Factor out inner current Earley item loop.

Change multiple functions to be non-public.
Change nameize to accept an iterator so as to avoid an allocation.

1 files changed, 153 insertions, 146 deletions

diff --git a/src/libsyntax/ext/tt/macro_parser.rs b/src/libsyntax/ext/tt/macro_parser.rs
index 64acce19c1c..3c57f7a05c2 100644
--- a/src/libsyntax/ext/tt/macro_parser.rs
+++ b/src/libsyntax/ext/tt/macro_parser.rs
@@ -130,7 +130,7 @@ struct MatcherTtFrame {
 }
 
 #[derive(Clone)]
-pub struct MatcherPos {
+struct MatcherPos {
     stack: Vec<MatcherTtFrame>,
     top_elts: TokenTreeOrTokenTreeVec,
     sep: Option<Token>,
@@ -162,14 +162,13 @@ pub fn count_names(ms: &[TokenTree]) -> usize {
     })
 }
 
-pub fn initial_matcher_pos(ms: Vec<TokenTree>, sep: Option<Token>, lo: BytePos)
-                           -> Box<MatcherPos> {
+fn initial_matcher_pos(ms: Vec<TokenTree>, lo: BytePos) -> Box<MatcherPos> {
     let match_idx_hi = count_names(&ms[..]);
-    let matches: Vec<_> = (0..match_idx_hi).map(|_| Vec::new()).collect();
+    let matches = create_matches(match_idx_hi);
     Box::new(MatcherPos {
         stack: vec![],
         top_elts: TtSeq(ms),
-        sep: sep,
+        sep: None,
         idx: 0,
         up: None,
         matches: matches,
@@ -202,26 +201,25 @@ pub enum NamedMatch {
     MatchedNonterminal(Rc<Nonterminal>)
 }
 
-fn nameize(ms: &[TokenTree], res: &[Rc<NamedMatch>]) -> NamedParseResult {
-    fn n_rec(m: &TokenTree, res: &[Rc<NamedMatch>],
-             ret_val: &mut HashMap<Ident, Rc<NamedMatch>>, idx: &mut usize)
+fn nameize<I: Iterator<Item=Rc<NamedMatch>>>(ms: &[TokenTree], mut res: I) -> NamedParseResult {
+    fn n_rec<I: Iterator<Item=Rc<NamedMatch>>>(m: &TokenTree, mut res: &mut I,
+             ret_val: &mut HashMap<Ident, Rc<NamedMatch>>)
              -> Result<(), (syntax_pos::Span, String)> {
         match *m {
             TokenTree::Sequence(_, ref seq) => {
                 for next_m in &seq.tts {
-                    n_rec(next_m, res, ret_val, idx)?
+                    n_rec(next_m, res.by_ref(), ret_val)?
                 }
             }
             TokenTree::Delimited(_, ref delim) => {
                 for next_m in &delim.tts {
-                    n_rec(next_m, res, ret_val, idx)?;
+                    n_rec(next_m, res.by_ref(), ret_val)?;
                 }
             }
             TokenTree::Token(sp, MatchNt(bind_name, _)) => {
                 match ret_val.entry(bind_name) {
                     Vacant(spot) => {
-                        spot.insert(res[*idx].clone());
-                        *idx += 1;
+                        spot.insert(res.next().unwrap());
                     }
                     Occupied(..) => {
                         return Err((sp, format!("duplicated bind name: {}", bind_name)))
@@ -238,9 +236,8 @@ fn nameize(ms: &[TokenTree], res: &[Rc<NamedMatch>]) -> NamedParseResult {
     }
 
     let mut ret_val = HashMap::new();
-    let mut idx = 0;
     for m in ms {
-        match n_rec(m, res, &mut ret_val, &mut idx) {
+        match n_rec(m, res.by_ref(), &mut ret_val) {
             Ok(_) => {},
             Err((sp, msg)) => return Error(sp, msg),
         }
@@ -266,9 +263,8 @@ pub fn parse_failure_msg(tok: Token) -> String {
     }
 }
 
