743 lines
23 KiB
Rust
743 lines
23 KiB
Rust
// Copyright 2012-2013 The Rust Project Developers. See the COPYRIGHT
|
|
// file at the top-level directory of this distribution and at
|
|
// http://rust-lang.org/COPYRIGHT.
|
|
//
|
|
// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
|
|
// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
|
|
// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
|
|
// option. This file may not be copied, modified, or distributed
|
|
// except according to those terms.
|
|
|
|
use ast;
|
|
use ast::{P, Ident, Name, Mrk};
|
|
use ast_util;
|
|
use ext::mtwt;
|
|
use parse::token;
|
|
use util::interner::{RcStr, StrInterner};
|
|
use util::interner;
|
|
|
|
use serialize::{Decodable, Decoder, Encodable, Encoder};
|
|
use std::cast;
|
|
use std::char;
|
|
use std::fmt;
|
|
use std::local_data;
|
|
use std::path::BytesContainer;
|
|
use std::rc::Rc;
|
|
use std::strbuf::StrBuf;
|
|
|
|
#[allow(non_camel_case_types)]
|
|
#[deriving(Clone, Encodable, Decodable, Eq, TotalEq, Hash, Show)]
|
|
pub enum BinOp {
|
|
PLUS,
|
|
MINUS,
|
|
STAR,
|
|
SLASH,
|
|
PERCENT,
|
|
CARET,
|
|
AND,
|
|
OR,
|
|
SHL,
|
|
SHR,
|
|
}
|
|
|
|
#[allow(non_camel_case_types)]
|
|
#[deriving(Clone, Encodable, Decodable, Eq, TotalEq, Hash, Show)]
|
|
pub enum Token {
|
|
/* Expression-operator symbols. */
|
|
EQ,
|
|
LT,
|
|
LE,
|
|
EQEQ,
|
|
NE,
|
|
GE,
|
|
GT,
|
|
ANDAND,
|
|
OROR,
|
|
NOT,
|
|
TILDE,
|
|
BINOP(BinOp),
|
|
BINOPEQ(BinOp),
|
|
|
|
/* Structural symbols */
|
|
AT,
|
|
DOT,
|
|
DOTDOT,
|
|
DOTDOTDOT,
|
|
COMMA,
|
|
SEMI,
|
|
COLON,
|
|
MOD_SEP,
|
|
RARROW,
|
|
LARROW,
|
|
DARROW,
|
|
FAT_ARROW,
|
|
LPAREN,
|
|
RPAREN,
|
|
LBRACKET,
|
|
RBRACKET,
|
|
LBRACE,
|
|
RBRACE,
|
|
POUND,
|
|
DOLLAR,
|
|
|
|
/* Literals */
|
|
LIT_CHAR(u32),
|
|
LIT_INT(i64, ast::IntTy),
|
|
LIT_UINT(u64, ast::UintTy),
|
|
LIT_INT_UNSUFFIXED(i64),
|
|
LIT_FLOAT(ast::Ident, ast::FloatTy),
|
|
LIT_FLOAT_UNSUFFIXED(ast::Ident),
|
|
LIT_STR(ast::Ident),
|
|
LIT_STR_RAW(ast::Ident, uint), /* raw str delimited by n hash symbols */
|
|
|
|
/* Name components */
|
|
// an identifier contains an "is_mod_name" boolean,
|
|
// indicating whether :: follows this token with no
|
|
// whitespace in between.
|
|
IDENT(ast::Ident, bool),
|
|
UNDERSCORE,
|
|
LIFETIME(ast::Ident),
|
|
|
|
/* For interpolation */
|
|
INTERPOLATED(Nonterminal),
|
|
|
|
DOC_COMMENT(ast::Ident),
|
|
EOF,
|
|
}
|
|
|
|
#[deriving(Clone, Encodable, Decodable, Eq, TotalEq, Hash)]
|
|
/// For interpolation during macro expansion.
