auto merge of #13316 : eddyb/rust/ast-ptr, r=brson

Replaces Gc<T> in the AST with a custom owned smart pointer, P<T>. Fixes #7929.

## Benefits
* **Identity** (affinity?): sharing AST nodes is bad for the various analysis passes (e.g. one could bypass borrowck with a shared `ExprAddrOf` node taking a mutable borrow), the only reason we haven't hit any serious issues with it is because of inefficient folding passes which will always deduplicate any such shared nodes. Even if we were to switch to an arena, this would still hold, i.e. we wouldn't just use `&'a T` in the AST, but rather an wrapper (`P<'a, T>`?).

* **Immutability**: `P<T>` disallows mutating its inner `T` (unless that contains an `Unsafe` interior, which won't happen in the AST), unlike `~T`.

* **Efficiency**: folding can reuse allocation space for `P<T>` and `Vec<T>`, the latter even when the input and output types differ (as it would be the case with arenas or an AST with type parameters to toggle macro support). Also, various algorithms have been changed from copying `Gc<T>` to using `&T` and iterators.

* **Maintainability**: there is another reason I didn't just replace `Gc<T>` with `~T`: `P<T>` provides a fixed interface (`Deref`, `and_then` and `map`) which can remain fully functional even if the implementation changes (using a special thread-local heap, for example). Moreover, switching to, e.g. `P<'a, T>` (for a contextual arena) is easy and mostly automated.

Author: bors

mk/crates.mk

@@ -70,7 +70,7 @@ DEPS_graphviz := std
 DEPS_green := std native:context_switch
 DEPS_rustuv := std native:uv native:uv_support
 DEPS_native := std
-DEPS_syntax := std term serialize log fmt_macros debug
+DEPS_syntax := std term serialize log fmt_macros debug arena
 DEPS_rustc := syntax flate arena serialize getopts rbml \
               time log graphviz debug rustc_llvm rustc_back
 DEPS_rustc_llvm := native:rustllvm libc std

src/libfourcc/lib.rs

@@ -63,10 +63,9 @@ use syntax::ext::base::{ExtCtxt, MacExpr};
 use syntax::ext::build::AstBuilder;
 use syntax::parse::token;
 use syntax::parse::token::InternedString;
+use syntax::ptr::P;
 use rustc::plugin::Registry;
 
-use std::gc::Gc;
-
 #[plugin_registrar]
 pub fn plugin_registrar(reg: &mut Registry) {
     reg.register_macro("fourcc", expand_syntax_ext);
@@ -135,7 +134,7 @@ struct Ident {
 }
 
 fn parse_tts(cx: &ExtCtxt,
-             tts: &[ast::TokenTree]) -> (Gc<ast::Expr>, Option<Ident>) {
+             tts: &[ast::TokenTree]) -> (P<ast::Expr>, Option<Ident>) {
     let p = &mut cx.new_parser_from_tts(tts);
     let ex = p.parse_expr();
     let id = if p.token == token::EOF {
@@ -156,7 +155,7 @@ fn parse_tts(cx: &ExtCtxt,
 fn target_endian_little(cx: &ExtCtxt, sp: Span) -> bool {
     let meta = cx.meta_name_value(sp, InternedString::new("target_endian"),
         ast::LitStr(InternedString::new("little"), ast::CookedStr));
-    contains(cx.cfg().as_slice(), meta)
+    contains(cx.cfg().as_slice(), &*meta)
 }
 
 // FIXME (10872): This is required to prevent an LLVM assert on Windows

src/libhexfloat/lib.rs

@@ -57,10 +57,9 @@ use syntax::ext::base;
 use syntax::ext::base::{ExtCtxt, MacExpr};
 use syntax::ext::build::AstBuilder;
 use syntax::parse::token;
+use syntax::ptr::P;
 use rustc::plugin::Registry;
 
