Add initial support for raw lifetimes

This commit is contained in:
Michael Goulet 2024-09-05 05:43:55 -04:00
parent 3b3e43a386
commit 97910580aa
18 changed files with 116 additions and 40 deletions

View File

@ -752,7 +752,7 @@ fn visit_lazy_tts<T: MutVisitor>(vis: &mut T, lazy_tts: &mut Option<LazyAttrToke
pub fn visit_token<T: MutVisitor>(vis: &mut T, t: &mut Token) {
let Token { kind, span } = t;
match kind {
token::Ident(name, _ /*raw*/) | token::Lifetime(name) => {
token::Ident(name, _is_raw) | token::Lifetime(name, _is_raw) => {
let mut ident = Ident::new(*name, *span);
vis.visit_ident(&mut ident);
*name = ident.name;
@ -762,7 +762,7 @@ pub fn visit_token<T: MutVisitor>(vis: &mut T, t: &mut Token) {
token::NtIdent(ident, _is_raw) => {
vis.visit_ident(ident);
}
token::NtLifetime(ident) => {
token::NtLifetime(ident, _is_raw) => {
vis.visit_ident(ident);
}
token::Interpolated(nt) => {

View File

@ -331,11 +331,11 @@ pub enum TokenKind {
/// Do not forget about `NtLifetime` when you want to match on lifetime identifiers.
/// It's recommended to use `Token::(lifetime,uninterpolate,uninterpolated_span)` to
/// treat regular and interpolated lifetime identifiers in the same way.
Lifetime(Symbol),
Lifetime(Symbol, IdentIsRaw),
/// This identifier (and its span) is the lifetime passed to the
/// declarative macro. The span in the surrounding `Token` is the span of
/// the `lifetime` metavariable in the macro's RHS.
NtLifetime(Ident),
NtLifetime(Ident, IdentIsRaw),
/// An embedded AST node, as produced by a macro. This only exists for
/// historical reasons. We'd like to get rid of it, for multiple reasons.
@ -458,7 +458,7 @@ impl Token {
/// if they keep spans or perform edition checks.
pub fn uninterpolated_span(&self) -> Span {
match self.kind {
NtIdent(ident, _) | NtLifetime(ident) => ident.span,
NtIdent(ident, _) | NtLifetime(ident, _) => ident.span,
Interpolated(ref nt) => nt.use_span(),
_ => self.span,
}
@ -661,7 +661,9 @@ impl Token {
pub fn uninterpolate(&self) -> Cow<'_, Token> {
match self.kind {
NtIdent(ident, is_raw) => Cow::Owned(Token::new(Ident(ident.name, is_raw), ident.span)),
NtLifetime(ident) => Cow::Owned(Token::new(Lifetime(ident.name), ident.span)),
NtLifetime(ident, is_raw) => {
Cow::Owned(Token::new(Lifetime(ident.name, is_raw), ident.span))
}
_ => Cow::Borrowed(self),
}
}
@ -679,11 +681,11 @@ impl Token {
/// Returns a lifetime identifier if this token is a lifetime.
#[inline]
pub fn lifetime(&self) -> Option<Ident> {
pub fn lifetime(&self) -> Option<(Ident, IdentIsRaw)> {
// We avoid using `Token::uninterpolate` here because it's slow.
match self.kind {
Lifetime(name) => Some(Ident::new(name, self.span)),
NtLifetime(ident) => Some(ident),
Lifetime(name, is_raw) => Some((Ident::new(name, self.span), is_raw)),
NtLifetime(ident, is_raw) => Some((ident, is_raw)),
_ => None,
}
}
@ -865,7 +867,7 @@ impl Token {
_ => return None,
},
SingleQuote => match joint.kind {
Ident(name, IdentIsRaw::No) => Lifetime(Symbol::intern(&format!("'{name}"))),
Ident(name, is_raw) => Lifetime(Symbol::intern(&format!("'{name}")), is_raw),
_ => return None,
},

View File

@ -482,11 +482,11 @@ impl TokenStream {
token::NtIdent(ident, is_raw) => {
TokenTree::Token(Token::new(token::Ident(ident.name, is_raw), ident.span), spacing)
}
token::NtLifetime(ident) => TokenTree::Delimited(
token::NtLifetime(ident, is_raw) => TokenTree::Delimited(
DelimSpan::from_single(token.span),
DelimSpacing::new(Spacing::JointHidden, spacing),
Delimiter::Invisible,
TokenStream::token_alone(token::Lifetime(ident.name), ident.span),
TokenStream::token_alone(token::Lifetime(ident.name, is_raw), ident.span),
),
token::Interpolated(ref nt) => TokenTree::Delimited(
DelimSpan::from_single(token.span),

