Re-inline ena

Cargo.lock

@@ -55,14 +55,6 @@ name = "either"
 version = "1.5.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 
-[[package]]
-name = "ena"
-version = "0.13.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-dependencies = [
- "log 0.4.6 (registry+https://github.com/rust-lang/crates.io-index)",
-]
-
 [[package]]
 name = "fixedbitset"
 version = "0.1.9"
@@ -123,7 +115,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 
 [[package]]
 name = "log"
-version = "0.4.6"
+version = "0.4.8"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 dependencies = [
  "cfg-if 0.1.6 (registry+https://github.com/rust-lang/crates.io-index)",
@@ -263,12 +255,11 @@ version = "0.1.0"
 dependencies = [
  "combine 3.8.1 (registry+https://github.com/rust-lang/crates.io-index)",
  "dogged 0.2.0 (registry+https://github.com/rust-lang/crates.io-index)",
- "ena 0.13.0 (registry+https://github.com/rust-lang/crates.io-index)",
  "fraction 0.6.2 (registry+https://github.com/rust-lang/crates.io-index)",
  "fxhash 0.2.1 (registry+https://github.com/rust-lang/crates.io-index)",
  "im-rc 13.0.0 (registry+https://github.com/rust-lang/crates.io-index)",
  "indoc 0.3.3 (registry+https://github.com/rust-lang/crates.io-index)",
- "log 0.4.6 (registry+https://github.com/rust-lang/crates.io-index)",
+ "log 0.4.8 (registry+https://github.com/rust-lang/crates.io-index)",
  "maplit 1.0.1 (registry+https://github.com/rust-lang/crates.io-index)",
  "num 0.2.0 (registry+https://github.com/rust-lang/crates.io-index)",
  "petgraph 0.4.13 (registry+https://github.com/rust-lang/crates.io-index)",
@@ -371,7 +362,6 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 "checksum difference 2.0.0 (registry+https://github.com/rust-lang/crates.io-index)" = "524cbf6897b527295dff137cec09ecf3a05f4fddffd7dfcd1585403449e74198"
 "checksum dogged 0.2.0 (registry+https://github.com/rust-lang/crates.io-index)" = "2638df109789fe360f0d9998c5438dd19a36678aaf845e46f285b688b1a1657a"
 "checksum either 1.5.2 (registry+https://github.com/rust-lang/crates.io-index)" = "5527cfe0d098f36e3f8839852688e63c8fff1c90b2b405aef730615f9a7bcf7b"
-"checksum ena 0.13.0 (registry+https://github.com/rust-lang/crates.io-index)" = "3dc01d68e08ca384955a3aeba9217102ca1aa85b6e168639bf27739f1d749d87"
 "checksum fixedbitset 0.1.9 (registry+https://github.com/rust-lang/crates.io-index)" = "86d4de0081402f5e88cdac65c8dcdcc73118c1a7a465e2a05f0da05843a8ea33"
 "checksum fraction 0.6.2 (registry+https://github.com/rust-lang/crates.io-index)" = "1055159ac82fb210c813303f716b6c8db57ace9d5ec2dbbc2e1d7a864c1dd74e"
 "checksum fxhash 0.2.1 (registry+https://github.com/rust-lang/crates.io-index)" = "c31b6d751ae2c7f11320402d34e41349dd1016f8d5d45e48c4312bc8625af50c"
@@ -379,7 +369,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 "checksum indoc 0.3.3 (registry+https://github.com/rust-lang/crates.io-index)" = "a1f59f228c76fda6ecd8dab79683039a7054c748587f682a911094f473647bd6"
 "checksum indoc-impl 0.3.3 (registry+https://github.com/rust-lang/crates.io-index)" = "63f070ef080db3601c1a0ecc75c7bb35104cc0ce2d7c4e049952a96a61d8933b"
 "checksum lazy_static 1.3.0 (registry+https://github.com/rust-lang/crates.io-index)" = "bc5729f27f159ddd61f4df6228e827e86643d4d3e7c32183cb30a1c08f604a14"
-"checksum log 0.4.6 (registry+https://github.com/rust-lang/crates.io-index)" = "c84ec4b527950aa83a329754b01dbe3f58361d1c5efacd1f6d68c494d08a17c6"
+"checksum log 0.4.8 (registry+https://github.com/rust-lang/crates.io-index)" = "14b6052be84e6b71ab17edffc2eeabf5c2c3ae1fdb464aae35ac50c67a44e1f7"
 "checksum maplit 1.0.1 (registry+https://github.com/rust-lang/crates.io-index)" = "08cbb6b4fef96b6d77bfc40ec491b1690c779e77b05cd9f07f787ed376fd4c43"
 "checksum memchr 2.2.1 (registry+https://github.com/rust-lang/crates.io-index)" = "88579771288728879b57485cc7d6b07d648c9f0141eb955f8ab7f9d45394468e"
 "checksum num 0.2.0 (registry+https://github.com/rust-lang/crates.io-index)" = "cf4825417e1e1406b3782a8ce92f4d53f26ec055e3622e1881ca8e9f5f9e08db"