|
|
pub enum Nonterminal {
|
|
NtItem(@ast::Item),
|
|
NtBlock(P<ast::Block>),
|
|
NtStmt(@ast::Stmt),
|
|
NtPat( @ast::Pat),
|
|
NtExpr(@ast::Expr),
|
|
NtTy( P<ast::Ty>),
|
|
NtIdent(~ast::Ident, bool),
|
|
NtMeta(@ast::MetaItem), // stuff inside brackets for attributes
|
|
NtPath(~ast::Path),
|
|
NtTT( @ast::TokenTree), // needs @ed to break a circularity
|
|
NtMatchers(Vec<ast::Matcher> )
|
|
}
|
|
|
|
impl fmt::Show for Nonterminal {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
match *self {
|
|
NtItem(..) => f.pad("NtItem(..)"),
|
|
NtBlock(..) => f.pad("NtBlock(..)"),
|
|
NtStmt(..) => f.pad("NtStmt(..)"),
|
|
NtPat(..) => f.pad("NtPat(..)"),
|
|
NtExpr(..) => f.pad("NtExpr(..)"),
|
|
NtTy(..) => f.pad("NtTy(..)"),
|
|
NtIdent(..) => f.pad("NtIdent(..)"),
|
|
NtMeta(..) => f.pad("NtMeta(..)"),
|
|
NtPath(..) => f.pad("NtPath(..)"),
|
|
NtTT(..) => f.pad("NtTT(..)"),
|
|
NtMatchers(..) => f.pad("NtMatchers(..)"),
|
|
}
|
|
}
|
|
}
|
|
|
|
pub fn binop_to_str(o: BinOp) -> ~str {
|
|
match o {
|
|
PLUS => "+".to_owned(),
|
|
MINUS => "-".to_owned(),
|
|
STAR => "*".to_owned(),
|
|
SLASH => "/".to_owned(),
|
|
PERCENT => "%".to_owned(),
|
|
CARET => "^".to_owned(),
|
|
AND => "&".to_owned(),
|
|
OR => "|".to_owned(),
|
|
SHL => "<<".to_owned(),
|
|
SHR => ">>".to_owned()
|
|
}
|
|
}
|
|
|
|
pub fn to_str(t: &Token) -> ~str {
|
|
match *t {
|
|
EQ => "=".to_owned(),
|
|
LT => "<".to_owned(),
|
|
LE => "<=".to_owned(),
|
|
EQEQ => "==".to_owned(),
|
|
NE => "!=".to_owned(),
|
|
GE => ">=".to_owned(),
|
|
GT => ">".to_owned(),
|
|
NOT => "!".to_owned(),
|
|
TILDE => "~".to_owned(),
|
|
OROR => "||".to_owned(),
|
|
ANDAND => "&&".to_owned(),
|
|
BINOP(op) => binop_to_str(op),
|
|
BINOPEQ(op) => binop_to_str(op) + "=",
|
|
|
|
/* Structural symbols */
|
|
AT => "@".to_owned(),
|
|
DOT => ".".to_owned(),
|
|
DOTDOT => "..".to_owned(),
|
|
DOTDOTDOT => "...".to_owned(),
|
|
COMMA => ",".to_owned(),
|
|
SEMI => ";".to_owned(),
|
|
COLON => ":".to_owned(),
|
|
MOD_SEP => "::".to_owned(),
|
|
RARROW => "->".to_owned(),
|
|
LARROW => "<-".to_owned(),
|
|
DARROW => "<->".to_owned(),
|
|
FAT_ARROW => "=>".to_owned(),
|
|
LPAREN => "(".to_owned(),
|
|
RPAREN => ")".to_owned(),
|
|
LBRACKET => "[".to_owned(),
|
|
RBRACKET => "]".to_owned(),
|
|
LBRACE => "{".to_owned(),
|
|
RBRACE => "}".to_owned(),
|
|
POUND => "#".to_owned(),
|
|
DOLLAR => "$".to_owned(),
|
|
|
|
/* Literals */
|
|
LIT_CHAR(c) => {
|
|
let mut res = StrBuf::from_str("'");
|
|
char::from_u32(c).unwrap().escape_default(|c| {
|
|
res.push_char(c);
|
|
});
|
|
res.push_char('\'');
|
|
res.into_owned()
|
|
}
|
|
LIT_INT(i, t) => ast_util::int_ty_to_str(t, Some(i)),
|
|
