diff options
| author | bors <bors@rust-lang.org> | 2019-05-10 00:51:14 +0000 |
|---|---|---|
| committer | bors <bors@rust-lang.org> | 2019-05-10 00:51:14 +0000 |
| commit | 03bd2f653f0d0adb69d862fbeec64663157e71e1 (patch) | |
| tree | 3067cec05b04e1d4e72aef71a9be0d1a5f83affc /src/libsyntax | |
| parent | adcd4aa8d07945d5ef09129b05e8e1ff1bcce58e (diff) | |
| parent | 45b09453dbf120cc23d889435aac3ed7d2ec8eb7 (diff) | |
| download | rust-03bd2f653f0d0adb69d862fbeec64663157e71e1.tar.gz rust-03bd2f653f0d0adb69d862fbeec64663157e71e1.zip | |
Auto merge of #60683 - Centril:rollup-p05qh5d, r=Centril
Rollup of 8 pull requests Successful merges: - #59348 (Clean up and add tests for slice drop shims) - #60188 (Identify when a stmt could have been parsed as an expr) - #60234 (std: Derive `Default` for `io::Cursor`) - #60618 (Comment ext::tt::transcribe) - #60648 (Skip codegen for one UI test with long file path) - #60671 (remove unneeded `extern crate`s from build tools) - #60675 (Remove the old await! macro) - #60676 (Fix async desugaring providing wrong input to procedural macros.) Failed merges: r? @ghost
Diffstat (limited to 'src/libsyntax')
| -rw-r--r-- | src/libsyntax/ext/tt/macro_parser.rs | 5 | ||||
| -rw-r--r-- | src/libsyntax/ext/tt/macro_rules.rs | 2 | ||||
| -rw-r--r-- | src/libsyntax/ext/tt/quoted.rs | 1 | ||||
| -rw-r--r-- | src/libsyntax/ext/tt/transcribe.rs | 255 | ||||
| -rw-r--r-- | src/libsyntax/parse/lexer/mod.rs | 3 | ||||
| -rw-r--r-- | src/libsyntax/parse/mod.rs | 27 | ||||
| -rw-r--r-- | src/libsyntax/parse/parser.rs | 80 | ||||
| -rw-r--r-- | src/libsyntax/source_map.rs | 21 | ||||
| -rw-r--r-- | src/libsyntax/util/parser.rs | 25 |
9 files changed, 333 insertions, 86 deletions
diff --git a/src/libsyntax/ext/tt/macro_parser.rs b/src/libsyntax/ext/tt/macro_parser.rs index ab5823eaca5..084a69f4cda 100644 --- a/src/libsyntax/ext/tt/macro_parser.rs +++ b/src/libsyntax/ext/tt/macro_parser.rs @@ -554,7 +554,10 @@ fn inner_parse_loop<'root, 'tt>( match item.top_elts.get_tt(idx) { // Need to descend into a sequence TokenTree::Sequence(sp, seq) => { - // Examine the case where there are 0 matches of this sequence + // Examine the case where there are 0 matches of this sequence. We are + // implicitly disallowing OneOrMore from having 0 matches here. Thus, that will + // result in a "no rules expected token" error by virtue of this matcher not + // working. if seq.op == quoted::KleeneOp::ZeroOrMore || seq.op == quoted::KleeneOp::ZeroOrOne { diff --git a/src/libsyntax/ext/tt/macro_rules.rs b/src/libsyntax/ext/tt/macro_rules.rs index b1b9d25b3d5..a53cc2fe661 100644 --- a/src/libsyntax/ext/tt/macro_rules.rs +++ b/src/libsyntax/ext/tt/macro_rules.rs @@ -151,7 +151,7 @@ fn generic_extension<'cx>(cx: &'cx mut ExtCtxt<'_>, let rhs_spans = rhs.iter().map(|t| t.span()).collect::<Vec<_>>(); // rhs has holes ( `$id` and `$(...)` that need filled) - let mut tts = transcribe(cx, Some(named_matches), rhs); + let mut tts = transcribe(cx, &named_matches, rhs); // Replace all the tokens for the corresponding positions in the macro, to maintain // proper positions in error reporting, while maintaining the macro_backtrace. diff --git a/src/libsyntax/ext/tt/quoted.rs b/src/libsyntax/ext/tt/quoted.rs index b24edb57e52..ed8395f11ad 100644 --- a/src/libsyntax/ext/tt/quoted.rs +++ b/src/libsyntax/ext/tt/quoted.rs @@ -73,6 +73,7 @@ pub enum KleeneOp { ZeroOrMore, /// Kleene plus (`+`) for one or more repetitions OneOrMore, + /// Kleene optional (`?