Point at unclosed delimiters as part of the primary MultiSpan
Both the place where the parser encounters a needed closed delimiter and
the unclosed opening delimiter are important, so they should get the
same level of highlighting in the output.
_Context: https://twitter.com/mwk4/status/1430631546432675840_
Both the place where the parser encounters a needed closed delimiter and
the unclosed opening delimiter are important, so they should get the
same level of highlighting in the output.
# Stabilization report
## Summary
This stabilizes using macro expansion in key-value attributes, like so:
```rust
#[doc = include_str!("my_doc.md")]
struct S;
#[path = concat!(env!("OUT_DIR"), "/generated.rs")]
mod m;
```
See the changes to the reference for details on what macros are allowed;
see Petrochenkov's excellent blog post [on internals](https://internals.rust-lang.org/t/macro-expansion-points-in-attributes/11455)
for alternatives that were considered and rejected ("why accept no more
and no less?")
This has been available on nightly since 1.50 with no major issues.
## Notes
### Accepted syntax
The parser accepts arbitrary Rust expressions in this position, but any expression other than a macro invocation will ultimately lead to an error because it is not expected by the built-in expression forms (e.g., `#[doc]`). Note that decorators and the like may be able to observe other expression forms.
### Expansion ordering
Expansion of macro expressions in "inert" attributes occurs after decorators have executed, analogously to macro expressions appearing in the function body or other parts of decorator input.
There is currently no way for decorators to accept macros in key-value position if macro expansion must be performed before the decorator executes (if the macro can simply be copied into the output for later expansion, that can work).
## Test cases
- https://github.com/rust-lang/rust/blob/master/src/test/ui/attributes/key-value-expansion-on-mac.rs
- https://github.com/rust-lang/rust/blob/master/src/test/rustdoc/external-doc.rs
The feature has also been dogfooded extensively in the compiler and
standard library:
- https://github.com/rust-lang/rust/pull/83329
- https://github.com/rust-lang/rust/pull/83230
- https://github.com/rust-lang/rust/pull/82641
- https://github.com/rust-lang/rust/pull/80534
## Implementation history
- Initial proposal: https://github.com/rust-lang/rust/issues/55414#issuecomment-554005412
- Experiment to see how much code it would break: https://github.com/rust-lang/rust/pull/67121
- Preliminary work to restrict expansion that would conflict with this
feature: https://github.com/rust-lang/rust/pull/77271
- Initial implementation: https://github.com/rust-lang/rust/pull/78837
- Fix for an ICE: https://github.com/rust-lang/rust/pull/80563
## Unresolved Questions
~~https://github.com/rust-lang/rust/pull/83366#issuecomment-805180738 listed some concerns, but they have been resolved as of this final report.~~
## Additional Information
There are two workarounds that have a similar effect for `#[doc]`
attributes on nightly. One is to emulate this behavior by using a limited version of this feature that was stabilized for historical reasons:
```rust
macro_rules! forward_inner_docs {
($e:expr => $i:item) => {
#[doc = $e]
$i
};
}
forward_inner_docs!(include_str!("lib.rs") => struct S {});
```
This also works for other attributes (like `#[path = concat!(...)]`).
The other is to use `doc(include)`:
```rust
#![feature(external_doc)]
#[doc(include = "lib.rs")]
struct S {}
```
The first works, but is non-trivial for people to discover, and
difficult to read and maintain. The second is a strange special-case for
a particular use of the macro. This generalizes it to work for any use
case, not just including files.
I plan to remove `doc(include)` when this is stabilized. The
`forward_inner_docs` workaround will still compile without warnings, but
I expect it to be used less once it's no longer necessary.
This PR modifies the macro expansion infrastructure to handle attributes
in a fully token-based manner. As a result:
* Derives macros no longer lose spans when their input is modified
by eager cfg-expansion. This is accomplished by performing eager
cfg-expansion on the token stream that we pass to the derive
proc-macro
* Inner attributes now preserve spans in all cases, including when we
have multiple inner attributes in a row.
This is accomplished through the following changes:
* New structs `AttrAnnotatedTokenStream` and `AttrAnnotatedTokenTree` are introduced.
These are very similar to a normal `TokenTree`, but they also track
the position of attributes and attribute targets within the stream.
They are built when we collect tokens during parsing.
An `AttrAnnotatedTokenStream` is converted to a regular `TokenStream` when
we invoke a macro.