View File

@ -11,7 +11,9 @@ use std::borrow::Cow;
use ast::TraitBoundModifiers;
use rustc_ast::attr::AttrIdGenerator;
use rustc_ast::ptr::P;
use rustc_ast::token::{self, BinOpToken, CommentKind, Delimiter, Nonterminal, Token, TokenKind};
use rustc_ast::token::{
self, BinOpToken, CommentKind, Delimiter, IdentIsRaw, Nonterminal, Token, TokenKind,
};
use rustc_ast::tokenstream::{Spacing, TokenStream, TokenTree};
use rustc_ast::util::classify;
use rustc_ast::util::comments::{Comment, CommentStyle};
@ -947,8 +949,13 @@ pub trait PrintState<'a>: std::ops::Deref<Target = pp::Printer> + std::ops::Dere
token::NtIdent(ident, is_raw) => {
IdentPrinter::for_ast_ident(ident, is_raw.into()).to_string().into()
}
token::Lifetime(name) => name.to_string().into(),
token::NtLifetime(ident) => ident.name.to_string().into(),
token::Lifetime(name, IdentIsRaw::No)
| token::NtLifetime(Ident { name, .. }, IdentIsRaw::No) => name.to_string().into(),
token::Lifetime(name, IdentIsRaw::Yes)
| token::NtLifetime(Ident { name, .. }, IdentIsRaw::Yes) => {
format!("'r#{}", &name.as_str()[1..]).into()
}
/* Other */
token::DocComment(comment_kind, attr_style, data) => {

View File

@ -398,8 +398,10 @@ pub(crate) enum NamedMatch {
fn token_name_eq(t1: &Token, t2: &Token) -> bool {
if let (Some((ident1, is_raw1)), Some((ident2, is_raw2))) = (t1.ident(), t2.ident()) {
ident1.name == ident2.name && is_raw1 == is_raw2
} else if let (Some(ident1), Some(ident2)) = (t1.lifetime(), t2.lifetime()) {
ident1.name == ident2.name
} else if let (Some((ident1, is_raw1)), Some((ident2, is_raw2))) =
(t1.lifetime(), t2.lifetime())
{
ident1.name == ident2.name && is_raw1 == is_raw2
} else {
t1.kind == t2.kind
}

View File

@ -283,9 +283,9 @@ pub(super) fn transcribe<'a>(
let kind = token::NtIdent(*ident, *is_raw);
TokenTree::token_alone(kind, sp)
}
MatchedSingle(ParseNtResult::Lifetime(ident)) => {
MatchedSingle(ParseNtResult::Lifetime(ident, is_raw)) => {
marker.visit_span(&mut sp);
let kind = token::NtLifetime(*ident);
let kind = token::NtLifetime(*ident, *is_raw);
TokenTree::token_alone(kind, sp)
}
MatchedSingle(ParseNtResult::Nt(nt)) => {

View File

@ -229,15 +229,16 @@ impl FromInternal<(TokenStream, &mut Rustc<'_, '_>)> for Vec<TokenTree<TokenStre
span: ident.span,
})),
Lifetime(name) => {
Lifetime(name, is_raw) => {
let ident = symbol::Ident::new(name, span).without_first_quote();
trees.extend([
TokenTree::Punct(Punct { ch: b'\'', joint: true, span }),
TokenTree::Ident(Ident { sym: ident.name, is_raw: false, span }),
TokenTree::Ident(Ident { sym: ident.name, is_raw: is_raw.into(), span }),
]);
}
NtLifetime(ident) => {
let stream = TokenStream::token_alone(token::Lifetime(ident.name), ident.span);
NtLifetime(ident, is_raw) => {
let stream =
TokenStream::token_alone(token::Lifetime(ident.name, is_raw), ident.span);
trees.push(TokenTree::Group(Group {
delimiter: pm::Delimiter::None,
stream: Some(stream),

View File

@ -97,6 +97,9 @@ pub enum TokenKind {
/// and not the separator.
UnknownPrefixLifetime,
/// `'r#lt`, which in edition < 2021 is split into several tokens: `'r # lt`.
RawLifetime,
/// Similar to the above, but *always* an error on every edition. This is used
/// for emoji identifier recovery, as those are not meant to be ever accepted.
InvalidPrefix,
@ -683,9 +686,17 @@ impl Cursor<'_> {
return Literal { kind, suffix_start };
}
if self.first() == 'r' && self.second() == '#' && is_id_start(self.third()) {
// Eat "r" and `#`, and identifier start characters.
self.bump();
self.bump();
self.bump();
self.eat_while(is_id_continue);
return RawLifetime;
}
// Either a lifetime or a character literal with
// length greater than 1.
let starts_with_number = self.first().is_ascii_digit();
// Skip the literal contents.