`) for zero or one reptitions ZeroOrOne, } diff --git a/src/libsyntax/ext/tt/transcribe.rs b/src/libsyntax/ext/tt/transcribe.rs index bd2adb5ac13..0cefcf1ce03 100644 --- a/src/libsyntax/ext/tt/transcribe.rs +++ b/src/libsyntax/ext/tt/transcribe.rs @@ -1,10 +1,10 @@ use crate::ast::Ident; use crate::ext::base::ExtCtxt; use crate::ext::expand::Marker; -use crate::ext::tt::macro_parser::{NamedMatch, MatchedSeq, MatchedNonterminal}; +use crate::ext::tt::macro_parser::{MatchedNonterminal, MatchedSeq, NamedMatch}; use crate::ext::tt::quoted; use crate::mut_visit::noop_visit_tt; -use crate::parse::token::{self, Token, NtTT}; +use crate::parse::token::{self, NtTT, Token}; use crate::tokenstream::{DelimSpan, TokenStream, TokenTree, TreeAndJoint}; use smallvec::{smallvec, SmallVec}; @@ -13,24 +13,16 @@ use syntax_pos::DUMMY_SP; use rustc_data_structures::fx::FxHashMap; use rustc_data_structures::sync::Lrc; use std::mem; -use std::ops::Add; use std::rc::Rc; -// An iterator over the token trees in a delimited token tree (`{ ... }`) or a sequence (`$(...)`). +/// An iterator over the token trees in a delimited token tree (`{ ... }`) or a sequence (`$(...)`). enum Frame { - Delimited { - forest: Lrc<quoted::Delimited>, - idx: usize, - span: DelimSpan, - }, - Sequence { - forest: Lrc<quoted::SequenceRepetition>, - idx: usize, - sep: Option<Token>, - }, + Delimited { forest: Lrc<quoted::Delimited>, idx: usize, span: DelimSpan }, + Sequence { forest: Lrc<quoted::SequenceRepetition>, idx: usize, sep: Option<Token> }, } impl Frame { + /// Construct a new frame around the delimited set of tokens. fn new(tts: Vec<quoted::TokenTree>) -> Frame { let forest = Lrc::new(quoted::Delimited { delim: token::NoDelim, tts: tts }); Frame::Delimited { forest: forest, idx: 0, span: DelimSpan::dummy() } @@ -54,84 +46,161 @@ impl Iterator for Frame { } } -/// This can do Macro-By-Example transcription. On the other hand, if -/// `src` contains no `TokenTree::{Sequence, MetaVar, MetaVarDecl}`s, `interp` can -/// (and should) be None. -pub fn transcribe(cx: &ExtCtxt<'_>, - interp: Option<FxHashMap<Ident, Rc<NamedMatch>>>, - src: Vec<quoted::TokenTree>) - -> TokenStream { +/// This can do Macro-By-Example transcription. +/// - `interp` is a map of meta-variables to the tokens (non-terminals) they matched in the +/// invocation. We are assuming we already know there is a match. +/// - `src` is the RHS of the MBE, that is, the "example" we are filling in. +/// +/// For example, +/// +/// ```rust +/// macro_rules! foo { +/// ($id:ident) => { println!("{}", stringify!($id)); } +/// } +/// +/// foo!(bar); +/// ``` +/// +/// `interp` would contain `$id => bar` and `src` would contain `println!("{}", stringify!($id));`. +/// +/// `transcribe` would return a `TokenStream` containing `println!("{}", stringify!(bar));`. +/// +/// Along the way, we do some additional error checking. +pub fn transcribe( + cx: &ExtCtxt<'_>, + interp: &FxHashMap<Ident, Rc<NamedMatch>>, + src: Vec<quoted::TokenTree>, +) -> TokenStream { + // Nothing for us to transcribe... + if src.is_empty() { + return TokenStream::empty(); + } + + // We descend into the RHS (`src`), expanding things as we go. This stack contains the things + // we have yet to expand/are still expanding. We start the stack off with the whole RHS. let mut stack: SmallVec<[Frame; 1]> = smallvec![Frame::new(src)]; - let interpolations = interp.unwrap_or_else(FxHashMap::default); /* just a convenience */ + + // As we descend in the RHS, we will need to be able to match nested sequences of matchers. + // `repeats` keeps track of where we are in matching at each level, with the last element being + // the most deeply nested sequence. This is used as a stack. let mut repeats = Vec::new(); + + // `result` contains resulting token stream from the TokenTree we just finished processing. At + // the end, this will contain the full result of transcription, but at arbitrary points during + // `transcribe`, `result` will contain subsets of the final result. + // + // Specifically, as we descend into each TokenTree, we will push the existing results onto the + // `result_stack` and clear `results`. We