-/// Perform a token equality check, ignoring syntax context (that is, an
-/// unhygienic comparison)
-pub fn token_name_eq(t1 : &Token, t2 : &Token) -> bool {
+/// Perform a token equality check, ignoring syntax context (that is, an unhygienic comparison)
+fn token_name_eq(t1 : &Token, t2 : &Token) -> bool {
     match (t1,t2) {
         (&token::Ident(id1),&token::Ident(id2))
         | (&token::Lifetime(id1),&token::Lifetime(id2)) =>
@@ -277,154 +273,165 @@ pub fn token_name_eq(t1 : &Token, t2 : &Token) -> bool {
     }
 }
 
-pub fn parse(sess: &ParseSess, rdr: TtReader, ms: &[TokenTree]) -> NamedParseResult {
-    let mut parser = Parser::new_with_doc_flag(sess, Box::new(rdr), true);
-    let mut cur_eis = SmallVector::one(initial_matcher_pos(ms.to_owned(), None, parser.span.lo));
-
-    loop {
-        let mut bb_eis = Vec::new(); // black-box parsed by parser.rs
-        let mut next_eis = Vec::new(); // or proceed normally
-        let mut eof_eis = Vec::new();
+fn create_matches(len: usize) -> Vec<Vec<Rc<NamedMatch>>> {
+    (0..len).into_iter().map(|_| Vec::new()).collect()
+}
 
-        // for each Earley item
-        while let Some(mut ei) = cur_eis.pop() {
-            // When unzipped trees end, remove them
-            while ei.idx >= ei.top_elts.len() {
-                match ei.stack.pop() {
-                    Some(MatcherTtFrame { elts, idx }) => {
-                        ei.top_elts = elts;
-                        ei.idx = idx + 1;
-                    }
-                    None => break
+fn inner_parse_loop(cur_eis: &mut SmallVector<Box<MatcherPos>>,
+                    next_eis: &mut Vec<Box<MatcherPos>>,
+                    eof_eis: &mut Vec<Box<MatcherPos>>,
+                    bb_eis: &mut Vec<Box<MatcherPos>>,
+                    token: &Token, span: &syntax_pos::Span) -> ParseResult<()> {
+    while let Some(mut ei) = cur_eis.pop() {
+        // When unzipped trees end, remove them
+        while ei.idx >= ei.top_elts.len() {
+            match ei.stack.pop() {
+                Some(MatcherTtFrame { elts, idx }) => {
+                    ei.top_elts = elts;
+                    ei.idx = idx + 1;
                 }
+                None => break
             }
+        }
 
-            let idx = ei.idx;
-            let len = ei.top_elts.len();
-
-            /* at end of sequence */
-            if idx >= len {
-                // can't move out of `match`es, so:
-                if ei.up.is_some() {
-                    // hack: a matcher sequence is repeating iff it has a
-                    // parent (the top level is just a container)
-
-                    // disregard separator, try to go up
-                    // (remove this condition to make trailing seps ok)
-                    if idx == len {
-                        // pop from the matcher position
-
-                        let mut new_pos = ei.up.clone().unwrap();
-
-                        // update matches (the MBE "parse tree") by appending
-                        // each tree as a subtree.
-
-                        // I bet this is a perf problem: we're preemptively
-                        // doing a lot of array work that will get thrown away
-                        // most of the time.
-
-                        // Only touch the binders we have actually bound
-                        for idx in ei.match_lo..ei.match_hi {
-                            let sub = ei.matches[idx].clone();
-                            new_pos.matches[idx]
-                                   .push(Rc::new(MatchedSeq(sub, mk_sp(ei.sp_lo,
-                                                                       parser.span.hi))));
-                        }
-
-                        new_pos.match_cur = ei.match_hi;
-                        new_pos.idx += 1;
-                        cur_eis.push(new_pos);
+        let idx = ei.idx;
+        let len = ei.top_elts.len();
+
+        // at end of sequence
+        if idx >= len {
+            // We are repeating iff there is a parent
+            if ei.up.is_some() {
+                // Disregarding the separator, add the "up" case to the tokens that should be
+                // examined.
+                // (remove this condition to make trailing seps ok)
+                if idx == len {
+                    let mut new_pos = ei.up.clone().unwrap();
+
+                    // update matches (the MBE "parse tree") by appending
+                    // each tree as a subtree.
+
+                    // I bet this is a perf problem: we're preemptively
+                    // doing a lot of array work that will get thrown away
+                    // most of the time.
+
+                    // Only touch the binders we have actually bound
+                    for idx in ei.match_lo..ei.match_hi {
+                        let sub = ei.matches[idx].clone();
+                        new_pos.matches[idx]
+                            .push(Rc::new(MatchedSeq(sub, mk_sp(ei.sp_lo,
+                                                                span.hi))));
                     }
 
