Bless NLL test

[rust.git] / compiler / rustc_expand / src / mbe / macro_parser.rs

//! This is an NFA-based parser, which calls out to the main rust parser for named non-terminals
//! (which it commits to fully when it hits one in a grammar). There's a set of current NFA threads
//! and a set of next ones. Instead of NTs, we have a special case for Kleene star. The big-O, in
//! pathological cases, is worse than traditional use of NFA or Earley parsing, but it's an easier
//! fit for Macro-by-Example-style rules.
//!
//! (In order to prevent the pathological case, we'd need to lazily construct the resulting
//! `NamedMatch`es at the very end. It'd be a pain, and require more memory to keep around old
//! items, but it would also save overhead)
//!
//! We don't say this parser uses the Earley algorithm, because it's unnecessarily inaccurate.
//! The macro parser restricts itself to the features of finite state automata. Earley parsers
//! can be described as an extension of NFAs with completion rules, prediction rules, and recursion.
//!
//! Quick intro to how the parser works:
//!
//! A 'position' is a dot in the middle of a matcher, usually represented as a
//! dot. For example `· a $( a )* a b` is a position, as is `a $( · a )* a b`.
//!
//! The parser walks through the input a character at a time, maintaining a list
//! of threads consistent with the current position in the input string: `cur_items`.
//!
//! As it processes them, it fills up `eof_items` with threads that would be valid if
//! the macro invocation is now over, `bb_items` with threads that are waiting on
//! a Rust non-terminal like `$e:expr`, and `next_items` with threads that are waiting
//! on a particular token. Most of the logic concerns moving the · through the
//! repetitions indicated by Kleene stars. The rules for moving the · without
//! consuming any input are called epsilon transitions. It only advances or calls
//! out to the real Rust parser when no `cur_items` threads remain.
//!
//! Example:
//!
//! ```text, ignore
//! Start parsing a a a a b against [· a $( a )* a b].
//!
//! Remaining input: a a a a b
//! next: [· a $( a )* a b]
//!
//! - - - Advance over an a. - - -
//!
//! Remaining input: a a a b
//! cur: [a · $( a )* a b]
//! Descend/Skip (first item).
//! next: [a $( · a )* a b]  [a $( a )* · a b].
//!
//! - - - Advance over an a. - - -
//!
//! Remaining input: a a b
//! cur: [a $( a · )* a b]  [a $( a )* a · b]
//! Follow epsilon transition: Finish/Repeat (first item)
//! next: [a $( a )* · a b]  [a $( · a )* a b]  [a $( a )* a · b]
//!
//! - - - Advance over an a. - - - (this looks exactly like the last step)
//!
//! Remaining input: a b
//! cur: [a $( a · )* a b]  [a $( a )* a · b]
//! Follow epsilon transition: Finish/Repeat (first item)
//! next: [a $( a )* · a b]  [a $( · a )* a b]  [a $( a )* a · b]
//!
//! - - - Advance over an a. - - - (this looks exactly like the last step)
//!
//! Remaining input: b
//! cur: [a $( a · )* a b]  [a $( a )* a · b]
//! Follow epsilon transition: Finish/Repeat (first item)
//! next: [a $( a )* · a b]  [a $( · a )* a b]  [a $( a )* a · b]
//!
//! - - - Advance over a b. - - -
//!
//! Remaining input: ''
//! eof: [a $( a )* a b ·]
//! ```

crate use NamedMatch::*;
crate use ParseResult::*;
use TokenTreeOrTokenTreeSlice::*;

use crate::mbe::{self, TokenTree};

use rustc_ast::token::{self, DocComment, Nonterminal, Token};
use rustc_parse::parser::Parser;
use rustc_session::parse::ParseSess;
use rustc_span::symbol::MacroRulesNormalizedIdent;

use smallvec::{smallvec, SmallVec};

use rustc_data_structures::fx::FxHashMap;
use rustc_data_structures::sync::Lrc;
use std::borrow::Cow;
use std::collections::hash_map::Entry::{Occupied, Vacant};
use std::mem;
use std::ops::{Deref, DerefMut};

// To avoid costly uniqueness checks, we require that `MatchSeq` always has a nonempty body.