* Token capturing and `LazyTokenStream` are modified to work with
`AttrAnnotatedTokenStream`. A new `ReplaceRange` type is introduced, which
is created during the parsing of a nested AST node to make the 'outer'
AST node aware of the attributes and attribute target stored deeper in the token stream.
* When we need to perform eager cfg-expansion (either due to `#[derive]` or `#[cfg_eval]`),
we tokenize and reparse our target, capturing additional information about the locations of
`#[cfg]` and `#[cfg_attr]` attributes at any depth within the target.
This is a performance optimization, allowing us to perform less work
in the typical case where captured tokens never have eager cfg-expansion run.
Those two recovery attempts have a very bad interaction that causes too
unnecessary output. Add a simple gate to avoid interpreting a `;` as a
`,` when there are unclosed braces.
Previously, we would silently remove any `None`-delimiters when
capturing a `TokenStream`, 'flattenting' them to their inner tokens.
This was not normally visible, since we usually have
`TokenKind::Interpolated` (which gets converted to a `None`-delimited
group during macro invocation) instead of an actual `None`-delimited
group.
However, there are a couple of cases where this becomes visible to
proc-macros:
1. A cross-crate `macro_rules!` macro has a `None`-delimited group
stored in its body (as a result of being produced by another
`macro_rules!` macro). The cross-crate `macro_rules!` invocation
can then expand to an attribute macro invocation, which needs
to be able to see the `None`-delimited group.
2. A proc-macro can invoke an attribute proc-macro with its re-collected
input. If there are any nonterminals present in the input, they will
get re-collected to `None`-delimited groups, which will then get
captured as part of the attribute macro invocation.
Both of these cases are incredibly obscure, so there hopefully won't be
any breakage. This change will allow more agressive 'flattenting' of
nonterminals in #82608 without losing `None`-delimited groups.
When token-based attribute handling is implemeneted in #80689,
we will need to access tokens from `HasAttrs` (to perform
cfg-stripping), and we will to access attributes from `HasTokens` (to
construct a `PreexpTokenStream`).
This PR merges the `HasAttrs` and `HasTokens` traits into a new
`AstLike` trait. The previous `HasAttrs` impls from `Vec<Attribute>` and `AttrVec`
are removed - they aren't attribute targets, so the impls never really
made sense.
Improve suggestion for tuple struct pattern matching errors.
Closes#80174
This change allows numbers to be parsed as field names when pattern matching on structs, which allows us to provide better error messages when tuple structs are matched using a struct pattern.
r? ``@estebank``
Along the way, we also implement a handful of diagnostics improvements
and fixes, particularly with respect to the special handling of `||` in
place of `|` and when there are leading verts in function params, which
don't allow top-level or-patterns anyway.
This is a pure refactoring split out from #80689.
It represents the most invasive part of that PR, requiring changes in
every caller of `parse_outer_attributes`
In order to eagerly expand `#[cfg]` attributes while preserving the
original `TokenStream`, we need to know the range of tokens that
corresponds to every attribute target. This is accomplished by making
`parse_outer_attributes` return an opaque `AttrWrapper` struct. An
`AttrWrapper` must be converted to a plain `AttrVec` by passing it to
`collect_tokens_trailing_token`. This makes it difficult to accidentally
construct an AST node with attributes without calling `collect_tokens_trailing_token`,
since AST nodes store an `AttrVec`, not an `AttrWrapper`.
As a result, we now call `collect_tokens_trailing_token` for attribute
targets which only support inert attributes, such as generic arguments
and struct fields. Currently, the constructed `LazyTokenStream` is
simply discarded. Future PRs will record the token range corresponding
to the attribute target, allowing those tokens to be removed from an
enclosing `collect_tokens_trailing_token` call if necessary.
Reverts PR #80830Fixestaiki-e/pin-project#312
We can have an arbitrary number of `None`-delimited group frames pushed
on the stack due to proc-macro invocations, which can legally be exited.
Attempting to account for this would add a lot of complexity for a tiny
performance gain, so let's just use the original strategy.
Improve diagnostics when parsing angle args
https://github.com/rust-lang/rust/pull/79266 introduced parsing of generic arguments in associated type constraints, this however resulted in possibly very confusing error messages in cases in which closing angle brackets were missing such as in `Vec<(u32, _, _) = vec![]`, which outputs an incorrectly parsed equality constraint error, as noted by `@cynecx.`
This PR tries to provide better error messages in such cases.
r? `@petrochenkov`
Currently, when a user uses a struct pattern to pattern match on
a tuple struct, the errors we emit generally suggest adding fields
using their field names, which are numbers. However, numbers are
not valid identifiers, so the suggestions, which use the shorthand
notation, are not valid syntax. This commit changes those errors
to suggest using the actual tuple struct pattern syntax instead,
which is a more actionable suggestion.