LIT_UINT(u, t) => ast_util::uint_ty_to_str(t, Some(u)),
|
|
LIT_INT_UNSUFFIXED(i) => { i.to_str() }
|
|
LIT_FLOAT(s, t) => {
|
|
let mut body = StrBuf::from_str(get_ident(s).get());
|
|
if body.as_slice().ends_with(".") {
|
|
body.push_char('0'); // `10.f` is not a float literal
|
|
}
|
|
body.push_str(ast_util::float_ty_to_str(t));
|
|
body.into_owned()
|
|
}
|
|
LIT_FLOAT_UNSUFFIXED(s) => {
|
|
let mut body = StrBuf::from_str(get_ident(s).get());
|
|
if body.as_slice().ends_with(".") {
|
|
body.push_char('0'); // `10.f` is not a float literal
|
|
}
|
|
body.into_owned()
|
|
}
|
|
LIT_STR(s) => {
|
|
format!("\"{}\"", get_ident(s).get().escape_default())
|
|
}
|
|
LIT_STR_RAW(s, n) => {
|
|
format!("r{delim}\"{string}\"{delim}",
|
|
delim="#".repeat(n), string=get_ident(s))
|
|
}
|
|
|
|
/* Name components */
|
|
IDENT(s, _) => get_ident(s).get().to_str(),
|
|
LIFETIME(s) => {
|
|
format!("'{}", get_ident(s))
|
|
}
|
|
UNDERSCORE => "_".to_owned(),
|
|
|
|
/* Other */
|
|
DOC_COMMENT(s) => get_ident(s).get().to_str(),
|
|
EOF => "<eof>".to_owned(),
|
|
INTERPOLATED(ref nt) => {
|
|
match nt {
|
|
&NtExpr(e) => ::print::pprust::expr_to_str(e),
|
|
&NtMeta(e) => ::print::pprust::meta_item_to_str(e),
|
|
_ => {
|
|
"an interpolated ".to_owned() +
|
|
match *nt {
|
|
NtItem(..) => "item".to_owned(),
|
|
NtBlock(..) => "block".to_owned(),
|
|
NtStmt(..) => "statement".to_owned(),
|
|
NtPat(..) => "pattern".to_owned(),
|
|
NtMeta(..) => fail!("should have been handled"),
|
|
NtExpr(..) => fail!("should have been handled above"),
|
|
NtTy(..) => "type".to_owned(),
|
|
NtIdent(..) => "identifier".to_owned(),
|
|
NtPath(..) => "path".to_owned(),
|
|
NtTT(..) => "tt".to_owned(),
|
|
NtMatchers(..) => "matcher sequence".to_owned()
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
pub fn can_begin_expr(t: &Token) -> bool {
|
|
match *t {
|
|
LPAREN => true,
|
|
LBRACE => true,
|
|
LBRACKET => true,
|
|
IDENT(_, _) => true,
|
|
UNDERSCORE => true,
|
|
TILDE => true,
|
|
LIT_CHAR(_) => true,
|
|
LIT_INT(_, _) => true,
|
|
LIT_UINT(_, _) => true,
|
|
LIT_INT_UNSUFFIXED(_) => true,
|
|
LIT_FLOAT(_, _) => true,
|
|
LIT_FLOAT_UNSUFFIXED(_) => true,
|
|
LIT_STR(_) => true,
|
|
LIT_STR_RAW(_, _) => true,
|
|
POUND => true,
|
|
AT => true,
|
|
NOT => true,
|
|
BINOP(MINUS) => true,
|
|
BINOP(STAR) => true,
|
|
BINOP(AND) => true,
|
|
BINOP(OR) => true, // in lambda syntax
|
|
OROR => true, // in lambda syntax
|
|
MOD_SEP => true,
|
|
INTERPOLATED(NtExpr(..))
|
|
| INTERPOLATED(NtIdent(..))
|
|
| INTERPOLATED(NtBlock(..))
|
|
| INTERPOLATED(NtPath(..)) => true,
|
|
_ => false
|
|
}
|
|
}
|
|
|
|
/// Returns the matching close delimiter if this is an open delimiter,
|
|
/// otherwise `None`.