will then produce the results of transcribing the + // TokenTree into `results`. Then, as we unwind back out of the `TokenTree`, we will pop the + // `result_stack` and append `results` too it to produce the new `results` up to that point. + // + // Thus, if we try to pop the `result_stack` and it is empty, we have reached the top-level + // again, and we are done transcribing. let mut result: Vec<TreeAndJoint> = Vec::new(); let mut result_stack = Vec::new(); loop { + // Look at the last frame on the stack. let tree = if let Some(tree) = stack.last_mut().unwrap().next() { + // If it still has a TokenTree we have not looked at yet, use that tree. tree - } else { + } + // The else-case never produces a value for `tree` (it `continue`s or `return`s). + else { + // Otherwise, if we have just reached the end of a sequence and we can keep repeating, + // go back to the beginning of the sequence. if let Frame::Sequence { ref mut idx, ref sep, .. } = *stack.last_mut().unwrap() { let (ref mut repeat_idx, repeat_len) = *repeats.last_mut().unwrap(); *repeat_idx += 1; if *repeat_idx < repeat_len { *idx = 0; if let Some(sep) = sep.clone() { - // repeat same span, I guess let prev_span = match result.last() { Some((tt, _)) => tt.span(), None => DUMMY_SP, }; result.push(TokenTree::Token(prev_span, sep).into()); } - continue + continue; } } + // We are done with the top of the stack. Pop it. Depending on what it was, we do + // different things. Note that the outermost item must be the delimited, wrapped RHS + // that was passed in originally to `transcribe`. match stack.pop().unwrap() { + // Done with a sequence. Pop from repeats. Frame::Sequence { .. } => { repeats.pop(); } + + // We are done processing a Delimited. If this is the top-level delimited, we are + // done. Otherwise, we unwind the result_stack to append what we have produced to + // any previous results. Frame::Delimited { forest, span, .. } => { if result_stack.is_empty() { + // No results left to compute! We are back at the top-level. return TokenStream::new(result); } - let tree = TokenTree::Delimited( - span, - forest.delim, - TokenStream::new(result).into(), - ); + + // Step back into the parent Delimited. + let tree = + TokenTree::Delimited(span, forest.delim, TokenStream::new(result).into()); result = result_stack.pop().unwrap(); result.push(tree.into()); } } - continue + continue; }; + // At this point, we know we are in the middle of a TokenTree (the last one on `stack`). + // `tree` contains the next `TokenTree` to be processed. match tree { - quoted::TokenTree::Sequence(sp, seq) => { - // FIXME(pcwalton): Bad copy. - match lockstep_iter_size("ed::TokenTree::Sequence(sp, seq.clone()), - &interpolations, - &repeats) { + // We are descending into a sequence. We first make sure that the matchers in the RHS + // and the matches in `interp` have the same shape. Otherwise, either the caller or the + // macro writer has made a mistake. + seq @ quoted::TokenTree::Sequence(..) => { + match lockstep_iter_size(&seq, interp, &repeats) { LockstepIterSize::Unconstrained => { - cx.span_fatal(sp.entire(), /* blame macro writer */ - "attempted to repeat an expression \ - containing no syntax \ - variables matched as repeating at this depth"); + cx.span_fatal( + seq.span(), /* blame macro writer */ + "attempted to repeat an expression containing no syntax variables \ + matched as repeating at this depth", + ); } + LockstepIterSize::Contradiction(ref msg) => { + // FIXME: this should be impossible. I (mark-i-m) believe it would + // represent a bug in the macro_parser. // FIXME #2887 blame macro invoker instead - cx.span_fatal(sp.entire(), &msg[..]); + cx.span_fatal(seq.span(), &msg[..]); } + LockstepIterSize::Constraint(len, _) => { + // We do this to avoid an extra clone above. We know that this is a + // sequence already. + let (sp, seq) = if let quoted::TokenTree::Sequence(sp, seq) = seq { + (sp, seq) + } else { + unreachable!