move syntax::parse -> librustc_parse

also move MACRO_ARGUMENTS -> librustc_parse

src/librustc_parse/Cargo.toml

@@ -0,0 +1,21 @@
+[package]
+authors = ["The Rust Project Developers"]
+name = "rustc_parse"
+version = "0.0.0"
+edition = "2018"
+
+[lib]
+name = "rustc_parse"
+path = "lib.rs"
+doctest = false
+
+[dependencies]
+bitflags = "1.0"
+log = "0.4"
+syntax_pos = { path = "../libsyntax_pos" }
+syntax = { path = "../libsyntax" }
+errors = { path = "../librustc_errors", package = "rustc_errors" }
+rustc_data_structures = { path = "../librustc_data_structures" }
+rustc_lexer = { path = "../librustc_lexer" }
+rustc_target = { path = "../librustc_target" }
+smallvec = { version = "1.0", features = ["union", "may_dangle"] }

src/librustc_parse/error_codes.rs

@@ -0,0 +1,174 @@
+// Error messages for EXXXX errors.
+// Each message should start and end with a new line, and be wrapped to 80
+// characters.  In vim you can `:set tw=80` and use `gq` to wrap paragraphs. Use
+// `:set tw=0` to disable.
+syntax::register_diagnostics! {
+
+E0178: r##"
+In types, the `+` type operator has low precedence, so it is often necessary
+to use parentheses.
+
+For example:
+
+```compile_fail,E0178
+trait Foo {}
+
+struct Bar<'a> {
+    w: &'a Foo + Copy,   // error, use &'a (Foo + Copy)
+    x: &'a Foo + 'a,     // error, use &'a (Foo + 'a)
+    y: &'a mut Foo + 'a, // error, use &'a mut (Foo + 'a)
+    z: fn() -> Foo + 'a, // error, use fn() -> (Foo + 'a)
+}
+```
+
+More details can be found in [RFC 438].
+
+[RFC 438]: https://github.com/rust-lang/rfcs/pull/438
+"##,
+
+E0583: r##"
+A file wasn't found for an out-of-line module.
+
+Erroneous code example:
+
+```ignore (compile_fail not working here; see Issue #43707)
+mod file_that_doesnt_exist; // error: file not found for module
+
+fn main() {}
+```
+
+Please be sure that a file corresponding to the module exists. If you
+want to use a module named `file_that_doesnt_exist`, you need to have a file
+named `file_that_doesnt_exist.rs` or `file_that_doesnt_exist/mod.rs` in the
+same directory.
+"##,
+
+E0584: r##"
+A doc comment that is not attached to anything has been encountered.
+
+Erroneous code example:
+
+```compile_fail,E0584
+trait Island {
+    fn lost();
+
+    /// I'm lost!
+}
+```
+
+A little reminder: a doc comment has to be placed before the item it's supposed
+to document. So if you want to document the `Island` trait, you need to put a
+doc comment before it, not inside it. Same goes for the `lost` method: the doc
+comment needs to be before it:
+
+```
+/// I'm THE island!
+trait Island {
+    /// I'm lost!
+    fn lost();
+}
+```
+"##,
+
+E0585: r##"
+A documentation comment that doesn't document anything was found.
+
+Erroneous code example:
+
+```compile_fail,E0585
+fn main() {
+    // The following doc comment will fail:
+    /// This is a useless doc comment!
+}
+```
+
+Documentation comments need to be followed by items, including functions,
+types, modules, etc. Examples:
+
+```
+/// I'm documenting the following struct:
+struct Foo;
+
+/// I'm documenting the following function:
+fn foo() {}
+```
+"##,
+
+E0586: r##"
+An inclusive range was used with no end.
+
+Erroneous code example:
+
+```compile_fail,E0586
+fn main() {
+    let tmp = vec![0, 1, 2, 3, 4, 4, 3, 3, 2, 1];
+    let x = &tmp[1..=]; // error: inclusive range was used with no end
+}
+```
+
+An inclusive range needs an end in order to *include* it. If you just need a
+start and no end, use a non-inclusive range (with `..`):
+
+```
+fn main() {
+    let tmp = vec![0, 1, 2, 3, 4, 4, 3, 3, 2, 1];
+    let x = &tmp[1..]; // ok!
+}
+```
+
+Or put an end to your inclusive range:
+
+```
+fn main() {
+    let tmp = vec![0, 1, 2, 3, 4, 4, 3, 3, 2, 1];
+    let x = &tmp[1..=3]; // ok!
+}
+```
+"##,
+
+E0704: r##"
+This error indicates that a incorrect visibility restriction was specified.
+
+Example of erroneous code:
+
+```compile_fail,E0704
+mod foo {
+    pub(foo) struct Bar {
+        x: i32
+    }
+}
+```
+
+To make struct `Bar` only visible in module `foo` the `in` keyword should be
+used:
+```
+mod foo {
+    pub(in crate::foo) struct Bar {
+        x: i32
+    }
+}
+# fn main() {}
+```
+
+For more information see the Rust Reference on [Visibility].
+
+[Visibility]: https://doc.rust-lang.org/reference/visibility-and-privacy.html
+"##,
+
+E0743: r##"
+C-variadic has been used on a non-foreign function.
+
+Erroneous code example:
+
+```compile_fail,E0743
+fn foo2(x: u8, ...) {} // error!
+```
+
+Only foreign functions can use C-variadic (`...`). It is used to give an
+undefined number of parameters to a given function (like `printf` in C). The
+equivalent in Rust would be to use macros directly.
+"##,
+
+;
+
+}

src/librustc_parse/lexer/mod.rs

@@ -0,0 +1,643 @@
+use syntax::token::{self, Token, TokenKind};
+use syntax::sess::ParseSess;
+use syntax::symbol::{sym, Symbol};
+use syntax::util::comments;
+
+use errors::{FatalError, DiagnosticBuilder};
+use syntax_pos::{BytePos, Pos, Span};
+use rustc_lexer::Base;
+use rustc_lexer::unescape;
+
+use std::char;
+use std::convert::TryInto;
+use rustc_data_structures::sync::Lrc;
+use log::debug;
+
+mod tokentrees;
+mod unicode_chars;
+mod unescape_error_reporting;
+use unescape_error_reporting::{emit_unescape_error, push_escaped_char};
+
+#[derive(Clone, Debug)]
+pub struct UnmatchedBrace {
+    pub expected_delim: token::DelimToken,
+    pub found_delim: Option<token::DelimToken>,
+    pub found_span: Span,
+    pub unclosed_span: Option<Span>,
+    pub candidate_span: Option<Span>,
+}
+
+pub struct StringReader<'a> {
+    sess: &'a ParseSess,
+    /// Initial position, read-only.
+    start_pos: BytePos,
+    /// The absolute offset within the source_map of the current character.
+    // FIXME(#64197): `pub` is needed by tests for now.
+    pub pos: BytePos,
+    /// Stop reading src at this index.
+    end_src_index: usize,
+    /// Source text to tokenize.
+    src: Lrc<String>,
+    override_span: Option<Span>,
+}
+
+impl<'a> StringReader<'a> {
+    pub fn new(sess: &'a ParseSess,
+               source_file: Lrc<syntax_pos::SourceFile>,
+               override_span: Option<Span>) -> Self {
+        if source_file.src.is_none() {
+            sess.span_diagnostic.bug(&format!("cannot lex `source_file` without source: {}",
+                                              source_file.name));
+        }
+
+        let src = (*source_file.src.as_ref().unwrap()).clone();
+
+        StringReader {
+            sess,
+            start_pos: source_file.start_pos,
+            pos: source_file.start_pos,
+            end_src_index: src.len(),
+            src,
+            override_span,
+        }
+    }
+
+    pub fn retokenize(sess: &'a ParseSess, mut span: Span) -> Self {
+        let begin = sess.source_map().lookup_byte_offset(span.lo());
+        let end = sess.source_map().lookup_byte_offset(span.hi());
+
+        // Make the range zero-length if the span is invalid.
+        if begin.sf.start_pos != end.sf.start_pos {
+            span = span.shrink_to_lo();
+        }
+
+        let mut sr = StringReader::new(sess, begin.sf, None);
+
+        // Seek the lexer to the right byte range.
+        sr.end_src_index = sr.src_index(span.hi());
+
+        sr
+    }
+
+
+    fn mk_sp(&self, lo: BytePos, hi: BytePos) -> Span {
+        self.override_span.unwrap_or_else(|| Span::with_root_ctxt(lo, hi))
+    }
+
+    /// Returns the next token, including trivia like whitespace or comments.
+    ///
+    /// `Err(())` means that some errors were encountered, which can be
+    /// retrieved using `buffer_fatal_errors`.
+    pub fn next_token(&mut self) -> Token {
+        let start_src_index = self.src_index(self.pos);
+        let text: &str = &self.src[start_src_index..self.end_src_index];
+
+        if text.is_empty() {
+            let span = self.mk_sp(self.pos, self.pos);
+            return Token::new(token::Eof, span);
+        }
+
+        {
+            let is_beginning_of_file = self.pos == self.start_pos;
+            if is_beginning_of_file {
+                if let Some(shebang_len) = rustc_lexer::strip_shebang(text) {
+                    let start = self.pos;
+                    self.pos = self.pos + BytePos::from_usize(shebang_len);
+
+                    let sym = self.symbol_from(start + BytePos::from_usize("#!".len()));
+                    let kind = token::Shebang(sym);
+
+                    let span = self.mk_sp(start, self.pos);
+                    return Token::new(kind, span);
+                }
+            }
+        }
+
+        let token = rustc_lexer::first_token(text);
+
+        let start = self.pos;
+        self.pos = self.pos + BytePos::from_usize(token.len);
+
+        debug!("try_next_token: {:?}({:?})", token.kind, self.str_from(start));
+
+        // This could use `?`, but that makes code significantly (10-20%) slower.
+        // https://github.com/rust-lang/rust/issues/37939
+        let kind = self.cook_lexer_token(token.kind, start);
+
+        let span = self.mk_sp(start, self.pos);
+        Token::new(kind, span)
+    }
+
+    /// Report a fatal lexical error with a given span.
+    fn fatal_span(&self, sp: Span, m: &str) -> FatalError {
+        self.sess.span_diagnostic.span_fatal(sp, m)
+    }
+
+    /// Report a lexical error with a given span.
+    fn err_span(&self, sp: Span, m: &str) {
+        self.sess.span_diagnostic.struct_span_err(sp, m).emit();
+    }
+
+
+    /// Report a fatal error spanning [`from_pos`, `to_pos`).
+    fn fatal_span_(&self, from_pos: BytePos, to_pos: BytePos, m: &str) -> FatalError {
+        self.fatal_span(self.mk_sp(from_pos, to_pos), m)
+    }
+
+    /// Report a lexical error spanning [`from_pos`, `to_pos`).
+    fn err_span_(&self, from_pos: BytePos, to_pos: BytePos, m: &str) {
+        self.err_span(self.mk_sp(from_pos, to_pos), m)
+    }
+
+    fn struct_span_fatal(&self, from_pos: BytePos, to_pos: BytePos, m: &str)
+        -> DiagnosticBuilder<'a>
+    {
+        self.sess.span_diagnostic.struct_span_fatal(self.mk_sp(from_pos, to_pos), m)
+    }
+
+    fn struct_fatal_span_char(&self, from_pos: BytePos, to_pos: BytePos, m: &str, c: char)
+        -> DiagnosticBuilder<'a>
+    {
+        let mut m = m.to_string();
+        m.push_str(": ");
+        push_escaped_char(&mut m, c);
+
+        self.sess.span_diagnostic.struct_span_fatal(self.mk_sp(from_pos, to_pos), &m[..])
+    }
+
+    /// Turns simple `rustc_lexer::TokenKind` enum into a rich
+    /// `libsyntax::TokenKind`. This turns strings into interned
+    /// symbols and runs additional validation.
+    fn cook_lexer_token(
+        &self,
+        token: rustc_lexer::TokenKind,
+        start: BytePos,
+    ) -> TokenKind {
+        match token {
+            rustc_lexer::TokenKind::LineComment => {
+                let string = self.str_from(start);
+                // comments with only more "/"s are not doc comments
+                let tok = if comments::is_line_doc_comment(string) {
+                    self.forbid_bare_cr(start, string, "bare CR not allowed in doc-comment");
+                    token::DocComment(Symbol::intern(string))
+                } else {
+                    token::Comment
+                };
+
+                tok
+            }
+            rustc_lexer::TokenKind::BlockComment { terminated } => {
+                let string = self.str_from(start);
+                // block comments starting with "/**" or "/*!" are doc-comments
+                // but comments with only "*"s between two "/"s are not
+                let is_doc_comment = comments::is_block_doc_comment(string);
+
+                if !terminated {
+                    let msg = if is_doc_comment {
+                        "unterminated block doc-comment"
+                    } else {
+                        "unterminated block comment"
+                    };
+                    let last_bpos = self.pos;
+                    self.fatal_span_(start, last_bpos, msg).raise();
+                }
+
+                let tok = if is_doc_comment {
+                    self.forbid_bare_cr(start,
+                                        string,
+                                        "bare CR not allowed in block doc-comment");
+                    token::DocComment(Symbol::intern(string))
+                } else {
+                    token::Comment
+                };
+
+                tok
+            }
+            rustc_lexer::TokenKind::Whitespace => token::Whitespace,
+            rustc_lexer::TokenKind::Ident | rustc_lexer::TokenKind::RawIdent => {
+                let is_raw_ident = token == rustc_lexer::TokenKind::RawIdent;
+                let mut ident_start = start;
+                if is_raw_ident {
+                    ident_start = ident_start + BytePos(2);
+                }
+                // FIXME: perform NFKC normalization here. (Issue #2253)
+                let sym = self.symbol_from(ident_start);
+                if is_raw_ident {
+                    let span = self.mk_sp(start, self.pos);
+                    if !sym.can_be_raw() {
+                        self.err_span(span, &format!("`{}` cannot be a raw identifier", sym));
+                    }
+                    self.sess.raw_identifier_spans.borrow_mut().push(span);
+                }
+                token::Ident(sym, is_raw_ident)
+            }
+            rustc_lexer::TokenKind::Literal { kind, suffix_start } => {
+                let suffix_start = start + BytePos(suffix_start as u32);
+                let (kind, symbol) = self.cook_lexer_literal(start, suffix_start, kind);
+                let suffix = if suffix_start < self.pos {
+                    let string = self.str_from(suffix_start);
+                    if string == "_" {
+                        self.sess.span_diagnostic
+                            .struct_span_warn(self.mk_sp(suffix_start, self.pos),
+                                              "underscore literal suffix is not allowed")
+                            .warn("this was previously accepted by the compiler but is \
+                                   being phased out; it will become a hard error in \
+                                   a future release!")
+                            .note("for more information, see issue #42326 \
+                                   <https://github.com/rust-lang/rust/issues/42326>")
+                            .emit();
+                        None
+                    } else {
+                        Some(Symbol::intern(string))
+                    }
+                } else {
+                    None
+                };
+                token::Literal(token::Lit { kind, symbol, suffix })
+            }
+            rustc_lexer::TokenKind::Lifetime { starts_with_number } => {
+                // Include the leading `'` in the real identifier, for macro
+                // expansion purposes. See #12512 for the gory details of why
+                // this is necessary.
+                let lifetime_name = self.str_from(start);
+                if starts_with_number {
+                    self.err_span_(
+                        start,
+                        self.pos,
+                        "lifetimes cannot start with a number",
+                    );
+                }
+                let ident = Symbol::intern(lifetime_name);
+                token::Lifetime(ident)
+            }
+            rustc_lexer::TokenKind::Semi => token::Semi,
+            rustc_lexer::TokenKind::Comma => token::Comma,
+            rustc_lexer::TokenKind::Dot => token::Dot,
+            rustc_lexer::TokenKind::OpenParen => token::OpenDelim(token::Paren),
+            rustc_lexer::TokenKind::CloseParen => token::CloseDelim(token::Paren),
+            rustc_lexer::TokenKind::OpenBrace => token::OpenDelim(token::Brace),
+            rustc_lexer::TokenKind::CloseBrace => token::CloseDelim(token::Brace),
+            rustc_lexer::TokenKind::OpenBracket => token::OpenDelim(token::Bracket),
+            rustc_lexer::TokenKind::CloseBracket => token::CloseDelim(token::Bracket),
+            rustc_lexer::TokenKind::At => token::At,
+            rustc_lexer::TokenKind::Pound => token::Pound,
+            rustc_lexer::TokenKind::Tilde => token::Tilde,
+            rustc_lexer::TokenKind::Question => token::Question,
+            rustc_lexer::TokenKind::Colon => token::Colon,
+            rustc_lexer::TokenKind::Dollar => token::Dollar,
+            rustc_lexer::TokenKind::Eq => token::Eq,
+            rustc_lexer::TokenKind::Not => token::Not,
+            rustc_lexer::TokenKind::Lt => token::Lt,
+            rustc_lexer::TokenKind::Gt => token::Gt,
+            rustc_lexer::TokenKind::Minus => token::BinOp(token::Minus),
+            rustc_lexer::TokenKind::And => token::BinOp(token::And),
+            rustc_lexer::TokenKind::Or => token::BinOp(token::Or),
+            rustc_lexer::TokenKind::Plus => token::BinOp(token::Plus),
+            rustc_lexer::TokenKind::Star => token::BinOp(token::Star),
+            rustc_lexer::TokenKind::Slash => token::BinOp(token::Slash),
+            rustc_lexer::TokenKind::Caret => token::BinOp(token::Caret),
+            rustc_lexer::TokenKind::Percent => token::BinOp(token::Percent),
+
+            rustc_lexer::TokenKind::Unknown => {
+                let c = self.str_from(start).chars().next().unwrap();
+                let mut err = self.struct_fatal_span_char(start,
+                                                          self.pos,
+                                                          "unknown start of token",
+                                                          c);
+                // FIXME: the lexer could be used to turn the ASCII version of unicode homoglyphs,
+                // instead of keeping a table in `check_for_substitution`into the token. Ideally,
+                // this should be inside `rustc_lexer`. However, we should first remove compound
+                // tokens like `<<` from `rustc_lexer`, and then add fancier error recovery to it,
+                // as there will be less overall work to do this way.
+                let token = unicode_chars::check_for_substitution(self, start, c, &mut err)
+                    .unwrap_or_else(|| token::Unknown(self.symbol_from(start)));
+                err.emit();
+                token
+            }
+        }
+    }
+
+    fn cook_lexer_literal(
+        &self,
+        start: BytePos,
+        suffix_start: BytePos,
+        kind: rustc_lexer::LiteralKind
+    ) -> (token::LitKind, Symbol) {
+        match kind {
+            rustc_lexer::LiteralKind::Char { terminated } => {
+                if !terminated {
+                    self.fatal_span_(start, suffix_start,
+                                     "unterminated character literal".into())
+                        .raise()
+                }
+                let content_start = start + BytePos(1);
+                let content_end = suffix_start - BytePos(1);
+                self.validate_char_escape(content_start, content_end);
+                let id = self.symbol_from_to(content_start, content_end);
+                (token::Char, id)
+            },
+            rustc_lexer::LiteralKind::Byte { terminated } => {
+                if !terminated {
+                    self.fatal_span_(start + BytePos(1), suffix_start,
+                                     "unterminated byte constant".into())
+                        .raise()
+                }
+                let content_start = start + BytePos(2);
+                let content_end = suffix_start - BytePos(1);
+                self.validate_byte_escape(content_start, content_end);
+                let id = self.symbol_from_to(content_start, content_end);
+                (token::Byte, id)
+            },
+            rustc_lexer::LiteralKind::Str { terminated } => {
+                if !terminated {
+                    self.fatal_span_(start, suffix_start,
+                                     "unterminated double quote string".into())
+                        .raise()
+                }
+                let content_start = start + BytePos(1);
+                let content_end = suffix_start - BytePos(1);
+                self.validate_str_escape(content_start, content_end);
+                let id = self.symbol_from_to(content_start, content_end);
+                (token::Str, id)
+            }
+            rustc_lexer::LiteralKind::ByteStr { terminated } => {
+                if !terminated {
+                    self.fatal_span_(start + BytePos(1), suffix_start,
+                                     "unterminated double quote byte string".into())
+                        .raise()
+                }
+                let content_start = start + BytePos(2);
+                let content_end = suffix_start - BytePos(1);
+                self.validate_byte_str_escape(content_start, content_end);
+                let id = self.symbol_from_to(content_start, content_end);
+                (token::ByteStr, id)
+            }
+            rustc_lexer::LiteralKind::RawStr { n_hashes, started, terminated } => {
+                if !started {
+                    self.report_non_started_raw_string(start);
+                }
+                if !terminated {
+                    self.report_unterminated_raw_string(start, n_hashes)
+                }
+                let n_hashes: u16 = self.restrict_n_hashes(start, n_hashes);
+                let n = u32::from(n_hashes);
+                let content_start = start + BytePos(2 + n);
+                let content_end = suffix_start - BytePos(1 + n);
+                self.validate_raw_str_escape(content_start, content_end);
+                let id = self.symbol_from_to(content_start, content_end);
+                (token::StrRaw(n_hashes), id)
+            }
+            rustc_lexer::LiteralKind::RawByteStr { n_hashes, started, terminated } => {
+                if !started {
+                    self.report_non_started_raw_string(start);
+                }
+                if !terminated {
+                    self.report_unterminated_raw_string(start, n_hashes)
+                }
+                let n_hashes: u16 = self.restrict_n_hashes(start, n_hashes);
+                let n = u32::from(n_hashes);
+                let content_start = start + BytePos(3 + n);
+                let content_end = suffix_start - BytePos(1 + n);
+                self.validate_raw_byte_str_escape(content_start, content_end);
+                let id = self.symbol_from_to(content_start, content_end);
+                (token::ByteStrRaw(n_hashes), id)
+            }
+            rustc_lexer::LiteralKind::Int { base, empty_int } => {
+                if empty_int {
+                    self.err_span_(start, suffix_start, "no valid digits found for number");
+                    (token::Integer, sym::integer(0))
+                } else {
+                    self.validate_int_literal(base, start, suffix_start);
+                    (token::Integer, self.symbol_from_to(start, suffix_start))
+                }
+            },
+            rustc_lexer::LiteralKind::Float { base, empty_exponent } => {
+                if empty_exponent {
+                    let mut err = self.struct_span_fatal(
+                        start, self.pos,
+                        "expected at least one digit in exponent"
+                    );
+                    err.emit();
+                }
+
+                match base {
+                    Base::Hexadecimal => {
+                        self.err_span_(start, suffix_start,
+                                       "hexadecimal float literal is not supported")
+                    }
+                    Base::Octal => {
+                        self.err_span_(start, suffix_start,
+                                       "octal float literal is not supported")
+                    }
+                    Base::Binary => {
+                        self.err_span_(start, suffix_start,
+                                       "binary float literal is not supported")
+                    }
+                    _ => ()
+                }
+
+                let id = self.symbol_from_to(start, suffix_start);
+                (token::Float, id)
+            },
+        }
+    }
+
+    #[inline]
+    fn src_index(&self, pos: BytePos) -> usize {
+        (pos - self.start_pos).to_usize()
+    }
+
+    /// Slice of the source text from `start` up to but excluding `self.pos`,
+    /// meaning the slice does not include the character `self.ch`.
+    fn str_from(&self, start: BytePos) -> &str
+    {
+        self.str_from_to(start, self.pos)
+    }
+
+    /// Creates a Symbol from a given offset to the current offset.
+    fn symbol_from(&self, start: BytePos) -> Symbol {
+        debug!("taking an ident from {:?} to {:?}", start, self.pos);
+        Symbol::intern(self.str_from(start))
+    }
+
+    /// As symbol_from, with an explicit endpoint.
+    fn symbol_from_to(&self, start: BytePos, end: BytePos) -> Symbol {
+        debug!("taking an ident from {:?} to {:?}", start, end);
+        Symbol::intern(self.str_from_to(start, end))
+    }
+
+    /// Slice of the source text spanning from `start` up to but excluding `end`.
+    fn str_from_to(&self, start: BytePos, end: BytePos) -> &str
+    {
+        &self.src[self.src_index(start)..self.src_index(end)]
+    }
+
+    fn forbid_bare_cr(&self, start: BytePos, s: &str, errmsg: &str) {
+        let mut idx = 0;
+        loop {
+            idx = match s[idx..].find('\r') {
+                None => break,
+                Some(it) => idx + it + 1
+            };
+            self.err_span_(start + BytePos(idx as u32 - 1),
+                           start + BytePos(idx as u32),
+                           errmsg);
+        }
+    }
+
+    fn report_non_started_raw_string(&self, start: BytePos) -> ! {
+        let bad_char = self.str_from(start).chars().last().unwrap();
+        self
+            .struct_fatal_span_char(
+                start,
+                self.pos,
+                "found invalid character; only `#` is allowed \
+                 in raw string delimitation",
+                bad_char,
+            )
+            .emit();
+        FatalError.raise()
+    }
+
+    fn report_unterminated_raw_string(&self, start: BytePos, n_hashes: usize) -> ! {
+        let mut err = self.struct_span_fatal(
+            start, start,
+            "unterminated raw string",
+        );
+        err.span_label(
+            self.mk_sp(start, start),
+            "unterminated raw string",
+        );
+
+        if n_hashes > 0 {
+            err.note(&format!("this raw string should be terminated with `\"{}`",
+                                "#".repeat(n_hashes as usize)));
+        }
+
+        err.emit();
+        FatalError.raise()
+    }
+
+    fn restrict_n_hashes(&self, start: BytePos, n_hashes: usize) -> u16 {
+        match n_hashes.try_into() {
+            Ok(n_hashes) => n_hashes,
+            Err(_) => {
+                self.fatal_span_(start,
+                                 self.pos,
+                                 "too many `#` symbols: raw strings may be \
+                                  delimited by up to 65535 `#` symbols").raise();
+            }
+        }
+    }
+
+    fn validate_char_escape(&self, content_start: BytePos, content_end: BytePos) {
+        let lit = self.str_from_to(content_start, content_end);
+        if let Err((off, err)) = unescape::unescape_char(lit) {
+            emit_unescape_error(
+                &self.sess.span_diagnostic,
+                lit,
+                self.mk_sp(content_start - BytePos(1), content_end + BytePos(1)),
+                unescape::Mode::Char,
+                0..off,
+                err,
+            )
+        }
+    }
+
+    fn validate_byte_escape(&self, content_start: BytePos, content_end: BytePos) {
+        let lit = self.str_from_to(content_start, content_end);
+        if let Err((off, err)) = unescape::unescape_byte(lit) {
+            emit_unescape_error(
+                &self.sess.span_diagnostic,
+                lit,
+                self.mk_sp(content_start - BytePos(1), content_end + BytePos(1)),
+                unescape::Mode::Byte,
+                0..off,
+                err,
+            )
+        }
+    }
+
+    fn validate_str_escape(&self, content_start: BytePos, content_end: BytePos) {
+        let lit = self.str_from_to(content_start, content_end);
+        unescape::unescape_str(lit, &mut |range, c| {
+            if let Err(err) = c {
+                emit_unescape_error(
+                    &self.sess.span_diagnostic,
+                    lit,
+                    self.mk_sp(content_start - BytePos(1), content_end + BytePos(1)),
+                    unescape::Mode::Str,
+                    range,
+                    err,
+                )
+            }
+        })
+    }
+
+    fn validate_raw_str_escape(&self, content_start: BytePos, content_end: BytePos) {
+        let lit = self.str_from_to(content_start, content_end);
+        unescape::unescape_raw_str(lit, &mut |range, c| {
+            if let Err(err) = c {
+                emit_unescape_error(
+                    &self.sess.span_diagnostic,
+                    lit,
+                    self.mk_sp(content_start - BytePos(1), content_end + BytePos(1)),
+                    unescape::Mode::Str,
+                    range,
+                    err,
+                )
+            }
+        })
+    }
+
+    fn validate_raw_byte_str_escape(&self, content_start: BytePos, content_end: BytePos) {
+        let lit = self.str_from_to(content_start, content_end);
+        unescape::unescape_raw_byte_str(lit, &mut |range, c| {
+            if let Err(err) = c {
+                emit_unescape_error(
+                    &self.sess.span_diagnostic,
+                    lit,
+                    self.mk_sp(content_start - BytePos(1), content_end + BytePos(1)),
+                    unescape::Mode::ByteStr,
+                    range,
+                    err,
+                )
+            }
+        })
+    }
+
+    fn validate_byte_str_escape(&self, content_start: BytePos, content_end: BytePos) {
+        let lit = self.str_from_to(content_start, content_end);
+        unescape::unescape_byte_str(lit, &mut |range, c| {
+            if let Err(err) = c {
+                emit_unescape_error(
+                    &self.sess.span_diagnostic,
+                    lit,
+                    self.mk_sp(content_start - BytePos(1), content_end + BytePos(1)),
+                    unescape::Mode::ByteStr,
+                    range,
+                    err,
+                )
+            }
+        })
+    }
+
+    fn validate_int_literal(&self, base: Base, content_start: BytePos, content_end: BytePos) {
+        let base = match base {
+            Base::Binary => 2,
+            Base::Octal => 8,
+            _ => return,
+        };
+        let s = self.str_from_to(content_start + BytePos(2), content_end);
+        for (idx, c) in s.char_indices() {
+            let idx = idx as u32;
+            if c != '_' && c.to_digit(base).is_none() {
+                let lo = content_start + BytePos(2 + idx);
+                let hi = content_start + BytePos(2 + idx + c.len_utf8() as u32);
+                self.err_span_(lo, hi,
+                               &format!("invalid digit for a base {} literal", base));
+
+            }
+        }
+    }
+}