|
|
pub fn close_delimiter_for(t: &Token) -> Option<Token> {
|
|
match *t {
|
|
LPAREN => Some(RPAREN),
|
|
LBRACE => Some(RBRACE),
|
|
LBRACKET => Some(RBRACKET),
|
|
_ => None
|
|
}
|
|
}
|
|
|
|
pub fn is_lit(t: &Token) -> bool {
|
|
match *t {
|
|
LIT_CHAR(_) => true,
|
|
LIT_INT(_, _) => true,
|
|
LIT_UINT(_, _) => true,
|
|
LIT_INT_UNSUFFIXED(_) => true,
|
|
LIT_FLOAT(_, _) => true,
|
|
LIT_FLOAT_UNSUFFIXED(_) => true,
|
|
LIT_STR(_) => true,
|
|
LIT_STR_RAW(_, _) => true,
|
|
_ => false
|
|
}
|
|
}
|
|
|
|
pub fn is_ident(t: &Token) -> bool {
|
|
match *t { IDENT(_, _) => true, _ => false }
|
|
}
|
|
|
|
pub fn is_ident_or_path(t: &Token) -> bool {
|
|
match *t {
|
|
IDENT(_, _) | INTERPOLATED(NtPath(..)) => true,
|
|
_ => false
|
|
}
|
|
}
|
|
|
|
pub fn is_plain_ident(t: &Token) -> bool {
|
|
match *t { IDENT(_, false) => true, _ => false }
|
|
}
|
|
|
|
pub fn is_bar(t: &Token) -> bool {
|
|
match *t { BINOP(OR) | OROR => true, _ => false }
|
|
}
|
|
|
|
// Get the first "argument"
|
|
macro_rules! first {
|
|
( $first:expr, $( $remainder:expr, )* ) => ( $first )
|
|
}
|
|
|
|
// Get the last "argument" (has to be done recursively to avoid phoney local ambiguity error)
|
|
macro_rules! last {
|
|
( $first:expr, $( $remainder:expr, )+ ) => ( last!( $( $remainder, )+ ) );
|
|
( $first:expr, ) => ( $first )
|
|
}
|
|
|
|
// In this macro, there is the requirement that the name (the number) must be monotonically
|
|
// increasing by one in the special identifiers, starting at 0; the same holds for the keywords,
|
|
// except starting from the next number instead of zero, and with the additional exception that
|
|
// special identifiers are *also* allowed (they are deduplicated in the important place, the
|
|
// interner), an exception which is demonstrated by "static" and "self".
|
|
macro_rules! declare_special_idents_and_keywords {(
|
|
// So now, in these rules, why is each definition parenthesised?
|
|
// Answer: otherwise we get a spurious local ambiguity bug on the "}"
|
|
pub mod special_idents {
|
|
$( ($si_name:expr, $si_static:ident, $si_str:expr); )*
|
|
}
|
|
|
|
pub mod keywords {
|
|
'strict:
|
|
$( ($sk_name:expr, $sk_variant:ident, $sk_str:expr); )*
|
|
'reserved:
|
|
$( ($rk_name:expr, $rk_variant:ident, $rk_str:expr); )*
|
|
}
|
|
) => {
|
|
static STRICT_KEYWORD_START: Name = first!($( $sk_name, )*);
|
|
static STRICT_KEYWORD_FINAL: Name = last!($( $sk_name, )*);
|
|
static RESERVED_KEYWORD_START: Name = first!($( $rk_name, )*);
|
|
static RESERVED_KEYWORD_FINAL: Name = last!($( $rk_name, )*);
|
|
|
|
pub mod special_idents {
|
|
use ast::Ident;
|
|
$( pub static $si_static: Ident = Ident { name: $si_name, ctxt: 0 }; )*
|
|
}
|
|
|
|
/**
|
|
* All the valid words that have meaning in the Rust language.
|
|
*
|
|
* Rust keywords are either 'strict' or 'reserved'. Strict keywords may not
|
|
* appear as identifiers at all. Reserved keywords are not used anywhere in
|
|
* the language and may not appear as identifiers.
|
|
*/
|
|
pub mod keywords {
|
|
use ast::Ident;
|
|
|
|
pub enum Keyword {
|
|
$( $sk_variant, )*
|
|
$( $rk_variant, )*
|
|
}
|
|
|
|
impl Keyword {
|
|
pub fn to_ident(&self) -> Ident {
|
|
match *self {
|
|
$( $sk_variant => Ident { name: $sk_name, ctxt: 0 }, )*
|
|
$( $rk_variant => Ident { name: $rk_name, ctxt: 0 }, )*
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
fn mk_fresh_ident_interner() -> IdentInterner {
|
|
// The indices here must correspond to the numbers in
|
|
// special_idents, in Keyword to_ident(), and in static
|
|
// constants below.