-                    // can we go around again?
-
-                    // Check if we need a separator
-                    if idx == len && ei.sep.is_some() {
-                        if ei.sep.as_ref().map(|ref sep| token_name_eq(&parser.token, sep))
-                            .unwrap_or(false) {
-                            // i'm conflicted about whether this should be hygienic....  though in
-                            // this case, if the separators are never legal idents, it shouldn't
-                            // matter.
-                            // ei.match_cur = ei.match_lo;
-                            ei.idx += 1;
-                            next_eis.push(ei);
-                        }
-                    } else { // we don't need a separator
-                        ei.match_cur = ei.match_lo;
-                        ei.idx = 0;
-                        cur_eis.push(ei);
-                    }
-                } else {
-                    eof_eis.push(ei);
+                    new_pos.match_cur = ei.match_hi;
+                    new_pos.idx += 1;
+                    cur_eis.push(new_pos);
                 }
-            } else {
-                match ei.top_elts.get_tt(idx) {
-                    /* need to descend into sequence */
-                    TokenTree::Sequence(sp, seq) => {
-                        if seq.op == tokenstream::KleeneOp::ZeroOrMore {
-                            let mut new_ei = ei.clone();
-                            new_ei.match_cur += seq.num_captures;
-                            new_ei.idx += 1;
-                            //we specifically matched zero repeats.
-                            for idx in ei.match_cur..ei.match_cur + seq.num_captures {
-                                new_ei.matches[idx].push(Rc::new(MatchedSeq(vec![], sp)));
-                            }
-
-                            cur_eis.push(new_ei);
-                        }
 