src/librustc_parse/lexer/tokentrees.rs

@@ -0,0 +1,280 @@
+use rustc_data_structures::fx::FxHashMap;
+use syntax_pos::Span;
+
+use super::{StringReader, UnmatchedBrace};
+
+use syntax::print::pprust::token_to_string;
+use syntax::token::{self, Token};
+use syntax::tokenstream::{DelimSpan, IsJoint::{self, *}, TokenStream, TokenTree, TreeAndJoint};
+
+use errors::PResult;
+
+impl<'a> StringReader<'a> {
+    crate fn into_token_trees(self) -> (PResult<'a, TokenStream>, Vec<UnmatchedBrace>) {
+        let mut tt_reader = TokenTreesReader {
+            string_reader: self,
+            token: Token::dummy(),
+            joint_to_prev: Joint,
+            open_braces: Vec::new(),
+            unmatched_braces: Vec::new(),
+            matching_delim_spans: Vec::new(),
+            last_unclosed_found_span: None,
+            last_delim_empty_block_spans: FxHashMap::default()
+        };
+        let res = tt_reader.parse_all_token_trees();
+        (res, tt_reader.unmatched_braces)
+    }
+}
+
+struct TokenTreesReader<'a> {
+    string_reader: StringReader<'a>,
+    token: Token,
+    joint_to_prev: IsJoint,
+    /// Stack of open delimiters and their spans. Used for error message.
+    open_braces: Vec<(token::DelimToken, Span)>,
+    unmatched_braces: Vec<UnmatchedBrace>,
+    /// The type and spans for all braces
+    ///
+    /// Used only for error recovery when arriving to EOF with mismatched braces.
+    matching_delim_spans: Vec<(token::DelimToken, Span, Span)>,
+    last_unclosed_found_span: Option<Span>,
+    last_delim_empty_block_spans: FxHashMap<token::DelimToken, Span>
+}
+
+impl<'a> TokenTreesReader<'a> {
+    // Parse a stream of tokens into a list of `TokenTree`s, up to an `Eof`.
+    fn parse_all_token_trees(&mut self) -> PResult<'a, TokenStream> {
+        let mut buf = TokenStreamBuilder::default();
+
+        self.real_token();
+        while self.token != token::Eof {
+            buf.push(self.parse_token_tree()?);
+        }
+
+        Ok(buf.into_token_stream())
+    }
+
+    // Parse a stream of tokens into a list of `TokenTree`s, up to a `CloseDelim`.
+    fn parse_token_trees_until_close_delim(&mut self) -> TokenStream {
+        let mut buf = TokenStreamBuilder::default();
+        loop {
+            if let token::CloseDelim(..) = self.token.kind {
+                return buf.into_token_stream();
+            }
+
+            match self.parse_token_tree() {
+                Ok(tree) => buf.push(tree),
+                Err(mut e) => {
+                    e.emit();
+                    return buf.into_token_stream();
+                }
+            }
+        }
+    }
+
+    fn parse_token_tree(&mut self) -> PResult<'a, TreeAndJoint> {
+        let sm = self.string_reader.sess.source_map();
+        match self.token.kind {
+            token::Eof => {
+                let msg = "this file contains an un-closed delimiter";
+                let mut err = self.string_reader.sess.span_diagnostic
+                    .struct_span_err(self.token.span, msg);
+                for &(_, sp) in &self.open_braces {
+                    err.span_label(sp, "un-closed delimiter");
+                    self.unmatched_braces.push(UnmatchedBrace {
+                        expected_delim: token::DelimToken::Brace,
+                        found_delim: None,
+                        found_span: self.token.span,
+                        unclosed_span: Some(sp),
+                        candidate_span: None,
+                    });
+                }
+
+                if let Some((delim, _)) = self.open_braces.last() {
+                    if let Some((_, open_sp, close_sp)) = self.matching_delim_spans.iter()
+                        .filter(|(d, open_sp, close_sp)| {
+                            if let Some(close_padding) = sm.span_to_margin(*close_sp) {
+                                if let Some(open_padding) = sm.span_to_margin(*open_sp) {
+                                    return delim == d && close_padding != open_padding;
+                                }
+                            }
+                            false
+                        }).next()  // these are in reverse order as they get inserted on close, but
+                    {              // we want the last open/first close
+                        err.span_label(
+                            *open_sp,
+                            "this delimiter might not be properly closed...",
+                        );
+                        err.span_label(
+                            *close_sp,
+                            "...as it matches this but it has different indentation",
+                        );
+                    }
+                }
+                Err(err)
+            },
+            token::OpenDelim(delim) => {
+                // The span for beginning of the delimited section
+                let pre_span = self.token.span;
+
+                // Parse the open delimiter.
+                self.open_braces.push((delim, self.token.span));
+                self.real_token();
+
+                // Parse the token trees within the delimiters.
+                // We stop at any delimiter so we can try to recover if the user
+                // uses an incorrect delimiter.
+                let tts = self.parse_token_trees_until_close_delim();
+
+                // Expand to cover the entire delimited token tree
+                let delim_span = DelimSpan::from_pair(pre_span, self.token.span);
+
+                match self.token.kind {
+                    // Correct delimiter.
+                    token::CloseDelim(d) if d == delim => {
+                        let (open_brace, open_brace_span) = self.open_braces.pop().unwrap();
+                        let close_brace_span = self.token.span;
+
+                        if tts.is_empty() {
+                            let empty_block_span = open_brace_span.to(close_brace_span);
+                            self.last_delim_empty_block_spans.insert(delim, empty_block_span);
+                        }
+
+                        if self.open_braces.len() == 0 {
+                            // Clear up these spans to avoid suggesting them as we've found
+                            // properly matched delimiters so far for an entire block.
+                            self.matching_delim_spans.clear();
+                        } else {
+                            self.matching_delim_spans.push(
+                                (open_brace, open_brace_span, close_brace_span),
+                            );
+                        }
+                        // Parse the close delimiter.
+                        self.real_token();
+                    }
+                    // Incorrect delimiter.
+                    token::CloseDelim(other) => {
+                        let mut unclosed_delimiter = None;
+                        let mut candidate = None;
+                        if self.last_unclosed_found_span != Some(self.token.span) {
+                            // do not complain about the same unclosed delimiter multiple times
+                            self.last_unclosed_found_span = Some(self.token.span);
+                            // This is a conservative error: only report the last unclosed
+                            // delimiter. The previous unclosed delimiters could actually be
+                            // closed! The parser just hasn't gotten to them yet.
+                            if let Some(&(_, sp)) = self.open_braces.last() {
+                                unclosed_delimiter = Some(sp);
+                            };
+                            if let Some(current_padding) = sm.span_to_margin(self.token.span) {
+                                for (brace, brace_span) in &self.open_braces {
+                                    if let Some(padding) = sm.span_to_margin(*brace_span) {
+                                        // high likelihood of these two corresponding
+                                        if current_padding == padding && brace == &other {
+                                            candidate = Some(*brace_span);
+                                        }
+                                    }
+                                }
+                            }
+                            let (tok, _) = self.open_braces.pop().unwrap();
+                            self.unmatched_braces.push(UnmatchedBrace {
+                                expected_delim: tok,
+                                found_delim: Some(other),
+                                found_span: self.token.span,
+                                unclosed_span: unclosed_delimiter,
+                                candidate_span: candidate,
+                            });
+                        } else {
+                            self.open_braces.pop();
+                        }
+
+                        // If the incorrect delimiter matches an earlier opening
+                        // delimiter, then don't consume it (it can be used to
+                        // close the earlier one). Otherwise, consume it.
+                        // E.g., we try to recover from:
+                        // fn foo() {
+                        //     bar(baz(
+                        // }  // Incorrect delimiter but matches the earlier `{`
+                        if !self.open_braces.iter().any(|&(b, _)| b == other) {
+                            self.real_token();
+                        }
+                    }
+                    token::Eof => {
+                        // Silently recover, the EOF token will be seen again
+                        // and an error emitted then. Thus we don't pop from
+                        // self.open_braces here.
+                    },
+                    _ => {}
+                }
+
+                Ok(TokenTree::Delimited(
+                    delim_span,
+                    delim,
+                    tts.into()
+                ).into())
+            },
+            token::CloseDelim(delim) => {
+                // An unexpected closing delimiter (i.e., there is no
+                // matching opening delimiter).
+                let token_str = token_to_string(&self.token);
+                let msg = format!("unexpected close delimiter: `{}`", token_str);
+                let mut err = self.string_reader.sess.span_diagnostic
+                    .struct_span_err(self.token.span, &msg);
+
+                if let Some(span) = self.last_delim_empty_block_spans.remove(&delim) {
+                    err.span_label(
+                        span,
+                        "this block is empty, you might have not meant to close it"
+                    );
+                }
+                err.span_label(self.token.span, "unexpected close delimiter");
+                Err(err)
+            },
+            _ => {
+                let tt = TokenTree::Token(self.token.take());
+                self.real_token();
+                let is_joint = self.joint_to_prev == Joint && self.token.is_op();
+                Ok((tt, if is_joint { Joint } else { NonJoint }))
+            }
+        }
+    }
+
+    fn real_token(&mut self) {
+        self.joint_to_prev = Joint;
+        loop {
+            let token = self.string_reader.next_token();
+            match token.kind {
+                token::Whitespace | token::Comment | token::Shebang(_) | token::Unknown(_) => {
+                    self.joint_to_prev = NonJoint;
+                }
+                _ => {
+                    self.token = token;
+                    return;
+                }
+            }
+        }
+    }
+}
+
+#[derive(Default)]
+struct TokenStreamBuilder {
+    buf: Vec<TreeAndJoint>,
+}
+
+impl TokenStreamBuilder {
+    fn push(&mut self, (tree, joint): TreeAndJoint) {
+        if let Some((TokenTree::Token(prev_token), Joint)) = self.buf.last() {
+            if let TokenTree::Token(token) = &tree {
+                if let Some(glued) = prev_token.glue(token) {
+                    self.buf.pop();
+                    self.buf.push((TokenTree::Token(glued), joint));
+                    return;
+                }
+            }
+        }
+        self.buf.push((tree, joint))
+    }
+
+    fn into_token_stream(self) -> TokenStream {
+        TokenStream::new(self.buf)
+    }
+}

src/librustc_parse/lexer/unescape_error_reporting.rs

@@ -0,0 +1,215 @@
+//! Utilities for rendering escape sequence errors as diagnostics.
+
+use std::ops::Range;
+use std::iter::once;
+
+use rustc_lexer::unescape::{EscapeError, Mode};
+use syntax_pos::{Span, BytePos};
+
+use syntax::errors::{Handler, Applicability};
+
+pub(crate) fn emit_unescape_error(
+    handler: &Handler,
+    // interior part of the literal, without quotes
+    lit: &str,
+    // full span of the literal, including quotes
+    span_with_quotes: Span,
+    mode: Mode,
+    // range of the error inside `lit`
+    range: Range<usize>,
+    error: EscapeError,
+) {
+    log::debug!("emit_unescape_error: {:?}, {:?}, {:?}, {:?}, {:?}",
+                lit, span_with_quotes, mode, range, error);
+    let span = {
+        let Range { start, end } = range;
+        let (start, end) = (start as u32, end as u32);
+        let lo = span_with_quotes.lo() + BytePos(start + 1);
+        let hi = lo + BytePos(end - start);
+            span_with_quotes
+            .with_lo(lo)
+            .with_hi(hi)
+    };
+    let last_char = || {
+        let c = lit[range.clone()].chars().rev().next().unwrap();
+        let span = span.with_lo(span.hi() - BytePos(c.len_utf8() as u32));
+        (c, span)
+    };
+    match error {
+        EscapeError::LoneSurrogateUnicodeEscape => {
+            handler.struct_span_err(span, "invalid unicode character escape")
+                .help("unicode escape must not be a surrogate")
+                .emit();
+        }
+        EscapeError::OutOfRangeUnicodeEscape => {
+            handler.struct_span_err(span, "invalid unicode character escape")
+                .help("unicode escape must be at most 10FFFF")
+                .emit();
+        }
+        EscapeError::MoreThanOneChar => {
+            let msg = if mode.is_bytes() {
+                "if you meant to write a byte string literal, use double quotes"
+            } else {
+                "if you meant to write a `str` literal, use double quotes"
+            };
+
+            handler
+                .struct_span_err(
+                    span_with_quotes,
+                    "character literal may only contain one codepoint",
+                )
+                .span_suggestion(
+                    span_with_quotes,
+                    msg,
+                    format!("\"{}\"", lit),
+                    Applicability::MachineApplicable,
+                ).emit()
+        }
+        EscapeError::EscapeOnlyChar => {
+            let (c, _span) = last_char();
+
+            let mut msg = if mode.is_bytes() {
+                "byte constant must be escaped: "
+            } else {
+                "character constant must be escaped: "
+            }.to_string();
+            push_escaped_char(&mut msg, c);
+
+            handler.span_err(span, msg.as_str())
+        }
+        EscapeError::BareCarriageReturn => {
+            let msg = if mode.in_double_quotes() {
+                "bare CR not allowed in string, use \\r instead"
+            } else {
+                "character constant must be escaped: \\r"
+            };
+            handler.span_err(span, msg);
+        }
+        EscapeError::BareCarriageReturnInRawString => {
+            assert!(mode.in_double_quotes());
+            let msg = "bare CR not allowed in raw string";
+            handler.span_err(span, msg);
+        }
+        EscapeError::InvalidEscape => {
+            let (c, span) = last_char();
+
+            let label = if mode.is_bytes() {
+                "unknown byte escape"
+            } else {
+                "unknown character escape"
+            };
+            let mut msg = label.to_string();
+            msg.push_str(": ");
+            push_escaped_char(&mut msg, c);
+
+            let mut diag = handler.struct_span_err(span, msg.as_str());
+            diag.span_label(span, label);
+            if c == '{' || c == '}' && !mode.is_bytes() {
+                diag.help("if used in a formatting string, \
+                           curly braces are escaped with `{{` and `}}`");
+            } else if c == '\r' {
+                diag.help("this is an isolated carriage return; \
+                           consider checking your editor and version control settings");
+            }
+            diag.emit();
+        }
+        EscapeError::TooShortHexEscape => {
+            handler.span_err(span, "numeric character escape is too short")
+        }
+        EscapeError::InvalidCharInHexEscape | EscapeError::InvalidCharInUnicodeEscape => {
+            let (c, span) = last_char();
+
+            let mut msg = if error == EscapeError::InvalidCharInHexEscape {
+                "invalid character in numeric character escape: "
+            } else {
+                "invalid character in unicode escape: "
+            }.to_string();
+            push_escaped_char(&mut msg, c);
+
+            handler.span_err(span, msg.as_str())
+        }
+        EscapeError::NonAsciiCharInByte => {
+            assert!(mode.is_bytes());
+            let (_c, span) = last_char();
+            handler.span_err(span, "byte constant must be ASCII. \
+                                    Use a \\xHH escape for a non-ASCII byte")
+        }
+        EscapeError::NonAsciiCharInByteString => {
+            assert!(mode.is_bytes());
+            let (_c, span) = last_char();
+            handler.span_err(span, "raw byte string must be ASCII")
+        }
+        EscapeError::OutOfRangeHexEscape => {
+            handler.span_err(span, "this form of character escape may only be used \
+                                    with characters in the range [\\x00-\\x7f]")
+        }
+        EscapeError::LeadingUnderscoreUnicodeEscape => {
+            let (_c, span) = last_char();
+            handler.span_err(span, "invalid start of unicode escape")
+        }
+        EscapeError::OverlongUnicodeEscape => {
+            handler.span_err(span, "overlong unicode escape (must have at most 6 hex digits)")
+        }
+        EscapeError::UnclosedUnicodeEscape => {
+            handler.span_err(span, "unterminated unicode escape (needed a `}`)")
+        }
+        EscapeError::NoBraceInUnicodeEscape => {
+            let msg = "incorrect unicode escape sequence";
+            let mut diag = handler.struct_span_err(span, msg);
+
+            let mut suggestion = "\\u{".to_owned();
+            let mut suggestion_len = 0;
+            let (c, char_span) = last_char();
+            let chars = once(c).chain(lit[range.end..].chars());
+            for c in chars.take(6).take_while(|c| c.is_digit(16)) {
+                suggestion.push(c);
+                suggestion_len += c.len_utf8();
+            }
+
+            if suggestion_len > 0 {
+                suggestion.push('}');
+                let lo = char_span.lo();
+                let hi = lo + BytePos(suggestion_len as u32);
+                diag.span_suggestion(
+                    span.with_lo(lo).with_hi(hi),
+                    "format of unicode escape sequences uses braces",
+                    suggestion,
+                    Applicability::MaybeIncorrect,
+                );
+            } else {
+                diag.span_label(span, msg);
+                diag.help(
+                    "format of unicode escape sequences is `\\u{...}`",
+                );
+            }
+
+            diag.emit();
+        }
+        EscapeError::UnicodeEscapeInByte => {
+            handler.span_err(span, "unicode escape sequences cannot be used \
+                                    as a byte or in a byte string")
+        }
+        EscapeError::EmptyUnicodeEscape => {
+            handler.span_err(span, "empty unicode escape (must have at least 1 hex digit)")
+        }
+        EscapeError::ZeroChars => {
+            handler.span_err(span, "empty character literal")
+        }
+        EscapeError::LoneSlash => {
+            handler.span_err(span, "invalid trailing slash in literal")
+        }
+    }
+}
+
+/// Pushes a character to a message string for error reporting
+pub(crate) fn push_escaped_char(msg: &mut String, c: char) {
+    match c {
+        '\u{20}'..='\u{7e}' => {
+            // Don't escape \, ' or " for user-facing messages
+            msg.push(c);
+        }
+        _ => {
+            msg.extend(c.escape_default());
+        }
+    }
+}