|
|
let mut init_vec = Vec::new();
|
|
$(init_vec.push($si_str);)*
|
|
$(init_vec.push($sk_str);)*
|
|
$(init_vec.push($rk_str);)*
|
|
interner::StrInterner::prefill(init_vec.as_slice())
|
|
}
|
|
}}
|
|
|
|
// If the special idents get renumbered, remember to modify these two as appropriate
|
|
static SELF_KEYWORD_NAME: Name = 1;
|
|
static STATIC_KEYWORD_NAME: Name = 2;
|
|
|
|
declare_special_idents_and_keywords! {
|
|
pub mod special_idents {
|
|
// These ones are statics
|
|
(0, invalid, "");
|
|
(super::SELF_KEYWORD_NAME, self_, "self");
|
|
(super::STATIC_KEYWORD_NAME, statik, "static");
|
|
|
|
// for matcher NTs
|
|
(3, tt, "tt");
|
|
(4, matchers, "matchers");
|
|
|
|
// outside of libsyntax
|
|
(5, clownshoe_abi, "__rust_abi");
|
|
(6, opaque, "<opaque>");
|
|
(7, unnamed_field, "<unnamed_field>");
|
|
(8, type_self, "Self");
|
|
}
|
|
|
|
pub mod keywords {
|
|
// These ones are variants of the Keyword enum
|
|
|
|
'strict:
|
|
(9, As, "as");
|
|
(10, Break, "break");
|
|
(11, Const, "const");
|
|
(12, Crate, "crate");
|
|
(13, Else, "else");
|
|
(14, Enum, "enum");
|
|
(15, Extern, "extern");
|
|
(16, False, "false");
|
|
(17, Fn, "fn");
|
|
(18, For, "for");
|
|
(19, If, "if");
|
|
(20, Impl, "impl");
|
|
(21, In, "in");
|
|
(22, Let, "let");
|
|
(23, Loop, "loop");
|
|
(24, Match, "match");
|
|
(25, Mod, "mod");
|
|
(26, Mut, "mut");
|
|
(27, Once, "once");
|
|
(28, Pub, "pub");
|
|
(29, Ref, "ref");
|
|
(30, Return, "return");
|
|
// Static and Self are also special idents (prefill de-dupes)
|
|
(super::STATIC_KEYWORD_NAME, Static, "static");
|
|
(super::SELF_KEYWORD_NAME, Self, "self");
|
|
(31, Struct, "struct");
|
|
(32, Super, "super");
|
|
(33, True, "true");
|
|
(34, Trait, "trait");
|
|
(35, Type, "type");
|
|
(36, Unsafe, "unsafe");
|
|
(37, Use, "use");
|
|
(38, While, "while");
|
|
(39, Continue, "continue");
|
|
(40, Proc, "proc");
|
|
(41, Box, "box");
|
|
|
|
'reserved:
|
|
(42, Alignof, "alignof");
|
|
(43, Be, "be");
|
|
(44, Offsetof, "offsetof");
|
|
(45, Priv, "priv");
|
|
(46, Pure, "pure");
|
|
(47, Sizeof, "sizeof");
|
|
(48, Typeof, "typeof");
|
|
(49, Unsized, "unsized");
|
|
(50, Yield, "yield");
|
|
(51, Do, "do");
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Maps a token to a record specifying the corresponding binary
|
|
* operator
|
|
*/
|
|
pub fn token_to_binop(tok: &Token) -> Option<ast::BinOp> {
|
|
match *tok {
|
|
BINOP(STAR) => Some(ast::BiMul),
|
|
BINOP(SLASH) => Some(ast::BiDiv),
|
|
BINOP(PERCENT) => Some(ast::BiRem),
|
|
BINOP(PLUS) => Some(ast::BiAdd),
|
|
BINOP(MINUS) => Some(ast::BiSub),
|
|
BINOP(SHL) => Some(ast::BiShl),
|
|
BINOP(SHR) => Some(ast::BiShr),
|
|
BINOP(AND) => Some(ast::BiBitAnd),
|
|
BINOP(CARET) => Some(ast::BiBitXor),
|
|
BINOP(OR) => Some(ast::BiBitOr),
|
|
LT => Some(ast::BiLt),
|
|
LE => Some(ast::BiLe),
|
|
GE => Some(ast::BiGe),
|
|
GT => Some(ast::BiGt),
|
|
EQEQ => Some(ast::BiEq),
|
|
NE => Some(ast::BiNe),
|
|
ANDAND => Some(ast::BiAnd),
|
|
OROR => Some(ast::BiOr),
|
|
_ => None
|
|
}
|
|
}
|
|
|
|
// looks like we can get rid of this completely...