/// Either a sequence of token trees or a single one. This is used as the representation of the
/// sequence of tokens that make up a matcher.
#[derive(Clone)]
enum TokenTreeOrTokenTreeSlice<'tt> {
    Tt(TokenTree),
    TtSeq(&'tt [TokenTree]),
}

impl<'tt> TokenTreeOrTokenTreeSlice<'tt> {
    /// Returns the number of constituent top-level token trees of `self` (top-level in that it
    /// will not recursively descend into subtrees).
    fn len(&self) -> usize {
        match *self {
            TtSeq(ref v) => v.len(),
            Tt(ref tt) => tt.len(),
        }
    }

    /// The `index`-th token tree of `self`.
    fn get_tt(&self, index: usize) -> TokenTree {
        match *self {
            TtSeq(ref v) => v[index].clone(),
            Tt(ref tt) => tt.get_tt(index),
        }
    }
}

/// An unzipping of `TokenTree`s... see the `stack` field of `MatcherPos`.
///
/// This is used by `inner_parse_loop` to keep track of delimited submatchers that we have
/// descended into.
#[derive(Clone)]
struct MatcherTtFrame<'tt> {
    /// The "parent" matcher that we are descending into.
    elts: TokenTreeOrTokenTreeSlice<'tt>,
    /// The position of the "dot" in `elts` at the time we descended.
    idx: usize,
}

type NamedMatchVec = SmallVec<[NamedMatch; 4]>;

/// Represents a single "position" (aka "matcher position", aka "item"), as
/// described in the module documentation.
///
/// Here:
///
/// - `'root` represents the lifetime of the stack slot that holds the root
///   `MatcherPos`. As described in `MatcherPosHandle`, the root `MatcherPos`
///   structure is stored on the stack, but subsequent instances are put into
///   the heap.
/// - `'tt` represents the lifetime of the token trees that this matcher
///   position refers to.
///
/// It is important to distinguish these two lifetimes because we have a
/// `SmallVec<TokenTreeOrTokenTreeSlice<'tt>>` below, and the destructor of
/// that is considered to possibly access the data from its elements (it lacks
/// a `#[may_dangle]` attribute). As a result, the compiler needs to know that
/// all the elements in that `SmallVec` strictly outlive the root stack slot
/// lifetime. By separating `'tt` from `'root`, we can show that.
#[derive(Clone)]
struct MatcherPos<'root, 'tt> {
    /// The token or sequence of tokens that make up the matcher
    top_elts: TokenTreeOrTokenTreeSlice<'tt>,

    /// The position of the "dot" in this matcher
    idx: usize,

    /// For each named metavar in the matcher, we keep track of token trees matched against the
    /// metavar by the black box parser. In particular, there may be more than one match per
    /// metavar if we are in a repetition (each repetition matches each of the variables).
    /// Moreover, matchers and repetitions can be nested; the `matches` field is shared (hence the
    /// `Rc`) among all "nested" matchers. `match_lo`, `match_cur`, and `match_hi` keep track of
    /// the current position of the `self` matcher position in the shared `matches` list.
    ///
    /// Also, note that while we are descending into a sequence, matchers are given their own
    /// `matches` vector. Only once we reach the end of a full repetition of the sequence do we add
    /// all bound matches from the submatcher into the shared top-level `matches` vector. If `sep`
    /// and `up` are `Some`, then `matches` is _not_ the shared top-level list. Instead, if one
    /// wants the shared `matches`, one should use `up.matches`.
    matches: Box<[Lrc<NamedMatchVec>]>,
    /// The position in `matches` corresponding to the first metavar in this matcher's sequence of
    /// token trees. In other words, the first metavar in the first token of `top_elts` corresponds
    /// to `matches[match_lo]`.
    match_lo: usize,
    /// The position in `matches` corresponding to the metavar we are currently trying to match
    /// against the source token stream. `match_lo <= match_cur <= match_hi`.
    match_cur: usize,
    /// Similar to `match_lo` except `match_hi` is the position in `matches` of the _last_ metavar
    /// in this matcher.
    match_hi: usize,

    // The following fields are used if we are matching a repetition. If we aren't, they should be
    // `None`.
    /// The KleeneOp of this sequence if we are in a repetition.
    seq_op: Option<mbe::KleeneOp>,

    /// The separator if we are in a repetition.
    sep: Option<Token>,

    /// The "parent" matcher position if we are in a repetition. That is, the matcher position just
    /// before we enter the sequence.
    up: Option<MatcherPosHandle<'root, 'tt>>,