src/librustc_parse/lexer/unicode_chars.rs

@@ -0,0 +1,392 @@
+// Characters and their corresponding confusables were collected from
+// http://www.unicode.org/Public/security/10.0.0/confusables.txt
+
+use super::StringReader;
+use errors::{Applicability, DiagnosticBuilder};
+use syntax_pos::{BytePos, Pos, Span, symbol::kw};
+use crate::token;
+
+#[rustfmt::skip] // for line breaks
+const UNICODE_ARRAY: &[(char, &str, char)] = &[
+    ('
', "Line Separator", ' '),
+    ('
', "Paragraph Separator", ' '),
+    (' ', "Ogham Space mark", ' '),
+    (' ', "En Quad", ' '),
+    (' ', "Em Quad", ' '),
+    (' ', "En Space", ' '),
+    (' ', "Em Space", ' '),
+    (' ', "Three-Per-Em Space", ' '),
+    (' ', "Four-Per-Em Space", ' '),
+    (' ', "Six-Per-Em Space", ' '),
+    (' ', "Punctuation Space", ' '),
+    (' ', "Thin Space", ' '),
+    (' ', "Hair Space", ' '),
+    (' ', "Medium Mathematical Space", ' '),
+    (' ', "No-Break Space", ' '),
+    (' ', "Figure Space", ' '),
+    (' ', "Narrow No-Break Space", ' '),
+    ('　', "Ideographic Space", ' '),
+
+    ('ߺ', "Nko Lajanyalan", '_'),
+    ('﹍', "Dashed Low Line", '_'),
+    ('﹎', "Centreline Low Line", '_'),
+    ('﹏', "Wavy Low Line", '_'),
+    ('＿', "Fullwidth Low Line", '_'),
+
+    ('‐', "Hyphen", '-'),
+    ('‑', "Non-Breaking Hyphen", '-'),
+    ('‒', "Figure Dash", '-'),
+    ('–', "En Dash", '-'),
+    ('—', "Em Dash", '-'),
+    ('﹘', "Small Em Dash", '-'),
+    ('۔', "Arabic Full Stop", '-'),
+    ('⁃', "Hyphen Bullet", '-'),
+    ('˗', "Modifier Letter Minus Sign", '-'),
+    ('−', "Minus Sign", '-'),
+    ('➖', "Heavy Minus Sign", '-'),
+    ('Ⲻ', "Coptic Letter Dialect-P Ni", '-'),
+    ('ー', "Katakana-Hiragana Prolonged Sound Mark", '-'),
+    ('－', "Fullwidth Hyphen-Minus", '-'),
+    ('―', "Horizontal Bar", '-'),
+    ('─', "Box Drawings Light Horizontal", '-'),
+    ('━', "Box Drawings Heavy Horizontal", '-'),
+    ('㇐', "CJK Stroke H", '-'),
+    ('ꟷ', "Latin Epigraphic Letter Sideways I", '-'),
+    ('ᅳ', "Hangul Jungseong Eu", '-'),
+    ('ㅡ', "Hangul Letter Eu", '-'),
+    ('一', "CJK Unified Ideograph-4E00", '-'),
+    ('⼀', "Kangxi Radical One", '-'),
+
+    ('؍', "Arabic Date Separator", ','),
+    ('٫', "Arabic Decimal Separator", ','),
+    ('‚', "Single Low-9 Quotation Mark", ','),
+    ('¸', "Cedilla", ','),
+    ('ꓹ', "Lisu Letter Tone Na Po", ','),
+    ('，', "Fullwidth Comma", ','),
+
+    (';', "Greek Question Mark", ';'),
+    ('；', "Fullwidth Semicolon", ';'),
+    ('︔', "Presentation Form For Vertical Semicolon", ';'),
+
+    ('ः', "Devanagari Sign Visarga", ':'),
+    ('ઃ', "Gujarati Sign Visarga", ':'),
+    ('：', "Fullwidth Colon", ':'),
+    ('։', "Armenian Full Stop", ':'),
+    ('܃', "Syriac Supralinear Colon", ':'),
+    ('܄', "Syriac Sublinear Colon", ':'),
+    ('᛬', "Runic Multiple Punctuation", ':'),
+    ('︰', "Presentation Form For Vertical Two Dot Leader", ':'),
+    ('᠃', "Mongolian Full Stop", ':'),
+    ('᠉', "Mongolian Manchu Full Stop", ':'),
+    ('⁚', "Two Dot Punctuation", ':'),
+    ('׃', "Hebrew Punctuation Sof Pasuq", ':'),
+    ('˸', "Modifier Letter Raised Colon", ':'),
+    ('꞉', "Modifier Letter Colon", ':'),
+    ('∶', "Ratio", ':'),
+    ('ː', "Modifier Letter Triangular Colon", ':'),
+    ('ꓽ', "Lisu Letter Tone Mya Jeu", ':'),
+    ('︓', "Presentation Form For Vertical Colon", ':'),
+
+    ('！', "Fullwidth Exclamation Mark", '!'),
+    ('ǃ', "Latin Letter Retroflex Click", '!'),
+    ('ⵑ', "Tifinagh Letter Tuareg Yang", '!'),
+    ('︕', "Presentation Form For Vertical Exclamation Mark", '!'),
+
+    ('ʔ', "Latin Letter Glottal Stop", '?'),
+    ('Ɂ', "Latin Capital Letter Glottal Stop", '?'),
+    ('ॽ', "Devanagari Letter Glottal Stop", '?'),
+    ('Ꭾ', "Cherokee Letter He", '?'),
+    ('ꛫ', "Bamum Letter Ntuu", '?'),
+    ('？', "Fullwidth Question Mark", '?'),
+    ('︖', "Presentation Form For Vertical Question Mark", '?'),
+
+    ('𝅭', "Musical Symbol Combining Augmentation Dot", '.'),
+    ('․', "One Dot Leader", '.'),
+    ('܁', "Syriac Supralinear Full Stop", '.'),
+    ('܂', "Syriac Sublinear Full Stop", '.'),
+    ('꘎', "Vai Full Stop", '.'),
+    ('𐩐', "Kharoshthi Punctuation Dot", '.'),
+    ('٠', "Arabic-Indic Digit Zero", '.'),
+    ('۰', "Extended Arabic-Indic Digit Zero", '.'),
+    ('ꓸ', "Lisu Letter Tone Mya Ti", '.'),
+    ('·', "Middle Dot", '.'),
+    ('・', "Katakana Middle Dot", '.'),
+    ('･', "Halfwidth Katakana Middle Dot", '.'),
+    ('᛫', "Runic Single Punctuation", '.'),
+    ('·', "Greek Ano Teleia", '.'),
+    ('⸱', "Word Separator Middle Dot", '.'),
+    ('𐄁', "Aegean Word Separator Dot", '.'),
+    ('•', "Bullet", '.'),
+    ('‧', "Hyphenation Point", '.'),
+    ('∙', "Bullet Operator", '.'),
+    ('⋅', "Dot Operator", '.'),
+    ('ꞏ', "Latin Letter Sinological Dot", '.'),
+    ('ᐧ', "Canadian Syllabics Final Middle Dot", '.'),
+    ('ᐧ', "Canadian Syllabics Final Middle Dot", '.'),
+    ('．', "Fullwidth Full Stop", '.'),
+    ('。', "Ideographic Full Stop", '.'),
+    ('︒', "Presentation Form For Vertical Ideographic Full Stop", '.'),
+
+    ('՝', "Armenian Comma", '\''),
+    ('＇', "Fullwidth Apostrophe", '\''),
+    ('‘', "Left Single Quotation Mark", '\''),
+    ('’', "Right Single Quotation Mark", '\''),
+    ('‛', "Single High-Reversed-9 Quotation Mark", '\''),
+    ('′', "Prime", '\''),
+    ('‵', "Reversed Prime", '\''),
+    ('՚', "Armenian Apostrophe", '\''),
+    ('׳', "Hebrew Punctuation Geresh", '\''),
+    ('`', "Grave Accent", '\''),
+    ('`', "Greek Varia", '\''),
+    ('｀', "Fullwidth Grave Accent", '\''),
+    ('´', "Acute Accent", '\''),
+    ('΄', "Greek Tonos", '\''),
+    ('´', "Greek Oxia", '\''),
+    ('᾽', "Greek Koronis", '\''),
+    ('᾿', "Greek Psili", '\''),
+    ('῾', "Greek Dasia", '\''),
+    ('ʹ', "Modifier Letter Prime", '\''),
+    ('ʹ', "Greek Numeral Sign", '\''),
+    ('ˈ', "Modifier Letter Vertical Line", '\''),
+    ('ˊ', "Modifier Letter Acute Accent", '\''),
+    ('ˋ', "Modifier Letter Grave Accent", '\''),
+    ('˴', "Modifier Letter Middle Grave Accent", '\''),
+    ('ʻ', "Modifier Letter Turned Comma", '\''),
+    ('ʽ', "Modifier Letter Reversed Comma", '\''),
+    ('ʼ', "Modifier Letter Apostrophe", '\''),
+    ('ʾ', "Modifier Letter Right Half Ring", '\''),
+    ('ꞌ', "Latin Small Letter Saltillo", '\''),
+    ('י', "Hebrew Letter Yod", '\''),
+    ('ߴ', "Nko High Tone Apostrophe", '\''),
+    ('ߵ', "Nko Low Tone Apostrophe", '\''),
+    ('ᑊ', "Canadian Syllabics West-Cree P", '\''),
+    ('ᛌ', "Runic Letter Short-Twig-Sol S", '\''),
+    ('𖽑', "Miao Sign Aspiration", '\''),
+    ('𖽒', "Miao Sign Reformed Voicing", '\''),
+
+    ('᳓', "Vedic Sign Nihshvasa", '"'),
+    ('＂', "Fullwidth Quotation Mark", '"'),
+    ('“', "Left Double Quotation Mark", '"'),
+    ('”', "Right Double Quotation Mark", '"'),
+    ('‟', "Double High-Reversed-9 Quotation Mark", '"'),
+    ('″', "Double Prime", '"'),
+    ('‶', "Reversed Double Prime", '"'),
+    ('〃', "Ditto Mark", '"'),
+    ('״', "Hebrew Punctuation Gershayim", '"'),
+    ('˝', "Double Acute Accent", '"'),
+    ('ʺ', "Modifier Letter Double Prime", '"'),
+    ('˶', "Modifier Letter Middle Double Acute Accent", '"'),
+    ('˵', "Modifier Letter Middle Double Grave Accent", '"'),
+    ('ˮ', "Modifier Letter Double Apostrophe", '"'),
+    ('ײ', "Hebrew Ligature Yiddish Double Yod", '"'),
+    ('❞', "Heavy Double Comma Quotation Mark Ornament", '"'),
+    ('❝', "Heavy Double Turned Comma Quotation Mark Ornament", '"'),
+
+    ('（', "Fullwidth Left Parenthesis", '('),
+    ('❨', "Medium Left Parenthesis Ornament", '('),
+    ('﴾', "Ornate Left Parenthesis", '('),
+
+    ('）', "Fullwidth Right Parenthesis", ')'),
+    ('❩', "Medium Right Parenthesis Ornament", ')'),
+    ('﴿', "Ornate Right Parenthesis", ')'),
+
+    ('［', "Fullwidth Left Square Bracket", '['),
+    ('❲', "Light Left Tortoise Shell Bracket Ornament", '['),
+    ('「', "Left Corner Bracket", '['),
+    ('『', "Left White Corner Bracket", '['),
+    ('【', "Left Black Lenticular Bracket", '['),
+    ('〔', "Left Tortoise Shell Bracket", '['),
+    ('〖', "Left White Lenticular Bracket", '['),
+    ('〘', "Left White Tortoise Shell Bracket", '['),
+    ('〚', "Left White Square Bracket", '['),
+
+    ('］', "Fullwidth Right Square Bracket", ']'),
+    ('❳', "Light Right Tortoise Shell Bracket Ornament", ']'),
+    ('」', "Right Corner Bracket", ']'),
+    ('』', "Right White Corner Bracket", ']'),
+    ('】', "Right Black Lenticular Bracket", ']'),
+    ('〕', "Right Tortoise Shell Bracket", ']'),
+    ('〗', "Right White Lenticular Bracket", ']'),
+    ('〙', "Right White Tortoise Shell Bracket", ']'),
+    ('〛', "Right White Square Bracket", ']'),
+
+    ('❴', "Medium Left Curly Bracket Ornament", '{'),
+    ('𝄔', "Musical Symbol Brace", '{'),
+    ('｛', "Fullwidth Left Curly Bracket", '{'),
+
+    ('❵', "Medium Right Curly Bracket Ornament", '}'),
+    ('｝', "Fullwidth Right Curly Bracket", '}'),
+
+    ('⁎', "Low Asterisk", '*'),
+    ('٭', "Arabic Five Pointed Star", '*'),
+    ('∗', "Asterisk Operator", '*'),
+    ('𐌟', "Old Italic Letter Ess", '*'),
+    ('＊', "Fullwidth Asterisk", '*'),
+
+    ('᜵', "Philippine Single Punctuation", '/'),
+    ('⁁', "Caret Insertion Point", '/'),
+    ('∕', "Division Slash", '/'),
+    ('⁄', "Fraction Slash", '/'),
+    ('╱', "Box Drawings Light Diagonal Upper Right To Lower Left", '/'),
+    ('⟋', "Mathematical Rising Diagonal", '/'),
+    ('⧸', "Big Solidus", '/'),
+    ('𝈺', "Greek Instrumental Notation Symbol-47", '/'),
+    ('㇓', "CJK Stroke Sp", '/'),
+    ('〳', "Vertical Kana Repeat Mark Upper Half", '/'),
+    ('Ⳇ', "Coptic Capital Letter Old Coptic Esh", '/'),
+    ('ノ', "Katakana Letter No", '/'),
+    ('丿', "CJK Unified Ideograph-4E3F", '/'),
+    ('⼃', "Kangxi Radical Slash", '/'),
+    ('／', "Fullwidth Solidus", '/'),
+
+    ('＼', "Fullwidth Reverse Solidus", '\\'),
+    ('﹨', "Small Reverse Solidus", '\\'),
+    ('∖', "Set Minus", '\\'),
+    ('⟍', "Mathematical Falling Diagonal", '\\'),
+    ('⧵', "Reverse Solidus Operator", '\\'),
+    ('⧹', "Big Reverse Solidus", '\\'),
+    ('⧹', "Greek Vocal Notation Symbol-16", '\\'),
+    ('⧹', "Greek Instrumental Symbol-48", '\\'),
+    ('㇔', "CJK Stroke D", '\\'),
+    ('丶', "CJK Unified Ideograph-4E36", '\\'),
+    ('⼂', "Kangxi Radical Dot", '\\'),
+    ('、', "Ideographic Comma", '\\'),
+    ('ヽ', "Katakana Iteration Mark", '\\'),
+
+    ('ꝸ', "Latin Small Letter Um", '&'),
+    ('＆', "Fullwidth Ampersand", '&'),
+
+    ('᛭', "Runic Cross Punctuation", '+'),
+    ('➕', "Heavy Plus Sign", '+'),
+    ('𐊛', "Lycian Letter H", '+'),
+    ('﬩', "Hebrew Letter Alternative Plus Sign", '+'),
+    ('＋', "Fullwidth Plus Sign", '+'),
+
+    ('‹', "Single Left-Pointing Angle Quotation Mark", '<'),
+    ('❮', "Heavy Left-Pointing Angle Quotation Mark Ornament", '<'),
+    ('˂', "Modifier Letter Left Arrowhead", '<'),
+    ('𝈶', "Greek Instrumental Symbol-40", '<'),
+    ('ᐸ', "Canadian Syllabics Pa", '<'),
+    ('ᚲ', "Runic Letter Kauna", '<'),
+    ('❬', "Medium Left-Pointing Angle Bracket Ornament", '<'),
+    ('⟨', "Mathematical Left Angle Bracket", '<'),
+    ('〈', "Left-Pointing Angle Bracket", '<'),
+    ('〈', "Left Angle Bracket", '<'),
+    ('㇛', "CJK Stroke Pd", '<'),
+    ('く', "Hiragana Letter Ku", '<'),
+    ('𡿨', "CJK Unified Ideograph-21FE8", '<'),
+    ('《', "Left Double Angle Bracket", '<'),
+    ('＜', "Fullwidth Less-Than Sign", '<'),
+
+    ('᐀', "Canadian Syllabics Hyphen", '='),
+    ('⹀', "Double Hyphen", '='),
+    ('゠', "Katakana-Hiragana Double Hyphen", '='),
+    ('꓿', "Lisu Punctuation Full Stop", '='),
+    ('＝', "Fullwidth Equals Sign", '='),
+
+    ('›', "Single Right-Pointing Angle Quotation Mark", '>'),
+    ('❯', "Heavy Right-Pointing Angle Quotation Mark Ornament", '>'),
+    ('˃', "Modifier Letter Right Arrowhead", '>'),
+    ('𝈷', "Greek Instrumental Symbol-42", '>'),
+    ('ᐳ', "Canadian Syllabics Po", '>'),
+    ('𖼿', "Miao Letter Archaic Zza", '>'),
+    ('❭', "Medium Right-Pointing Angle Bracket Ornament", '>'),
+    ('⟩', "Mathematical Right Angle Bracket", '>'),
+    ('〉', "Right-Pointing Angle Bracket", '>'),
+    ('〉', "Right Angle Bracket", '>'),
+    ('》', "Right Double Angle Bracket", '>'),
+    ('＞', "Fullwidth Greater-Than Sign", '>'),
+];
+
+// FIXME: the lexer could be used to turn the ASCII version of unicode homoglyphs, instead of
+// keeping the substitution token in this table. Ideally, this should be inside `rustc_lexer`.
+// However, we should first remove compound tokens like `<<` from `rustc_lexer`, and then add
+// fancier error recovery to it, as there will be less overall work to do this way.
+const ASCII_ARRAY: &[(char, &str, Option<token::TokenKind>)] = &[
+    (' ', "Space", Some(token::Whitespace)),
+    ('_', "Underscore", Some(token::Ident(kw::Underscore, false))),
+    ('-', "Minus/Hyphen", Some(token::BinOp(token::Minus))),
+    (',', "Comma", Some(token::Comma)),
+    (';', "Semicolon", Some(token::Semi)),
+    (':', "Colon", Some(token::Colon)),
+    ('!', "Exclamation Mark", Some(token::Not)),
+    ('?', "Question Mark", Some(token::Question)),
+    ('.', "Period", Some(token::Dot)),
+    ('(', "Left Parenthesis", Some(token::OpenDelim(token::Paren))),
+    (')', "Right Parenthesis", Some(token::CloseDelim(token::Paren))),
+    ('[', "Left Square Bracket", Some(token::OpenDelim(token::Bracket))),
+    (']', "Right Square Bracket", Some(token::CloseDelim(token::Bracket))),
+    ('{', "Left Curly Brace", Some(token::OpenDelim(token::Brace))),
+    ('}', "Right Curly Brace", Some(token::CloseDelim(token::Brace))),
+    ('*', "Asterisk", Some(token::BinOp(token::Star))),
+    ('/', "Slash", Some(token::BinOp(token::Slash))),
+    ('\\', "Backslash", None),
+    ('&', "Ampersand", Some(token::BinOp(token::And))),
+    ('+', "Plus Sign", Some(token::BinOp(token::Plus))),
+    ('<', "Less-Than Sign", Some(token::Lt)),
+    ('=', "Equals Sign", Some(token::Eq)),
+    ('>', "Greater-Than Sign", Some(token::Gt)),
+    // FIXME: Literals are already lexed by this point, so we can't recover gracefully just by
+    // spitting the correct token out.
+    ('\'', "Single Quote", None),
+    ('"', "Quotation Mark", None),
+];
+
+crate fn check_for_substitution<'a>(
+    reader: &StringReader<'a>,
+    pos: BytePos,
+    ch: char,
+    err: &mut DiagnosticBuilder<'a>,
+) -> Option<token::TokenKind> {
+    let (u_name, ascii_char) = match UNICODE_ARRAY.iter().find(|&&(c, _, _)| c == ch) {
+        Some(&(_u_char, u_name, ascii_char)) => (u_name, ascii_char),
+        None => return None,
+    };
+
+    let span = Span::with_root_ctxt(pos, pos + Pos::from_usize(ch.len_utf8()));
+
+    let (ascii_name, token) = match ASCII_ARRAY.iter().find(|&&(c, _, _)| c == ascii_char) {
+        Some((_ascii_char, ascii_name, token)) => (ascii_name, token),
+        None => {
+            let msg = format!("substitution character not found for '{}'", ch);
+            reader.sess.span_diagnostic.span_bug_no_panic(span, &msg);
+            return None;
+        }
+    };
+
+    // special help suggestion for "directed" double quotes
+    if let Some(s) = peek_delimited(&reader.src[reader.src_index(pos)..], '“', '”') {
+        let msg = format!(
+            "Unicode characters '“' (Left Double Quotation Mark) and \
+             '”' (Right Double Quotation Mark) look like '{}' ({}), but are not",
+            ascii_char, ascii_name
+        );
+        err.span_suggestion(
+            Span::with_root_ctxt(
+                pos,
+                pos + Pos::from_usize('“'.len_utf8() + s.len() + '”'.len_utf8()),
+            ),
+            &msg,
+            format!("\"{}\"", s),
+            Applicability::MaybeIncorrect,
+        );
+    } else {
+        let msg = format!(
+            "Unicode character '{}' ({}) looks like '{}' ({}), but it is not",
+            ch, u_name, ascii_char, ascii_name
+        );
+        err.span_suggestion(span, &msg, ascii_char.to_string(), Applicability::MaybeIncorrect);
+    }
+    token.clone()
+}
+
+/// Extract string if found at current position with given delimiters
+fn peek_delimited(text: &str, from_ch: char, to_ch: char) -> Option<&str> {
+    let mut chars = text.chars();
+    let first_char = chars.next()?;
+    if first_char != from_ch {
+        return None;
+    }
+    let last_char_idx = chars.as_str().find(to_ch)?;
+    Some(&chars.as_str()[..last_char_idx])
+}