|
|
pub type IdentInterner = StrInterner;
|
|
|
|
// if an interner exists in TLS, return it. Otherwise, prepare a
|
|
// fresh one.
|
|
// FIXME(eddyb) #8726 This should probably use a task-local reference.
|
|
pub fn get_ident_interner() -> Rc<IdentInterner> {
|
|
local_data_key!(key: Rc<::parse::token::IdentInterner>)
|
|
match local_data::get(key, |k| k.map(|k| k.clone())) {
|
|
Some(interner) => interner,
|
|
None => {
|
|
let interner = Rc::new(mk_fresh_ident_interner());
|
|
local_data::set(key, interner.clone());
|
|
interner
|
|
}
|
|
}
|
|
}
|
|
|
|
/// Represents a string stored in the task-local interner. Because the
|
|
/// interner lives for the life of the task, this can be safely treated as an
|
|
/// immortal string, as long as it never crosses between tasks.
|
|
///
|
|
/// FIXME(pcwalton): You must be careful about what you do in the destructors
|
|
/// of objects stored in TLS, because they may run after the interner is
|
|
/// destroyed. In particular, they must not access string contents. This can
|
|
/// be fixed in the future by just leaking all strings until task death
|
|
/// somehow.
|
|
#[deriving(Clone, Eq, Hash, Ord, TotalEq, TotalOrd)]
|
|
pub struct InternedString {
|
|
string: RcStr,
|
|
}
|
|
|
|
impl InternedString {
|
|
#[inline]
|
|
pub fn new(string: &'static str) -> InternedString {
|
|
InternedString {
|
|
string: RcStr::new(string),
|
|
}
|
|
}
|
|
|
|
#[inline]
|
|
fn new_from_rc_str(string: RcStr) -> InternedString {
|
|
InternedString {
|
|
string: string,
|
|
}
|
|
}
|
|
|
|
#[inline]
|
|
pub fn get<'a>(&'a self) -> &'a str {
|
|
self.string.as_slice()
|
|
}
|
|
}
|
|
|
|
impl BytesContainer for InternedString {
|
|
fn container_as_bytes<'a>(&'a self) -> &'a [u8] {
|
|
// FIXME(pcwalton): This is a workaround for the incorrect signature
|
|
// of `BytesContainer`, which is itself a workaround for the lack of
|
|
// DST.
|
|
unsafe {
|
|
let this = self.get();
|
|
cast::transmute(this.container_as_bytes())
|
|
}
|
|
}
|
|
}
|
|
|
|
impl fmt::Show for InternedString {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
write!(f.buf, "{}", self.string.as_slice())
|
|
}
|
|
}
|
|
|
|
impl<'a> Equiv<&'a str> for InternedString {
|
|
fn equiv(&self, other: & &'a str) -> bool {
|
|
(*other) == self.string.as_slice()
|
|
}
|
|
}
|
|
|
|
impl<D:Decoder<E>, E> Decodable<D, E> for InternedString {
|
|
fn decode(d: &mut D) -> Result<InternedString, E> {
|
|
Ok(get_name(get_ident_interner().intern(try!(d.read_str()))))
|
|
}
|
|
}
|
|
|
|
impl<S:Encoder<E>, E> Encodable<S, E> for InternedString {
|
|
fn encode(&self, s: &mut S) -> Result<(), E> {
|
|
s.emit_str(self.string.as_slice())
|
|
}
|
|
}
|
|
|
|
/// Returns the string contents of a name, using the task-local interner.
|
|
#[inline]
|
|
pub fn get_name(name: Name) -> InternedString {
|
|
let interner = get_ident_interner();
|
|
InternedString::new_from_rc_str(interner.get(name))
|
|
}
|
|
|
|
/// Returns the string contents of an identifier, using the task-local
|
|
/// interner.
|
|
#[inline]
|
|
pub fn get_ident(ident: Ident) -> InternedString {
|
|
get_name(ident.name)
|
|
}
|
|
|
|
/// Interns and returns the string contents of an identifier, using the
|
|
/// task-local interner.
|
|
#[inline]
|
|
pub fn intern_and_get_ident(s: &str) -> InternedString {
|
|
get_name(intern(s))
|
|
}
|
|
|
|
/// Maps a string to its interned representation.
|
|
#[inline]
|
|
pub fn intern(s: &str) -> Name {
|
|
get_ident_interner().intern(s)
|
|
}
|
|
|
|
/// gensym's a new uint, using the current interner.