    /// Specifically used to "unzip" token trees. By "unzip", we mean to unwrap the delimiters from
    /// a delimited token tree (e.g., something wrapped in `(` `)`) or to get the contents of a doc
    /// comment...
    ///
    /// When matching against matchers with nested delimited submatchers (e.g., `pat ( pat ( .. )
    /// pat ) pat`), we need to keep track of the matchers we are descending into. This stack does
    /// that where the bottom of the stack is the outermost matcher.
    /// Also, throughout the comments, this "descent" is often referred to as "unzipping"...
    stack: SmallVec<[MatcherTtFrame<'tt>; 1]>,
}

impl<'root, 'tt> MatcherPos<'root, 'tt> {
    /// Adds `m` as a named match for the `idx`-th metavar.
    fn push_match(&mut self, idx: usize, m: NamedMatch) {
        let matches = Lrc::make_mut(&mut self.matches[idx]);
        matches.push(m);
    }
}

// Lots of MatcherPos instances are created at runtime. Allocating them on the
// heap is slow. Furthermore, using SmallVec<MatcherPos> to allocate them all
// on the stack is also slow, because MatcherPos is quite a large type and
// instances get moved around a lot between vectors, which requires lots of
// slow memcpy calls.
//
// Therefore, the initial MatcherPos is always allocated on the stack,
// subsequent ones (of which there aren't that many) are allocated on the heap,
// and this type is used to encapsulate both cases.
enum MatcherPosHandle<'root, 'tt> {
    Ref(&'root mut MatcherPos<'root, 'tt>),
    Box(Box<MatcherPos<'root, 'tt>>),
}

impl<'root, 'tt> Clone for MatcherPosHandle<'root, 'tt> {
    // This always produces a new Box.
    fn clone(&self) -> Self {
        MatcherPosHandle::Box(match *self {
            MatcherPosHandle::Ref(ref r) => Box::new((**r).clone()),
            MatcherPosHandle::Box(ref b) => b.clone(),
        })
    }
}

impl<'root, 'tt> Deref for MatcherPosHandle<'root, 'tt> {
    type Target = MatcherPos<'root, 'tt>;
    fn deref(&self) -> &Self::Target {
        match *self {
            MatcherPosHandle::Ref(ref r) => r,
            MatcherPosHandle::Box(ref b) => b,
        }
    }
}

impl<'root, 'tt> DerefMut for MatcherPosHandle<'root, 'tt> {
    fn deref_mut(&mut self) -> &mut MatcherPos<'root, 'tt> {
        match *self {
            MatcherPosHandle::Ref(ref mut r) => r,
            MatcherPosHandle::Box(ref mut b) => b,
        }
    }
}

/// Represents the possible results of an attempted parse.
crate enum ParseResult<T> {
    /// Parsed successfully.
    Success(T),
    /// Arm failed to match. If the second parameter is `token::Eof`, it indicates an unexpected
    /// end of macro invocation. Otherwise, it indicates that no rules expected the given token.
    Failure(Token, &'static str),
    /// Fatal error (malformed macro?). Abort compilation.
    Error(rustc_span::Span, String),
    ErrorReported,
}

/// A `ParseResult` where the `Success` variant contains a mapping of
/// `MacroRulesNormalizedIdent`s to `NamedMatch`es. This represents the mapping
/// of metavars to the token trees they bind to.
crate type NamedParseResult = ParseResult<FxHashMap<MacroRulesNormalizedIdent, NamedMatch>>;

/// Count how many metavars are named in the given matcher `ms`.
pub(super) fn count_names(ms: &[TokenTree]) -> usize {
    ms.iter().fold(0, |count, elt| {
        count
            + match *elt {
                TokenTree::Sequence(_, ref seq) => seq.num_captures,
                TokenTree::Delimited(_, ref delim) => count_names(&delim.tts),
                TokenTree::MetaVar(..) => 0,
                TokenTree::MetaVarDecl(..) => 1,
                TokenTree::Token(..) => 0,
            }
    })
}

/// `len` `Vec`s (initially shared and empty) that will store matches of metavars.
fn create_matches(len: usize) -> Box<[Lrc<NamedMatchVec>]> {
    if len == 0 {
        vec![]
    } else {
        let empty_matches = Lrc::new(SmallVec::new());
        vec![empty_matches; len]
    }
    .into_boxed_slice()
}