src/librustc_parse/lib.rs

@@ -0,0 +1,423 @@
+//! The main parser interface.
+
+#![feature(crate_visibility_modifier)]
+
+use syntax::ast;
+use syntax::print::pprust;
+use syntax::sess::ParseSess;
+use syntax::token::{self, Nonterminal};
+use syntax::tokenstream::{self, TokenStream, TokenTree};
+
+use errors::{PResult, FatalError, Level, Diagnostic};
+use rustc_data_structures::sync::Lrc;
+use syntax_pos::{Span, SourceFile, FileName};
+
+use std::borrow::Cow;
+use std::path::Path;
+use std::str;
+
+use log::info;
+
+pub const MACRO_ARGUMENTS: Option<&'static str> = Some("macro arguments");
+
+#[macro_use]
+pub mod parser;
+use parser::{Parser, emit_unclosed_delims, make_unclosed_delims_error};
+pub mod lexer;
+pub mod validate_attr;
+pub mod error_codes;
+
+#[derive(Clone)]
+pub struct Directory<'a> {
+    pub path: Cow<'a, Path>,
+    pub ownership: DirectoryOwnership,
+}
+
+#[derive(Copy, Clone)]
+pub enum DirectoryOwnership {
+    Owned {
+        // None if `mod.rs`, `Some("foo")` if we're in `foo.rs`.
+        relative: Option<ast::Ident>,
+    },
+    UnownedViaBlock,
+    UnownedViaMod,
+}
+
+// A bunch of utility functions of the form `parse_<thing>_from_<source>`
+// where <thing> includes crate, expr, item, stmt, tts, and one that
+// uses a HOF to parse anything, and <source> includes file and
+// `source_str`.
+
+/// A variant of 'panictry!' that works on a Vec<Diagnostic> instead of a single DiagnosticBuilder.
+macro_rules! panictry_buffer {
+    ($handler:expr, $e:expr) => ({
+        use std::result::Result::{Ok, Err};
+        use errors::FatalError;
+        match $e {
+            Ok(e) => e,
+            Err(errs) => {
+                for e in errs {
+                    $handler.emit_diagnostic(&e);
+                }
+                FatalError.raise()
+            }
+        }
+    })
+}
+
+pub fn parse_crate_from_file<'a>(input: &Path, sess: &'a ParseSess) -> PResult<'a, ast::Crate> {
+    let mut parser = new_parser_from_file(sess, input);
+    parser.parse_crate_mod()
+}
+
+pub fn parse_crate_attrs_from_file<'a>(input: &Path, sess: &'a ParseSess)
+                                       -> PResult<'a, Vec<ast::Attribute>> {
+    let mut parser = new_parser_from_file(sess, input);
+    parser.parse_inner_attributes()
+}
+
+pub fn parse_crate_from_source_str(name: FileName, source: String, sess: &ParseSess)
+                                       -> PResult<'_, ast::Crate> {
+    new_parser_from_source_str(sess, name, source).parse_crate_mod()
+}
+
+pub fn parse_crate_attrs_from_source_str(name: FileName, source: String, sess: &ParseSess)
+                                             -> PResult<'_, Vec<ast::Attribute>> {
+    new_parser_from_source_str(sess, name, source).parse_inner_attributes()
+}
+
+pub fn parse_stream_from_source_str(
+    name: FileName,
+    source: String,
+    sess: &ParseSess,
+    override_span: Option<Span>,
+) -> TokenStream {
+    let (stream, mut errors) = source_file_to_stream(
+        sess,
+        sess.source_map().new_source_file(name, source),
+        override_span,
+    );
+    emit_unclosed_delims(&mut errors, &sess);
+    stream
+}
+
+/// Creates a new parser from a source string.
+pub fn new_parser_from_source_str(sess: &ParseSess, name: FileName, source: String) -> Parser<'_> {
+    panictry_buffer!(&sess.span_diagnostic, maybe_new_parser_from_source_str(sess, name, source))
+}
+
+/// Creates a new parser from a source string. Returns any buffered errors from lexing the initial
+/// token stream.
+pub fn maybe_new_parser_from_source_str(sess: &ParseSess, name: FileName, source: String)
+    -> Result<Parser<'_>, Vec<Diagnostic>>
+{
+    let mut parser = maybe_source_file_to_parser(sess,
+                                                 sess.source_map().new_source_file(name, source))?;
+    parser.recurse_into_file_modules = false;
+    Ok(parser)
+}
+
+/// Creates a new parser, handling errors as appropriate if the file doesn't exist.
+pub fn new_parser_from_file<'a>(sess: &'a ParseSess, path: &Path) -> Parser<'a> {
+    source_file_to_parser(sess, file_to_source_file(sess, path, None))
+}
+
+/// Creates a new parser, returning buffered diagnostics if the file doesn't exist,
+/// or from lexing the initial token stream.
+pub fn maybe_new_parser_from_file<'a>(sess: &'a ParseSess, path: &Path)
+    -> Result<Parser<'a>, Vec<Diagnostic>> {
+    let file = try_file_to_source_file(sess, path, None).map_err(|db| vec![db])?;
+    maybe_source_file_to_parser(sess, file)
+}
+
+/// Given a session, a crate config, a path, and a span, add
+/// the file at the given path to the `source_map`, and returns a parser.
+/// On an error, uses the given span as the source of the problem.
+pub fn new_sub_parser_from_file<'a>(sess: &'a ParseSess,
+                                    path: &Path,
+                                    directory_ownership: DirectoryOwnership,
+                                    module_name: Option<String>,
+                                    sp: Span) -> Parser<'a> {
+    let mut p = source_file_to_parser(sess, file_to_source_file(sess, path, Some(sp)));
+    p.directory.ownership = directory_ownership;
+    p.root_module_name = module_name;
+    p
+}
+
+/// Given a `source_file` and config, returns a parser.
+fn source_file_to_parser(sess: &ParseSess, source_file: Lrc<SourceFile>) -> Parser<'_> {
+    panictry_buffer!(&sess.span_diagnostic,
+                     maybe_source_file_to_parser(sess, source_file))
+}
+
+/// Given a `source_file` and config, return a parser. Returns any buffered errors from lexing the
+/// initial token stream.
+fn maybe_source_file_to_parser(
+    sess: &ParseSess,
+    source_file: Lrc<SourceFile>,
+) -> Result<Parser<'_>, Vec<Diagnostic>> {
+    let end_pos = source_file.end_pos;
+    let (stream, unclosed_delims) = maybe_file_to_stream(sess, source_file, None)?;
+    let mut parser = stream_to_parser(sess, stream, None);
+    parser.unclosed_delims = unclosed_delims;
+    if parser.token == token::Eof && parser.token.span.is_dummy() {
+        parser.token.span = Span::new(end_pos, end_pos, parser.token.span.ctxt());
+    }
+
+    Ok(parser)
+}
+
+// Must preserve old name for now, because `quote!` from the *existing*
+// compiler expands into it.
+pub fn new_parser_from_tts(sess: &ParseSess, tts: Vec<TokenTree>) -> Parser<'_> {
+    stream_to_parser(sess, tts.into_iter().collect(), crate::MACRO_ARGUMENTS)
+}
+
+
+// Base abstractions
+
+/// Given a session and a path and an optional span (for error reporting),
+/// add the path to the session's source_map and return the new source_file or
+/// error when a file can't be read.
+fn try_file_to_source_file(sess: &ParseSess, path: &Path, spanopt: Option<Span>)
+                   -> Result<Lrc<SourceFile>, Diagnostic> {
+    sess.source_map().load_file(path)
+    .map_err(|e| {
+        let msg = format!("couldn't read {}: {}", path.display(), e);
+        let mut diag = Diagnostic::new(Level::Fatal, &msg);
+        if let Some(sp) = spanopt {
+            diag.set_span(sp);
+        }
+        diag
+    })
+}
+
+/// Given a session and a path and an optional span (for error reporting),
+/// adds the path to the session's `source_map` and returns the new `source_file`.
+fn file_to_source_file(sess: &ParseSess, path: &Path, spanopt: Option<Span>)
+                   -> Lrc<SourceFile> {
+    match try_file_to_source_file(sess, path, spanopt) {
+        Ok(source_file) => source_file,
+        Err(d) => {
+            sess.span_diagnostic.emit_diagnostic(&d);
+            FatalError.raise();
+        }
+    }
+}
+
+/// Given a `source_file`, produces a sequence of token trees.
+pub fn source_file_to_stream(
+    sess: &ParseSess,
+    source_file: Lrc<SourceFile>,
+    override_span: Option<Span>,
+) -> (TokenStream, Vec<lexer::UnmatchedBrace>) {
+    panictry_buffer!(&sess.span_diagnostic, maybe_file_to_stream(sess, source_file, override_span))
+}
+
+/// Given a source file, produces a sequence of token trees. Returns any buffered errors from
+/// parsing the token stream.
+pub fn maybe_file_to_stream(
+    sess: &ParseSess,
+    source_file: Lrc<SourceFile>,
+    override_span: Option<Span>,
+) -> Result<(TokenStream, Vec<lexer::UnmatchedBrace>), Vec<Diagnostic>> {
+    let srdr = lexer::StringReader::new(sess, source_file, override_span);
+    let (token_trees, unmatched_braces) = srdr.into_token_trees();
+
+    match token_trees {
+        Ok(stream) => Ok((stream, unmatched_braces)),
+        Err(err) => {
+            let mut buffer = Vec::with_capacity(1);
+            err.buffer(&mut buffer);
+            // Not using `emit_unclosed_delims` to use `db.buffer`
+            for unmatched in unmatched_braces {
+                if let Some(err) = make_unclosed_delims_error(unmatched, &sess) {
+                    err.buffer(&mut buffer);
+                }
+            }
+            Err(buffer)
+        }
+    }
+}
+
+/// Given a stream and the `ParseSess`, produces a parser.
+pub fn stream_to_parser<'a>(
+    sess: &'a ParseSess,
+    stream: TokenStream,
+    subparser_name: Option<&'static str>,
+) -> Parser<'a> {
+    Parser::new(sess, stream, None, true, false, subparser_name)
+}
+
+/// Given a stream, the `ParseSess` and the base directory, produces a parser.
+///
+/// Use this function when you are creating a parser from the token stream
+/// and also care about the current working directory of the parser (e.g.,
+/// you are trying to resolve modules defined inside a macro invocation).
+///
+/// # Note
+///
+/// The main usage of this function is outside of rustc, for those who uses
+/// libsyntax as a library. Please do not remove this function while refactoring
+/// just because it is not used in rustc codebase!
+pub fn stream_to_parser_with_base_dir<'a>(
+    sess: &'a ParseSess,
+    stream: TokenStream,
+    base_dir: Directory<'a>,
+) -> Parser<'a> {
+    Parser::new(sess, stream, Some(base_dir), true, false, None)
+}
+
+/// Runs the given subparser `f` on the tokens of the given `attr`'s item.
+pub fn parse_in_attr<'a, T>(
+    sess: &'a ParseSess,
+    attr: &ast::Attribute,
+    mut f: impl FnMut(&mut Parser<'a>) -> PResult<'a, T>,
+) -> PResult<'a, T> {
+    let mut parser = Parser::new(
+        sess,
+        attr.get_normal_item().tokens.clone(),
+        None,
+        false,
+        false,
+        Some("attribute"),
+    );
+    let result = f(&mut parser)?;
+    if parser.token != token::Eof {
+        parser.unexpected()?;
+    }
+    Ok(result)
+}
+
+// NOTE(Centril): The following probably shouldn't be here but it acknowledges the
+// fact that architecturally, we are using parsing (read on below to understand why).
+
+pub fn nt_to_tokenstream(nt: &Nonterminal, sess: &ParseSess, span: Span) -> TokenStream {
+    // A `Nonterminal` is often a parsed AST item. At this point we now
+    // need to convert the parsed AST to an actual token stream, e.g.
+    // un-parse it basically.
+    //
+    // Unfortunately there's not really a great way to do that in a
+    // guaranteed lossless fashion right now. The fallback here is to just
+    // stringify the AST node and reparse it, but this loses all span
+    // information.
+    //
+    // As a result, some AST nodes are annotated with the token stream they
+    // came from. Here we attempt to extract these lossless token streams
+    // before we fall back to the stringification.
+    let tokens = match *nt {
+        Nonterminal::NtItem(ref item) => {
+            prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span)
+        }
+        Nonterminal::NtTraitItem(ref item) => {
+            prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span)
+        }
+        Nonterminal::NtImplItem(ref item) => {
+            prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span)
+        }
+        Nonterminal::NtIdent(ident, is_raw) => {
+            Some(tokenstream::TokenTree::token(token::Ident(ident.name, is_raw), ident.span).into())
+        }
+        Nonterminal::NtLifetime(ident) => {
+            Some(tokenstream::TokenTree::token(token::Lifetime(ident.name), ident.span).into())
+        }
+        Nonterminal::NtTT(ref tt) => {
+            Some(tt.clone().into())
+        }
+        _ => None,
+    };
+
+    // FIXME(#43081): Avoid this pretty-print + reparse hack
+    let source = pprust::nonterminal_to_string(nt);
+    let filename = FileName::macro_expansion_source_code(&source);
+    let tokens_for_real = parse_stream_from_source_str(filename, source, sess, Some(span));
+
+    // During early phases of the compiler the AST could get modified
+    // directly (e.g., attributes added or removed) and the internal cache
+    // of tokens my not be invalidated or updated. Consequently if the
+    // "lossless" token stream disagrees with our actual stringification
+    // (which has historically been much more battle-tested) then we go
+    // with the lossy stream anyway (losing span information).
+    //
+    // Note that the comparison isn't `==` here to avoid comparing spans,
+    // but it *also* is a "probable" equality which is a pretty weird
+    // definition. We mostly want to catch actual changes to the AST
+    // like a `#[cfg]` being processed or some weird `macro_rules!`
+    // expansion.
+    //
+    // What we *don't* want to catch is the fact that a user-defined
+    // literal like `0xf` is stringified as `15`, causing the cached token
+    // stream to not be literal `==` token-wise (ignoring spans) to the
+    // token stream we got from stringification.
+    //
+    // Instead the "probably equal" check here is "does each token
+    // recursively have the same discriminant?" We basically don't look at
+    // the token values here and assume that such fine grained token stream
+    // modifications, including adding/removing typically non-semantic
+    // tokens such as extra braces and commas, don't happen.
+    if let Some(tokens) = tokens {
+        if tokens.probably_equal_for_proc_macro(&tokens_for_real) {
+            return tokens
+        }
+        info!("cached tokens found, but they're not \"probably equal\", \
+                going with stringified version");
+    }
+    return tokens_for_real
+}
+
+fn prepend_attrs(
+    sess: &ParseSess,
+    attrs: &[ast::Attribute],
+    tokens: Option<&tokenstream::TokenStream>,
+    span: syntax_pos::Span
+) -> Option<tokenstream::TokenStream> {
+    let tokens = tokens?;
+    if attrs.len() == 0 {
+        return Some(tokens.clone())
+    }
+    let mut builder = tokenstream::TokenStreamBuilder::new();
+    for attr in attrs {
+        assert_eq!(attr.style, ast::AttrStyle::Outer,
+                   "inner attributes should prevent cached tokens from existing");
+
+        let source = pprust::attribute_to_string(attr);
+        let macro_filename = FileName::macro_expansion_source_code(&source);
+
+        let item = match attr.kind {
+            ast::AttrKind::Normal(ref item) => item,
+            ast::AttrKind::DocComment(_) => {
+                let stream = parse_stream_from_source_str(macro_filename, source, sess, Some(span));
+                builder.push(stream);
+                continue
+            }
+        };
+
+        // synthesize # [ $path $tokens ] manually here
+        let mut brackets = tokenstream::TokenStreamBuilder::new();
+
+        // For simple paths, push the identifier directly
+        if item.path.segments.len() == 1 && item.path.segments[0].args.is_none() {
+            let ident = item.path.segments[0].ident;
+            let token = token::Ident(ident.name, ident.as_str().starts_with("r#"));
+            brackets.push(tokenstream::TokenTree::token(token, ident.span));
+
+        // ... and for more complicated paths, fall back to a reparse hack that
+        // should eventually be removed.
+        } else {
+            let stream = parse_stream_from_source_str(macro_filename, source, sess, Some(span));
+            brackets.push(stream);
+        }
+
+        brackets.push(item.tokens.clone());
+
+        // The span we list here for `#` and for `[ ... ]` are both wrong in
+        // that it encompasses more than each token, but it hopefully is "good
+        // enough" for now at least.
+        builder.push(tokenstream::TokenTree::token(token::Pound, attr.span));
+        let delim_span = tokenstream::DelimSpan::from_single(attr.span);
+        builder.push(tokenstream::TokenTree::Delimited(
+            delim_span, token::DelimToken::Bracket, brackets.build().into()));
+    }
+    builder.push(tokens.clone());
+    Some(builder.build())
+}

