Fix rustbuild on 32 bit Linux
This is cherry-picked from #37817 which seems to be stalled and currently needs to be rebased anyway.
r? @alexcrichton (who authored this change)
For a given file
```
trait Foo {
fn bar(&self);
}
pub struct FooConstForMethod;
impl Foo for FooConstForMethod {
const bar: u64 = 1;
}
```
show
```
error[E0323]: item `bar` is an associated const, which doesn't match its trait `Foo`
```
instead of
```
error[E0323]: item `bar` is an associated const, which doesn't match its trait `<FooConstForMethod as Foo>`
```
This is an annoying gotcha with the projection cache's handling of
nested obligations.
Nested projection obligations enter the issue in this case:
```
DEBUG:rustc::traits::project: AssociatedTypeNormalizer: depth=3
normalized
<std::iter::Map<std::ops::Range<i32>,
[closure@not-a-recursion-error.rs:5:30: 5:53]> as
std::iter::IntoIterator>::Item to _#7t with 12 add'l obligations
```
Here the normalization result is the result of the nested impl
`<[closure@not-a-recursion-error.rs:5:30: 5:53] as FnMut(i32)>::Output`,
which is an additional obligation that is a part of "add'l obligations".
By itself, this is proper behaviour - the additional obligation is
returned, and the RFC 447 rules ensure that it is processed before the
output `#_7t` is used in any way.
However, the projection cache breaks this - it caches the
`<std::iter::Map<std::ops::Range<i32>,[closure@not-a-recursion-error.rs:5:30:
5:53]> as std::iter::IntoIterator>::Item = #_7t` resolution. Now
everybody else that attempts to look up the projection will just get
`#_7t` *without* any additional obligations. This obviously causes all
sorts of trouble (here a spurious `EvaluatedToAmbig` results in
specializations not being discarded
[here](9ca50bd4d5/src/librustc/traits/select.rs (L1705))).
The compiler works even with this projection cache gotcha because in most
cases during "one-pass evaluation". we tend to process obligations in LIFO
order - after an obligation is added to the cache, we process its nested
obligations before we do anything else (and if we have a cycle, we handle
it specifically) - which makes sure the inference variables are resolved
before they are used.
That "LIFO" order That was not done when projecting out of a closure, so
let's just fix that for the time being.
Fixes#38033.
Support `?Sized` in where clauses
Implemented as described in https://github.com/rust-lang/rust/issues/20503#issuecomment-258677026 - `?Trait` bounds are moved on type parameter definitions when possible, reported as errors otherwise.
(It'd be nice to unify bounds and where clauses in HIR, but this is mostly blocked by rustdoc now - it needs to render bounds in pleasant way and the best way to do it so far is to mirror what was written in source code.)
Fixes https://github.com/rust-lang/rust/issues/20503
r? @nikomatsakis
StringRefs have a length and their contents are not usually null-terminated.
The solution is to either copy the string data (in rustc_llvm::diagnostic) or take the size into account (in LLVMRustPrintPasses).
I couldn't trigger a bug caused by this (apparently all the strings returned in practice are actually null-terminated) but this is more correct and more future-proof.
Avoid loading needless proc-macro dependencies
Fixes#37958 when no proc-macros are exported; in particular, without `pub extern crate proc_macros;`, `#![feature(macro_reexport)]`, or `#![feature(use_extern_macros)]`.
I opened https://github.com/rust-lang/cargo/issues/3334 for exported proc macros.
r? @alexcrichton
[7/n] rustc: desugar UFCS in HIR and don't use DefMap for associated resolutions.
_This is part of a series ([prev](https://github.com/rust-lang/rust/pull/37412) | [next](https://github.com/rust-lang/rust/pull/37688)) of patches designed to rework rustc into an out-of-order on-demand pipeline model for both better feature support (e.g. [MIR-based](https://github.com/solson/miri) early constant evaluation) and incremental execution of compiler passes (e.g. type-checking), with beneficial consequences to IDE support as well.
If any motivation is unclear, please ask for additional PR description clarifications or code comments._
<hr>
Previously, a path like `T::Assoc::method`, while equivalent to `<<T>::Assoc>::method`, wasn't desugared in any way at the HIR level and everything inspecting it had to either deal with knowing only `T` (before typeck) or knowing only the definition of `method` (after typeck).
Such a path also had only one `NodeId` and associated resolution during typeck modified `DefMap`, in a way that would be hard for incremental recompilation to track, and inconvenient for partial type conversions from HIR to `Ty`, which are required to break faux-cycles in on-demand type collection.
The desugarings performed by this PR are as follows:
* `use a::{b,c};` is flattened to `use a as _; use a::b; use a::c;`
* as resolution is complete, `use a as _;` doesn't do anything, except get checked for stability
* `Vec::new` (an expression) becomes `Vec<..>::new<..>`, to distinguish it from `<Vec>::new<..>`
* the "infer all parameters" `<..>` form is internal and not even pretty-printed
* used when there are no type parameters at all, in an expression or pattern path segment
* `T::A::B` becomes `<<T>::A>::B` in a type, and `<<T<..>>::A<..>>::B<..>` in an expression/pattern
* one additional `hir::Ty` node is created for each prefix, starting with the fully-resolved type (`T`) and extending it with each segment (e.g. `<T>::A`)
* fully-resolved paths contain their `Def` in HIR, getting rid of the `DefMap` and absolving incremental recompilation of needing to manually look up nodes to handle that side information
Not keeping the `DefMap` around meant that associated resolutions had to be stored somewhere else:
* expressions and patterns use a new `NodeId -> Def` map in `ty::Tables`
* compatible with the future per-body (constant / `fn` / closure) `Tables`
* types are accessible via `Ty` and the usual per-item generics / predicates / type
* `rustdoc` and `save-analysis` are the only situations which insist on mapping syntactical types to semantical ones, or at least understand the resolution of associated types, therefore the type conversion cache, i.e. a `NodeId -> Ty` map, is exposed by typeck for this purpose
* stability had to be split into a pass that runs on HIR and checks the results of name resolution, and impromptu checks triggered by `typeck` for associated paths, methods, fields, etc.
* privacy using semantic types results in accurate reachability for `impl Trait`, which fixes#35870, and thorough introspection of associated types, which may allow relaxing private-in-public checking on bounds, while keeping the intended ban on projections with private type parameters
cc @petrochenkov
Use displacement instead of initial bucket in HashMap code
Use displacement instead of initial bucket in HashMap code. It makes the code a bit cleaner and also saves a few instructions (handy since it'll be using some to do some sort of adaptive behavior soon).