/// Generates the top-level matcher position in which the "dot" is before the first token of the
/// matcher `ms`.
fn initial_matcher_pos<'root, 'tt>(ms: &'tt [TokenTree]) -> MatcherPos<'root, 'tt> {
    let match_idx_hi = count_names(ms);
    let matches = create_matches(match_idx_hi);
    MatcherPos {
        // Start with the top level matcher given to us
        top_elts: TtSeq(ms), // "elts" is an abbr. for "elements"
        // The "dot" is before the first token of the matcher
        idx: 0,

        // Initialize `matches` to a bunch of empty `Vec`s -- one for each metavar in `top_elts`.
        // `match_lo` for `top_elts` is 0 and `match_hi` is `matches.len()`. `match_cur` is 0 since
        // we haven't actually matched anything yet.
        matches,
        match_lo: 0,
        match_cur: 0,
        match_hi: match_idx_hi,

        // Haven't descended into any delimiters, so empty stack
        stack: smallvec![],

        // Haven't descended into any sequences, so both of these are `None`.
        seq_op: None,
        sep: None,
        up: None,
    }
}

/// `NamedMatch` is a pattern-match result for a single `token::MATCH_NONTERMINAL`:
/// so it is associated with a single ident in a parse, and all
/// `MatchedNonterminal`s in the `NamedMatch` have the same non-terminal type
/// (expr, item, etc). Each leaf in a single `NamedMatch` corresponds to a
/// single `token::MATCH_NONTERMINAL` in the `TokenTree` that produced it.
///
/// The in-memory structure of a particular `NamedMatch` represents the match
/// that occurred when a particular subset of a matcher was applied to a
/// particular token tree.
///
/// The width of each `MatchedSeq` in the `NamedMatch`, and the identity of
/// the `MatchedNonterminal`s, will depend on the token tree it was applied
/// to: each `MatchedSeq` corresponds to a single `TTSeq` in the originating
/// token tree. The depth of the `NamedMatch` structure will therefore depend
/// only on the nesting depth of `ast::TTSeq`s in the originating
/// token tree it was derived from.
#[derive(Debug, Clone)]
crate enum NamedMatch {
    MatchedSeq(Lrc<NamedMatchVec>),
    MatchedNonterminal(Lrc<Nonterminal>),
}

/// Takes a sequence of token trees `ms` representing a matcher which successfully matched input
/// and an iterator of items that matched input and produces a `NamedParseResult`.
fn nameize<I: Iterator<Item = NamedMatch>>(
    sess: &ParseSess,
    ms: &[TokenTree],
    mut res: I,
) -> NamedParseResult {
    // Recursively descend into each type of matcher (e.g., sequences, delimited, metavars) and make
    // sure that each metavar has _exactly one_ binding. If a metavar does not have exactly one
    // binding, then there is an error. If it does, then we insert the binding into the
    // `NamedParseResult`.
    fn n_rec<I: Iterator<Item = NamedMatch>>(
        sess: &ParseSess,
        m: &TokenTree,
        res: &mut I,
        ret_val: &mut FxHashMap<MacroRulesNormalizedIdent, NamedMatch>,
    ) -> Result<(), (rustc_span::Span, String)> {
        match *m {
            TokenTree::Sequence(_, ref seq) => {
                for next_m in &seq.tts {
                    n_rec(sess, next_m, res.by_ref(), ret_val)?
                }
            }
            TokenTree::Delimited(_, ref delim) => {
                for next_m in &delim.tts {
                    n_rec(sess, next_m, res.by_ref(), ret_val)?;
                }
            }
            TokenTree::MetaVarDecl(sp, bind_name, _) => match ret_val
                .entry(MacroRulesNormalizedIdent::new(bind_name))
            {
                Vacant(spot) => {
                    spot.insert(res.next().unwrap());
                }
                Occupied(..) => return Err((sp, format!("duplicated bind name: {}", bind_name))),
            },
            TokenTree::MetaVar(..) | TokenTree::Token(..) => (),
        }

        Ok(())
    }

    let mut ret_val = FxHashMap::default();
    for m in ms {
        match n_rec(sess, m, res.by_ref(), &mut ret_val) {
            Ok(_) => {}
            Err((sp, msg)) => return Error(sp, msg),
        }
    }

