I no longer remember why I needed this (or thought I did). The way
that the `BitDenotation` is passed around in all existing use cases
(and planned future ones), the thing that were in the `Ctxt` can just
be part of `Self` instead.
(I think ariel had been pushing me to do this back when I first put in
this infrastructure; it took me a while to see how much of pain the
`Ctxt` was causing.)
More systematic error reporting in path resolution
Path resolution for types, expressions and patterns used various heuristics to give more helpful messages on unresolved or incorrectly resolved paths.
This PR combines these heuristics and applies them to all non-import paths.
First a path is resolved in all namespaces, starting from its primary namespace (to give messages like "expected function, found macro, you probably forgot `!`").
If this resolution doesn't give a desired result we create a base error - either "path is not resolved" or "path is resolved, but the resolution is not acceptable in this context".
Other helps and notes are applied to this base error using heuristics.
Here's the list of heuristics for a path with a last segment `name` in order.
First we issue special messages for unresolved `Self` and `self`.
Second we try to find free items named `name` in other modules and suggest to import them.
Then we try to find fields and associated items named `name` and suggest `self.name` or `Self::name`.
After that we try several deterministic context dependent heuristics like "expected value, found struct, you probably forgot `{}`".
If nothing of the above works we try to find candidates with other names using Levenshtein distance.
---
Some alternatives/notes/unresolved questions:
- ~~I had a strong desire to migrate all affected tests to `test/ui`, diagnostics comparison becomes much more meaningful, but I did this only for few tests so far.~~ (Done)
- ~~Labels for "unresolved path" errors are mostly useless now, it may make sense to move some help/notes to these labels, help becomes closer to the error this way.~~ (Done)
- ~~Currently only the first successful heuristic results in additional message shown to the user, it may make sense to print them all, they are rarely compatible, so the diagnostics bloat is unlikely.~~ (Done)
- Now when https://github.com/rust-lang/rust/pull/38014 landed `resolve_path` can potentially be replaced with `smart_resolve_path` in couple more places - e.g. ~~visibilities~~ (done), ~~import prefixes~~ (done), HIR paths.
---
Some additional fixes:
- Associated suggestions and typo suggestions are filtered with a context specific predicate to avoid inapplicable suggestions.
- `adjust_local_def` works properly in speculative resolution.
- I also fixed a recently introduced ICE in partially resolved UFCS paths (see test `ufcs-partially-resolved.rs`). Minimal reproduction:
```
enum E {}
fn main() {
<u8 as E>::A;
}
```
Fixes https://github.com/rust-lang/rust/issues/38409, fixes https://github.com/rust-lang/rust/issues/38504 (duplicates).
- Some bugs in resolution of visibilities are fixed - `pub(Enum)`, `pub(Trait)`, `pub(non::local::path)`.
- Fixes https://github.com/rust-lang/rust/issues/38012.
---
r? @jseyfried for technical details + @jonathandturner for diagnostics changes
How to read the patch: `smart_resolve_path(_fragment)/resolve_qpath_anywhere` are written anew and replace `resolve_trait_reference`/`resolve_type`/`resolve_pattern_path`/`resolve_struct_path`/`resolve_expr` for `ExprKind::Path`, everything else can be read as a diff.
I timed this locally and plain old `gzip` took 2m06s while `gzip -9` took a
whopping 6m23s to save a mere 4MB out of 1.2GB. Let's shave a few minutes off
the Android builder by turning down the compression level.
Use `DefId`s instead of `NodeId`s for `pub(restricted)` visibilities
This is groundwork for hygiene 2.0, specifically privacy checking hygienic intercrate name resolutions.
r? @nrc
Fix Markdown list formatting.
The Markdown engine used by the book can cope with a single leading space on the list marker:
Like this:
* List item
Rather than like this:
* List item
… but it’s not the typical convention employed in the book, and moreover the Markdown engine used for producing the error index *can’t* cope with it (its behaviour looks like a bug, as it appears to lose one of the two line breaks as well, but that’s immaterial here).
So, we shift to a single convention which doesn’t trigger bugs in the Markdown renderer.
----
See https://doc.rust-lang.org/error-index.html#E0458 and https://doc.rust-lang.org/error-index.html#E0101 for the bad current rendering in the error index.
Removes magenta build warning.
Small bug fix to remove an unused type in the magenta process code that causes build failures for magenta's rustc.
r? @alexcrichton
@tedsta @raphlinus
rustdoc: properly calculate line length for where clauses
Apparently, while I was cleaning up #37190, I regressed the formatting for long where clauses, where it wouldn't take the "prefix" length into account when deciding whether to break the line up. This patch fixes that.
Advertise Vec in LinkedList docs
r? @steveklabnik
Hi! We already [advise](https://doc.rust-lang.org/std/collections/#use-a-linkedlist-when) to use `Vec` instead of `LinkedList` in the top-level collections documentation. But I think it may be missed by someone who just directly finds `LinkedList`.
What do you feel about advertising `Vec` directly in `LinkedList` docs as well?
Simplify notes on testing and concurrency
The start of the notes on tests running concurrently, added in https://github.com/rust-lang/rust/pull/37766 read a little awkwardly. This PR fixes that and simplifies the wording a bit.
r? @steveklabnik
std: Fix partial writes in LineWriter
Previously the `LineWriter` could successfully write some bytes but then fail to
report that it has done so. Additionally, an erroneous flush after a successful
write was permanently ignored. This commit fixes these two issues by (a)
maintaining a `need_flush` flag to indicate whether a flush should be the first
operation in `LineWriter::write` and (b) avoiding returning an error once some
bytes have been successfully written.
Closes#37807
Rationale: some ISAs, e.g. OR1K, do not have atomic instructions
for byte and halfword access, and at the same time do not have
a fixed endianness, which makes it unreasonable to implement these
through word-sized atomic accesses.
Prevent where < ident > from parsing.
In order to be forward compatible with `where<'a>` syntax for higher
rank parameters, prevent potential conflicts with UFCS from parsing
correctly for the near term.
Disable field reordering
This was decided via IRC and needs a backport to beta. Basically, #37429 broke servo, and probably needs an announcement and opt-in flag. I didn't run all tests locally but think I've already reverted all the ones that need to be reverted.
r? @nikomatsakis
Allow legacy custom derive authors to disable warnings in downstream crates
This PR allows legacy custom derive authors to use a pre-deprecated method `registry.register_custom_derive()` instead of `registry.register_syntax_extension()` to avoid downstream deprecation warnings.
r? @nrc
Redox Cross Compilation
I will admit - there are things here that I wish I did not have to do. This completes the ability to create a cross compiler from the rust repository for `x86_64-unknown-redox`. I will document this PR with inline comments explaining some things.
[View this gist to see how a cross compiler is built](https://gist.github.com/jackpot51/6680ad973986e84d69c79854249f2b7e)
Prior discussion of a smaller change is here: https://github.com/rust-lang/rust/pull/38366
Refactor global paths
This PR removes the field `global: bool` from `ast::Path` and `hir::Path`, instead representing a global path `::foo::bar` as `{{root}}::foo::bar`, where `{{root}}` is a virtual keyword `keywords::CrateRoot`.
Also, fixes#38016.
r? @nrc