src/librustc_parse/parser/attr.rs

@@ -0,0 +1,351 @@
+use super::{SeqSep, Parser, TokenType, PathStyle};
+use syntax::attr;
+use syntax::ast;
+use syntax::util::comments;
+use syntax::token::{self, Nonterminal, DelimToken};
+use syntax::tokenstream::{TokenStream, TokenTree};
+use syntax_pos::{Span, Symbol};
+use errors::PResult;
+
+use log::debug;
+
+#[derive(Debug)]
+enum InnerAttributeParsePolicy<'a> {
+    Permitted,
+    NotPermitted { reason: &'a str, saw_doc_comment: bool, prev_attr_sp: Option<Span> },
+}
+
+const DEFAULT_UNEXPECTED_INNER_ATTR_ERR_MSG: &str = "an inner attribute is not \
+                                                     permitted in this context";
+
+impl<'a> Parser<'a> {
+    /// Parses attributes that appear before an item.
+    pub(super) fn parse_outer_attributes(&mut self) -> PResult<'a, Vec<ast::Attribute>> {
+        let mut attrs: Vec<ast::Attribute> = Vec::new();
+        let mut just_parsed_doc_comment = false;
+        loop {
+            debug!("parse_outer_attributes: self.token={:?}", self.token);
+            match self.token.kind {
+                token::Pound => {
+                    let inner_error_reason = if just_parsed_doc_comment {
+                        "an inner attribute is not permitted following an outer doc comment"
+                    } else if !attrs.is_empty() {
+                        "an inner attribute is not permitted following an outer attribute"
+                    } else {
+                        DEFAULT_UNEXPECTED_INNER_ATTR_ERR_MSG
+                    };
+                    let inner_parse_policy =
+                        InnerAttributeParsePolicy::NotPermitted {
+                            reason: inner_error_reason,
+                            saw_doc_comment: just_parsed_doc_comment,
+                            prev_attr_sp: attrs.last().and_then(|a| Some(a.span))
+                        };
+                    let attr = self.parse_attribute_with_inner_parse_policy(inner_parse_policy)?;
+                    attrs.push(attr);
+                    just_parsed_doc_comment = false;
+                }
+                token::DocComment(s) => {
+                    let attr = self.mk_doc_comment(s);
+                    if attr.style != ast::AttrStyle::Outer {
+                        let mut err = self.fatal("expected outer doc comment");
+                        err.note("inner doc comments like this (starting with \
+                                  `//!` or `/*!`) can only appear before items");
+                        return Err(err);
+                    }
+                    attrs.push(attr);
+                    self.bump();
+                    just_parsed_doc_comment = true;
+                }
+                _ => break,
+            }
+        }
+        Ok(attrs)
+    }
+
+    fn mk_doc_comment(&self, s: Symbol) -> ast::Attribute {
+        let style = comments::doc_comment_style(&s.as_str());
+        attr::mk_doc_comment(style, s, self.token.span)
+    }
+
+    /// Matches `attribute = # ! [ meta_item ]`.
+    ///
+    /// If `permit_inner` is `true`, then a leading `!` indicates an inner
+    /// attribute.
+    pub fn parse_attribute(&mut self, permit_inner: bool) -> PResult<'a, ast::Attribute> {
+        debug!("parse_attribute: permit_inner={:?} self.token={:?}",
+               permit_inner,
+               self.token);
+        let inner_parse_policy = if permit_inner {
+            InnerAttributeParsePolicy::Permitted
+        } else {
+            InnerAttributeParsePolicy::NotPermitted {
+                reason: DEFAULT_UNEXPECTED_INNER_ATTR_ERR_MSG,
+                saw_doc_comment: false,
+                prev_attr_sp: None
+            }
+        };
+        self.parse_attribute_with_inner_parse_policy(inner_parse_policy)
+    }
+
+    /// The same as `parse_attribute`, except it takes in an `InnerAttributeParsePolicy`
+    /// that prescribes how to handle inner attributes.
+    fn parse_attribute_with_inner_parse_policy(
+        &mut self,
+        inner_parse_policy: InnerAttributeParsePolicy<'_>
+    ) -> PResult<'a, ast::Attribute> {
+        debug!("parse_attribute_with_inner_parse_policy: inner_parse_policy={:?} self.token={:?}",
+               inner_parse_policy,
+               self.token);
+        let (span, item, style) = match self.token.kind {
+            token::Pound => {
+                let lo = self.token.span;
+                self.bump();
+
+                if let InnerAttributeParsePolicy::Permitted = inner_parse_policy {
+                    self.expected_tokens.push(TokenType::Token(token::Not));
+                }
+
+                let style = if self.token == token::Not {
+                    self.bump();
+                    ast::AttrStyle::Inner
+                } else {
+                    ast::AttrStyle::Outer
+                };
+
+                self.expect(&token::OpenDelim(token::Bracket))?;
+                let item = self.parse_attr_item()?;
+                self.expect(&token::CloseDelim(token::Bracket))?;
+                let hi = self.prev_span;
+
+                let attr_sp = lo.to(hi);
+
+                // Emit error if inner attribute is encountered and not permitted
+                if style == ast::AttrStyle::Inner {
+                    if let InnerAttributeParsePolicy::NotPermitted { reason,
+                        saw_doc_comment, prev_attr_sp } = inner_parse_policy {
+                        let prev_attr_note = if saw_doc_comment {
+                            "previous doc comment"
+                        } else {
+                            "previous outer attribute"
+                        };
+
+                        let mut diagnostic = self
+                            .diagnostic()
+                            .struct_span_err(attr_sp, reason);
+
+                        if let Some(prev_attr_sp) = prev_attr_sp {
+                            diagnostic
+                                .span_label(attr_sp, "not permitted following an outer attibute")
+                                .span_label(prev_attr_sp, prev_attr_note);
+                        }
+
+                        diagnostic
+                            .note("inner attributes, like `#![no_std]`, annotate the item \
+                                   enclosing them, and are usually found at the beginning of \
+                                   source files. Outer attributes, like `#[test]`, annotate the \
+                                   item following them.")
+                            .emit()
+                    }
+                }
+
+                (attr_sp, item, style)
+            }
+            _ => {
+                let token_str = self.this_token_to_string();
+                return Err(self.fatal(&format!("expected `#`, found `{}`", token_str)));
+            }
+        };
+
+        Ok(attr::mk_attr_from_item(style, item, span))
+    }
+
+    /// Parses an inner part of an attribute (the path and following tokens).
+    /// The tokens must be either a delimited token stream, or empty token stream,
+    /// or the "legacy" key-value form.
+    ///     PATH `(` TOKEN_STREAM `)`
+    ///     PATH `[` TOKEN_STREAM `]`
+    ///     PATH `{` TOKEN_STREAM `}`
+    ///     PATH
+    ///     PATH `=` UNSUFFIXED_LIT
+    /// The delimiters or `=` are still put into the resulting token stream.
+    pub fn parse_attr_item(&mut self) -> PResult<'a, ast::AttrItem> {
+        let item = match self.token.kind {
+            token::Interpolated(ref nt) => match **nt {
+                Nonterminal::NtMeta(ref item) => Some(item.clone()),
+                _ => None,
+            },
+            _ => None,
+        };
+        Ok(if let Some(item) = item {
+            self.bump();
+            item
+        } else {
+            let path = self.parse_path(PathStyle::Mod)?;
+            let tokens = if self.check(&token::OpenDelim(DelimToken::Paren)) ||
+               self.check(&token::OpenDelim(DelimToken::Bracket)) ||
+               self.check(&token::OpenDelim(DelimToken::Brace)) {
+                   self.parse_token_tree().into()
+            } else if self.eat(&token::Eq) {
+                let eq = TokenTree::token(token::Eq, self.prev_span);
+                let mut is_interpolated_expr = false;
+                if let token::Interpolated(nt) = &self.token.kind {
+                    if let token::NtExpr(..) = **nt {
+                        is_interpolated_expr = true;
+                    }
+                }
+                let token_tree = if is_interpolated_expr {
+                    // We need to accept arbitrary interpolated expressions to continue
+                    // supporting things like `doc = $expr` that work on stable.
+                    // Non-literal interpolated expressions are rejected after expansion.
+                    self.parse_token_tree()
+                } else {
+                    self.parse_unsuffixed_lit()?.token_tree()
+                };
+                TokenStream::new(vec![eq.into(), token_tree.into()])
+            } else {
+                TokenStream::default()
+            };
+            ast::AttrItem { path, tokens }
+        })
+    }
+
+    /// Parses attributes that appear after the opening of an item. These should
+    /// be preceded by an exclamation mark, but we accept and warn about one
+    /// terminated by a semicolon.
+    ///
+    /// Matches `inner_attrs*`.
+    crate fn parse_inner_attributes(&mut self) -> PResult<'a, Vec<ast::Attribute>> {
+        let mut attrs: Vec<ast::Attribute> = vec![];
+        loop {
+            match self.token.kind {
+                token::Pound => {
+                    // Don't even try to parse if it's not an inner attribute.
+                    if !self.look_ahead(1, |t| t == &token::Not) {
+                        break;
+                    }
+
+                    let attr = self.parse_attribute(true)?;
+                    assert_eq!(attr.style, ast::AttrStyle::Inner);
+                    attrs.push(attr);
+                }
+                token::DocComment(s) => {
+                    // We need to get the position of this token before we bump.
+                    let attr = self.mk_doc_comment(s);
+                    if attr.style == ast::AttrStyle::Inner {
+                        attrs.push(attr);
+                        self.bump();
+                    } else {
+                        break;
+                    }
+                }
+                _ => break,
+            }
+        }
+        Ok(attrs)
+    }
+
+    fn parse_unsuffixed_lit(&mut self) -> PResult<'a, ast::Lit> {
+        let lit = self.parse_lit()?;
+        debug!("checking if {:?} is unusuffixed", lit);
+
+        if !lit.kind.is_unsuffixed() {
+            let msg = "suffixed literals are not allowed in attributes";
+            self.diagnostic().struct_span_err(lit.span, msg)
+                             .help("instead of using a suffixed literal \
+                                    (1u8, 1.0f32, etc.), use an unsuffixed version \
+                                    (1, 1.0, etc.).")
+                             .emit()
+        }
+
+        Ok(lit)
+    }
+
+    /// Parses `cfg_attr(pred, attr_item_list)` where `attr_item_list` is comma-delimited.
+    pub fn parse_cfg_attr(&mut self) -> PResult<'a, (ast::MetaItem, Vec<(ast::AttrItem, Span)>)> {
+        self.expect(&token::OpenDelim(token::Paren))?;
+
+        let cfg_predicate = self.parse_meta_item()?;
+        self.expect(&token::Comma)?;
+
+        // Presumably, the majority of the time there will only be one attr.
+        let mut expanded_attrs = Vec::with_capacity(1);
+
+        while !self.check(&token::CloseDelim(token::Paren)) {
+            let lo = self.token.span.lo();
+            let item = self.parse_attr_item()?;
+            expanded_attrs.push((item, self.prev_span.with_lo(lo)));
+            self.expect_one_of(&[token::Comma], &[token::CloseDelim(token::Paren)])?;
+        }
+
+        self.expect(&token::CloseDelim(token::Paren))?;
+        Ok((cfg_predicate, expanded_attrs))
+    }
+
+    /// Matches the following grammar (per RFC 1559).
+    ///
+    ///     meta_item : PATH ( '=' UNSUFFIXED_LIT | '(' meta_item_inner? ')' )? ;
+    ///     meta_item_inner : (meta_item | UNSUFFIXED_LIT) (',' meta_item_inner)? ;
+    pub fn parse_meta_item(&mut self) -> PResult<'a, ast::MetaItem> {
+        let nt_meta = match self.token.kind {
+            token::Interpolated(ref nt) => match **nt {
+                token::NtMeta(ref e) => Some(e.clone()),
+                _ => None,
+            },
+            _ => None,
+        };
+
+        if let Some(item) = nt_meta {
+            return match item.meta(item.path.span) {
+                Some(meta) => {
+                    self.bump();
+                    Ok(meta)
+                }
+                None => self.unexpected(),
+            }
+        }
+
+        let lo = self.token.span;
+        let path = self.parse_path(PathStyle::Mod)?;
+        let kind = self.parse_meta_item_kind()?;
+        let span = lo.to(self.prev_span);
+        Ok(ast::MetaItem { path, kind, span })
+    }
+
+    crate fn parse_meta_item_kind(&mut self) -> PResult<'a, ast::MetaItemKind> {
+        Ok(if self.eat(&token::Eq) {
+            ast::MetaItemKind::NameValue(self.parse_unsuffixed_lit()?)
+        } else if self.eat(&token::OpenDelim(token::Paren)) {
+            ast::MetaItemKind::List(self.parse_meta_seq()?)
+        } else {
+            ast::MetaItemKind::Word
+        })
+    }
+
+    /// Matches `meta_item_inner : (meta_item | UNSUFFIXED_LIT) ;`.
+    fn parse_meta_item_inner(&mut self) -> PResult<'a, ast::NestedMetaItem> {
+        match self.parse_unsuffixed_lit() {
+            Ok(lit) => {
+                return Ok(ast::NestedMetaItem::Literal(lit))
+            }
+            Err(ref mut err) => err.cancel(),
+        }
+
+        match self.parse_meta_item() {
+            Ok(mi) => {
+                return Ok(ast::NestedMetaItem::MetaItem(mi))
+            }
+            Err(ref mut err) => err.cancel(),
+        }
+
+        let found = self.this_token_to_string();
+        let msg = format!("expected unsuffixed literal or identifier, found `{}`", found);
+        Err(self.diagnostic().struct_span_err(self.token.span, &msg))
+    }
+
+    /// Matches `meta_seq = ( COMMASEP(meta_item_inner) )`.
+    fn parse_meta_seq(&mut self) -> PResult<'a, Vec<ast::NestedMetaItem>> {
+        self.parse_seq_to_end(&token::CloseDelim(token::Paren),
+                              SeqSep::trailing_allowed(token::Comma),
+                              |p: &mut Parser<'a>| p.parse_meta_item_inner())
+    }
+}

src/librustc_parse/parser/generics.rs

@@ -0,0 +1,308 @@
+use super::Parser;
+
+use syntax::ast::{self, WhereClause, GenericParam, GenericParamKind, GenericBounds, Attribute};
+use syntax::token;
+use syntax::source_map::DUMMY_SP;
+use syntax_pos::symbol::{kw, sym};
+
+use errors::PResult;
+
+impl<'a> Parser<'a> {
+    /// Parses bounds of a lifetime parameter `BOUND + BOUND + BOUND`, possibly with trailing `+`.
+    ///
+    /// ```
+    /// BOUND = LT_BOUND (e.g., `'a`)
+    /// ```
+    fn parse_lt_param_bounds(&mut self) -> GenericBounds {
+        let mut lifetimes = Vec::new();
+        while self.check_lifetime() {
+            lifetimes.push(ast::GenericBound::Outlives(self.expect_lifetime()));
+
+            if !self.eat_plus() {
+                break
+            }
+        }
+        lifetimes
+    }
+
+    /// Matches `typaram = IDENT (`?` unbound)? optbounds ( EQ ty )?`.
+    fn parse_ty_param(&mut self,
+                      preceding_attrs: Vec<Attribute>)
+                      -> PResult<'a, GenericParam> {
+        let ident = self.parse_ident()?;
+
+        // Parse optional colon and param bounds.
+        let bounds = if self.eat(&token::Colon) {
+            self.parse_generic_bounds(Some(self.prev_span))?
+        } else {
+            Vec::new()
+        };
+
+        let default = if self.eat(&token::Eq) {
+            Some(self.parse_ty()?)
+        } else {
+            None
+        };
+
+        Ok(GenericParam {
+            ident,
+            id: ast::DUMMY_NODE_ID,
+            attrs: preceding_attrs.into(),
+            bounds,
+            kind: GenericParamKind::Type {
+                default,
+            },
+            is_placeholder: false
+        })
+    }
+
+    fn parse_const_param(&mut self, preceding_attrs: Vec<Attribute>) -> PResult<'a, GenericParam> {
+        let lo = self.token.span;
+
+        self.expect_keyword(kw::Const)?;
+        let ident = self.parse_ident()?;
+        self.expect(&token::Colon)?;
+        let ty = self.parse_ty()?;
+
+        self.sess.gated_spans.gate(sym::const_generics, lo.to(self.prev_span));
+
+        Ok(GenericParam {
+            ident,
+            id: ast::DUMMY_NODE_ID,
+            attrs: preceding_attrs.into(),
+            bounds: Vec::new(),
+            kind: GenericParamKind::Const {
+                ty,
+            },
+            is_placeholder: false
+        })
+    }
+
+    /// Parses a (possibly empty) list of lifetime and type parameters, possibly including
+    /// a trailing comma and erroneous trailing attributes.
+    pub(super) fn parse_generic_params(&mut self) -> PResult<'a, Vec<ast::GenericParam>> {
+        let mut params = Vec::new();
+        loop {
+            let attrs = self.parse_outer_attributes()?;
+            if self.check_lifetime() {
+                let lifetime = self.expect_lifetime();
+                // Parse lifetime parameter.
+                let bounds = if self.eat(&token::Colon) {
+                    self.parse_lt_param_bounds()
+                } else {
+                    Vec::new()
+                };
+                params.push(ast::GenericParam {
+                    ident: lifetime.ident,
+                    id: lifetime.id,
+                    attrs: attrs.into(),
+                    bounds,
+                    kind: ast::GenericParamKind::Lifetime,
+                    is_placeholder: false
+                });
+            } else if self.check_keyword(kw::Const) {
+                // Parse const parameter.
+                params.push(self.parse_const_param(attrs)?);
+            } else if self.check_ident() {
+                // Parse type parameter.
+                params.push(self.parse_ty_param(attrs)?);
+            } else if self.token.can_begin_type() {
+                // Trying to write an associated type bound? (#26271)
+                let snapshot = self.clone();
+                match self.parse_ty_where_predicate() {
+                    Ok(where_predicate) => {
+                        self.struct_span_err(
+                            where_predicate.span(),
+                            "bounds on associated types do not belong here",
+                        )
+                        .span_label(where_predicate.span(), "belongs in `where` clause")
+                        .emit();
+                    }
+                    Err(mut err) => {
+                        err.cancel();
+                        std::mem::replace(self, snapshot);
+                        break
+                    }
+                }
+            } else {
+                // Check for trailing attributes and stop parsing.
+                if !attrs.is_empty() {
+                    if !params.is_empty() {
+                        self.struct_span_err(
+                            attrs[0].span,
+                            "trailing attribute after generic parameter",
+                        )
+                        .span_label(attrs[0].span, "attributes must go before parameters")
+                        .emit();
+                    } else {
+                        self.struct_span_err(
+                            attrs[0].span,
+                            &format!("attribute without generic parameters"),
+                        )
+                        .span_label(
+                            attrs[0].span,
+                            "attributes are only permitted when preceding parameters",
+                        )
+                        .emit();
+                    }
+                }
+                break
+            }
+
+            if !self.eat(&token::Comma) {
+                break
+            }
+        }
+        Ok(params)
+    }
+
+    /// Parses a set of optional generic type parameter declarations. Where
+    /// clauses are not parsed here, and must be added later via
+    /// `parse_where_clause()`.
+    ///
+    /// matches generics = ( ) | ( < > ) | ( < typaramseq ( , )? > ) | ( < lifetimes ( , )? > )
+    ///                  | ( < lifetimes , typaramseq ( , )? > )
+    /// where   typaramseq = ( typaram ) | ( typaram , typaramseq )
+    pub(super) fn parse_generics(&mut self) -> PResult<'a, ast::Generics> {
+        let span_lo = self.token.span;
+        let (params, span) = if self.eat_lt() {
+            let params = self.parse_generic_params()?;
+            self.expect_gt()?;
+            (params, span_lo.to(self.prev_span))
+        } else {
+            (vec![], self.prev_span.between(self.token.span))
+        };
+        Ok(ast::Generics {
+            params,
+            where_clause: WhereClause {
+                predicates: Vec::new(),
+                span: DUMMY_SP,
+            },
+            span,
+        })
+    }
+
+    /// Parses an optional where-clause and places it in `generics`.
+    ///
+    /// ```ignore (only-for-syntax-highlight)
+    /// where T : Trait<U, V> + 'b, 'a : 'b
+    /// ```
+    pub(super) fn parse_where_clause(&mut self) -> PResult<'a, WhereClause> {
+        let mut where_clause = WhereClause {
+            predicates: Vec::new(),
+            span: self.prev_span.to(self.prev_span),
+        };
+
+        if !self.eat_keyword(kw::Where) {
+            return Ok(where_clause);
+        }
+        let lo = self.prev_span;
+
+        // We are considering adding generics to the `where` keyword as an alternative higher-rank
+        // parameter syntax (as in `where<'a>` or `where<T>`. To avoid that being a breaking
+        // change we parse those generics now, but report an error.
+        if self.choose_generics_over_qpath() {
+            let generics = self.parse_generics()?;
+            self.struct_span_err(
+                generics.span,
+                "generic parameters on `where` clauses are reserved for future use",
+            )
+                .span_label(generics.span, "currently unsupported")
+                .emit();
+        }
+
+        loop {
+            let lo = self.token.span;
+            if self.check_lifetime() && self.look_ahead(1, |t| !t.is_like_plus()) {
+                let lifetime = self.expect_lifetime();
+                // Bounds starting with a colon are mandatory, but possibly empty.
+                self.expect(&token::Colon)?;
+                let bounds = self.parse_lt_param_bounds();
+                where_clause.predicates.push(ast::WherePredicate::RegionPredicate(
+                    ast::WhereRegionPredicate {
+                        span: lo.to(self.prev_span),
+                        lifetime,
+                        bounds,
+                    }
+                ));
+            } else if self.check_type() {
+                where_clause.predicates.push(self.parse_ty_where_predicate()?);
+            } else {
+                break
+            }
+
+            if !self.eat(&token::Comma) {
+                break
+            }
+        }
+
+        where_clause.span = lo.to(self.prev_span);
+        Ok(where_clause)
+    }
+
+    fn parse_ty_where_predicate(&mut self) -> PResult<'a, ast::WherePredicate> {
+        let lo = self.token.span;
+        // Parse optional `for<'a, 'b>`.
+        // This `for` is parsed greedily and applies to the whole predicate,
+        // the bounded type can have its own `for` applying only to it.
+        // Examples:
+        // * `for<'a> Trait1<'a>: Trait2<'a /* ok */>`
+        // * `(for<'a> Trait1<'a>): Trait2<'a /* not ok */>`
+        // * `for<'a> for<'b> Trait1<'a, 'b>: Trait2<'a /* ok */, 'b /* not ok */>`
+        let lifetime_defs = self.parse_late_bound_lifetime_defs()?;
+
+        // Parse type with mandatory colon and (possibly empty) bounds,
+        // or with mandatory equality sign and the second type.
+        let ty = self.parse_ty()?;
+        if self.eat(&token::Colon) {
+            let bounds = self.parse_generic_bounds(Some(self.prev_span))?;
+            Ok(ast::WherePredicate::BoundPredicate(
+                ast::WhereBoundPredicate {
+                    span: lo.to(self.prev_span),
+                    bound_generic_params: lifetime_defs,
+                    bounded_ty: ty,
+                    bounds,
+                }
+            ))
+        // FIXME: Decide what should be used here, `=` or `==`.
+        // FIXME: We are just dropping the binders in lifetime_defs on the floor here.
+        } else if self.eat(&token::Eq) || self.eat(&token::EqEq) {
+            let rhs_ty = self.parse_ty()?;
+            Ok(ast::WherePredicate::EqPredicate(
+                ast::WhereEqPredicate {
+                    span: lo.to(self.prev_span),
+                    lhs_ty: ty,
+                    rhs_ty,
+                    id: ast::DUMMY_NODE_ID,
+                }
+            ))
+        } else {
+            self.unexpected()
+        }
+    }
+
+    pub(super) fn choose_generics_over_qpath(&self) -> bool {
+        // There's an ambiguity between generic parameters and qualified paths in impls.
+        // If we see `<` it may start both, so we have to inspect some following tokens.
+        // The following combinations can only start generics,
+        // but not qualified paths (with one exception):
+        //     `<` `>` - empty generic parameters
+        //     `<` `#` - generic parameters with attributes
+        //     `<` (LIFETIME|IDENT) `>` - single generic parameter
+        //     `<` (LIFETIME|IDENT) `,` - first generic parameter in a list
+        //     `<` (LIFETIME|IDENT) `:` - generic parameter with bounds
+        //     `<` (LIFETIME|IDENT) `=` - generic parameter with a default
+        //     `<` const                - generic const parameter
+        // The only truly ambiguous case is
+        //     `<` IDENT `>` `::` IDENT ...
+        // we disambiguate it in favor of generics (`impl<T> ::absolute::Path<T> { ... }`)
+        // because this is what almost always expected in practice, qualified paths in impls
+        // (`impl <Type>::AssocTy { ... }`) aren't even allowed by type checker at the moment.
+        self.token == token::Lt &&
+            (self.look_ahead(1, |t| t == &token::Pound || t == &token::Gt) ||
+             self.look_ahead(1, |t| t.is_lifetime() || t.is_ident()) &&
+                self.look_ahead(2, |t| t == &token::Gt || t == &token::Comma ||
+                                       t == &token::Colon || t == &token::Eq) ||
+            self.is_keyword_ahead(1, &[kw::Const]))
+    }
+}