-use std::gc::Gc;
-
 #[plugin_registrar]
 pub fn plugin_registrar(reg: &mut Registry) {
     reg.register_macro("hexfloat", expand_syntax_ext);
@@ -122,7 +121,7 @@ pub fn expand_syntax_ext(cx: &mut ExtCtxt, sp: Span, tts: &[ast::TokenTree])
 
     let s = match expr.node {
         // expression is a literal
-        ast::ExprLit(lit) => match lit.node {
+        ast::ExprLit(ref lit) => match lit.node {
             // string literal
             ast::LitStr(ref s, _) => {
                 s.clone()
@@ -167,7 +166,7 @@ struct Ident {
 }
 
 fn parse_tts(cx: &ExtCtxt,
-             tts: &[ast::TokenTree]) -> (Gc<ast::Expr>, Option<Ident>) {
+             tts: &[ast::TokenTree]) -> (P<ast::Expr>, Option<Ident>) {
     let p = &mut cx.new_parser_from_tts(tts);
     let ex = p.parse_expr();
     let id = if p.token == token::EOF {

src/libregex_macros/lib.rs

@@ -26,7 +26,6 @@ extern crate syntax;
 extern crate rustc;
 
 use std::rc::Rc;
-use std::gc::{Gc, GC};
 
 use syntax::ast;
 use syntax::codemap;
@@ -35,6 +34,7 @@ use syntax::ext::base::{ExtCtxt, MacResult, MacExpr, DummyResult};
 use syntax::parse::token;
 use syntax::print::pprust;
 use syntax::fold::Folder;
+use syntax::ptr::P;
 
 use rustc::plugin::Registry;
 
@@ -111,7 +111,7 @@ struct NfaGen<'a> {
 }
 
 impl<'a> NfaGen<'a> {
-    fn code(&mut self) -> Gc<ast::Expr> {
+    fn code(&mut self) -> P<ast::Expr> {
         // Most or all of the following things are used in the quasiquoted
         // expression returned.
         let num_cap_locs = 2 * self.prog.num_captures();
@@ -332,7 +332,7 @@ fn exec<'t>(which: ::regex::native::MatchKind, input: &'t str,
 
     // Generates code for the `add` method, which is responsible for adding
     // zero-width states to the next queue of states to visit.
-    fn add_insts(&self) -> Gc<ast::Expr> {
+    fn add_insts(&self) -> P<ast::Expr> {
         let arms = self.prog.insts.iter().enumerate().map(|(pc, inst)| {
             let nextpc = pc + 1;
             let body = match *inst {
@@ -433,7 +433,7 @@ fn exec<'t>(which: ::regex::native::MatchKind, input: &'t str,
 
     // Generates the code for the `step` method, which processes all states
     // in the current queue that consume a single character.
-    fn step_insts(&self) -> Gc<ast::Expr> {
+    fn step_insts(&self) -> P<ast::Expr> {
         let arms = self.prog.insts.iter().enumerate().map(|(pc, inst)| {
             let nextpc = pc + 1;
             let body = match *inst {
@@ -524,17 +524,15 @@ fn exec<'t>(which: ::regex::native::MatchKind, input: &'t str,
     // Translates a character class into a match expression.
     // This avoids a binary search (and is hopefully replaced by a jump
     // table).
-    fn match_class(&self, casei: bool, ranges: &[(char, char)]) -> Gc<ast::Expr> {
-        let expr_true = quote_expr!(self.cx, true);
-
+    fn match_class(&self, casei: bool, ranges: &[(char, char)]) -> P<ast::Expr> {
         let mut arms = ranges.iter().map(|&(mut start, mut end)| {
             if casei {
                 start = start.to_uppercase();
                 end = end.to_uppercase();
             }
             let pat = self.cx.pat(self.sp, ast::PatRange(quote_expr!(self.cx, $start),
                                                          quote_expr!(self.cx, $end)));
-            self.cx.arm(self.sp, vec!(pat), expr_true)
+            self.cx.arm(self.sp, vec!(pat), quote_expr!(self.cx, true))
         }).collect::<Vec<ast::Arm>>();
 