|
|
#[inline]
|
|
pub fn gensym(s: &str) -> Name {
|
|
get_ident_interner().gensym(s)
|
|
}
|
|
|
|
/// Maps a string to an identifier with an empty syntax context.
|
|
#[inline]
|
|
pub fn str_to_ident(s: &str) -> ast::Ident {
|
|
ast::Ident::new(intern(s))
|
|
}
|
|
|
|
/// Maps a string to a gensym'ed identifier.
|
|
#[inline]
|
|
pub fn gensym_ident(s: &str) -> ast::Ident {
|
|
ast::Ident::new(gensym(s))
|
|
}
|
|
|
|
// create a fresh name that maps to the same string as the old one.
|
|
// note that this guarantees that str_ptr_eq(ident_to_str(src),interner_get(fresh_name(src)));
|
|
// that is, that the new name and the old one are connected to ptr_eq strings.
|
|
pub fn fresh_name(src: &ast::Ident) -> Name {
|
|
let interner = get_ident_interner();
|
|
interner.gensym_copy(src.name)
|
|
// following: debug version. Could work in final except that it's incompatible with
|
|
// good error messages and uses of struct names in ambiguous could-be-binding
|
|
// locations. Also definitely destroys the guarantee given above about ptr_eq.
|
|
/*let num = rand::task_rng().gen_uint_range(0,0xffff);
|
|
gensym(format!("{}_{}",ident_to_str(src),num))*/
|
|
}
|
|
|
|
// create a fresh mark.
|
|
pub fn fresh_mark() -> Mrk {
|
|
gensym("mark")
|
|
}
|
|
|
|
// See the macro above about the types of keywords
|
|
|
|
pub fn is_keyword(kw: keywords::Keyword, tok: &Token) -> bool {
|
|
match *tok {
|
|
token::IDENT(sid, false) => { kw.to_ident().name == sid.name }
|
|
_ => { false }
|
|
}
|
|
}
|
|
|
|
pub fn is_any_keyword(tok: &Token) -> bool {
|
|
match *tok {
|
|
token::IDENT(sid, false) => match sid.name {
|
|
SELF_KEYWORD_NAME | STATIC_KEYWORD_NAME |
|
|
STRICT_KEYWORD_START .. RESERVED_KEYWORD_FINAL => true,
|
|
_ => false,
|
|
},
|
|
_ => false
|
|
}
|
|
}
|
|
|
|
pub fn is_strict_keyword(tok: &Token) -> bool {
|
|
match *tok {
|
|
token::IDENT(sid, false) => match sid.name {
|
|
SELF_KEYWORD_NAME | STATIC_KEYWORD_NAME |
|
|
STRICT_KEYWORD_START .. STRICT_KEYWORD_FINAL => true,
|
|
_ => false,
|
|
},
|
|
_ => false,
|
|
}
|
|
}
|
|
|
|
pub fn is_reserved_keyword(tok: &Token) -> bool {
|
|
match *tok {
|
|
token::IDENT(sid, false) => match sid.name {
|
|
RESERVED_KEYWORD_START .. RESERVED_KEYWORD_FINAL => true,
|
|
_ => false,
|
|
},
|
|
_ => false,
|
|
}
|
|
}
|
|
|
|
pub fn mtwt_token_eq(t1 : &Token, t2 : &Token) -> bool {
|
|
match (t1,t2) {
|
|
(&IDENT(id1,_),&IDENT(id2,_)) | (&LIFETIME(id1),&LIFETIME(id2)) =>
|
|
mtwt::resolve(id1) == mtwt::resolve(id2),
|
|
_ => *t1 == *t2
|
|
}
|
|
}
|
|
|
|
|
|
#[cfg(test)]
|
|
mod test {
|
|
use super::*;
|
|
use ast;
|
|
use ext::mtwt;
|
|
|
|
fn mark_ident(id : ast::Ident, m : ast::Mrk) -> ast::Ident {
|
|
ast::Ident{name:id.name,ctxt:mtwt::new_mark(m,id.ctxt)}
|
|
}
|
|
|
|
#[test] fn mtwt_token_eq_test() {
|
|
assert!(mtwt_token_eq(>,>));
|
|
let a = str_to_ident("bac");
|
|
let a1 = mark_ident(a,92);
|
|
assert!(mtwt_token_eq(&IDENT(a,true),&IDENT(a1,false)));
|
|
}
|
|
}
|