Fixes#81007
Previously, we would fail to collect tokens in the proper place when
only builtin attributes were present. As a result, we would end up with
attribute tokens in the collected `TokenStream`, leading to duplication
when we attempted to prepend the attributes from the AST node.
We now explicitly track when token collection must be performed due to
nomterminal parsing.
A new `HasTokens` trait is introduced, which is used to move logic from
the callers of `collect_tokens` into the body of `collect_tokens`.
In addition to reducing duplication, this paves the way for PR #80689,
which needs to perform additional logic during token collection.
We will never need to pop past our starting frame during token
capturing. Using an empty stack allows us to avoid pointless heap
allocations/deallocations.
If we try to capture the `Vec<u8>` in `Option<Vec<u8>>`, we'll
need to capture a `>` token which was 'unglued' from a `>>` token.
The processing of unglueing a token for parsing purposes bypasses the
usual capturing infrastructure, so we currently lose the trailing `>`.
As a result, we fall back to the reparsed `TokenStream`, causing us to
lose spans.
This commit makes token capturing keep track of a trailing 'unglued'
token. Note that we don't need to care about unglueing except at the end
of the captured tokens - if we capture both the first and second unglued
tokens, then we'll end up capturing the full 'glued' token, which
already works correctly.
rustc_parse: fix ConstBlock expr span
The span for a ConstBlock expression should presumably run through the end of the block it contains and not stop at the keyword, just like is done with similar block-containing expression kinds, such as a TryBlock
We now collect tokens for the underlying node wrapped by `StmtKind`
instead of storing tokens directly in `Stmt`.
`LazyTokenStream` now supports capturing a trailing semicolon after it
is initially constructed. This allows us to avoid refactoring statement
parsing to wrap the parsing of the semicolon in `parse_tokens`.
Attributes on item statements
(e.g. `fn foo() { #[bar] struct MyStruct; }`) are now treated as
item attributes, not statement attributes, which is consistent with how
we handle attributes on other kinds of statements. The feature-gating
code is adjusted so that proc-macro attributes are still allowed on item
statements on stable.
Two built-in macros (`#[global_allocator]` and `#[test]`) needed to be
adjusted to support being passed `Annotatable::Stmt`.
The optimization conflates empty token streams with unknown token stream, which is at least suspicious, and doesn't affect performance because 0-length token streams are very rare.
Suggest that expressions that look like const generic arguments should be enclosed in brackets
I pulled out the changes for const expressions from https://github.com/rust-lang/rust/pull/71592 (without the trait object diagnostic changes) and made some small changes; the implementation is `@estebank's.`
We're also going to want to make some changes separately to account for trait objects (they result in poor diagnostics, as is evident from one of the test cases here), such as an adaption of https://github.com/rust-lang/rust/pull/72273.
Fixes https://github.com/rust-lang/rust/issues/70753.
r? `@petrochenkov`
Unconditionally capture tokens for attributes.
This allows us to avoid synthesizing tokens in `prepend_attr`, since we
have the original tokens available.
We still need to synthesize tokens when expanding `cfg_attr`,
but this is an unavoidable consequence of the syntax of `cfg_attr` -
the user does not supply the `#` and `[]` tokens that a `cfg_attr`
expands to.
This is based on PR https://github.com/rust-lang/rust/pull/77250 - this PR exposes a bug in the current `collect_tokens` implementation, which is fixed by the rewrite.
Rewrite `collect_tokens` implementations to use a flattened buffer
Instead of trying to collect tokens at each depth, we 'flatten' the
stream as we go allong, pushing open/close delimiters to our buffer
just like regular tokens. One capturing is complete, we reconstruct a
nested `TokenTree::Delimited` structure, producing a normal
`TokenStream`.
The reconstructed `TokenStream` is not created immediately - instead, it is
produced on-demand by a closure (wrapped in a new `LazyTokenStream` type). This
closure stores a clone of the original `TokenCursor`, plus a record of the
number of calls to `next()/next_desugared()`. This is sufficient to reconstruct
the tokenstream seen by the callback without storing any additional state. If
the tokenstream is never used (e.g. when a captured `macro_rules!` argument is
never passed to a proc macro), we never actually create a `TokenStream`.