src/librustc_parse/parser/module.rs

@@ -0,0 +1,316 @@
+use super::Parser;
+use super::item::ItemInfo;
+use super::diagnostics::Error;
+
+use crate::{new_sub_parser_from_file, DirectoryOwnership};
+
+use syntax::attr;
+use syntax::ast::{self, Ident, Attribute, ItemKind, Mod, Crate};
+use syntax::token::{self, TokenKind};
+use syntax::source_map::{SourceMap, Span, DUMMY_SP, FileName};
+
+use syntax_pos::symbol::sym;
+use errors::PResult;
+
+use std::path::{self, Path, PathBuf};
+
+/// Information about the path to a module.
+pub(super) struct ModulePath {
+    name: String,
+    path_exists: bool,
+    pub result: Result<ModulePathSuccess, Error>,
+}
+
+pub(super) struct ModulePathSuccess {
+    pub path: PathBuf,
+    pub directory_ownership: DirectoryOwnership,
+}
+
+impl<'a> Parser<'a> {
+    /// Parses a source module as a crate. This is the main entry point for the parser.
+    pub fn parse_crate_mod(&mut self) -> PResult<'a, Crate> {
+        let lo = self.token.span;
+        let krate = Ok(ast::Crate {
+            attrs: self.parse_inner_attributes()?,
+            module: self.parse_mod_items(&token::Eof, lo)?,
+            span: lo.to(self.token.span),
+        });
+        krate
+    }
+
+    /// Parses a `mod <foo> { ... }` or `mod <foo>;` item.
+    pub(super) fn parse_item_mod(&mut self, outer_attrs: &[Attribute]) -> PResult<'a, ItemInfo> {
+        // HACK(Centril): See documentation on `ParseSess::process_cfg_mod`.
+        let (in_cfg, outer_attrs) = (self.sess.process_cfg_mod)(
+            self.sess,
+            self.cfg_mods,
+            outer_attrs,
+        );
+
+        let id_span = self.token.span;
+        let id = self.parse_ident()?;
+        if self.eat(&token::Semi) {
+            if in_cfg && self.recurse_into_file_modules {
+                // This mod is in an external file. Let's go get it!
+                let ModulePathSuccess { path, directory_ownership } =
+                    self.submod_path(id, &outer_attrs, id_span)?;
+                let (module, attrs) =
+                    self.eval_src_mod(path, directory_ownership, id.to_string(), id_span)?;
+                Ok((id, ItemKind::Mod(module), Some(attrs)))
+            } else {
+                let placeholder = ast::Mod {
+                    inner: DUMMY_SP,
+                    items: Vec::new(),
+                    inline: false
+                };
+                Ok((id, ItemKind::Mod(placeholder), None))
+            }
+        } else {
+            let old_directory = self.directory.clone();
+            self.push_directory(id, &outer_attrs);
+
+            self.expect(&token::OpenDelim(token::Brace))?;
+            let mod_inner_lo = self.token.span;
+            let attrs = self.parse_inner_attributes()?;
+            let module = self.parse_mod_items(&token::CloseDelim(token::Brace), mod_inner_lo)?;
+
+            self.directory = old_directory;
+            Ok((id, ItemKind::Mod(module), Some(attrs)))
+        }
+    }
+
+    /// Given a termination token, parses all of the items in a module.
+    fn parse_mod_items(&mut self, term: &TokenKind, inner_lo: Span) -> PResult<'a, Mod> {
+        let mut items = vec![];
+        while let Some(item) = self.parse_item()? {
+            items.push(item);
+            self.maybe_consume_incorrect_semicolon(&items);
+        }
+
+        if !self.eat(term) {
+            let token_str = self.this_token_descr();
+            if !self.maybe_consume_incorrect_semicolon(&items) {
+                let mut err = self.fatal(&format!("expected item, found {}", token_str));
+                err.span_label(self.token.span, "expected item");
+                return Err(err);
+            }
+        }
+
+        let hi = if self.token.span.is_dummy() {
+            inner_lo
+        } else {
+            self.prev_span
+        };
+
+        Ok(Mod {
+            inner: inner_lo.to(hi),
+            items,
+            inline: true
+        })
+    }
+
+    fn submod_path(
+        &mut self,
+        id: ast::Ident,
+        outer_attrs: &[Attribute],
+        id_sp: Span
+    ) -> PResult<'a, ModulePathSuccess> {
+        if let Some(path) = Parser::submod_path_from_attr(outer_attrs, &self.directory.path) {
+            return Ok(ModulePathSuccess {
+                directory_ownership: match path.file_name().and_then(|s| s.to_str()) {
+                    // All `#[path]` files are treated as though they are a `mod.rs` file.
+                    // This means that `mod foo;` declarations inside `#[path]`-included
+                    // files are siblings,
+                    //
+                    // Note that this will produce weirdness when a file named `foo.rs` is
+                    // `#[path]` included and contains a `mod foo;` declaration.
+                    // If you encounter this, it's your own darn fault :P
+                    Some(_) => DirectoryOwnership::Owned { relative: None },
+                    _ => DirectoryOwnership::UnownedViaMod,
+                },
+                path,
+            });
+        }
+
+        let relative = match self.directory.ownership {
+            DirectoryOwnership::Owned { relative } => relative,
+            DirectoryOwnership::UnownedViaBlock |
+            DirectoryOwnership::UnownedViaMod => None,
+        };
+        let paths = Parser::default_submod_path(
+                        id, relative, &self.directory.path, self.sess.source_map());
+
+        match self.directory.ownership {
+            DirectoryOwnership::Owned { .. } => {
+                paths.result.map_err(|err| self.span_fatal_err(id_sp, err))
+            },
+            DirectoryOwnership::UnownedViaBlock => {
+                let msg =
+                    "Cannot declare a non-inline module inside a block \
+                    unless it has a path attribute";
+                let mut err = self.diagnostic().struct_span_err(id_sp, msg);
+                if paths.path_exists {
+                    let msg = format!("Maybe `use` the module `{}` instead of redeclaring it",
+                                      paths.name);
+                    err.span_note(id_sp, &msg);
+                }
+                Err(err)
+            }
+            DirectoryOwnership::UnownedViaMod => {
+                let mut err = self.diagnostic().struct_span_err(id_sp,
+                    "cannot declare a new module at this location");
+                if !id_sp.is_dummy() {
+                    let src_path = self.sess.source_map().span_to_filename(id_sp);
+                    if let FileName::Real(src_path) = src_path {
+                        if let Some(stem) = src_path.file_stem() {
+                            let mut dest_path = src_path.clone();
+                            dest_path.set_file_name(stem);
+                            dest_path.push("mod.rs");
+                            err.span_note(id_sp,
+                                    &format!("maybe move this module `{}` to its own \
+                                                directory via `{}`", src_path.display(),
+                                            dest_path.display()));
+                        }
+                    }
+                }
+                if paths.path_exists {
+                    err.span_note(id_sp,
+                                  &format!("... or maybe `use` the module `{}` instead \
+                                            of possibly redeclaring it",
+                                           paths.name));
+                }
+                Err(err)
+            }
+        }
+    }
+
+    pub(super) fn submod_path_from_attr(attrs: &[Attribute], dir_path: &Path) -> Option<PathBuf> {
+        if let Some(s) = attr::first_attr_value_str_by_name(attrs, sym::path) {
+            let s = s.as_str();
+
+            // On windows, the base path might have the form
+            // `\\?\foo\bar` in which case it does not tolerate
+            // mixed `/` and `\` separators, so canonicalize
+            // `/` to `\`.
+            #[cfg(windows)]
+            let s = s.replace("/", "\\");
+            Some(dir_path.join(&*s))
+        } else {
+            None
+        }
+    }
+
+    /// Returns a path to a module.
+    pub(super) fn default_submod_path(
+        id: ast::Ident,
+        relative: Option<ast::Ident>,
+        dir_path: &Path,
+        source_map: &SourceMap) -> ModulePath
+    {
+        // If we're in a foo.rs file instead of a mod.rs file,
+        // we need to look for submodules in
+        // `./foo/<id>.rs` and `./foo/<id>/mod.rs` rather than
+        // `./<id>.rs` and `./<id>/mod.rs`.
+        let relative_prefix_string;
+        let relative_prefix = if let Some(ident) = relative {
+            relative_prefix_string = format!("{}{}", ident, path::MAIN_SEPARATOR);
+            &relative_prefix_string
+        } else {
+            ""
+        };
+
+        let mod_name = id.to_string();
+        let default_path_str = format!("{}{}.rs", relative_prefix, mod_name);
+        let secondary_path_str = format!("{}{}{}mod.rs",
+                                         relative_prefix, mod_name, path::MAIN_SEPARATOR);
+        let default_path = dir_path.join(&default_path_str);
+        let secondary_path = dir_path.join(&secondary_path_str);
+        let default_exists = source_map.file_exists(&default_path);
+        let secondary_exists = source_map.file_exists(&secondary_path);
+
+        let result = match (default_exists, secondary_exists) {
+            (true, false) => Ok(ModulePathSuccess {
+                path: default_path,
+                directory_ownership: DirectoryOwnership::Owned {
+                    relative: Some(id),
+                },
+            }),
+            (false, true) => Ok(ModulePathSuccess {
+                path: secondary_path,
+                directory_ownership: DirectoryOwnership::Owned {
+                    relative: None,
+                },
+            }),
+            (false, false) => Err(Error::FileNotFoundForModule {
+                mod_name: mod_name.clone(),
+                default_path: default_path_str,
+                secondary_path: secondary_path_str,
+                dir_path: dir_path.display().to_string(),
+            }),
+            (true, true) => Err(Error::DuplicatePaths {
+                mod_name: mod_name.clone(),
+                default_path: default_path_str,
+                secondary_path: secondary_path_str,
+            }),
+        };
+
+        ModulePath {
+            name: mod_name,
+            path_exists: default_exists || secondary_exists,
+            result,
+        }
+    }
+
+    /// Reads a module from a source file.
+    fn eval_src_mod(
+        &mut self,
+        path: PathBuf,
+        directory_ownership: DirectoryOwnership,
+        name: String,
+        id_sp: Span,
+    ) -> PResult<'a, (Mod, Vec<Attribute>)> {
+        let mut included_mod_stack = self.sess.included_mod_stack.borrow_mut();
+        if let Some(i) = included_mod_stack.iter().position(|p| *p == path) {
+            let mut err = String::from("circular modules: ");
+            let len = included_mod_stack.len();
+            for p in &included_mod_stack[i.. len] {
+                err.push_str(&p.to_string_lossy());
+                err.push_str(" -> ");
+            }
+            err.push_str(&path.to_string_lossy());
+            return Err(self.span_fatal(id_sp, &err[..]));
+        }
+        included_mod_stack.push(path.clone());
+        drop(included_mod_stack);
+
+        let mut p0 =
+            new_sub_parser_from_file(self.sess, &path, directory_ownership, Some(name), id_sp);
+        p0.cfg_mods = self.cfg_mods;
+        let mod_inner_lo = p0.token.span;
+        let mod_attrs = p0.parse_inner_attributes()?;
+        let mut m0 = p0.parse_mod_items(&token::Eof, mod_inner_lo)?;
+        m0.inline = false;
+        self.sess.included_mod_stack.borrow_mut().pop();
+        Ok((m0, mod_attrs))
+    }
+
+    fn push_directory(&mut self, id: Ident, attrs: &[Attribute]) {
+        if let Some(path) = attr::first_attr_value_str_by_name(attrs, sym::path) {
+            self.directory.path.to_mut().push(&*path.as_str());
+            self.directory.ownership = DirectoryOwnership::Owned { relative: None };
+        } else {
+            // We have to push on the current module name in the case of relative
+            // paths in order to ensure that any additional module paths from inline
+            // `mod x { ... }` come after the relative extension.
+            //
+            // For example, a `mod z { ... }` inside `x/y.rs` should set the current
+            // directory path to `/x/y/z`, not `/x/z` with a relative offset of `y`.
+            if let DirectoryOwnership::Owned { relative } = &mut self.directory.ownership {
+                if let Some(ident) = relative.take() { // remove the relative offset
+                    self.directory.path.to_mut().push(&*ident.as_str());
+                }
+            }
+            self.directory.path.to_mut().push(&*id.as_str());
+        }
+    }
+}

src/librustc_parse/parser/path.rs

@@ -0,0 +1,497 @@
+use super::{Parser, TokenType};
+use crate::maybe_whole;
+use syntax::ast::{self, QSelf, Path, PathSegment, Ident, ParenthesizedArgs, AngleBracketedArgs};
+use syntax::ast::{AnonConst, GenericArg, AssocTyConstraint, AssocTyConstraintKind, BlockCheckMode};
+use syntax::ThinVec;
+use syntax::token::{self, Token};
+use syntax::source_map::{Span, BytePos};
+use syntax_pos::symbol::{kw, sym};
+
+use std::mem;
+use log::debug;
+use errors::{PResult, Applicability, pluralize};
+
+/// Specifies how to parse a path.
+#[derive(Copy, Clone, PartialEq)]
+pub enum PathStyle {
+    /// In some contexts, notably in expressions, paths with generic arguments are ambiguous
+    /// with something else. For example, in expressions `segment < ....` can be interpreted
+    /// as a comparison and `segment ( ....` can be interpreted as a function call.
+    /// In all such contexts the non-path interpretation is preferred by default for practical
+    /// reasons, but the path interpretation can be forced by the disambiguator `::`, e.g.
+    /// `x<y>` - comparisons, `x::<y>` - unambiguously a path.
+    Expr,
+    /// In other contexts, notably in types, no ambiguity exists and paths can be written
+    /// without the disambiguator, e.g., `x<y>` - unambiguously a path.
+    /// Paths with disambiguators are still accepted, `x::<Y>` - unambiguously a path too.
+    Type,
+    /// A path with generic arguments disallowed, e.g., `foo::bar::Baz`, used in imports,
+    /// visibilities or attributes.
+    /// Technically, this variant is unnecessary and e.g., `Expr` can be used instead
+    /// (paths in "mod" contexts have to be checked later for absence of generic arguments
+    /// anyway, due to macros), but it is used to avoid weird suggestions about expected
+    /// tokens when something goes wrong.
+    Mod,
+}
+
+impl<'a> Parser<'a> {
+    /// Parses a qualified path.
+    /// Assumes that the leading `<` has been parsed already.
+    ///
+    /// `qualified_path = <type [as trait_ref]>::path`
+    ///
+    /// # Examples
+    /// `<T>::default`
+    /// `<T as U>::a`
+    /// `<T as U>::F::a<S>` (without disambiguator)
+    /// `<T as U>::F::a::<S>` (with disambiguator)
+    pub(super) fn parse_qpath(&mut self, style: PathStyle) -> PResult<'a, (QSelf, Path)> {
+        let lo = self.prev_span;
+        let ty = self.parse_ty()?;
+
+        // `path` will contain the prefix of the path up to the `>`,
+        // if any (e.g., `U` in the `<T as U>::*` examples
+        // above). `path_span` has the span of that path, or an empty
+        // span in the case of something like `<T>::Bar`.
+        let (mut path, path_span);
+        if self.eat_keyword(kw::As) {
+            let path_lo = self.token.span;
+            path = self.parse_path(PathStyle::Type)?;
+            path_span = path_lo.to(self.prev_span);
+        } else {
+            path_span = self.token.span.to(self.token.span);
+            path = ast::Path { segments: Vec::new(), span: path_span };
+        }
+
+        // See doc comment for `unmatched_angle_bracket_count`.
+        self.expect(&token::Gt)?;
+        if self.unmatched_angle_bracket_count > 0 {
+            self.unmatched_angle_bracket_count -= 1;
+            debug!("parse_qpath: (decrement) count={:?}", self.unmatched_angle_bracket_count);
+        }
+
+        self.expect(&token::ModSep)?;
+
+        let qself = QSelf { ty, path_span, position: path.segments.len() };
+        self.parse_path_segments(&mut path.segments, style)?;
+
+        Ok((qself, Path { segments: path.segments, span: lo.to(self.prev_span) }))
+    }
+
+    /// Parses simple paths.
+    ///
+    /// `path = [::] segment+`
+    /// `segment = ident | ident[::]<args> | ident[::](args) [-> type]`
+    ///
+    /// # Examples
+    /// `a::b::C<D>` (without disambiguator)
+    /// `a::b::C::<D>` (with disambiguator)
+    /// `Fn(Args)` (without disambiguator)
+    /// `Fn::(Args)` (with disambiguator)
+    pub fn parse_path(&mut self, style: PathStyle) -> PResult<'a, Path> {
+        maybe_whole!(self, NtPath, |path| {
+            if style == PathStyle::Mod &&
+               path.segments.iter().any(|segment| segment.args.is_some()) {
+                self.diagnostic().span_err(path.span, "unexpected generic arguments in path");
+            }
+            path
+        });
+
+        let lo = self.meta_var_span.unwrap_or(self.token.span);
+        let mut segments = Vec::new();
+        let mod_sep_ctxt = self.token.span.ctxt();
+        if self.eat(&token::ModSep) {
+            segments.push(PathSegment::path_root(lo.shrink_to_lo().with_ctxt(mod_sep_ctxt)));
+        }
+        self.parse_path_segments(&mut segments, style)?;
+
+        Ok(Path { segments, span: lo.to(self.prev_span) })
+    }
+
+    /// Like `parse_path`, but also supports parsing `Word` meta items into paths for
+    /// backwards-compatibility. This is used when parsing derive macro paths in `#[derive]`
+    /// attributes.
+    fn parse_path_allowing_meta(&mut self, style: PathStyle) -> PResult<'a, Path> {
+        let meta_ident = match self.token.kind {
+            token::Interpolated(ref nt) => match **nt {
+                token::NtMeta(ref item) => match item.tokens.is_empty() {
+                    true => Some(item.path.clone()),
+                    false => None,
+                },
+                _ => None,
+            },
+            _ => None,
+        };
+        if let Some(path) = meta_ident {
+            self.bump();
+            return Ok(path);
+        }
+        self.parse_path(style)
+    }
+
+    /// Parse a list of paths inside `#[derive(path_0, ..., path_n)]`.
+    pub fn parse_derive_paths(&mut self) -> PResult<'a, Vec<Path>> {
+        self.expect(&token::OpenDelim(token::Paren))?;
+        let mut list = Vec::new();
+        while !self.eat(&token::CloseDelim(token::Paren)) {
+            let path = self.parse_path_allowing_meta(PathStyle::Mod)?;
+            list.push(path);
+            if !self.eat(&token::Comma) {
+                self.expect(&token::CloseDelim(token::Paren))?;
+                break
+            }
+        }
+        Ok(list)
+    }
+
+    pub(super) fn parse_path_segments(
+        &mut self,
+        segments: &mut Vec<PathSegment>,
+        style: PathStyle,
+    ) -> PResult<'a, ()> {
+        loop {
+            let segment = self.parse_path_segment(style)?;
+            if style == PathStyle::Expr {
+                // In order to check for trailing angle brackets, we must have finished
+                // recursing (`parse_path_segment` can indirectly call this function),
+                // that is, the next token must be the highlighted part of the below example:
+                //
+                // `Foo::<Bar as Baz<T>>::Qux`
+                //                      ^ here
+                //
+                // As opposed to the below highlight (if we had only finished the first
+                // recursion):
+                //
+                // `Foo::<Bar as Baz<T>>::Qux`
+                //                     ^ here
+                //
+                // `PathStyle::Expr` is only provided at the root invocation and never in
+                // `parse_path_segment` to recurse and therefore can be checked to maintain
+                // this invariant.
+                self.check_trailing_angle_brackets(&segment, token::ModSep);
+            }
+            segments.push(segment);
+
+            if self.is_import_coupler() || !self.eat(&token::ModSep) {
+                return Ok(());
+            }
+        }
+    }
+
+    pub(super) fn parse_path_segment(&mut self, style: PathStyle) -> PResult<'a, PathSegment> {
+        let ident = self.parse_path_segment_ident()?;
+
+        let is_args_start = |token: &Token| match token.kind {
+            token::Lt | token::BinOp(token::Shl) | token::OpenDelim(token::Paren)
+            | token::LArrow => true,
+            _ => false,
+        };
+        let check_args_start = |this: &mut Self| {
+            this.expected_tokens.extend_from_slice(
+                &[TokenType::Token(token::Lt), TokenType::Token(token::OpenDelim(token::Paren))]
+            );
+            is_args_start(&this.token)
+        };
+
+        Ok(if style == PathStyle::Type && check_args_start(self) ||
+              style != PathStyle::Mod && self.check(&token::ModSep)
+                                      && self.look_ahead(1, |t| is_args_start(t)) {
+            // We use `style == PathStyle::Expr` to check if this is in a recursion or not. If
+            // it isn't, then we reset the unmatched angle bracket count as we're about to start
+            // parsing a new path.
+            if style == PathStyle::Expr {
+                self.unmatched_angle_bracket_count = 0;
+                self.max_angle_bracket_count = 0;
+            }
+
+            // Generic arguments are found - `<`, `(`, `::<` or `::(`.
+            self.eat(&token::ModSep);
+            let lo = self.token.span;
+            let args = if self.eat_lt() {
+                // `<'a, T, A = U>`
+                let (args, constraints) =
+                    self.parse_generic_args_with_leaning_angle_bracket_recovery(style, lo)?;
+                self.expect_gt()?;
+                let span = lo.to(self.prev_span);
+                AngleBracketedArgs { args, constraints, span }.into()
+            } else {
+                // `(T, U) -> R`
+                let (inputs, _) = self.parse_paren_comma_seq(|p| p.parse_ty())?;
+                let span = ident.span.to(self.prev_span);
+                let output = if self.eat(&token::RArrow) {
+                    Some(self.parse_ty_common(false, false, false)?)
+                } else {
+                    None
+                };
+                ParenthesizedArgs { inputs, output, span }.into()
+            };
+
+            PathSegment { ident, args, id: ast::DUMMY_NODE_ID }
+        } else {
+            // Generic arguments are not found.
+            PathSegment::from_ident(ident)
+        })
+    }
+
+    pub(super) fn parse_path_segment_ident(&mut self) -> PResult<'a, Ident> {
+        match self.token.kind {
+            token::Ident(name, _) if name.is_path_segment_keyword() => {
+                let span = self.token.span;
+                self.bump();
+                Ok(Ident::new(name, span))
+            }
+            _ => self.parse_ident(),
+        }
+    }
+
+    /// Parses generic args (within a path segment) with recovery for extra leading angle brackets.
+    /// For the purposes of understanding the parsing logic of generic arguments, this function
+    /// can be thought of being the same as just calling `self.parse_generic_args()` if the source
+    /// had the correct amount of leading angle brackets.
+    ///
+    /// ```ignore (diagnostics)
+    /// bar::<<<<T as Foo>::Output>();
+    ///      ^^ help: remove extra angle brackets
+    /// ```
+    fn parse_generic_args_with_leaning_angle_bracket_recovery(
+        &mut self,
+        style: PathStyle,
+        lo: Span,
+    ) -> PResult<'a, (Vec<GenericArg>, Vec<AssocTyConstraint>)> {
+        // We need to detect whether there are extra leading left angle brackets and produce an
+        // appropriate error and suggestion. This cannot be implemented by looking ahead at
+        // upcoming tokens for a matching `>` character - if there are unmatched `<` tokens
+        // then there won't be matching `>` tokens to find.
+        //
+        // To explain how this detection works, consider the following example:
+        //
+        // ```ignore (diagnostics)
+        // bar::<<<<T as Foo>::Output>();
+        //      ^^ help: remove extra angle brackets
+        // ```
+        //
+        // Parsing of the left angle brackets starts in this function. We start by parsing the
+        // `<` token (incrementing the counter of unmatched angle brackets on `Parser` via
+        // `eat_lt`):
+        //
+        // *Upcoming tokens:* `<<<<T as Foo>::Output>;`
+        // *Unmatched count:* 1
+        // *`parse_path_segment` calls deep:* 0
+        //
+        // This has the effect of recursing as this function is called if a `<` character
+        // is found within the expected generic arguments:
+        //
+        // *Upcoming tokens:* `<<<T as Foo>::Output>;`
+        // *Unmatched count:* 2
+        // *`parse_path_segment` calls deep:* 1
+        //
+        // Eventually we will have recursed until having consumed all of the `<` tokens and
+        // this will be reflected in the count:
+        //
+        // *Upcoming tokens:* `T as Foo>::Output>;`
+        // *Unmatched count:* 4
+        // `parse_path_segment` calls deep:* 3
+        //
+        // The parser will continue until reaching the first `>` - this will decrement the
+        // unmatched angle bracket count and return to the parent invocation of this function
+        // having succeeded in parsing:
+        //
+        // *Upcoming tokens:* `::Output>;`
+        // *Unmatched count:* 3
+        // *`parse_path_segment` calls deep:* 2
+        //
+        // This will continue until the next `>` character which will also return successfully
+        // to the parent invocation of this function and decrement the count:
+        //
+        // *Upcoming tokens:* `;`
+        // *Unmatched count:* 2
+        // *`parse_path_segment` calls deep:* 1
+        //
+        // At this point, this function will expect to find another matching `>` character but
+        // won't be able to and will return an error. This will continue all the way up the
+        // call stack until the first invocation:
+        //
+        // *Upcoming tokens:* `;`
+        // *Unmatched count:* 2
+        // *`parse_path_segment` calls deep:* 0
+        //
+        // In doing this, we have managed to work out how many unmatched leading left angle
+        // brackets there are, but we cannot recover as the unmatched angle brackets have
+        // already been consumed. To remedy this, we keep a snapshot of the parser state
+        // before we do the above. We can then inspect whether we ended up with a parsing error
+        // and unmatched left angle brackets and if so, restore the parser state before we
+        // consumed any `<` characters to emit an error and consume the erroneous tokens to
+        // recover by attempting to parse again.
+        //
+        // In practice, the recursion of this function is indirect and there will be other
+        // locations that consume some `<` characters - as long as we update the count when
+        // this happens, it isn't an issue.
+
+        let is_first_invocation = style == PathStyle::Expr;
+        // Take a snapshot before attempting to parse - we can restore this later.
+        let snapshot = if is_first_invocation {
+            Some(self.clone())
+        } else {
+            None
+        };
+
+        debug!("parse_generic_args_with_leading_angle_bracket_recovery: (snapshotting)");
+        match self.parse_generic_args() {
+            Ok(value) => Ok(value),
+            Err(ref mut e) if is_first_invocation && self.unmatched_angle_bracket_count > 0 => {
+                // Cancel error from being unable to find `>`. We know the error
+                // must have been this due to a non-zero unmatched angle bracket
+                // count.
+                e.cancel();
+
+                // Swap `self` with our backup of the parser state before attempting to parse
+                // generic arguments.
+                let snapshot = mem::replace(self, snapshot.unwrap());
+
+                debug!(
+                    "parse_generic_args_with_leading_angle_bracket_recovery: (snapshot failure) \
+                     snapshot.count={:?}",
+                    snapshot.unmatched_angle_bracket_count,
+                );
+
+                // Eat the unmatched angle brackets.
+                for _ in 0..snapshot.unmatched_angle_bracket_count {
+                    self.eat_lt();
+                }
+
+                // Make a span over ${unmatched angle bracket count} characters.
+                let span = lo.with_hi(
+                    lo.lo() + BytePos(snapshot.unmatched_angle_bracket_count)
+                );
+                self.diagnostic()
+                    .struct_span_err(
+                        span,
+                        &format!(
+                            "unmatched angle bracket{}",
+                            pluralize!(snapshot.unmatched_angle_bracket_count)
+                        ),
+                    )
+                    .span_suggestion(
+                        span,
+                        &format!(
+                            "remove extra angle bracket{}",
+                            pluralize!(snapshot.unmatched_angle_bracket_count)
+                        ),
+                        String::new(),
+                        Applicability::MachineApplicable,
+                    )
+                    .emit();
+
+                // Try again without unmatched angle bracket characters.
+                self.parse_generic_args()
+            },
+            Err(e) => Err(e),
+        }
+    }
+
+    /// Parses (possibly empty) list of lifetime and type arguments and associated type bindings,
+    /// possibly including trailing comma.
+    fn parse_generic_args(&mut self) -> PResult<'a, (Vec<GenericArg>, Vec<AssocTyConstraint>)> {
+        let mut args = Vec::new();
+        let mut constraints = Vec::new();
+        let mut misplaced_assoc_ty_constraints: Vec<Span> = Vec::new();
+        let mut assoc_ty_constraints: Vec<Span> = Vec::new();
+
+        let args_lo = self.token.span;
+
+        loop {
+            if self.check_lifetime() && self.look_ahead(1, |t| !t.is_like_plus()) {
+                // Parse lifetime argument.
+                args.push(GenericArg::Lifetime(self.expect_lifetime()));
+                misplaced_assoc_ty_constraints.append(&mut assoc_ty_constraints);
+            } else if self.check_ident()
+                && self.look_ahead(1, |t| t == &token::Eq || t == &token::Colon)
+            {
+                // Parse associated type constraint.
+                let lo = self.token.span;
+                let ident = self.parse_ident()?;
+                let kind = if self.eat(&token::Eq) {
+                    AssocTyConstraintKind::Equality {
+                        ty: self.parse_ty()?,
+                    }
+                } else if self.eat(&token::Colon) {
+                    AssocTyConstraintKind::Bound {
+                        bounds: self.parse_generic_bounds(Some(self.prev_span))?,
+                    }
+                } else {
+                    unreachable!();
+                };
+
+                let span = lo.to(self.prev_span);
+
+                // Gate associated type bounds, e.g., `Iterator<Item: Ord>`.
+                if let AssocTyConstraintKind::Bound { .. } = kind {
+                    self.sess.gated_spans.gate(sym::associated_type_bounds, span);
+                }
+
+                constraints.push(AssocTyConstraint {
+                    id: ast::DUMMY_NODE_ID,
+                    ident,
+                    kind,
+                    span,
+                });
+                assoc_ty_constraints.push(span);
+            } else if self.check_const_arg() {
+                // Parse const argument.
+                let expr = if let token::OpenDelim(token::Brace) = self.token.kind {
+                    self.parse_block_expr(
+                        None, self.token.span, BlockCheckMode::Default, ThinVec::new()
+                    )?
+                } else if self.token.is_ident() {
+                    // FIXME(const_generics): to distinguish between idents for types and consts,
+                    // we should introduce a GenericArg::Ident in the AST and distinguish when
+                    // lowering to the HIR. For now, idents for const args are not permitted.
+                    if self.token.is_bool_lit() {
+                        self.parse_literal_maybe_minus()?
+                    } else {
+                        return Err(
+                            self.fatal("identifiers may currently not be used for const generics")
+                        );
+                    }
+                } else {
+                    self.parse_literal_maybe_minus()?
+                };
+                let value = AnonConst {
+                    id: ast::DUMMY_NODE_ID,
+                    value: expr,
+                };
+                args.push(GenericArg::Const(value));
+                misplaced_assoc_ty_constraints.append(&mut assoc_ty_constraints);
+            } else if self.check_type() {
+                // Parse type argument.
+                args.push(GenericArg::Type(self.parse_ty()?));
+                misplaced_assoc_ty_constraints.append(&mut assoc_ty_constraints);
+            } else {
+                break
+            }
+
+            if !self.eat(&token::Comma) {
+                break
+            }
+        }
+
+        // FIXME: we would like to report this in ast_validation instead, but we currently do not
+        // preserve ordering of generic parameters with respect to associated type binding, so we
+        // lose that information after parsing.
+        if misplaced_assoc_ty_constraints.len() > 0 {
+            let mut err = self.struct_span_err(
+                args_lo.to(self.prev_span),
+                "associated type bindings must be declared after generic parameters",
+            );
+            for span in misplaced_assoc_ty_constraints {
+                err.span_label(
+                    span,
+                    "this associated type binding should be moved after the generic parameters",
+                );
+            }
+            err.emit();
+        }
+
+        Ok((args, constraints))
+    }
+}