() + }; + + // Is the repetition empty? if len == 0 { if seq.op == quoted::KleeneOp::OneOrMore { + // FIXME: this should be impossible because we check for this in + // macro_parser.rs // FIXME #2887 blame invoker cx.span_fatal(sp.entire(), "this must repeat at least once"); } } else { + // 0 is the initial counter (we have done 0 repretitions so far). `len` + // is the total number of reptitions we should generate. repeats.push((0, len)); + + // The first time we encounter the sequence we push it to the stack. It + // then gets reused (see the beginning of the loop) until we are done + // repeating. stack.push(Frame::Sequence { idx: 0, sep: seq.separator.clone(), @@ -141,10 +210,16 @@ pub fn transcribe(cx: &ExtCtxt<'_>, } } } - // FIXME #2887: think about span stuff here + + // Replace the meta-var with the matched token tree from the invocation. quoted::TokenTree::MetaVar(mut sp, ident) => { - if let Some(cur_matched) = lookup_cur_matched(ident, &interpolations, &repeats) { + // Find the matched nonterminal from the macro invocation, and use it to replace + // the meta-var. + if let Some(cur_matched) = lookup_cur_matched(ident, interp, &repeats) { if let MatchedNonterminal(ref nt) = *cur_matched { + // FIXME #2887: why do we apply a mark when matching a token tree meta-var + // (e.g. `$x:tt`), but not when we are matching any other type of token + // tree? if let NtTT(ref tt) = **nt { result.push(tt.clone().into()); } else { @@ -153,10 +228,15 @@ pub fn transcribe(cx: &ExtCtxt<'_>, result.push(token.into()); } } else { - cx.span_fatal(sp, /* blame the macro writer */ - &format!("variable '{}' is still repeating at this depth", ident)); + // We were unable to descend far enough. This is an error. + cx.span_fatal( + sp, /* blame the macro writer */ + &format!("variable '{}' is still repeating at this depth", ident), + ); } } else { + // If we aren't able to match the meta-var, we push it back into the result but + // with modified syntax context. (I believe this supports nested macros). let ident = Ident::new(ident.name, ident.span.apply_mark(cx.current_expansion.mark)); sp = sp.apply_mark(cx.current_expansion.mark); @@ -164,26 +244,44 @@ pub fn transcribe(cx: &ExtCtxt<'_>, result.push(TokenTree::Token(sp, token::Token::from_ast_ident(ident)).into()); } } + + // If we are entering a new delimiter, we push its contents to the `stack` to be + // processed, and we push all of the currently produced results to the `result_stack`. + // We will produce all of the results of the inside of the `Delimited` and then we will + // jump back out of the Delimited, pop the result_stack and add the new results back to + // the previous results (from outside the Delimited). quoted::TokenTree::Delimited(mut span, delimited) => { span = span.apply_mark(cx.current_expansion.mark); stack.push(Frame::Delimited { forest: delimited, idx: 0, span: span }); result_stack.push(mem::replace(&mut result, Vec::new())); } + + // Nothing much to do here. Just push the token to the result, being careful to + // preserve syntax context. quoted::TokenTree::Token(sp, tok) => { let mut marker = Marker(cx.current_expansion.mark); let mut tt = TokenTree::Token(sp, tok); noop_visit_tt(&mut tt, &mut marker); result.push(tt.into()); } + + // There should be no meta-var declarations in the invocation of a macro. quoted::TokenTree::MetaVarDecl(..) => panic!