View File

@ -707,6 +707,10 @@ lint_range_endpoint_out_of_range = range endpoint is out of range for `{$ty}`
lint_range_use_inclusive_range = use an inclusive range instead
lint_raw_prefix = prefix `'r` is reserved
.label = reserved prefix
.suggestion = insert whitespace here to avoid this being parsed as a prefix in Rust 2021
lint_reason_must_be_string_literal = reason must be a string literal
lint_reason_must_come_last = reason in lint attribute must come last

View File

@ -172,6 +172,10 @@ pub(super) fn decorate_lint(sess: &Session, diagnostic: BuiltinLintDiag, diag: &
}
.decorate_lint(diag);
}
BuiltinLintDiag::RawPrefix(label_span) => {
lints::RawPrefix { label: label_span, suggestion: label_span.shrink_to_hi() }
.decorate_lint(diag);
}
BuiltinLintDiag::UnusedBuiltinAttribute { attr_name, macro_name, invoc_span } => {
lints::UnusedBuiltinAttribute { invoc_span, attr_name, macro_name }.decorate_lint(diag);
}

View File

@ -2772,6 +2772,15 @@ pub(crate) struct ReservedPrefix {
pub prefix: String,
}
#[derive(LintDiagnostic)]
#[diag(lint_raw_prefix)]
pub(crate) struct RawPrefix {
#[label]
pub label: Span,
#[suggestion(code = " ", applicability = "machine-applicable")]
pub suggestion: Span,
}
#[derive(LintDiagnostic)]
#[diag(lint_unused_builtin_attribute)]
pub(crate) struct UnusedBuiltinAttribute {

View File

@ -609,6 +609,8 @@ pub enum BuiltinLintDiag {
LegacyDeriveHelpers(Span),
OrPatternsBackCompat(Span, String),
ReservedPrefix(Span, String),
/// `'r#` in edition < 2021.
RawPrefix(Span),
TrailingMacro(bool, Ident),
BreakWithLabelAndLoop(Span),
UnicodeTextFlow(Span, String),

View File

@ -13,7 +13,6 @@ use rustc_session::lint::builtin::{
};
use rustc_session::lint::BuiltinLintDiag;
use rustc_session::parse::ParseSess;
use rustc_span::edition::Edition;
use rustc_span::symbol::Symbol;
use rustc_span::{BytePos, Pos, Span};
use tracing::debug;
@ -284,7 +283,41 @@ impl<'psess, 'src> StringReader<'psess, 'src> {
.stash(span, StashKey::LifetimeIsChar);
}
let ident = Symbol::intern(lifetime_name);
token::Lifetime(ident)
token::Lifetime(ident, IdentIsRaw::No)
}
rustc_lexer::TokenKind::RawLifetime => {
self.last_lifetime = Some(self.mk_sp(start, start + BytePos(1)));
let ident_start = start + BytePos(3);
let prefix_span = self.mk_sp(start, ident_start);
if prefix_span.at_least_rust_2021() {
let lifetime_name_without_tick = self.str_from(ident_start);
// Put the `'` back onto the lifetime name.
let mut lifetime_name = String::with_capacity(lifetime_name_without_tick.len() + 1);
lifetime_name.push('\'');
lifetime_name += lifetime_name_without_tick;
let sym = Symbol::intern(&lifetime_name);
token::Lifetime(sym, IdentIsRaw::Yes)
} else {
// Otherwise, this is just `'r`. Warn about it though.
self.psess.buffer_lint(
RUST_2021_PREFIXES_INCOMPATIBLE_SYNTAX,
prefix_span,
ast::CRATE_NODE_ID,
BuiltinLintDiag::RawPrefix(prefix_span),
);
// Reset the state so we just lex the `'r`.
let lt_start = start + BytePos(2);
self.pos = lt_start;
self.cursor = Cursor::new(&str_before[2 as usize..]);
let lifetime_name = self.str_from(start);
let ident = Symbol::intern(lifetime_name);
token::Lifetime(ident, IdentIsRaw::No)
}
}
rustc_lexer::TokenKind::Semi => token::Semi,
rustc_lexer::TokenKind::Comma => token::Comma,
@ -712,7 +745,7 @@ impl<'psess, 'src> StringReader<'psess, 'src> {
let expn_data = prefix_span.ctxt().outer_expn_data();
if expn_data.edition >= Edition::Edition2021 {
if expn_data.edition.at_least_rust_2021() {
// In Rust 2021, this is a hard error.
let sugg = if prefix == "rb" {
Some(errors::UnknownPrefixSugg::UseBr(prefix_span))