This implementation has a number of advantages over the previous one:
* It is significantly simpler, with no edge cases around capturing the
start/end of a delimited group.
* It can be easily extended to allow replacing tokens an an arbitrary
'depth' by just using `Vec::splice` at the proper position. This is
important for PR #76130, which requires us to track information about
attributes along with tokens.
* The lazy approach to `TokenStream` construction allows us to easily
parse an AST struct, and then decide after the fact whether we need a
`TokenStream`. This will be useful when we start collecting tokens for
`Attribute` - we can discard the `LazyTokenStream` if the parsed
attribute doesn't need tokens (e.g. is a builtin attribute).
The performance impact seems to be neglibile (see
https://github.com/rust-lang/rust/pull/77250#issuecomment-703960604). There is a
small slowdown on a few benchmarks, but it only rises above 1% for incremental
builds, where it represents a larger fraction of the much smaller instruction
count. There a ~1% speedup on a few other incremental benchmarks - my guess is
that the speedups and slowdowns will usually cancel out in practice.
Instead of trying to collect tokens at each depth, we 'flatten' the
stream as we go allong, pushing open/close delimiters to our buffer
just like regular tokens. One capturing is complete, we reconstruct a
nested `TokenTree::Delimited` structure, producing a normal
`TokenStream`.
The reconstructed `TokenStream` is not created immediately - instead, it is
produced on-demand by a closure (wrapped in a new `LazyTokenStream` type). This
closure stores a clone of the original `TokenCursor`, plus a record of the
number of calls to `next()/next_desugared()`. This is sufficient to reconstruct
the tokenstream seen by the callback without storing any additional state. If
the tokenstream is never used (e.g. when a captured `macro_rules!` argument is
never passed to a proc macro), we never actually create a `TokenStream`.
This implementation has a number of advantages over the previous one:
* It is significantly simpler, with no edge cases around capturing the
start/end of a delimited group.
* It can be easily extended to allow replacing tokens an an arbitrary
'depth' by just using `Vec::splice` at the proper position. This is
important for PR #76130, which requires us to track information about
attributes along with tokens.
* The lazy approach to `TokenStream` construction allows us to easily
parse an AST struct, and then decide after the fact whether we need a
`TokenStream`. This will be useful when we start collecting tokens for
`Attribute` - we can discard the `LazyTokenStream` if the parsed
attribute doesn't need tokens (e.g. is a builtin attribute).
The performance impact seems to be neglibile (see
https://github.com/rust-lang/rust/pull/77250#issuecomment-703960604). There is a
small slowdown on a few benchmarks, but it only rises above 1% for incremental
builds, where it represents a larger fraction of the much smaller instruction
count. There a ~1% speedup on a few other incremental benchmarks - my guess is
that the speedups and slowdowns will usually cancel out in practice.
This approach lives exclusively in the parser, so struct expr bodies
that are syntactically correct on their own but are otherwise incorrect
will still emit confusing errors, like in the following case:
```rust
fn foo() -> Foo {
bar: Vec::new()
}
```
```
error[E0425]: cannot find value `bar` in this scope
--> src/file.rs:5:5
|
5 | bar: Vec::new()
| ^^^ expecting a type here because of type ascription
error[E0214]: parenthesized type parameters may only be used with a `Fn` trait
--> src/file.rs:5:15
|
5 | bar: Vec::new()
| ^^^^^ only `Fn` traits may use parentheses
error[E0107]: wrong number of type arguments: expected 1, found 0
--> src/file.rs:5:10
|
5 | bar: Vec::new()
| ^^^^^^^^^^ expected 1 type argument
```
If that field had a trailing comma, that would be a parse error and it
would trigger the new, more targetted, error:
```
error: struct literal body without path
--> file.rs:4:17
|
4 | fn foo() -> Foo {
| _________________^
5 | | bar: Vec::new(),
6 | | }
| |_^
|
help: you might have forgotten to add the struct literal inside the block
|
4 | fn foo() -> Foo { Path {
5 | bar: Vec::new(),
6 | } }
|
```
Partially address last part of #34255.
Improve recovery on malformed format call
The token following a format expression should be a comma. However, when it is replaced with a similar token (such as a dot), then the corresponding error is emitted, but the token is treated as a comma, and the parsing step continues.
r? @petrochenkov