("unexpected `TokenTree::MetaVarDecl"), } } } -fn lookup_cur_matched(ident: Ident, - interpolations: &FxHashMap<Ident, Rc<NamedMatch>>, - repeats: &[(usize, usize)]) - -> Option<Rc<NamedMatch>> { +/// Lookup the meta-var named `ident` and return the matched token tree from the invocation using +/// the set of matches `interpolations`. +/// +/// See the definition of `repeats` in the `transcribe` function. `repeats` is used to descend +/// into the right place in nested matchers. If we attempt to descend too far, the macro writer has +/// made a mistake, and we return `None`. +fn lookup_cur_matched( + ident: Ident, + interpolations: &FxHashMap<Ident, Rc<NamedMatch>>, + repeats: &[(usize, usize)], +) -> Option<Rc<NamedMatch>> { interpolations.get(&ident).map(|matched| { let mut matched = matched.clone(); for &(idx, _) in repeats { @@ -198,17 +296,30 @@ fn lookup_cur_matched(ident: Ident, }) } +/// An accumulator over a TokenTree to be used with `fold`. During transcription, we need to make +/// sure that the size of each sequence and all of its nested sequences are the same as the sizes +/// of all the matched (nested) sequences in the macro invocation. If they don't match, somebody +/// has made a mistake (either the macro writer or caller). #[derive(Clone)] enum LockstepIterSize { + /// No constraints on length of matcher. This is true for any TokenTree variants except a + /// `MetaVar` with an actual `MatchedSeq` (as opposed to a `MatchedNonterminal`). Unconstrained, + + /// A `MetaVar` with an actual `MatchedSeq`. The length of the match and the name of the + /// meta-var are returned. Constraint(usize, Ident), + + /// Two `Constraint`s on the same sequence had different lengths. This is an error. Contradiction(String), } -impl Add for LockstepIterSize { - type Output = LockstepIterSize; - - fn add(self, other: LockstepIterSize) -> LockstepIterSize { +impl LockstepIterSize { + /// Find incompatibilities in matcher/invocation sizes. + /// - `Unconstrained` is compatible with everything. + /// - `Contradiction` is incompatible with everything. + /// - `Constraint(len)` is only compatible with other constraints of the same length. + fn with(self, other: LockstepIterSize) -> LockstepIterSize { match self { LockstepIterSize::Unconstrained => other, LockstepIterSize::Contradiction(_) => self, @@ -217,9 +328,11 @@ impl Add for LockstepIterSize { LockstepIterSize::Contradiction(_) => other, LockstepIterSize::Constraint(r_len, _) if l_len == r_len => self, LockstepIterSize::Constraint(r_len, r_id) => { - let msg = format!("inconsistent lockstep iteration: \ - '{}' has {} items, but '{}' has {}", - l_id, l_len, r_id, r_len); + let msg = format!( + "inconsistent lockstep iteration: \ + '{}' has {} items, but '{}' has {}", + l_id, l_len, r_id, r_len + ); LockstepIterSize::Contradiction(msg) } }, @@ -227,30 +340,38 @@ impl Add for LockstepIterSize { } } -fn lockstep_iter_size(tree: "ed::TokenTree, - interpolations: &FxHashMap<Ident, Rc<NamedMatch>>, - repeats: &[(usize, usize)]) - -> LockstepIterSize { +/// Given a `tree`, make sure that all sequences have the same length as the matches for the +/// appropriate meta-vars in `interpolations`. +/// +/// Note that if `repeats` does not match the exact correct depth of a meta-var, +/// `lookup_cur_matched` will return `None`, which is why this still works even in the presnece of +/// multiple nested matcher sequences. +fn lockstep_iter_size( + tree: "ed::TokenTree, + interpolations: &FxHashMap<Ident, Rc<NamedMatch>>, + repeats: &[(usize, usize)], +) -> LockstepIterSize { use quoted::TokenTree; match *tree { TokenTree::Delimited(_, ref delimed) => { delimed.tts.iter().fold(LockstepIterSize::Unconstrained, |size, tt| { - size + lockstep_iter_size(tt, interpolations, repeats) + size.with(lockstep_iter_size(tt, interpolations, repeats)) }) - }, + } TokenTree::Sequence(_, ref seq) => { seq.tts.iter().fold(LockstepIterSize::Unconstrained, |size, tt| { - size + lockstep_iter_size(tt, interpolations, repeats) + size.with(lockstep_iter_size(tt, interpolations, repeats)) }) - }, - TokenTree::MetaVar(_, name) | TokenTree::MetaVarDecl(_, name, _) => + } + TokenTree::MetaVar(_, name) | TokenTree::MetaVarDecl(_, name, _) => { match