         arms.push(self.wild_arm_expr(quote_expr!(self.cx, false)));
@@ -546,7 +544,7 @@ fn exec<'t>(which: ::regex::native::MatchKind, input: &'t str,
     // Generates code for checking a literal prefix of the search string.
     // The code is only generated if the regex *has* a literal prefix.
     // Otherwise, a no-op is returned.
-    fn check_prefix(&self) -> Gc<ast::Expr> {
+    fn check_prefix(&self) -> P<ast::Expr> {
         if self.prog.prefix.len() == 0 {
             self.empty_block()
         } else {
@@ -570,32 +568,32 @@ fn exec<'t>(which: ::regex::native::MatchKind, input: &'t str,
     // A wild-card arm is automatically added that executes a no-op. It will
     // never be used, but is added to satisfy the compiler complaining about
     // non-exhaustive patterns.
-    fn match_insts(&self, mut arms: Vec<ast::Arm>) -> Gc<ast::Expr> {
+    fn match_insts(&self, mut arms: Vec<ast::Arm>) -> P<ast::Expr> {
         arms.push(self.wild_arm_expr(self.empty_block()));
         self.cx.expr_match(self.sp, quote_expr!(self.cx, pc), arms)
     }
 
-    fn empty_block(&self) -> Gc<ast::Expr> {
+    fn empty_block(&self) -> P<ast::Expr> {
         quote_expr!(self.cx, {})
     }
 
     // Creates a match arm for the instruction at `pc` with the expression
     // `body`.
-    fn arm_inst(&self, pc: uint, body: Gc<ast::Expr>) -> ast::Arm {
+    fn arm_inst(&self, pc: uint, body: P<ast::Expr>) -> ast::Arm {
         let pc_pat = self.cx.pat_lit(self.sp, quote_expr!(self.cx, $pc));
 
         self.cx.arm(self.sp, vec!(pc_pat), body)
     }
 
     // Creates a wild-card match arm with the expression `body`.
-    fn wild_arm_expr(&self, body: Gc<ast::Expr>) -> ast::Arm {
+    fn wild_arm_expr(&self, body: P<ast::Expr>) -> ast::Arm {
         ast::Arm {
             attrs: vec!(),
-            pats: vec!(box(GC) ast::Pat{
+            pats: vec!(P(ast::Pat{
                 id: ast::DUMMY_NODE_ID,
                 span: self.sp,
                 node: ast::PatWild(ast::PatWildSingle),
-            }),
+            })),
             guard: None,
             body: body,
         }
@@ -605,8 +603,8 @@ fn exec<'t>(which: ::regex::native::MatchKind, input: &'t str,
     // Converts `xs` to a `[x1, x2, .., xN]` expression by calling `to_expr`
     // on each element in `xs`.
     fn vec_expr<T, It: Iterator<T>>(&self, xs: It,
-                                    to_expr: |&ExtCtxt, T| -> Gc<ast::Expr>)
-                  -> Gc<ast::Expr> {
+                                    to_expr: |&ExtCtxt, T| -> P<ast::Expr>)
+                  -> P<ast::Expr> {
         let exprs = xs.map(|x| to_expr(self.cx, x)).collect();
         self.cx.expr_vec(self.sp, exprs)
     }
@@ -618,13 +616,13 @@ fn parse(cx: &mut ExtCtxt, tts: &[ast::TokenTree]) -> Option<String> {
     let mut parser = cx.new_parser_from_tts(tts);
     let entry = cx.expander().fold_expr(parser.parse_expr());
     let regex = match entry.node {
-        ast::ExprLit(lit) => {
+        ast::ExprLit(ref lit) => {
             match lit.node {
                 ast::LitStr(ref s, _) => s.to_string(),
                 _ => {
                     cx.span_err(entry.span, format!(
                         "expected string literal but got `{}`",
-                        pprust::lit_to_string(&*lit)).as_slice());
+                        pprust::lit_to_string(&**lit)).as_slice());
                     return None
                 }
             }