Cargo.toml

@@ -5,14 +5,13 @@ authors = ["Richard Feldman <oss@rtfeldman.com>"]
 
 [dependencies]
 dogged = { version = "0.2.0", optional = true }
-log = "0.4"
+log = "0.4.8"
 petgraph = { version = "0.4.5", optional = true }
 combine = "3.8.1"
 im-rc = "13.0.0"
 fraction = "0.6.2"
 num = "0.2.0"
 fxhash = "0.2.1"
-ena = "0.13.0"
 
 [dev-dependencies]
 pretty_assertions = "0.5.1"

src/ena/bitvec.rs

@@ -0,0 +1,301 @@
+// Copyright 2015 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.
+
+/// A very simple BitVector type.
+pub struct BitVector {
+    data: Vec<u64>,
+}
+
+impl BitVector {
+    pub fn new(num_bits: usize) -> BitVector {
+        let num_words = u64s(num_bits);
+        BitVector { data: vec![0; num_words] }
+    }
+
+    pub fn contains(&self, bit: usize) -> bool {
+        let (word, mask) = word_mask(bit);
+        (self.data[word] & mask) != 0
+    }
+
+    /// Returns true if the bit has changed.
+    pub fn insert(&mut self, bit: usize) -> bool {
+        let (word, mask) = word_mask(bit);
+        let data = &mut self.data[word];
+        let value = *data;
+        let new_value = value | mask;
+        *data = new_value;
+        new_value != value
+    }
+
+    pub fn insert_all(&mut self, all: &BitVector) -> bool {
+        assert!(self.data.len() == all.data.len());
+        let mut changed = false;
+        for (i, j) in self.data.iter_mut().zip(&all.data) {
+            let value = *i;
+            *i = value | *j;
+            if value != *i {
+                changed = true;
+            }
+        }
+        changed
+    }
+
+    pub fn grow(&mut self, num_bits: usize) {
+        let num_words = u64s(num_bits);
+        let extra_words = self.data.len() - num_words;
+        self.data.extend((0..extra_words).map(|_| 0));
+    }
+
+    /// Iterates over indexes of set bits in a sorted order
+    pub fn iter<'a>(&'a self) -> BitVectorIter<'a> {
+        BitVectorIter {
+            iter: self.data.iter(),
+            current: 0,
+            idx: 0,
+        }
+    }
+}
+
+pub struct BitVectorIter<'a> {
+    iter: ::std::slice::Iter<'a, u64>,
+    current: u64,
+    idx: usize,
+}
+
+impl<'a> Iterator for BitVectorIter<'a> {
+    type Item = usize;
+    fn next(&mut self) -> Option<usize> {
+        while self.current == 0 {
+            self.current = if let Some(&i) = self.iter.next() {
+                if i == 0 {
+                    self.idx += 64;
+                    continue;
+                } else {
+                    self.idx = u64s(self.idx) * 64;
+                    i
+                }
+            } else {
+                return None;
+            }
+        }
+        let offset = self.current.trailing_zeros() as usize;
+        self.current >>= offset;
+        self.current >>= 1; // shift otherwise overflows for 0b1000_0000_…_0000
+        self.idx += offset + 1;
+        return Some(self.idx - 1);
+    }
+}
+
+/// A "bit matrix" is basically a square matrix of booleans
+/// represented as one gigantic bitvector. In other words, it is as if