lookup_cur_matched(name, interpolations, repeats) { Some(matched) => match *matched { MatchedNonterminal(_) => LockstepIterSize::Unconstrained, MatchedSeq(ref ads, _) => LockstepIterSize::Constraint(ads.len(), name), }, - _ => LockstepIterSize::Unconstrained - }, + _ => LockstepIterSize::Unconstrained, + } + } TokenTree::Token(..) => LockstepIterSize::Unconstrained, } } diff --git a/src/libsyntax/parse/lexer/mod.rs b/src/libsyntax/parse/lexer/mod.rs index 3c342c2ff26..2882acb0e78 100644 --- a/src/libsyntax/parse/lexer/mod.rs +++ b/src/libsyntax/parse/lexer/mod.rs @@ -1598,7 +1598,7 @@ mod tests { use std::io; use std::path::PathBuf; use syntax_pos::{BytePos, Span, NO_EXPANSION}; - use rustc_data_structures::fx::FxHashSet; + use rustc_data_structures::fx::{FxHashSet, FxHashMap}; use rustc_data_structures::sync::Lock; fn mk_sess(sm: Lrc<SourceMap>) -> ParseSess { @@ -1617,6 +1617,7 @@ mod tests { raw_identifier_spans: Lock::new(Vec::new()), registered_diagnostics: Lock::new(ErrorMap::new()), buffered_lints: Lock::new(vec![]), + ambiguous_block_expr_parse: Lock::new(FxHashMap::default()), } } diff --git a/src/libsyntax/parse/mod.rs b/src/libsyntax/parse/mod.rs index 8c2fab9ada7..be44b964ba5 100644 --- a/src/libsyntax/parse/mod.rs +++ b/src/libsyntax/parse/mod.rs @@ -11,12 +11,12 @@ use crate::tokenstream::{TokenStream, TokenTree}; use crate::diagnostics::plugin::ErrorMap; use crate::print::pprust::token_to_string; -use errors::{FatalError, Level, Handler, ColorConfig, Diagnostic, DiagnosticBuilder}; +use errors::{Applicability, FatalError, Level, Handler, ColorConfig, Diagnostic, DiagnosticBuilder}; use rustc_data_structures::sync::{Lrc, Lock}; use syntax_pos::{Span, SourceFile, FileName, MultiSpan}; use log::debug; -use rustc_data_structures::fx::FxHashSet; +use rustc_data_structures::fx::{FxHashSet, FxHashMap}; use std::borrow::Cow; use std::path::{Path, PathBuf}; use std::str; @@ -52,6 +52,10 @@ pub struct ParseSess { included_mod_stack: Lock<Vec<PathBuf>>, source_map: Lrc<SourceMap>, pub buffered_lints: Lock<Vec<BufferedEarlyLint>>, + /// Contains the spans of block expressions that could have been incomplete based on the + /// operation token that followed it, but that the parser cannot identify without further + /// analysis. + pub ambiguous_block_expr_parse: Lock<FxHashMap<Span, Span>>, } impl ParseSess { @@ -75,6 +79,7 @@ impl ParseSess { included_mod_stack: Lock::new(vec![]), source_map, buffered_lints: Lock::new(vec![]), + ambiguous_block_expr_parse: Lock::new(FxHashMap::default()), } } @@ -98,6 +103,24 @@ impl ParseSess { }); }); } + + /// Extend an error with a suggestion to wrap an expression with parentheses to allow the + /// parser to continue parsing the following operation as part of the same expression. + pub fn expr_parentheses_needed( + &self, + err: &mut DiagnosticBuilder<'_>, + span: Span, + alt_snippet: Option<String>, + ) { + if let Some(snippet) = self.source_map().span_to_snippet(span).ok().or(alt_snippet) { + err.span_suggestion( + span, + "parentheses are required to parse this as an expression", + format!("({})", snippet), + Applicability::MachineApplicable, + ); + } + } } #[derive(Clone)] diff --git a/src/libsyntax/parse/parser.rs b/src/libsyntax/parse/parser.rs index f3eac71ee77..d97d1e2f0f4 100644 --- a/src/libsyntax/parse/parser.rs +++ b/src/libsyntax/parse/parser.rs @@ -50,7 +50,10 @@ use crate::symbol::{Symbol, keywords}; use errors::{Applicability, DiagnosticBuilder, DiagnosticId, FatalError}; use rustc_target::spec::abi::{self, Abi}; -use syntax_pos::{Span, MultiSpan, BytePos, FileName}; +use syntax_pos::{ + Span, MultiSpan, BytePos, FileName, + hygiene::CompilerDesugaringKind, +}; use log::{debug, trace}; use std::borrow::Cow; @@ -186,6 +189,7 @@ enum PrevTokenKind { Interpolated, Eof, Ident, + BitOr, Other, } @@ -1375,6 +1379,7 @@ impl<'a> Parser<'a> { token::DocComment(..) => PrevTokenKind::DocComment, token::Comma => PrevTokenKind::Comma, token::BinOp(token::Plus) => PrevTokenKind::Plus, + token::BinOp(token::Or) => PrevTokenKind::BitOr, token::Interpolated(..) => PrevTokenKind::Interpolated, token::Eof => PrevTokenKind::Eof, token::Ident(..) => PrevTokenKind::Ident, @@ -2806,6 +2811,12 @@ impl<'a> Parser<'a> { let msg = format!