    Success(ret_val)
}

/// Performs a token equality check, ignoring syntax context (that is, an unhygienic comparison)
fn token_name_eq(t1: &Token, t2: &Token) -> bool {
    if let (Some((ident1, is_raw1)), Some((ident2, is_raw2))) = (t1.ident(), t2.ident()) {
        ident1.name == ident2.name && is_raw1 == is_raw2
    } else if let (Some(ident1), Some(ident2)) = (t1.lifetime(), t2.lifetime()) {
        ident1.name == ident2.name
    } else {
        t1.kind == t2.kind
    }
}

/// Process the matcher positions of `cur_items` until it is empty. In the process, this will
/// produce more items in `next_items`, `eof_items`, and `bb_items`.
///
/// For more info about the how this happens, see the module-level doc comments and the inline
/// comments of this function.
///
/// # Parameters
///
/// - `sess`: the parsing session into which errors are emitted.
/// - `cur_items`: the set of current items to be processed. This should be empty by the end of a
///   successful execution of this function.
/// - `next_items`: the set of newly generated items. These are used to replenish `cur_items` in
///   the function `parse`.
/// - `eof_items`: the set of items that would be valid if this was the EOF.
/// - `bb_items`: the set of items that are waiting for the black-box parser.
/// - `token`: the current token of the parser.
/// - `span`: the `Span` in the source code corresponding to the token trees we are trying to match
///   against the matcher positions in `cur_items`.
///
/// # Returns
///
/// A `ParseResult`. Note that matches are kept track of through the items generated.
fn inner_parse_loop<'root, 'tt>(
    cur_items: &mut SmallVec<[MatcherPosHandle<'root, 'tt>; 1]>,
    next_items: &mut Vec<MatcherPosHandle<'root, 'tt>>,
    eof_items: &mut SmallVec<[MatcherPosHandle<'root, 'tt>; 1]>,
    bb_items: &mut SmallVec<[MatcherPosHandle<'root, 'tt>; 1]>,
    token: &Token,
) -> ParseResult<()> {
    // Pop items from `cur_items` until it is empty.
    while let Some(mut item) = cur_items.pop() {
        // When unzipped trees end, remove them. This corresponds to backtracking out of a
        // delimited submatcher into which we already descended. In backtracking out again, we need
        // to advance the "dot" past the delimiters in the outer matcher.
        while item.idx >= item.top_elts.len() {
            match item.stack.pop() {
                Some(MatcherTtFrame { elts, idx }) => {
                    item.top_elts = elts;
                    item.idx = idx + 1;
                }
                None => break,
            }
        }

        // Get the current position of the "dot" (`idx`) in `item` and the number of token trees in
        // the matcher (`len`).
        let idx = item.idx;
        let len = item.top_elts.len();

        // If `idx >= len`, then we are at or past the end of the matcher of `item`.
        if idx >= len {
            // We are repeating iff there is a parent. If the matcher is inside of a repetition,
            // then we could be at the end of a sequence or at the beginning of the next
            // repetition.
            if item.up.is_some() {
                // At this point, regardless of whether there is a separator, we should add all
                // matches from the complete repetition of the sequence to the shared, top-level
                // `matches` list (actually, `up.matches`, which could itself not be the top-level,
                // but anyway...). Moreover, we add another item to `cur_items` in which the "dot"
                // is at the end of the `up` matcher. This ensures that the "dot" in the `up`
                // matcher is also advanced sufficiently.
                //
                // NOTE: removing the condition `idx == len` allows trailing separators.
                if idx == len {
                    // Get the `up` matcher
                    let mut new_pos = item.up.clone().unwrap();

                    // Add matches from this repetition to the `matches` of `up`
                    for idx in item.match_lo..item.match_hi {
                        let sub = item.matches[idx].clone();
                        new_pos.push_match(idx, MatchedSeq(sub));
                    }

                    // Move the "dot" past the repetition in `up`
                    new_pos.match_cur = item.match_hi;
                    new_pos.idx += 1;
                    cur_items.push(new_pos);
                }