+/// you have N bitvectors, each of length N. Note that `elements` here is `N`/
+#[derive(Clone)]
+pub struct BitMatrix {
+    elements: usize,
+    vector: Vec<u64>,
+}
+
+impl BitMatrix {
+    // Create a new `elements x elements` matrix, initially empty.
+    pub fn new(elements: usize) -> BitMatrix {
+        // For every element, we need one bit for every other
+        // element. Round up to an even number of u64s.
+        let u64s_per_elem = u64s(elements);
+        BitMatrix {
+            elements: elements,
+            vector: vec![0; elements * u64s_per_elem],
+        }
+    }
+
+    /// The range of bits for a given element.
+    fn range(&self, element: usize) -> (usize, usize) {
+        let u64s_per_elem = u64s(self.elements);
+        let start = element * u64s_per_elem;
+        (start, start + u64s_per_elem)
+    }
+
+    pub fn add(&mut self, source: usize, target: usize) -> bool {
+        let (start, _) = self.range(source);
+        let (word, mask) = word_mask(target);
+        let mut vector = &mut self.vector[..];
+        let v1 = vector[start + word];
+        let v2 = v1 | mask;
+        vector[start + word] = v2;
+        v1 != v2
+    }
+
+    /// Do the bits from `source` contain `target`?
+    ///
+    /// Put another way, if the matrix represents (transitive)
+    /// reachability, can `source` reach `target`?
+    pub fn contains(&self, source: usize, target: usize) -> bool {
+        let (start, _) = self.range(source);
+        let (word, mask) = word_mask(target);
+        (self.vector[start + word] & mask) != 0
+    }
+
+    /// Returns those indices that are reachable from both `a` and
+    /// `b`. This is an O(n) operation where `n` is the number of
+    /// elements (somewhat independent from the actual size of the
+    /// intersection, in particular).
+    pub fn intersection(&self, a: usize, b: usize) -> Vec<usize> {
+        let (a_start, a_end) = self.range(a);
+        let (b_start, b_end) = self.range(b);
+        let mut result = Vec::with_capacity(self.elements);
+        for (base, (i, j)) in (a_start..a_end).zip(b_start..b_end).enumerate() {
+            let mut v = self.vector[i] & self.vector[j];
+            for bit in 0..64 {
+                if v == 0 {
+                    break;
+                }
+                if v & 0x1 != 0 {
+                    result.push(base * 64 + bit);
+                }
+                v >>= 1;
+            }
+        }
+        result
+    }
+
+    /// Add the bits from `read` to the bits from `write`,
+    /// return true if anything changed.
+    ///
+    /// This is used when computing transitive reachability because if
+    /// you have an edge `write -> read`, because in that case
+    /// `write` can reach everything that `read` can (and
+    /// potentially more).
+    pub fn merge(&mut self, read: usize, write: usize) -> bool {
+        let (read_start, read_end) = self.range(read);
+        let (write_start, write_end) = self.range(write);
+        let vector = &mut self.vector[..];
+        let mut changed = false;
