want to implement label-break-value,
we can't really decide whether to emit ScopeTarget::Loop or
ScopeTarget::Block in the code that is supposed to create it.
So we get rid of it and reconstruct the information when
needed.
Avoid many `cmt` allocations.
`cmt` is a ref-counted wrapper around `cmt_` The use of refcounting
keeps `cmt` handling simple, but a lot of `cmt` instances are very
short-lived, and heap-allocating the short-lived ones takes up time.
This patch changes things in the following ways.
- Most of the functions that produced `cmt` instances now produce `cmt_`
instances. The `Rc::new` calls that occurred within those functions
now occur at their call sites (but only when necessary, which isn't
that often).
- Many of the functions that took `cmt` arguments now take `&cmt_`
arguments. This includes all the methods in the `Delegate` trait.
As a result, the vast majority of the heap allocations are avoided. In
an extreme case, the number of calls to malloc in tuple-stress drops
from 9.9M to 7.9M, a drop of 20%. And the compile times for many runs of
coercions, deep-vector, and tuple-stress drop by 1--2%.
`cmt` is a ref-counted wrapper around `cmt_` The use of refcounting
keeps `cmt` handling simple, but a lot of `cmt` instances are very
short-lived, and heap-allocating the short-lived ones takes up time.
This patch changes things in the following ways.
- Most of the functions that produced `cmt` instances now produce `cmt_`
instances. The `Rc::new` calls that occurred within those functions
now occur at their call sites (but only when necessary, which isn't
that often).
- Many of the functions that took `cmt` arguments now take `&cmt_`
arguments. This includes all the methods in the `Delegate` trait.
As a result, the vast majority of the heap allocations are avoided. In
an extreme case, the number of calls to malloc in tuple-stress drops
from 9.9M to 7.9M, a drop of 20%. And the compile times for many runs of
coercions, deep-vector, and tuple-stress drop by 1--2%.
Better error message when trying to write default impls
Previously, if you tried to write this (using the specialization
feature flag):
default impl PartialEq<MyType> {
...
}
The compiler would give you the mysterious warning "inherent impls
cannot be default". What it really means is that you're trying to
write an impl for a Structure or *Trait Object*, and that cannot
be "default". However, one of the ways to encounter this error
(as shown by the above example) is when you forget to write "for
MyType".
This PR adds a help message that reads "maybe missing a `for`
keyword?" This is useful, actionable advice that will help any user
identify their mistake, and doesn't get in the way or mislead any
user that really meant to use the "default" keyword for this weird
purpose. In particular, this help message will be useful for any
users who don't know the "inherent impl" terminology, and/or users
who forget that inherent impls CAN be written for traits (they apply
to the trait objects). Both of these are somewhat confusing, seldom-
used concepts; a one-line error message without any error number for
longer explanation is NOT the place to introduce these ideas.
I wasn't quite sure what grammar / wording to use. I'm open to suggestions. CC @rust-lang/docs (I hope I'm doing that notation right)
(Apparently not. :( )
Previously, if you tried to write this (using the specialization
feature flag):
default impl PartialEq<MyType> {
...
}
The compiler would give you the mysterious warning "inherent impls
cannot be default". What it really means is that you're trying to
write an impl for a Structure or *Trait Object*, and that cannot
be "default". However, one of the ways to encounter this error
(as shown by the above example) is when you forget to write "for
MyType".
This PR adds a help message that reads "maybe missing a `for`
keyword?" This is useful, actionable advice that will help any user
identify their mistake, and doesn't get in the way or mislead any
user that really meant to use the "default" keyword for this weird
purpose. In particular, this help message will be useful for any
users who don't know the "inherent impl" terminology, and/or users
who forget that inherent impls CAN be written for traits (they apply
to the trait objects). Both of these are somewhat confusing, seldom-
used concepts; a one-line error message without any error number for
longer explanation is NOT the place to introduce these ideas.
- `ParamEnv::empty()` -- does not reveal all, good for typeck
- `ParamEnv::reveal_all()` -- does, good for trans
- `param_env.with_reveal_all()` -- converts an existing parameter environment
Warn about ignored generic bounds in `for`
This adds a new lint to fix#42181. For consistency and to avoid code duplication, I also moved the existing "bounds in type aliases are ignored" here.
Questions to the reviewer:
* Is it okay to just remove a diagnostic error code like this? Should I instead keep the warning about type aliases where it is? The old code provided a detailed explanation of what's going on when asked, that information is now lost. On the other hand, `span_warn!` seems deprecated (after this patch, it has exactly one user left!).
* Did I miss any syntactic construct that can appear as `for` in the surface syntax? I covered function types (`for<'a> fn(...)`), generic traits (`for <'a> Fn(...)`, can appear both as bounds as as trait objects) and bounds (`for<'a> F: ...`).
* For the sake of backwards compatibility, this adds a warning, not an error. @nikomatsakis suggested an error in https://github.com/rust-lang/rust/issues/42181#issuecomment-306924389, but I feel that can only happen in a new epoch -- right?
Cc @eddyb
Also move the check for not having type parameters into ast_validation.
I was not sure what to do with compile-fail/issue-23046.rs: The issue looks like
maybe the bounds actually played a role in triggering the ICE, but that seems
unlikely given that the compiler seems to entirely ignore them. However, I
couldn't find a testcase without the bounds, so I figured the best I could do is
to just remove the bounds and make sure at least that keeps working.
Remove "static item recursion checking" in favor of relying on cycle checks in the query engine
Tests are changed to use the cycle check error message instead. Some duplicate tests are removed.
r? @eddyb
Fix span of visibility
This PR
1. adds a closing parenthesis to the span of `Visibility::Crate` (e.g. `pub(crate)`). The current span only covers `pub(crate`.
2. adds a `span` field to `Visibility::Restricted`. This span covers the entire visibility expression (e.g. `pub (in self)`). Currently all we can have is a span for `Path`.
This PR is motivated by the bug found in rustfmt (https://github.com/rust-lang-nursery/rustfmt/issues/2398).
The first change is a strict improvement IMHO. The second change may not be desirable, as it adds a field which is currently not used by the compiler.
Don't promote to 'static the result of dereferences.
This is a **breaking change**, removing copies out of dereferences from rvalue-to-`'static` promotion.
With miri we won't easily know whether the dereference itself would see the same value at runtime as miri (e.g. after mutating a `static`) or even if it can be interpreted (e.g. integer pointers).
One alternative to this ban is defining at least *some* of those situations as UB, i.e. you shouldn't have a reference in the first place, and you should work through raw pointers instead, to avoid promotion.
**EDIT**: The other *may seem* to be to add some analysis which whitelists references-to-constant-values and assume any values produced by arbitrary computation to not be safe to promote dereferences thereof - but that means producing a reference from an associated constant or `const fn` would necessarily obscure it, and in the former case, this could still impact code that runs on stable today. What we do today to track "references to statics" only works because we restrict taking a reference to a `static` at all to other `static`s (which, again, are currently limited in that they can't be read at compile-time) and to runtime-only `fn`s (*not* `const fn`s).
I'm primarily opening this PR with a conservative first approximation (e.g. `&(*r).a` is not allowed, only reborrows are, and in the old borrow only implicit ones from adjustments, at that) for cratering.
r? @nikomatsakis