("expected expression, found {}", self.this_token_descr()); let mut err = self.fatal(&msg); + let sp = self.sess.source_map().start_point(self.span); + if let Some(sp) = self.sess.ambiguous_block_expr_parse.borrow() + .get(&sp) + { + self.sess.expr_parentheses_needed(&mut err, *sp, None); + } err.span_label(self.span, "expected expression"); return Err(err); } @@ -2845,7 +2856,7 @@ impl<'a> Parser<'a> { "struct literals are not allowed here", ); err.multipart_suggestion( - "surround the struct literal with parenthesis", + "surround the struct literal with parentheses", vec![ (lo.shrink_to_lo(), "(".to_string()), (expr.span.shrink_to_hi(), ")".to_string()), @@ -3506,9 +3517,42 @@ impl<'a> Parser<'a> { } }; - if self.expr_is_complete(&lhs) { - // Semi-statement forms are odd. See https://github.com/rust-lang/rust/issues/29071 - return Ok(lhs); + match (self.expr_is_complete(&lhs), AssocOp::from_token(&self.token)) { + (true, None) => { + // Semi-statement forms are odd. See https://github.com/rust-lang/rust/issues/29071 + return Ok(lhs); + } + (false, _) => {} // continue parsing the expression + // An exhaustive check is done in the following block, but these are checked first + // because they *are* ambiguous but also reasonable looking incorrect syntax, so we + // want to keep their span info to improve diagnostics in these cases in a later stage. + (true, Some(AssocOp::Multiply)) | // `{ 42 } *foo = bar;` or `{ 42 } * 3` + (true, Some(AssocOp::Subtract)) | // `{ 42 } -5` + (true, Some(AssocOp::Add)) => { // `{ 42 } + 42 + // These cases are ambiguous and can't be identified in the parser alone + let sp = self.sess.source_map().start_point(self.span); + self.sess.ambiguous_block_expr_parse.borrow_mut().insert(sp, lhs.span); + return Ok(lhs); + } + (true, Some(ref op)) if !op.can_continue_expr_unambiguously() => { + return Ok(lhs); + } + (true, Some(_)) => { + // We've found an expression that would be parsed as a statement, but the next + // token implies this should be parsed as an expression. + // For example: `if let Some(x) = x { x } else { 0 } / 2` + let mut err = self.sess.span_diagnostic.struct_span_err(self.span, &format!( + "expected expression, found `{}`", + pprust::token_to_string(&self.token), + )); + err.span_label(self.span, "expected expression"); + self.sess.expr_parentheses_needed( + &mut err, + lhs.span, + Some(pprust::expr_to_string(&lhs), + )); + err.emit(); + } } self.expected_tokens.push(TokenType::Operator); while let Some(op) = AssocOp::from_token(&self.token) { @@ -4819,6 +4863,10 @@ impl<'a> Parser<'a> { ); let mut err = self.fatal(&msg); err.span_label(self.span, format!("expected {}", expected)); + let sp = self.sess.source_map().start_point(self.span); + if let Some(sp) = self.sess.ambiguous_block_expr_parse.borrow().get(&sp) { + self.sess.expr_parentheses_needed(&mut err, *sp, None); + } return Err(err); } } @@ -8727,6 +8775,10 @@ impl<'a> Parser<'a> { /// The arguments of the function are replaced in HIR lowering with the arguments created by /// this function and the statements created here are inserted at the top of the closure body. fn construct_async_arguments(&mut self, asyncness: &mut Spanned<IsAsync>, decl: &mut FnDecl) { + // FIXME(davidtwco): This function should really live in the HIR lowering but because + // the types constructed here