+        for (read_index, write_index) in (read_start..read_end).zip(write_start..write_end) {
+            let v1 = vector[write_index];
+            let v2 = v1 | vector[read_index];
+            vector[write_index] = v2;
+            changed = changed | (v1 != v2);
+        }
+        changed
+    }
+}
+
+fn u64s(elements: usize) -> usize {
+    (elements + 63) / 64
+}
+
+fn word_mask(index: usize) -> (usize, u64) {
+    let word = index / 64;
+    let mask = 1 << (index % 64);
+    (word, mask)
+}
+
+#[test]
+fn bitvec_iter_works() {
+    let mut bitvec = BitVector::new(100);
+    bitvec.insert(1);
+    bitvec.insert(10);
+    bitvec.insert(19);
+    bitvec.insert(62);
+    bitvec.insert(63);
+    bitvec.insert(64);
+    bitvec.insert(65);
+    bitvec.insert(66);
+    bitvec.insert(99);
+    assert_eq!(bitvec.iter().collect::<Vec<_>>(),
+               [1, 10, 19, 62, 63, 64, 65, 66, 99]);
+}
+
+#[test]
+fn bitvec_iter_works_2() {
+    let mut bitvec = BitVector::new(300);
+    bitvec.insert(1);
+    bitvec.insert(10);
+    bitvec.insert(19);
+    bitvec.insert(62);
+    bitvec.insert(66);
+    bitvec.insert(99);
+    bitvec.insert(299);
+    assert_eq!(bitvec.iter().collect::<Vec<_>>(),
+               [1, 10, 19, 62, 66, 99, 299]);
+
+}
+
+#[test]
+fn bitvec_iter_works_3() {
+    let mut bitvec = BitVector::new(319);
+    bitvec.insert(0);
+    bitvec.insert(127);
+    bitvec.insert(191);
+    bitvec.insert(255);
+    bitvec.insert(319);
+    assert_eq!(bitvec.iter().collect::<Vec<_>>(), [0, 127, 191, 255, 319]);
+}
+
+#[test]
+fn union_two_vecs() {
+    let mut vec1 = BitVector::new(65);
+    let mut vec2 = BitVector::new(65);
+    assert!(vec1.insert(3));
+    assert!(!vec1.insert(3));
+    assert!(vec2.insert(5));
+    assert!(vec2.insert(64));
+    assert!(vec1.insert_all(&vec2));
+    assert!(!vec1.insert_all(&vec2));
+    assert!(vec1.contains(3));
+    assert!(!vec1.contains(4));
+    assert!(vec1.contains(5));
+    assert!(!vec1.contains(63));
+    assert!(vec1.contains(64));
+}
+
+#[test]
+fn grow() {
+    let mut vec1 = BitVector::new(65);
+    assert!(vec1.insert(3));
+    assert!(!vec1.insert(3));
+    assert!(vec1.insert(5));
+    assert!(vec1.insert(64));
+    vec1.grow(128);
+    assert!(vec1.contains(3));
+    assert!(vec1.contains(5));
+    assert!(vec1.contains(64));
+    assert!(!vec1.contains(126));
+}
+
+#[test]
+fn matrix_intersection() {
+    let mut vec1 = BitMatrix::new(200);
+
+    // (*) Elements reachable from both 2 and 65.
+
+    vec1.add(2, 3);
+    vec1.add(2, 6);
+    vec1.add(2, 10); // (*)
+    vec1.add(2, 64); // (*)
+    vec1.add(2, 65);
+    vec1.add(2, 130);
+    vec1.add(2, 160); // (*)
+
+    vec1.add(64, 133);
+
+    vec1.add(65, 2);
+    vec1.add(65, 8);
+    vec1.add(65, 10); // (*)
+    vec1.add(65, 64); // (*)
+    vec1.add(65, 68);
+    vec1.add(65, 133);
+    vec1.add(65, 160); // (*)
+
+    let intersection = vec1.intersection(2, 64);
+    assert!(intersection.is_empty());
+
+    let intersection = vec1.intersection(2, 65);
+    assert_eq!(intersection, &[10, 64, 160]);
+}