src/librustc_parse/parser/stmt.rs

@@ -0,0 +1,482 @@
+use super::{Parser, Restrictions, PrevTokenKind, SemiColonMode, BlockMode};
+use super::expr::LhsExpr;
+use super::path::PathStyle;
+use super::pat::GateOr;
+use super::diagnostics::Error;
+use crate::maybe_whole;
+use crate::DirectoryOwnership;
+
+use syntax::ThinVec;
+use syntax::ptr::P;
+use syntax::ast;
+use syntax::ast::{DUMMY_NODE_ID, Stmt, StmtKind, Local, Block, BlockCheckMode, Expr, ExprKind};
+use syntax::ast::{Attribute, AttrStyle, VisibilityKind, MacStmtStyle, Mac, MacDelimiter};
+use syntax::util::classify;
+use syntax::token;
+use syntax::source_map::{respan, Span};
+use syntax::symbol::{kw, sym};
+
+use std::mem;
+use errors::{PResult, Applicability};
+
+impl<'a> Parser<'a> {
+    /// Parses a statement. This stops just before trailing semicolons on everything but items.
+    /// e.g., a `StmtKind::Semi` parses to a `StmtKind::Expr`, leaving the trailing `;` unconsumed.
+    pub fn parse_stmt(&mut self) -> PResult<'a, Option<Stmt>> {
+        Ok(self.parse_stmt_(true))
+    }
+
+    fn parse_stmt_(&mut self, macro_legacy_warnings: bool) -> Option<Stmt> {
+        self.parse_stmt_without_recovery(macro_legacy_warnings).unwrap_or_else(|mut e| {
+            e.emit();
+            self.recover_stmt_(SemiColonMode::Break, BlockMode::Ignore);
+            None
+        })
+    }
+
+    fn parse_stmt_without_recovery(
+        &mut self,
+        macro_legacy_warnings: bool,
+    ) -> PResult<'a, Option<Stmt>> {
+        maybe_whole!(self, NtStmt, |x| Some(x));
+
+        let attrs = self.parse_outer_attributes()?;
+        let lo = self.token.span;
+
+        Ok(Some(if self.eat_keyword(kw::Let) {
+            Stmt {
+                id: DUMMY_NODE_ID,
+                kind: StmtKind::Local(self.parse_local(attrs.into())?),
+                span: lo.to(self.prev_span),
+            }
+        } else if let Some(macro_def) = self.eat_macro_def(
+            &attrs,
+            &respan(lo, VisibilityKind::Inherited),
+            lo,
+        )? {
+            Stmt {
+                id: DUMMY_NODE_ID,
+                kind: StmtKind::Item(macro_def),
+                span: lo.to(self.prev_span),
+            }
+        // Starts like a simple path, being careful to avoid contextual keywords
+        // such as a union items, item with `crate` visibility or auto trait items.
+        // Our goal here is to parse an arbitrary path `a::b::c` but not something that starts
+        // like a path (1 token), but it fact not a path.
+        // `union::b::c` - path, `union U { ... }` - not a path.
+        // `crate::b::c` - path, `crate struct S;` - not a path.
+        } else if self.token.is_path_start() &&
+                  !self.token.is_qpath_start() &&
+                  !self.is_union_item() &&
+                  !self.is_crate_vis() &&
+                  !self.is_auto_trait_item() &&
+                  !self.is_async_fn() {
+            let path = self.parse_path(PathStyle::Expr)?;
+
+            if !self.eat(&token::Not) {
+                let expr = if self.check(&token::OpenDelim(token::Brace)) {
+                    self.parse_struct_expr(lo, path, ThinVec::new())?
+                } else {
+                    let hi = self.prev_span;
+                    self.mk_expr(lo.to(hi), ExprKind::Path(None, path), ThinVec::new())
+                };
+
+                let expr = self.with_res(Restrictions::STMT_EXPR, |this| {
+                    let expr = this.parse_dot_or_call_expr_with(expr, lo, attrs.into())?;
+                    this.parse_assoc_expr_with(0, LhsExpr::AlreadyParsed(expr))
+                })?;
+
+                return Ok(Some(Stmt {
+                    id: DUMMY_NODE_ID,
+                    kind: StmtKind::Expr(expr),
+                    span: lo.to(self.prev_span),
+                }));
+            }
+
+            let (delim, tts) = self.expect_delimited_token_tree()?;
+            let hi = self.prev_span;
+
+            let style = if delim == MacDelimiter::Brace {
+                MacStmtStyle::Braces
+            } else {
+                MacStmtStyle::NoBraces
+            };
+
+            let mac = Mac {
+                path,
+                tts,
+                delim,
+                span: lo.to(hi),
+                prior_type_ascription: self.last_type_ascription,
+            };
+            let kind = if delim == MacDelimiter::Brace ||
+                          self.token == token::Semi || self.token == token::Eof {
+                StmtKind::Mac(P((mac, style, attrs.into())))
+            }
+            // We used to incorrectly stop parsing macro-expanded statements here.
+            // If the next token will be an error anyway but could have parsed with the
+            // earlier behavior, stop parsing here and emit a warning to avoid breakage.
+            else if macro_legacy_warnings && self.token.can_begin_expr() &&
+                match self.token.kind {
+                    // These can continue an expression, so we can't stop parsing and warn.
+                    token::OpenDelim(token::Paren) | token::OpenDelim(token::Bracket) |
+                    token::BinOp(token::Minus) | token::BinOp(token::Star) |
+                    token::BinOp(token::And) | token::BinOp(token::Or) |
+                    token::AndAnd | token::OrOr |
+                    token::DotDot | token::DotDotDot | token::DotDotEq => false,
+                    _ => true,
+                }
+            {
+                self.warn_missing_semicolon();
+                StmtKind::Mac(P((mac, style, attrs.into())))
+            } else {
+                let e = self.mk_expr(mac.span, ExprKind::Mac(mac), ThinVec::new());
+                let e = self.maybe_recover_from_bad_qpath(e, true)?;
+                let e = self.parse_dot_or_call_expr_with(e, lo, attrs.into())?;
+                let e = self.parse_assoc_expr_with(0, LhsExpr::AlreadyParsed(e))?;
+                StmtKind::Expr(e)
+            };
+            Stmt {
+                id: DUMMY_NODE_ID,
+                span: lo.to(hi),
+                kind,
+            }
+        } else {
+            // FIXME: Bad copy of attrs
+            let old_directory_ownership =
+                mem::replace(&mut self.directory.ownership, DirectoryOwnership::UnownedViaBlock);
+            let item = self.parse_item_(attrs.clone(), false, true)?;
+            self.directory.ownership = old_directory_ownership;
+
+            match item {
+                Some(i) => Stmt {
+                    id: DUMMY_NODE_ID,
+                    span: lo.to(i.span),
+                    kind: StmtKind::Item(i),
+                },
+                None => {
+                    let unused_attrs = |attrs: &[Attribute], s: &mut Self| {
+                        if !attrs.is_empty() {
+                            if s.prev_token_kind == PrevTokenKind::DocComment {
+                                s.span_fatal_err(s.prev_span, Error::UselessDocComment).emit();
+                            } else if attrs.iter().any(|a| a.style == AttrStyle::Outer) {
+                                s.span_err(
+                                    s.token.span, "expected statement after outer attribute"
+                                );
+                            }
+                        }
+                    };
+
+                    // Do not attempt to parse an expression if we're done here.
+                    if self.token == token::Semi {
+                        unused_attrs(&attrs, self);
+                        self.bump();
+                        let mut last_semi = lo;
+                        while self.token == token::Semi {
+                            last_semi = self.token.span;
+                            self.bump();
+                        }
+                        // We are encoding a string of semicolons as an
+                        // an empty tuple that spans the excess semicolons
+                        // to preserve this info until the lint stage
+                        return Ok(Some(Stmt {
+                            id: DUMMY_NODE_ID,
+                            span: lo.to(last_semi),
+                            kind: StmtKind::Semi(self.mk_expr(lo.to(last_semi),
+                                ExprKind::Tup(Vec::new()),
+                                ThinVec::new()
+                            )),
+                        }));
+                    }
+
+                    if self.token == token::CloseDelim(token::Brace) {
+                        unused_attrs(&attrs, self);
+                        return Ok(None);
+                    }
+
+                    // Remainder are line-expr stmts.
+                    let e = self.parse_expr_res(
+                        Restrictions::STMT_EXPR, Some(attrs.into()))?;
+                    Stmt {
+                        id: DUMMY_NODE_ID,
+                        span: lo.to(e.span),
+                        kind: StmtKind::Expr(e),
+                    }
+                }
+            }
+        }))
+    }
+
+    /// Parses a local variable declaration.
+    fn parse_local(&mut self, attrs: ThinVec<Attribute>) -> PResult<'a, P<Local>> {
+        let lo = self.prev_span;
+        let pat = self.parse_top_pat(GateOr::Yes)?;
+
+        let (err, ty) = if self.eat(&token::Colon) {
+            // Save the state of the parser before parsing type normally, in case there is a `:`
+            // instead of an `=` typo.
+            let parser_snapshot_before_type = self.clone();
+            let colon_sp = self.prev_span;
+            match self.parse_ty() {
+                Ok(ty) => (None, Some(ty)),
+                Err(mut err) => {
+                    // Rewind to before attempting to parse the type and continue parsing.
+                    let parser_snapshot_after_type = self.clone();
+                    mem::replace(self, parser_snapshot_before_type);
+
+                    let snippet = self.span_to_snippet(pat.span).unwrap();
+                    err.span_label(pat.span, format!("while parsing the type for `{}`", snippet));
+                    (Some((parser_snapshot_after_type, colon_sp, err)), None)
+                }
+            }
+        } else {
+            (None, None)
+        };
+        let init = match (self.parse_initializer(err.is_some()), err) {
+            (Ok(init), None) => {  // init parsed, ty parsed
+                init
+            }
+            (Ok(init), Some((_, colon_sp, mut err))) => {  // init parsed, ty error
+                // Could parse the type as if it were the initializer, it is likely there was a
+                // typo in the code: `:` instead of `=`. Add suggestion and emit the error.
+                err.span_suggestion_short(
+                    colon_sp,
+                    "use `=` if you meant to assign",
+                    " =".to_string(),
+                    Applicability::MachineApplicable
+                );
+                err.emit();
+                // As this was parsed successfully, continue as if the code has been fixed for the
+                // rest of the file. It will still fail due to the emitted error, but we avoid
+                // extra noise.
+                init
+            }
+            (Err(mut init_err), Some((snapshot, _, ty_err))) => {  // init error, ty error
+                init_err.cancel();
+                // Couldn't parse the type nor the initializer, only raise the type error and
+                // return to the parser state before parsing the type as the initializer.
+                // let x: <parse_error>;
+                mem::replace(self, snapshot);
+                return Err(ty_err);
+            }
+            (Err(err), None) => {  // init error, ty parsed
+                // Couldn't parse the initializer and we're not attempting to recover a failed
+                // parse of the type, return the error.
+                return Err(err);
+            }
+        };
+        let hi = if self.token == token::Semi {
+            self.token.span
+        } else {
+            self.prev_span
+        };
+        Ok(P(ast::Local {
+            ty,
+            pat,
+            init,
+            id: DUMMY_NODE_ID,
+            span: lo.to(hi),
+            attrs,
+        }))
+    }
+
+    /// Parses the RHS of a local variable declaration (e.g., '= 14;').
+    fn parse_initializer(&mut self, skip_eq: bool) -> PResult<'a, Option<P<Expr>>> {
+        if self.eat(&token::Eq) {
+            Ok(Some(self.parse_expr()?))
+        } else if skip_eq {
+            Ok(Some(self.parse_expr()?))
+        } else {
+            Ok(None)
+        }
+    }
+
+    fn is_auto_trait_item(&self) -> bool {
+        // auto trait
+        (self.token.is_keyword(kw::Auto) &&
+            self.is_keyword_ahead(1, &[kw::Trait]))
+        || // unsafe auto trait
+        (self.token.is_keyword(kw::Unsafe) &&
+         self.is_keyword_ahead(1, &[kw::Auto]) &&
+         self.is_keyword_ahead(2, &[kw::Trait]))
+    }
+
+    /// Parses a block. No inner attributes are allowed.
+    pub fn parse_block(&mut self) -> PResult<'a, P<Block>> {
+        maybe_whole!(self, NtBlock, |x| x);
+
+        let lo = self.token.span;
+
+        if !self.eat(&token::OpenDelim(token::Brace)) {
+            let sp = self.token.span;
+            let tok = self.this_token_descr();
+            let mut e = self.span_fatal(sp, &format!("expected `{{`, found {}", tok));
+            let do_not_suggest_help =
+                self.token.is_keyword(kw::In) || self.token == token::Colon;
+
+            if self.token.is_ident_named(sym::and) {
+                e.span_suggestion_short(
+                    self.token.span,
+                    "use `&&` instead of `and` for the boolean operator",
+                    "&&".to_string(),
+                    Applicability::MaybeIncorrect,
+                );
+            }
+            if self.token.is_ident_named(sym::or) {
+                e.span_suggestion_short(
+                    self.token.span,
+                    "use `||` instead of `or` for the boolean operator",
+                    "||".to_string(),
+                    Applicability::MaybeIncorrect,
+                );
+            }
+
+            // Check to see if the user has written something like
+            //
+            //    if (cond)
+            //      bar;
+            //
+            // which is valid in other languages, but not Rust.
+            match self.parse_stmt_without_recovery(false) {
+                Ok(Some(stmt)) => {
+                    if self.look_ahead(1, |t| t == &token::OpenDelim(token::Brace))
+                        || do_not_suggest_help {
+                        // If the next token is an open brace (e.g., `if a b {`), the place-
+                        // inside-a-block suggestion would be more likely wrong than right.
+                        e.span_label(sp, "expected `{`");
+                        return Err(e);
+                    }
+                    let mut stmt_span = stmt.span;
+                    // Expand the span to include the semicolon, if it exists.
+                    if self.eat(&token::Semi) {
+                        stmt_span = stmt_span.with_hi(self.prev_span.hi());
+                    }
+                    if let Ok(snippet) = self.span_to_snippet(stmt_span) {
+                        e.span_suggestion(
+                            stmt_span,
+                            "try placing this code inside a block",
+                            format!("{{ {} }}", snippet),
+                            // Speculative; has been misleading in the past (#46836).
+                            Applicability::MaybeIncorrect,
+                        );
+                    }
+                }
+                Err(mut e) => {
+                    self.recover_stmt_(SemiColonMode::Break, BlockMode::Ignore);
+                    e.cancel();
+                }
+                _ => ()
+            }
+            e.span_label(sp, "expected `{`");
+            return Err(e);
+        }
+
+        self.parse_block_tail(lo, BlockCheckMode::Default)
+    }
+
+    /// Parses a block. Inner attributes are allowed.
+    pub(super) fn parse_inner_attrs_and_block(
+        &mut self
+    ) -> PResult<'a, (Vec<Attribute>, P<Block>)> {
+        maybe_whole!(self, NtBlock, |x| (Vec::new(), x));
+
+        let lo = self.token.span;
+        self.expect(&token::OpenDelim(token::Brace))?;
+        Ok((self.parse_inner_attributes()?,
+            self.parse_block_tail(lo, BlockCheckMode::Default)?))
+    }
+
+    /// Parses the rest of a block expression or function body.
+    /// Precondition: already parsed the '{'.
+    pub(super) fn parse_block_tail(
+        &mut self,
+        lo: Span,
+        s: BlockCheckMode
+    ) -> PResult<'a, P<Block>> {
+        let mut stmts = vec![];
+        while !self.eat(&token::CloseDelim(token::Brace)) {
+            if self.token == token::Eof {
+                break;
+            }
+            let stmt = match self.parse_full_stmt(false) {
+                Err(mut err) => {
+                    self.maybe_annotate_with_ascription(&mut err, false);
+                    err.emit();
+                    self.recover_stmt_(SemiColonMode::Ignore, BlockMode::Ignore);
+                    Some(Stmt {
+                        id: DUMMY_NODE_ID,
+                        kind: StmtKind::Expr(self.mk_expr_err(self.token.span)),
+                        span: self.token.span,
+                    })
+                }
+                Ok(stmt) => stmt,
+            };
+            if let Some(stmt) = stmt {
+                stmts.push(stmt);
+            } else {
+                // Found only `;` or `}`.
+                continue;
+            };
+        }
+        Ok(P(ast::Block {
+            stmts,
+            id: DUMMY_NODE_ID,
+            rules: s,
+            span: lo.to(self.prev_span),
+        }))
+    }
+
+    /// Parses a statement, including the trailing semicolon.
+    pub fn parse_full_stmt(&mut self, macro_legacy_warnings: bool) -> PResult<'a, Option<Stmt>> {
+        // Skip looking for a trailing semicolon when we have an interpolated statement.
+        maybe_whole!(self, NtStmt, |x| Some(x));
+
+        let mut stmt = match self.parse_stmt_without_recovery(macro_legacy_warnings)? {
+            Some(stmt) => stmt,
+            None => return Ok(None),
+        };
+
+        let mut eat_semi = true;
+        match stmt.kind {
+            StmtKind::Expr(ref expr) if self.token != token::Eof => {
+                // expression without semicolon
+                if classify::expr_requires_semi_to_be_stmt(expr) {
+                    // Just check for errors and recover; do not eat semicolon yet.
+                    if let Err(mut e) =
+                        self.expect_one_of(&[], &[token::Semi, token::CloseDelim(token::Brace)])
+                    {
+                        e.emit();
+                        self.recover_stmt();
+                        // Don't complain about type errors in body tail after parse error (#57383).
+                        let sp = expr.span.to(self.prev_span);
+                        stmt.kind = StmtKind::Expr(self.mk_expr_err(sp));
+                    }
+                }
+            }
+            StmtKind::Local(..) => {
+                // We used to incorrectly allow a macro-expanded let statement to lack a semicolon.
+                if macro_legacy_warnings && self.token != token::Semi {
+                    self.warn_missing_semicolon();
+                } else {
+                    self.expect_semi()?;
+                    eat_semi = false;
+                }
+            }
+            _ => {}
+        }
+
+        if eat_semi && self.eat(&token::Semi) {
+            stmt = stmt.add_trailing_semicolon();
+        }
+        stmt.span = stmt.span.to(self.prev_span);
+        Ok(Some(stmt))
+    }
+
+    fn warn_missing_semicolon(&self) {
+        self.diagnostic().struct_span_warn(self.token.span, {
+            &format!("expected `;`, found {}", self.this_token_descr())
+        }).note({
+            "this was erroneously allowed and will become a hard error in a future release"
+        }).emit();
+    }
+}