View File

@ -2050,7 +2050,7 @@ impl<'a> Parser<'a> {
};
// On an error path, eagerly consider a lifetime to be an unclosed character lit, if that
// makes sense.
if let Some(ident) = self.token.lifetime()
if let Some((ident, IdentIsRaw::No)) = self.token.lifetime()
&& could_be_unclosed_char_literal(ident)
{
let lt = self.expect_lifetime();
@ -2925,9 +2925,9 @@ impl<'a> Parser<'a> {
}
pub(crate) fn eat_label(&mut self) -> Option<Label> {
if let Some(ident) = self.token.lifetime() {
if let Some((ident, is_raw)) = self.token.lifetime() {
// Disallow `'fn`, but with a better error message than `expect_lifetime`.
if ident.without_first_quote().is_reserved() {
if matches!(is_raw, IdentIsRaw::No) && ident.without_first_quote().is_reserved() {
self.dcx().emit_err(errors::InvalidLabel { span: ident.span, name: ident.name });
}

View File

@ -1666,7 +1666,7 @@ enum FlatToken {
pub enum ParseNtResult {
Tt(TokenTree),
Ident(Ident, IdentIsRaw),
Lifetime(Ident),
Lifetime(Ident, IdentIsRaw),
/// This case will eventually be removed, along with `Token::Interpolate`.
Nt(Lrc<Nonterminal>),

View File

@ -88,7 +88,7 @@ impl<'a> Parser<'a> {
},
NonterminalKind::Pat(pat_kind) => token.can_begin_pattern(pat_kind),
NonterminalKind::Lifetime => match &token.kind {
token::Lifetime(_) | token::NtLifetime(..) => true,
token::Lifetime(..) | token::NtLifetime(..) => true,
_ => false,
},
NonterminalKind::TT | NonterminalKind::Item | NonterminalKind::Stmt => {
@ -171,9 +171,9 @@ impl<'a> Parser<'a> {
NonterminalKind::Lifetime => {
// We want to keep `'keyword` parsing, just like `keyword` is still
// an ident for nonterminal purposes.
return if let Some(ident) = self.token.lifetime() {
return if let Some((ident, is_raw)) = self.token.lifetime() {
self.bump();
Ok(ParseNtResult::Lifetime(ident))
Ok(ParseNtResult::Lifetime(ident, is_raw))
} else {
Err(self.dcx().create_err(UnexpectedNonterminal::Lifetime {
span: self.token.span,

View File

@ -1,6 +1,6 @@
use rustc_ast::mut_visit::{walk_pat, MutVisitor};
use rustc_ast::ptr::P;
use rustc_ast::token::{self, BinOpToken, Delimiter, Token};
use rustc_ast::token::{self, BinOpToken, Delimiter, IdentIsRaw, Token};
use rustc_ast::{
self as ast, AttrVec, BindingMode, ByRef, Expr, ExprKind, MacCall, Mutability, Pat, PatField,
PatFieldsRest, PatKind, Path, QSelf, RangeEnd, RangeSyntax,
@ -548,7 +548,7 @@ impl<'a> Parser<'a> {
None => PatKind::Path(qself, path),
}
}
} else if let Some(lt) = self.token.lifetime()
} else if let Some((lt, IdentIsRaw::No)) = self.token.lifetime()
// In pattern position, we're totally fine with using "next token isn't colon"
// as a heuristic. We could probably just always try to recover if it's a lifetime,
// because we never have `'a: label {}` in a pattern position anyways, but it does
@ -689,7 +689,7 @@ impl<'a> Parser<'a> {
/// Parse `&pat` / `&mut pat`.
fn parse_pat_deref(&mut self, expected: Option<Expected>) -> PResult<'a, PatKind> {
self.expect_and()?;
if let Some(lifetime) = self.token.lifetime() {
if let Some((lifetime, _)) = self.token.lifetime() {
self.bump(); // `'a`
self.dcx().emit_err(UnexpectedLifetimeInPattern {

View File

@ -1,5 +1,5 @@
use rustc_ast::ptr::P;
use rustc_ast::token::{self, BinOpToken, Delimiter, Token, TokenKind};
use rustc_ast::token::{self, BinOpToken, Delimiter, IdentIsRaw, Token, TokenKind};
use rustc_ast::util::case::Case;
use rustc_ast::{
self as ast, BareFnTy, BoundAsyncness, BoundConstness, BoundPolarity, FnRetTy, GenericBound,
@ -1285,8 +1285,9 @@ impl<'a> Parser<'a> {
/// Parses a single lifetime `'a` or panics.
pub(super) fn expect_lifetime(&mut self) -> Lifetime {
if let Some(ident) = self.token.lifetime() {
if ident.without_first_quote().is_reserved()
if let Some((ident, is_raw)) = self.token.lifetime() {
if matches!(is_raw, IdentIsRaw::No)
&& ident.without_first_quote().is_reserved()
&& ![kw::UnderscoreLifetime, kw::StaticLifetime].contains(&ident.name)
{
self.dcx().emit_err(errors::KeywordLifetime { span: ident.span });