-                        let matches: Vec<_> = (0..ei.matches.len())
-                            .map(|_| Vec::new()).collect();
-                        cur_eis.push(Box::new(MatcherPos {
-                            stack: vec![],
-                            sep: seq.separator.clone(),
-                            idx: 0,
-                            matches: matches,
-                            match_lo: ei.match_cur,
-                            match_cur: ei.match_cur,
-                            match_hi: ei.match_cur + seq.num_captures,
-                            up: Some(ei),
-                            sp_lo: sp.lo,
-                            top_elts: Tt(TokenTree::Sequence(sp, seq)),
-                        }));
+                // Check if we need a separator
+                if idx == len && ei.sep.is_some() {
+                    // We have a separator, and it is the current token.
+                    if ei.sep.as_ref().map(|ref sep| token_name_eq(&token, sep)).unwrap_or(false) {
+                        ei.idx += 1;
+                        next_eis.push(ei);
                     }
-                    TokenTree::Token(_, MatchNt(..)) => {
-                        // Built-in nonterminals never start with these tokens,
-                        // so we can eliminate them from consideration.
-                        match parser.token {
-                            token::CloseDelim(_) => {},
-                            _ => bb_eis.push(ei),
+                } else { // we don't need a separator
+                    ei.match_cur = ei.match_lo;
+                    ei.idx = 0;
+                    cur_eis.push(ei);
+                }
+            } else {
+                // We aren't repeating, so we must be potentially at the end of the input.
+                eof_eis.push(ei);
+            }
+        } else {
+            match ei.top_elts.get_tt(idx) {
+                /* need to descend into sequence */
+                TokenTree::Sequence(sp, seq) => {
+                    if seq.op == tokenstream::KleeneOp::ZeroOrMore {
+                        // Examine the case where there are 0 matches of this sequence
+                        let mut new_ei = ei.clone();
+                        new_ei.match_cur += seq.num_captures;
+                        new_ei.idx += 1;
+                        for idx in ei.match_cur..ei.match_cur + seq.num_captures {
+                            new_ei.matches[idx].push(Rc::new(MatchedSeq(vec![], sp)));
                         }
+                        cur_eis.push(new_ei);
                     }
-                    TokenTree::Token(sp, SubstNt(..)) => {
-                        return Error(sp, "missing fragment specifier".to_string())
-                    }
-                    seq @ TokenTree::Delimited(..) | seq @ TokenTree::Token(_, DocComment(..)) => {
-                        let lower_elts = mem::replace(&mut ei.top_elts, Tt(seq));
-                        let idx = ei.idx;
-                        ei.stack.push(MatcherTtFrame {
-                            elts: lower_elts,
-                            idx: idx,
-                        });
-                        ei.idx = 0;
-                        cur_eis.push(ei);
+
+                    // Examine the case where there is at least one match of this sequence
+                    let matches = create_matches(ei.matches.len());
+                    cur_eis.push(Box::new(MatcherPos {
+                        stack: vec![],
+                        sep: seq.separator.clone(),
+                        idx: 0,
+                        matches: matches,
+                        match_lo: ei.match_cur,
+                        match_cur: ei.match_cur,
+                        match_hi: ei.match_cur + seq.num_captures,
+                        up: Some(ei),
+                        sp_lo: sp.lo,
+                        top_elts: Tt(TokenTree::Sequence(sp, seq)),
+                    }));
+                }
+                TokenTree::Token(_, MatchNt(..)) => {
+                    // Built-in nonterminals never start with these tokens,
+                    // so we can eliminate them from consideration.
+                    match *token {
+                        token::CloseDelim(_) => {},
+                        _ => bb_eis.push(ei),
                     }
-                    TokenTree::Token(_, ref t) => {
-                        if token_name_eq(t, &parser.token) {
-                            ei.idx += 1;
-                            next_eis.push(ei);
-                        }
+                }
+                TokenTree::Token(sp, SubstNt(..)) => {
+                    return Error(sp, "missing fragment specifier".to_string())
+                }
+                seq @ TokenTree::Delimited(..) | seq @ TokenTree::Token(_, DocComment(..)) => {
+                    let lower_elts = mem::replace(&mut ei.top_elts, Tt(seq));
+                    let idx = ei.idx;
+                    ei.stack.push(MatcherTtFrame {
+                        elts: lower_elts,
+                        idx: idx,
+                    });
+                    ei.idx = 0;
+                    cur_eis.push(ei);
+                }
+                TokenTree::Token(_, ref t) => {
+                    if token_name_eq(t, &token) {
+                        ei.idx += 1;
+                        next_eis.push(ei);
                     }
                 }
             }
         }
+    }
+
+    Success(())
+}
+
+pub fn parse(sess: &ParseSess, rdr: TtReader, ms: &[TokenTree]) -> NamedParseResult {
+    let mut parser = Parser::new_with_doc_flag(sess, Box::new(rdr), true);
+    let mut cur_eis = SmallVector::one(initial_matcher_pos(ms.to_owned(), parser.span.lo));
+
+    loop {
+        let mut bb_eis = Vec::new(); // black-box parsed by parser.rs
+        let mut next_eis = Vec::new(); // or proceed normally
+
+        // FIXME: Use SmallVector since in the successful case we will only have one
+        let mut eof_eis = Vec::new();
+
+        match inner_parse_loop(&mut cur_eis, &mut next_eis, &mut eof_eis, &mut bb_eis,
+                               &parser.token, &parser.span) {
+            Success(_) => {},
+            Failure(sp, tok) => return Failure(sp, tok),
+            Error(sp, msg) => return Error(sp, msg),
+        }
+
+        // inner parse loop handled all cur_eis, so it's empty
+        assert!(cur_eis.is_empty());
 
         /* error messages here could be improved with links to orig. rules */
         if token_name_eq(&parser.token, &token::Eof) {
             if eof_eis.len() == 1 {
-                let v = eof_eis[0].matches.iter_mut()
-                    .map(|dv| dv.pop().unwrap()).collect::<Vec<_>>();
-                return nameize(ms, &v[..]);
+                return nameize(ms, eof_eis[0].matches.iter_mut().map(|mut dv| dv.pop().unwrap()));
             } else if eof_eis.len() > 1 {
                 return Error(parser.span, "ambiguity: multiple successful parses".to_string());
             } else {
@@ -473,7 +480,7 @@ pub fn parse(sess: &ParseSess, rdr: TtReader, ms: &[TokenTree]) -> NamedParseRes
     }
 }
 
-pub fn parse_nt<'a>(p: &mut Parser<'a>, sp: Span, name: &str) -> Nonterminal {
+fn parse_nt<'a>(p: &mut Parser<'a>, sp: Span, name: &str) -> Nonterminal {
     match name {
         "tt" => {
             p.quote_depth += 1; //but in theory, non-quoted tts might be useful