src/librustc_parse/parser/ty.rs

@@ -0,0 +1,460 @@
+use super::{Parser, PathStyle, PrevTokenKind, TokenType};
+use super::item::ParamCfg;
+
+use crate::{maybe_whole, maybe_recover_from_interpolated_ty_qpath};
+
+use syntax::ptr::P;
+use syntax::ast::{self, Ty, TyKind, MutTy, BareFnTy, FunctionRetTy, GenericParam, Lifetime, Ident};
+use syntax::ast::{TraitBoundModifier, TraitObjectSyntax, GenericBound, GenericBounds, PolyTraitRef};
+use syntax::ast::{Mutability, AnonConst, Mac};
+use syntax::token::{self, Token};
+use syntax::source_map::Span;
+use syntax::struct_span_fatal;
+use syntax_pos::symbol::kw;
+
+use errors::{PResult, Applicability, pluralize};
+
+/// Returns `true` if `IDENT t` can start a type -- `IDENT::a::b`, `IDENT<u8, u8>`,
+/// `IDENT<<u8 as Trait>::AssocTy>`.
+///
+/// Types can also be of the form `IDENT(u8, u8) -> u8`, however this assumes
+/// that `IDENT` is not the ident of a fn trait.
+fn can_continue_type_after_non_fn_ident(t: &Token) -> bool {
+    t == &token::ModSep || t == &token::Lt ||
+    t == &token::BinOp(token::Shl)
+}
+
+impl<'a> Parser<'a> {
+    /// Parses a type.
+    pub fn parse_ty(&mut self) -> PResult<'a, P<Ty>> {
+        self.parse_ty_common(true, true, false)
+    }
+
+    /// Parses a type in restricted contexts where `+` is not permitted.
+    ///
+    /// Example 1: `&'a TYPE`
+    ///     `+` is prohibited to maintain operator priority (P(+) < P(&)).
+    /// Example 2: `value1 as TYPE + value2`
+    ///     `+` is prohibited to avoid interactions with expression grammar.
+    pub(super) fn parse_ty_no_plus(&mut self) -> PResult<'a, P<Ty>> {
+        self.parse_ty_common(false, true, false)
+    }
+
+    /// Parses an optional return type `[ -> TY ]` in a function declaration.
+    pub(super) fn parse_ret_ty(&mut self, allow_plus: bool) -> PResult<'a, FunctionRetTy> {
+        if self.eat(&token::RArrow) {
+            Ok(FunctionRetTy::Ty(self.parse_ty_common(allow_plus, true, false)?))
+        } else {
+            Ok(FunctionRetTy::Default(self.token.span.shrink_to_lo()))
+        }
+    }
+
+    pub(super) fn parse_ty_common(&mut self, allow_plus: bool, allow_qpath_recovery: bool,
+                       allow_c_variadic: bool) -> PResult<'a, P<Ty>> {
+        maybe_recover_from_interpolated_ty_qpath!(self, allow_qpath_recovery);
+        maybe_whole!(self, NtTy, |x| x);
+
+        let lo = self.token.span;
+        let mut impl_dyn_multi = false;
+        let kind = if self.eat(&token::OpenDelim(token::Paren)) {
+            // `(TYPE)` is a parenthesized type.
+            // `(TYPE,)` is a tuple with a single field of type TYPE.
+            let mut ts = vec![];
+            let mut last_comma = false;
+            while self.token != token::CloseDelim(token::Paren) {
+                ts.push(self.parse_ty()?);
+                if self.eat(&token::Comma) {
+                    last_comma = true;
+                } else {
+                    last_comma = false;
+                    break;
+                }
+            }
+            let trailing_plus = self.prev_token_kind == PrevTokenKind::Plus;
+            self.expect(&token::CloseDelim(token::Paren))?;
+
+            if ts.len() == 1 && !last_comma {
+                let ty = ts.into_iter().nth(0).unwrap().into_inner();
+                let maybe_bounds = allow_plus && self.token.is_like_plus();
+                match ty.kind {
+                    // `(TY_BOUND_NOPAREN) + BOUND + ...`.
+                    TyKind::Path(None, ref path) if maybe_bounds => {
+                        self.parse_remaining_bounds(Vec::new(), path.clone(), lo, true)?
+                    }
+                    TyKind::TraitObject(ref bounds, TraitObjectSyntax::None)
+                            if maybe_bounds && bounds.len() == 1 && !trailing_plus => {
+                        let path = match bounds[0] {
+                            GenericBound::Trait(ref pt, ..) => pt.trait_ref.path.clone(),
+                            GenericBound::Outlives(..) => self.bug("unexpected lifetime bound"),
+                        };
+                        self.parse_remaining_bounds(Vec::new(), path, lo, true)?
+                    }
+                    // `(TYPE)`
+                    _ => TyKind::Paren(P(ty))
+                }
+            } else {
+                TyKind::Tup(ts)
+            }
+        } else if self.eat(&token::Not) {
+            // Never type `!`
+            TyKind::Never
+        } else if self.eat(&token::BinOp(token::Star)) {
+            // Raw pointer
+            TyKind::Ptr(self.parse_ptr()?)
+        } else if self.eat(&token::OpenDelim(token::Bracket)) {
+            // Array or slice
+            let t = self.parse_ty()?;
+            // Parse optional `; EXPR` in `[TYPE; EXPR]`
+            let t = match self.maybe_parse_fixed_length_of_vec()? {
+                None => TyKind::Slice(t),
+                Some(length) => TyKind::Array(t, AnonConst {
+                    id: ast::DUMMY_NODE_ID,
+                    value: length,
+                }),
+            };
+            self.expect(&token::CloseDelim(token::Bracket))?;
+            t
+        } else if self.check(&token::BinOp(token::And)) || self.check(&token::AndAnd) {
+            // Reference
+            self.expect_and()?;
+            self.parse_borrowed_pointee()?
+        } else if self.eat_keyword_noexpect(kw::Typeof) {
+            // `typeof(EXPR)`
+            // In order to not be ambiguous, the type must be surrounded by parens.
+            self.expect(&token::OpenDelim(token::Paren))?;
+            let e = AnonConst {
+                id: ast::DUMMY_NODE_ID,
+                value: self.parse_expr()?,
+            };
+            self.expect(&token::CloseDelim(token::Paren))?;
+            TyKind::Typeof(e)
+        } else if self.eat_keyword(kw::Underscore) {
+            // A type to be inferred `_`
+            TyKind::Infer
+        } else if self.token_is_bare_fn_keyword() {
+            // Function pointer type
+            self.parse_ty_bare_fn(Vec::new())?
+        } else if self.check_keyword(kw::For) {
+            // Function pointer type or bound list (trait object type) starting with a poly-trait.
+            //   `for<'lt> [unsafe] [extern "ABI"] fn (&'lt S) -> T`
+            //   `for<'lt> Trait1<'lt> + Trait2 + 'a`
+            let lo = self.token.span;
+            let lifetime_defs = self.parse_late_bound_lifetime_defs()?;
+            if self.token_is_bare_fn_keyword() {
+                self.parse_ty_bare_fn(lifetime_defs)?
+            } else {
+                let path = self.parse_path(PathStyle::Type)?;
+                let parse_plus = allow_plus && self.check_plus();
+                self.parse_remaining_bounds(lifetime_defs, path, lo, parse_plus)?
+            }
+        } else if self.eat_keyword(kw::Impl) {
+            // Always parse bounds greedily for better error recovery.
+            let bounds = self.parse_generic_bounds(None)?;
+            impl_dyn_multi = bounds.len() > 1 || self.prev_token_kind == PrevTokenKind::Plus;
+            TyKind::ImplTrait(ast::DUMMY_NODE_ID, bounds)
+        } else if self.check_keyword(kw::Dyn) &&
+                  (self.token.span.rust_2018() ||
+                   self.look_ahead(1, |t| t.can_begin_bound() &&
+                                          !can_continue_type_after_non_fn_ident(t))) {
+            self.bump(); // `dyn`
+            // Always parse bounds greedily for better error recovery.
+            let bounds = self.parse_generic_bounds(None)?;
+            impl_dyn_multi = bounds.len() > 1 || self.prev_token_kind == PrevTokenKind::Plus;
+            TyKind::TraitObject(bounds, TraitObjectSyntax::Dyn)
+        } else if self.check(&token::Question) ||
+                  self.check_lifetime() && self.look_ahead(1, |t| t.is_like_plus()) {
+            // Bound list (trait object type)
+            TyKind::TraitObject(self.parse_generic_bounds_common(allow_plus, None)?,
+                                TraitObjectSyntax::None)
+        } else if self.eat_lt() {
+            // Qualified path
+            let (qself, path) = self.parse_qpath(PathStyle::Type)?;
+            TyKind::Path(Some(qself), path)
+        } else if self.token.is_path_start() {
+            // Simple path
+            let path = self.parse_path(PathStyle::Type)?;
+            if self.eat(&token::Not) {
+                // Macro invocation in type position
+                let (delim, tts) = self.expect_delimited_token_tree()?;
+                let mac = Mac {
+                    path,
+                    tts,
+                    delim,
+                    span: lo.to(self.prev_span),
+                    prior_type_ascription: self.last_type_ascription,
+                };
+                TyKind::Mac(mac)
+            } else {
+                // Just a type path or bound list (trait object type) starting with a trait.
+                //   `Type`
+                //   `Trait1 + Trait2 + 'a`
+                if allow_plus && self.check_plus() {
+                    self.parse_remaining_bounds(Vec::new(), path, lo, true)?
+                } else {
+                    TyKind::Path(None, path)
+                }
+            }
+        } else if self.check(&token::DotDotDot) {
+            if allow_c_variadic {
+                self.eat(&token::DotDotDot);
+                TyKind::CVarArgs
+            } else {
+                return Err(struct_span_fatal!(
+                    self.sess.span_diagnostic,
+                    self.token.span,
+                    E0743,
+                    "only foreign functions are allowed to be C-variadic",
+                ));
+            }
+        } else {
+            let msg = format!("expected type, found {}", self.this_token_descr());
+            let mut err = self.fatal(&msg);
+            err.span_label(self.token.span, "expected type");
+            self.maybe_annotate_with_ascription(&mut err, true);
+            return Err(err);
+        };
+
+        let span = lo.to(self.prev_span);
+        let ty = self.mk_ty(span, kind);
+
+        // Try to recover from use of `+` with incorrect priority.
+        self.maybe_report_ambiguous_plus(allow_plus, impl_dyn_multi, &ty);
+        self.maybe_recover_from_bad_type_plus(allow_plus, &ty)?;
+        self.maybe_recover_from_bad_qpath(ty, allow_qpath_recovery)
+    }
+
+    fn parse_remaining_bounds(&mut self, generic_params: Vec<GenericParam>, path: ast::Path,
+                              lo: Span, parse_plus: bool) -> PResult<'a, TyKind> {
+        let poly_trait_ref = PolyTraitRef::new(generic_params, path, lo.to(self.prev_span));
+        let mut bounds = vec![GenericBound::Trait(poly_trait_ref, TraitBoundModifier::None)];
+        if parse_plus {
+            self.eat_plus(); // `+`, or `+=` gets split and `+` is discarded
+            bounds.append(&mut self.parse_generic_bounds(Some(self.prev_span))?);
+        }
+        Ok(TyKind::TraitObject(bounds, TraitObjectSyntax::None))
+    }
+
+    fn parse_ptr(&mut self) -> PResult<'a, MutTy> {
+        let mutbl = self.parse_const_or_mut().unwrap_or_else(|| {
+            let span = self.prev_span;
+            let msg = "expected mut or const in raw pointer type";
+            self.struct_span_err(span, msg)
+                .span_label(span, msg)
+                .help("use `*mut T` or `*const T` as appropriate")
+                .emit();
+            Mutability::Immutable
+        });
+        let t = self.parse_ty_no_plus()?;
+        Ok(MutTy { ty: t, mutbl })
+    }
+
+    fn maybe_parse_fixed_length_of_vec(&mut self) -> PResult<'a, Option<P<ast::Expr>>> {
+        if self.eat(&token::Semi) {
+            Ok(Some(self.parse_expr()?))
+        } else {
+            Ok(None)
+        }
+    }
+
+    fn parse_borrowed_pointee(&mut self) -> PResult<'a, TyKind> {
+        let opt_lifetime = if self.check_lifetime() { Some(self.expect_lifetime()) } else { None };
+        let mutbl = self.parse_mutability();
+        let ty = self.parse_ty_no_plus()?;
+        return Ok(TyKind::Rptr(opt_lifetime, MutTy { ty, mutbl }));
+    }
+
+    /// Is the current token one of the keywords that signals a bare function type?
+    fn token_is_bare_fn_keyword(&mut self) -> bool {
+        self.check_keyword(kw::Fn) ||
+            self.check_keyword(kw::Unsafe) ||
+            self.check_keyword(kw::Extern)
+    }
+
+    /// Parses a `TyKind::BareFn` type.
+    fn parse_ty_bare_fn(&mut self, generic_params: Vec<GenericParam>) -> PResult<'a, TyKind> {
+        /*
+
+        [unsafe] [extern "ABI"] fn (S) -> T
+         ^~~~^           ^~~~^     ^~^    ^
+           |               |        |     |
+           |               |        |   Return type
+           |               |      Argument types
+           |               |
+           |              ABI
+        Function Style
+        */
+
+        let unsafety = self.parse_unsafety();
+        let abi = self.parse_extern_abi()?;
+        self.expect_keyword(kw::Fn)?;
+        let cfg = ParamCfg {
+            is_self_allowed: false,
+            allow_c_variadic: true,
+            is_name_required: |_| false,
+        };
+        let decl = self.parse_fn_decl(cfg, false)?;
+        Ok(TyKind::BareFn(P(BareFnTy {
+            abi,
+            unsafety,
+            generic_params,
+            decl,
+        })))
+    }
+
+    pub(super) fn parse_generic_bounds(&mut self,
+                                  colon_span: Option<Span>) -> PResult<'a, GenericBounds> {
+        self.parse_generic_bounds_common(true, colon_span)
+    }
+
+    /// Parses bounds of a type parameter `BOUND + BOUND + ...`, possibly with trailing `+`.
+    ///
+    /// ```
+    /// BOUND = TY_BOUND | LT_BOUND
+    /// LT_BOUND = LIFETIME (e.g., `'a`)
+    /// TY_BOUND = TY_BOUND_NOPAREN | (TY_BOUND_NOPAREN)
+    /// TY_BOUND_NOPAREN = [?] [for<LT_PARAM_DEFS>] SIMPLE_PATH (e.g., `?for<'a: 'b> m::Trait<'a>`)
+    /// ```
+    fn parse_generic_bounds_common(&mut self,
+                                   allow_plus: bool,
+                                   colon_span: Option<Span>) -> PResult<'a, GenericBounds> {
+        let mut bounds = Vec::new();
+        let mut negative_bounds = Vec::new();
+        let mut last_plus_span = None;
+        let mut was_negative = false;
+        loop {
+            // This needs to be synchronized with `TokenKind::can_begin_bound`.
+            let is_bound_start = self.check_path() || self.check_lifetime() ||
+                                 self.check(&token::Not) || // used for error reporting only
+                                 self.check(&token::Question) ||
+                                 self.check_keyword(kw::For) ||
+                                 self.check(&token::OpenDelim(token::Paren));
+            if is_bound_start {
+                let lo = self.token.span;
+                let has_parens = self.eat(&token::OpenDelim(token::Paren));
+                let inner_lo = self.token.span;
+                let is_negative = self.eat(&token::Not);
+                let question = if self.eat(&token::Question) { Some(self.prev_span) } else { None };
+                if self.token.is_lifetime() {
+                    if let Some(question_span) = question {
+                        self.span_err(question_span,
+                                      "`?` may only modify trait bounds, not lifetime bounds");
+                    }
+                    bounds.push(GenericBound::Outlives(self.expect_lifetime()));
+                    if has_parens {
+                        let inner_span = inner_lo.to(self.prev_span);
+                        self.expect(&token::CloseDelim(token::Paren))?;
+                        let mut err = self.struct_span_err(
+                            lo.to(self.prev_span),
+                            "parenthesized lifetime bounds are not supported"
+                        );
+                        if let Ok(snippet) = self.span_to_snippet(inner_span) {
+                            err.span_suggestion_short(
+                                lo.to(self.prev_span),
+                                "remove the parentheses",
+                                snippet.to_owned(),
+                                Applicability::MachineApplicable
+                            );
+                        }
+                        err.emit();
+                    }
+                } else {
+                    let lifetime_defs = self.parse_late_bound_lifetime_defs()?;
+                    let path = self.parse_path(PathStyle::Type)?;
+                    if has_parens {
+                        self.expect(&token::CloseDelim(token::Paren))?;
+                    }
+                    let poly_span = lo.to(self.prev_span);
+                    if is_negative {
+                        was_negative = true;
+                        if let Some(sp) = last_plus_span.or(colon_span) {
+                            negative_bounds.push(sp.to(poly_span));
+                        }
+                    } else {
+                        let poly_trait = PolyTraitRef::new(lifetime_defs, path, poly_span);
+                        let modifier = if question.is_some() {
+                            TraitBoundModifier::Maybe
+                        } else {
+                            TraitBoundModifier::None
+                        };
+                        bounds.push(GenericBound::Trait(poly_trait, modifier));
+                    }
+                }
+            } else {
+                break
+            }
+
+            if !allow_plus || !self.eat_plus() {
+                break
+            } else {
+                last_plus_span = Some(self.prev_span);
+            }
+        }
+
+        if !negative_bounds.is_empty() || was_negative {
+            let negative_bounds_len = negative_bounds.len();
+            let last_span = negative_bounds.last().map(|sp| *sp);
+            let mut err = self.struct_span_err(
+                negative_bounds,
+                "negative trait bounds are not supported",
+            );
+            if let Some(sp) = last_span {
+                err.span_label(sp, "negative trait bounds are not supported");
+            }
+            if let Some(bound_list) = colon_span {
+                let bound_list = bound_list.to(self.prev_span);
+                let mut new_bound_list = String::new();
+                if !bounds.is_empty() {
+                    let mut snippets = bounds.iter().map(|bound| bound.span())
+                        .map(|span| self.span_to_snippet(span));
+                    while let Some(Ok(snippet)) = snippets.next() {
+                        new_bound_list.push_str(" + ");
+                        new_bound_list.push_str(&snippet);
+                    }
+                    new_bound_list = new_bound_list.replacen(" +", ":", 1);
+                }
+                err.span_suggestion_hidden(
+                    bound_list,
+                    &format!("remove the trait bound{}", pluralize!(negative_bounds_len)),
+                    new_bound_list,
+                    Applicability::MachineApplicable,
+                );
+            }
+            err.emit();
+        }
+
+        return Ok(bounds);
+    }
+
+    pub(super) fn parse_late_bound_lifetime_defs(&mut self) -> PResult<'a, Vec<GenericParam>> {
+        if self.eat_keyword(kw::For) {
+            self.expect_lt()?;
+            let params = self.parse_generic_params()?;
+            self.expect_gt()?;
+            // We rely on AST validation to rule out invalid cases: There must not be type
+            // parameters, and the lifetime parameters must not have bounds.
+            Ok(params)
+        } else {
+            Ok(Vec::new())
+        }
+    }
+
+    pub fn check_lifetime(&mut self) -> bool {
+        self.expected_tokens.push(TokenType::Lifetime);
+        self.token.is_lifetime()
+    }
+
+    /// Parses a single lifetime `'a` or panics.
+    pub fn expect_lifetime(&mut self) -> Lifetime {
+        if let Some(ident) = self.token.lifetime() {
+            let span = self.token.span;
+            self.bump();
+            Lifetime { ident: Ident::new(ident.name, span), id: ast::DUMMY_NODE_ID }
+        } else {
+            self.span_bug(self.token.span, "not a lifetime")
+        }
+    }
+
+    pub(super) fn mk_ty(&self, span: Span, kind: TyKind) -> P<Ty> {
+        P(Ty { kind, span, id: ast::DUMMY_NODE_ID })
+    }
+}

src/librustc_parse/validate_attr.rs

@@ -0,0 +1,111 @@
+//! Meta-syntax validation logic of attributes for post-expansion.
+
+use errors::{PResult, Applicability};
+use syntax::ast::{self, Attribute, AttrKind, Ident, MetaItem};
+use syntax::attr::{AttributeTemplate, mk_name_value_item_str};
+use syntax::early_buffered_lints::BufferedEarlyLintId;
+use syntax::feature_gate::BUILTIN_ATTRIBUTE_MAP;
+use syntax::token;
+use syntax::tokenstream::TokenTree;
+use syntax::sess::ParseSess;
+use syntax_pos::{Symbol, sym};
+
+pub fn check_meta(sess: &ParseSess, attr: &Attribute) {
+    let attr_info =
+        attr.ident().and_then(|ident| BUILTIN_ATTRIBUTE_MAP.get(&ident.name)).map(|a| **a);
+
+    // Check input tokens for built-in and key-value attributes.
+    match attr_info {
+        // `rustc_dummy` doesn't have any restrictions specific to built-in attributes.
+        Some((name, _, template, _)) if name != sym::rustc_dummy =>
+            check_builtin_attribute(sess, attr, name, template),
+        _ => if let Some(TokenTree::Token(token)) = attr.get_normal_item().tokens.trees().next() {
+            if token == token::Eq {
+                // All key-value attributes are restricted to meta-item syntax.
+                parse_meta(sess, attr).map_err(|mut err| err.emit()).ok();
+            }
+        }
+    }
+}
+
+pub fn parse_meta<'a>(sess: &'a ParseSess, attr: &Attribute) -> PResult<'a, MetaItem> {
+    Ok(match attr.kind {
+        AttrKind::Normal(ref item) => MetaItem {
+            path: item.path.clone(),
+            kind: super::parse_in_attr(sess, attr, |p| p.parse_meta_item_kind())?,
+            span: attr.span,
+        },
+        AttrKind::DocComment(comment) => {
+            mk_name_value_item_str(Ident::new(sym::doc, attr.span), comment, attr.span)
+        }
+    })
+}
+
+pub fn check_builtin_attribute(
+    sess: &ParseSess,
+    attr: &Attribute,
+    name: Symbol,
+    template: AttributeTemplate,
+) {
+    // Some special attributes like `cfg` must be checked
+    // before the generic check, so we skip them here.
+    let should_skip = |name| name == sym::cfg;
+    // Some of previously accepted forms were used in practice,
+    // report them as warnings for now.
+    let should_warn = |name| name == sym::doc || name == sym::ignore ||
+                             name == sym::inline || name == sym::link ||
+                             name == sym::test || name == sym::bench;
+
+    match parse_meta(sess, attr) {
+        Ok(meta) => if !should_skip(name) && !template.compatible(&meta.kind) {
+            let error_msg = format!("malformed `{}` attribute input", name);
+            let mut msg = "attribute must be of the form ".to_owned();
+            let mut suggestions = vec![];
+            let mut first = true;
+            if template.word {
+                first = false;
+                let code = format!("#[{}]", name);
+                msg.push_str(&format!("`{}`", &code));
+                suggestions.push(code);
+            }
+            if let Some(descr) = template.list {
+                if !first {
+                    msg.push_str(" or ");
+                }
+                first = false;
+                let code = format!("#[{}({})]", name, descr);
+                msg.push_str(&format!("`{}`", &code));
+                suggestions.push(code);
+            }
+            if let Some(descr) = template.name_value_str {
+                if !first {
+                    msg.push_str(" or ");
+                }
+                let code = format!("#[{} = \"{}\"]", name, descr);
+                msg.push_str(&format!("`{}`", &code));
+                suggestions.push(code);
+            }
+            if should_warn(name) {
+                sess.buffer_lint(
+                    BufferedEarlyLintId::IllFormedAttributeInput,
+                    meta.span,
+                    ast::CRATE_NODE_ID,
+                    &msg,
+                );
+            } else {
+                sess.span_diagnostic.struct_span_err(meta.span, &error_msg)
+                    .span_suggestions(
+                        meta.span,
+                        if suggestions.len() == 1 {
+                            "must be of the form"
+                        } else {
+                            "the following are the possible correct uses"
+                        },
+                        suggestions.into_iter(),
+                        Applicability::HasPlaceholders,
+                    ).emit();
+            }
+        }
+        Err(mut err) => err.emit(),
+    }
+}