                // Check if we need a separator.
                if idx == len && item.sep.is_some() {
                    // We have a separator, and it is the current token. We can advance past the
                    // separator token.
                    if item.sep.as_ref().map(|sep| token_name_eq(token, sep)).unwrap_or(false) {
                        item.idx += 1;
                        next_items.push(item);
                    }
                }
                // We don't need a separator. Move the "dot" back to the beginning of the matcher
                // and try to match again UNLESS we are only allowed to have _one_ repetition.
                else if item.seq_op != Some(mbe::KleeneOp::ZeroOrOne) {
                    item.match_cur = item.match_lo;
                    item.idx = 0;
                    cur_items.push(item);
                }
            }
            // If we are not in a repetition, then being at the end of a matcher means that we have
            // reached the potential end of the input.
            else {
                eof_items.push(item);
            }
        }
        // We are in the middle of a matcher.
        else {
            // Look at what token in the matcher we are trying to match the current token (`token`)
            // against. Depending on that, we may generate new items.
            match item.top_elts.get_tt(idx) {
                // Need to descend into a sequence
                TokenTree::Sequence(sp, seq) => {
                    // Examine the case where there are 0 matches of this sequence. We are
                    // implicitly disallowing OneOrMore from having 0 matches here. Thus, that will
                    // result in a "no rules expected token" error by virtue of this matcher not
                    // working.
                    if seq.kleene.op == mbe::KleeneOp::ZeroOrMore
                        || seq.kleene.op == mbe::KleeneOp::ZeroOrOne
                    {
                        let mut new_item = item.clone();
                        new_item.match_cur += seq.num_captures;
                        new_item.idx += 1;
                        for idx in item.match_cur..item.match_cur + seq.num_captures {
                            new_item.push_match(idx, MatchedSeq(Lrc::new(smallvec![])));
                        }
                        cur_items.push(new_item);
                    }

                    let matches = create_matches(item.matches.len());
                    cur_items.push(MatcherPosHandle::Box(Box::new(MatcherPos {
                        stack: smallvec![],
                        sep: seq.separator.clone(),
                        seq_op: Some(seq.kleene.op),
                        idx: 0,
                        matches,
                        match_lo: item.match_cur,
                        match_cur: item.match_cur,
                        match_hi: item.match_cur + seq.num_captures,
                        up: Some(item),
                        top_elts: Tt(TokenTree::Sequence(sp, seq)),
                    })));
                }

                // We need to match a metavar with a valid ident... call out to the black-box
                // parser by adding an item to `bb_items`.
                TokenTree::MetaVarDecl(_, _, kind) => {
                    // Built-in nonterminals never start with these tokens,
                    // so we can eliminate them from consideration.
                    if Parser::nonterminal_may_begin_with(kind, token) {
                        bb_items.push(item);
                    }
                }

                // We need to descend into a delimited submatcher or a doc comment. To do this, we
                // push the current matcher onto a stack and push a new item containing the
                // submatcher onto `cur_items`.
                //
                // At the beginning of the loop, if we reach the end of the delimited submatcher,
                // we pop the stack to backtrack out of the descent.
                seq
                @
                (TokenTree::Delimited(..)
                | TokenTree::Token(Token { kind: DocComment(..), .. })) => {
                    let lower_elts = mem::replace(&mut item.top_elts, Tt(seq));
                    let idx = item.idx;
                    item.stack.push(MatcherTtFrame { elts: lower_elts, idx });
                    item.idx = 0;
                    cur_items.push(item);
                }

                // We just matched a normal token. We can just advance the parser.
                TokenTree::Token(t) if token_name_eq(&t, token) => {
                    item.idx += 1;
                    next_items.push(item);
                }

                // There was another token that was not `token`... This means we can't add any
                // rules. NOTE that this is not necessarily an error unless _all_ items in
                // `cur_items` end up doing this. There may still be some other matchers that do
                // end up working out.
                TokenTree::Token(..) | TokenTree::MetaVar(..) => {}
            }
        }
    }

    // Yay a successful parse (so far)!
    Success(())
}

/// Use the given sequence of token trees (`ms`) as a matcher. Match the token
/// stream from the given `parser` against it and return the match.
pub(super) fn parse_tt(parser: &mut Cow<'_, Parser<'_>>, ms: &[TokenTree]) -> NamedParseResult {
    // A queue of possible matcher positions. We initialize it with the matcher position in which
    // the "dot" is before the first token of the first token tree in `ms`. `inner_parse_loop` then
    // processes all of these possible matcher positions and produces possible next positions into
    // `next_items`. After some post-processing, the contents of `next_items` replenish `cur_items`
    // and we start over again.
    //
    // This MatcherPos instance is allocated on the stack. All others -- and
    // there are frequently *no* others! -- are allocated on the heap.
    let mut initial = initial_matcher_pos(ms);
    let mut cur_items = smallvec![MatcherPosHandle::Ref(&mut initial)];
    let mut next_items = Vec::new();

    loop {
        // Matcher positions black-box parsed by parser.rs (`parser`)
        let mut bb_items = SmallVec::new();