need to be used in parts of resolve so that the correct + // locals are considered upvars, it is currently easier for it to live here in the parser, + // where it can be constructed once. if let IsAsync::Async { ref mut arguments, .. } = asyncness.node { for (index, input) in decl.inputs.iter_mut().enumerate() { let id = ast::DUMMY_NODE_ID; @@ -8741,6 +8793,15 @@ impl<'a> Parser<'a> { // statement. let (binding_mode, ident, is_simple_pattern) = match input.pat.node { PatKind::Ident(binding_mode @ BindingMode::ByValue(_), ident, _) => { + // Simple patterns like this don't have a generated argument, but they are + // moved into the closure with a statement, so any `mut` bindings on the + // argument will be unused. This binding mode can't be removed, because + // this would affect the input to procedural macros, but they can have + // their span marked as being the result of a compiler desugaring so + // that they aren't linted against. + input.pat.span = self.sess.source_map().mark_span_with_reason( + CompilerDesugaringKind::Async, span, None); + (binding_mode, ident, true) } _ => (BindingMode::ByValue(Mutability::Mutable), ident, false), @@ -8810,15 +8871,6 @@ impl<'a> Parser<'a> { }) }; - // Remove mutability from arguments. If this is not a simple pattern, - // those arguments are replaced by `__argN`, so there is no need to do this. - if let PatKind::Ident(BindingMode::ByValue(mutability @ Mutability::Mutable), ..) = - &mut input.pat.node - { - assert!(is_simple_pattern); - *mutability = Mutability::Immutable; - } - let move_stmt = Stmt { id, node: StmtKind::Local(P(move_local)), span }; arguments.push(AsyncArgument { ident, arg, pat_stmt, move_stmt }); } diff --git a/src/libsyntax/source_map.rs b/src/libsyntax/source_map.rs index 08abbf5e8a4..215618bd09c 100644 --- a/src/libsyntax/source_map.rs +++ b/src/libsyntax/source_map.rs @@ -930,6 +930,27 @@ impl SourceMap { None } + + /// Reuses the span but adds information like the kind of the desugaring and features that are + /// allowed inside this span. + pub fn mark_span_with_reason( + &self, + reason: hygiene::CompilerDesugaringKind, + span: Span, + allow_internal_unstable: Option<Lrc<[symbol::Symbol]>>, + ) -> Span { + let mark = Mark::fresh(Mark::root()); + mark.set_expn_info(ExpnInfo { + call_site: span, + def_site: Some(span), + format: CompilerDesugaring(reason), + allow_internal_unstable, + allow_internal_unsafe: false, + local_inner_macros: false, + edition: hygiene::default_edition(), + }); + span.with_ctxt(SyntaxContext::empty().apply_mark(mark)) + } } impl SourceMapper for SourceMap { diff --git a/src/libsyntax/util/parser.rs b/src/libsyntax/util/parser.rs index 80dabffaba9..86e89945afe 100644 --- a/src/libsyntax/util/parser.rs +++ b/src/libsyntax/util/parser.rs @@ -207,6 +207,31 @@ impl AssocOp { ObsoleteInPlace | Assign | AssignOp(_) | As | DotDot | DotDotEq | Colon => None } } + + /// This operator could be used to follow a block unambiguously. + /// + /// This is used for error recovery at the moment, providing a suggestion to wrap blocks with + /// parentheses while having a high degree of confidence on the correctness of the suggestion. + pub fn can_continue_expr_unambiguously(&self) -> bool { + use AssocOp::*; + match self { + BitXor | // `{ 42 } ^ 3` + Assign | // `{ 42 } = { 42 }` + Divide | // `{ 42 } / 42` + Modulus | // `{ 42 } % 2` + ShiftRight | // `{ 42 } >> 2` + LessEqual | // `{ 42 } <= 3` + Greater | // `{ 42 } > 3` + GreaterEqual | // `{ 42 } >= 3` + AssignOp(_) | // `{ 42 } +=` + LAnd | // `{ 42 } &&foo` + As | // `{ 42 } as usize` + // Equal | // `{ 42 } == { 42 }` Accepting these here would regress incorrect + // NotEqual | // `{ 42 } != { 42 } struct literals parser recovery. + Colon => true, // `{ 42 }: usize` + _ => false, + } + } } pub const PREC_RESET: i8 = -100; |
