Use crossbeam to get safety & stability.

This converts `map`, `unordered_map`, `Pool::map` and
`Pool::unordered_map` to take a `crossbeam::Scope` argument, instead of
using the API of the old `thread::scoped` to return a guard (where the
destructor was required to run to ensure safety, something not
guaranteed). See rust-lang/rust#24292 for more
discussion of the issues with that API.

Author: huonw

Cargo.toml

@@ -20,6 +20,8 @@ operations, including parallel maps, for loops and thread pools.
 strided = "*"
 num = "*"
 num_cpus = "0.2"
+[dependencies]
+crossbeam = "0.1"
 
 [features]
 unstable = []

src/lib.rs

@@ -211,26 +211,19 @@
 //! contains more intricate example(s), such as a parallel fast
 //! Fourier transform implementation (it really works, and the
 //! parallelism does buy something... when tuned).
-#![cfg_attr(feature = "unstable", feature(scoped))]
 
-#[cfg(feature = "unstable")]
-use std::thread;
-#[cfg(feature = "unstable")]
-use std::iter::IntoIterator;
+extern crate crossbeam;
 
-#[cfg(feature = "unstable")]
 mod maps;
 
 mod fnbox;
 
 pub mod pool;
 
 pub mod one_to_one {
-    #[cfg(feature = "unstable")]
     pub use maps::{unordered_map, UnorderedParMap, map, ParMap};
 }
 
-#[cfg(feature = "unstable")]
 pub use one_to_one::{map, unordered_map};
 
 pub use pool::Pool;
@@ -241,15 +234,14 @@ pub use pool::Pool;
 /// If `f` panics, so does `for_`. If this occurs, the number of
 /// elements of `iter` that have had `f` called on them is
 /// unspecified.
-#[cfg(feature = "unstable")]
 pub fn for_<I: IntoIterator, F>(iter: I, ref f: F)
     where I::Item: Send, F: Fn(I::Item) + Sync
 {
-    let _guards: Vec<_> = iter.into_iter().map(|elem| {
-        thread::scoped(move || {
-            f(elem)
-        })
-    }).collect();
+    crossbeam::scope(|scope| {
+        for elem in iter {
+            scope.spawn(move || f(elem));
+        }
+    });
 }
 
 /// Execute `f` on both `x` and `y`, in parallel, returning the
@@ -258,14 +250,16 @@ pub fn for_<I: IntoIterator, F>(iter: I, ref f: F)
 /// This is the same (including panic semantics) as `(f(x), f(y))`, up
 /// to ordering. It is designed to be used for divide-and-conquer
 /// algorithms.
-#[cfg(feature = "unstable")]
 pub fn both<T, U, F>(x: T, y: T, ref f: F) -> (U, U)
     where T: Send,
           U: Send,
           F: Sync + Fn(T) -> U
 {
-    let guard = thread::scoped(move || f(y));
-    let a = f(x);
-    let b = guard.join();
-    (a, b)
+    crossbeam::scope(|scope| {
+        let guard = scope.spawn(move || f(y));
+        let a = f(x);
+        let b = guard.join();
+        (a, b)
+
+    })
 }

src/maps.rs

@@ -1,9 +1,10 @@
 use std::sync::mpsc::{self, Sender, Receiver};
-use std::thread::{self,JoinGuard};
 use std::cmp::Ordering;
 use std::collections::BinaryHeap;
 use std::iter::IntoIterator;
 
+use crossbeam::{Scope, ScopedJoinHandle};
+
 struct Packet<T> {
     // this should be unique for a given instance of `*ParMap`
     idx: usize,
@@ -25,12 +26,12 @@ impl<T> Eq for Packet<T> {}
 
 /// A parallel-mapping iterator that doesn't care about the order in
 /// which elements come out.
-pub struct UnorderedParMap<'a, T: 'a + Send> {
+pub struct UnorderedParMap<T: Send> {
     rx: Receiver<Packet<T>>,
-    _guards: Vec<JoinGuard<'a, ()>>
+    _guards: Vec<ScopedJoinHandle<()>>
 }
 
-impl<'a,T: 'static + Send> Iterator for UnorderedParMap<'a, T> {
+impl<T: 'static + Send> Iterator for UnorderedParMap<T> {
     type Item = (usize, T);
 
     fn next(&mut self) -> Option<(usize, T)> {
@@ -62,9 +63,9 @@ impl<T: Send + 'static> Drop for Panicker<T> {
 /// This behaves like `simple_parallel::map`, but does not make
 /// efforts to ensure that the elements are returned in the order of
 /// `iter`, hence this is cheaper.
-pub fn unordered_map<'a, I: IntoIterator, F, T>(iter: I, f: &'a F) -> UnorderedParMap<'a, T>
+pub fn unordered_map<'a, I: IntoIterator, F, T>(scope: &Scope<'a>, iter: I, f: &'a F) -> UnorderedParMap<T>
     where I::Item: Send + 'a,
-          F: 'a + Sync + Fn(I::Item) -> T,
+          F: Sync + Fn(I::Item) -> T,
           T: Send + 'static
 {
     let (tx, rx) = mpsc::channel();
@@ -73,7 +74,7 @@ pub fn unordered_map<'a, I: IntoIterator, F, T>(iter: I, f: &'a F) -> UnorderedP
         let tx = tx.clone();
         let f = f.clone();
 
-        thread::scoped(move || {
+        scope.spawn(move || {
             let mut p = Panicker { tx: tx, idx: idx, all_ok: false };
             let val = f(elem);
             let _ = p.tx.send(Packet { idx: idx, data: Some(val) });
@@ -88,13 +89,13 @@ pub fn unordered_map<'a, I: IntoIterator, F, T>(iter: I, f: &'a F) -> UnorderedP
 }
 
 /// A parallel-mapping iterator.
-pub struct ParMap<'a, T: 'a + Send> {
-    unordered: UnorderedParMap<'a, T>,
+pub struct ParMap<T:  Send> {
+    unordered: UnorderedParMap<T>,
     looking_for: usize,
     queue: BinaryHeap<Packet<T>>
 }
 
-impl<'a, T: Send + 'static> Iterator for ParMap<'a, T> {
+impl<T: Send + 'static> Iterator for ParMap<T> {
     type Item = T;
 
     fn next(&mut self) -> Option<T> {
@@ -127,13 +128,13 @@ impl<'a, T: Send + 'static> Iterator for ParMap<'a, T> {
 /// This is a drop-in replacement for `iter.map(f)`, that runs in
 /// parallel, and eagerly consumes `iter` spawning a thread for each
 /// element.
-pub fn map<'a, I: IntoIterator, F, T>(iter: I, f: &'a F) -> ParMap<'a, T>
-    where I::Item: Send + 'a,
-          F: 'a + Sync + Fn(I::Item) -> T,
+pub fn map<'a, I: IntoIterator, F, T>(scope: &Scope<'a>, iter: I, f: &'a F) -> ParMap<T>
+    where I::Item: 'a + Send,
+          F: Sync + Fn(I::Item) -> T,
           T: Send + 'static
 {
     ParMap {
-        unordered: unordered_map(iter, f),
+        unordered: unordered_map(scope, iter, f),
         looking_for: 0,
         queue: BinaryHeap::new(),
     }