        // Matcher positions that would be valid if the macro invocation was over now
        let mut eof_items = SmallVec::new();
        assert!(next_items.is_empty());

        // Process `cur_items` until either we have finished the input or we need to get some
        // parsing from the black-box parser done. The result is that `next_items` will contain a
        // bunch of possible next matcher positions in `next_items`.
        match inner_parse_loop(
            &mut cur_items,
            &mut next_items,
            &mut eof_items,
            &mut bb_items,
            &parser.token,
        ) {
            Success(_) => {}
            Failure(token, msg) => return Failure(token, msg),
            Error(sp, msg) => return Error(sp, msg),
            ErrorReported => return ErrorReported,
        }

        // inner parse loop handled all cur_items, so it's empty
        assert!(cur_items.is_empty());

        // We need to do some post processing after the `inner_parser_loop`.
        //
        // Error messages here could be improved with links to original rules.

        // If we reached the EOF, check that there is EXACTLY ONE possible matcher. Otherwise,
        // either the parse is ambiguous (which should never happen) or there is a syntax error.
        if parser.token == token::Eof {
            if eof_items.len() == 1 {
                let matches =
                    eof_items[0].matches.iter_mut().map(|dv| Lrc::make_mut(dv).pop().unwrap());
                return nameize(parser.sess, ms, matches);
            } else if eof_items.len() > 1 {
                return Error(
                    parser.token.span,
                    "ambiguity: multiple successful parses".to_string(),
                );
            } else {
                return Failure(
                    Token::new(
                        token::Eof,
                        if parser.token.span.is_dummy() {
                            parser.token.span
                        } else {
                            parser.token.span.shrink_to_hi()
                        },
                    ),
                    "missing tokens in macro arguments",
                );
            }
        }
        // Performance hack: eof_items may share matchers via Rc with other things that we want
        // to modify. Dropping eof_items now may drop these refcounts to 1, preventing an
        // unnecessary implicit clone later in Rc::make_mut.
        drop(eof_items);

        // If there are no possible next positions AND we aren't waiting for the black-box parser,
        // then there is a syntax error.
        if bb_items.is_empty() && next_items.is_empty() {
            return Failure(parser.token.clone(), "no rules expected this token in macro call");
        }
        // Another possibility is that we need to call out to parse some rust nonterminal
        // (black-box) parser. However, if there is not EXACTLY ONE of these, something is wrong.
        else if (!bb_items.is_empty() && !next_items.is_empty()) || bb_items.len() > 1 {
            let nts = bb_items
                .iter()
                .map(|item| match item.top_elts.get_tt(item.idx) {
                    TokenTree::MetaVarDecl(_, bind, kind) => format!("{} ('{}')", kind, bind),
                    _ => panic!(),
                })
                .collect::<Vec<String>>()
                .join(" or ");

            return Error(
                parser.token.span,
                format!(
                    "local ambiguity: multiple parsing options: {}",
                    match next_items.len() {
                        0 => format!("built-in NTs {}.", nts),
                        1 => format!("built-in NTs {} or 1 other option.", nts),
                        n => format!("built-in NTs {} or {} other options.", nts, n),
                    }
                ),
            );
        }
        // Dump all possible `next_items` into `cur_items` for the next iteration.
        else if !next_items.is_empty() {
            // Now process the next token
            cur_items.extend(next_items.drain(..));
            parser.to_mut().bump();
        }
        // Finally, we have the case where we need to call the black-box parser to get some
        // nonterminal.
        else {
            assert_eq!(bb_items.len(), 1);

            let mut item = bb_items.pop().unwrap();
            if let TokenTree::MetaVarDecl(span, _, kind) = item.top_elts.get_tt(item.idx) {
                let match_cur = item.match_cur;
                let nt = match parser.to_mut().parse_nonterminal(kind) {
                    Err(mut err) => {
                        err.span_label(
                            span,
                            format!("while parsing argument for this `{}` macro fragment", kind),
                        )
                        .emit();
                        return ErrorReported;
                    }
                    Ok(nt) => nt,
                };
                item.push_match(match_cur, MatchedNonterminal(Lrc::new(nt)));
                item.idx += 1;
                item.match_cur += 1;
            } else {
                unreachable!()
            }
            cur_items.push(item);
        }

        assert!(!cur_items.is_empty());
    }
}