src/ena/mod.rs

@@ -0,0 +1,15 @@
+// Copyright 2015 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.
+
+//! An implementation of union-find. See the `unify` module for more
+//! details.
+
+pub mod snapshot_vec;
+pub mod unify;

src/ena/snapshot_vec.rs

@@ -0,0 +1,374 @@
+// Copyright 2014 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.
+
+//! A utility class for implementing "snapshottable" things; a snapshottable data structure permits
+//! you to take a snapshot (via `start_snapshot`) and then, after making some changes, elect either
+//! to rollback to the start of the snapshot or commit those changes.
+//!
+//! This vector is intended to be used as part of an abstraction, not serve as a complete
+//! abstraction on its own. As such, while it will roll back most changes on its own, it also
+//! supports a `get_mut` operation that gives you an arbitrary mutable pointer into the vector. To
+//! ensure that any changes you make this with this pointer are rolled back, you must invoke
+//! `record` to record any changes you make and also supplying a delegate capable of reversing
+//! those changes.
+
+use self::UndoLog::*;
+
+use std::fmt;
+use std::mem;
+use std::ops;
+
+#[derive(Debug)]
+pub enum UndoLog<D: SnapshotVecDelegate> {
+    /// New variable with given index was created.
+    NewElem(usize),
+
+    /// Variable with given index was changed *from* the given value.
+    SetElem(usize, D::Value),
+
+    /// Extensible set of actions
+    Other(D::Undo),
+}
+
+pub struct SnapshotVec<D: SnapshotVecDelegate> {
+    values: Vec<D::Value>,
+    undo_log: Vec<UndoLog<D>>,
+    num_open_snapshots: usize,
+}
+
+impl<D> fmt::Debug for SnapshotVec<D>
+    where D: SnapshotVecDelegate,
+          D: fmt::Debug,
+          D::Undo: fmt::Debug,
+          D::Value: fmt::Debug
+{
+    fn fmt(&self, fmt: &mut fmt::Formatter) -> fmt::Result {
+        fmt.debug_struct("SnapshotVec")
+            .field("values", &self.values)
+            .field("undo_log", &self.undo_log)
+            .field("num_open_snapshots", &self.num_open_snapshots)
+            .finish()
+    }
+}
+
+// Snapshots are tokens that should be created/consumed linearly.
+pub struct Snapshot {
+    // Number of values at the time the snapshot was taken.
+    pub(crate) value_count: usize,
+    // Length of the undo log at the time the snapshot was taken.
+    undo_len: usize,
+}
+
+pub trait SnapshotVecDelegate {
+    type Value;
+    type Undo;
+
+    fn reverse(values: &mut Vec<Self::Value>, action: Self::Undo);
+}
+
+// HACK(eddyb) manual impl avoids `Default` bound on `D`.
+impl<D: SnapshotVecDelegate> Default for SnapshotVec<D> {
+    fn default() -> Self {
+        SnapshotVec {
+            values: Vec::new(),
+            undo_log: Vec::new(),
+            num_open_snapshots: 0,
+        }
+    }
+}
+
+impl<D: SnapshotVecDelegate> SnapshotVec<D> {
+    pub fn new() -> Self {
+        Self::default()
+    }
+
+    pub fn with_capacity(c: usize) -> SnapshotVec<D> {
+        SnapshotVec {
+            values: Vec::with_capacity(c),
+            undo_log: Vec::new(),
+            num_open_snapshots: 0,
+        }
+    }
+
+    fn in_snapshot(&self) -> bool {
+        self.num_open_snapshots > 0
+    }
+
+    pub fn record(&mut self, action: D::Undo) {
+        if self.in_snapshot() {
+            self.undo_log.push(Other(action));
+        }
+    }
+
+    pub fn len(&self) -> usize {
+        self.values.len()
+    }
+
+    pub fn push(&mut self, elem: D::Value) -> usize {
+        let len = self.values.len();
+        self.values.push(elem);
+
+        if self.in_snapshot() {
+            self.undo_log.push(NewElem(len));
+        }
+
+        len
+    }
+
+    pub fn get(&self, index: usize) -> &D::Value {
+        &self.values[index]
+    }
+
+    /// Reserve space for new values, just like an ordinary vec.
+    pub fn reserve(&mut self, additional: usize) {
+        // This is not affected by snapshots or anything.
+        self.values.reserve(additional);
+    }
+
+    /// Returns a mutable pointer into the vec; whatever changes you make here cannot be undone
+    /// automatically, so you should be sure call `record()` with some sort of suitable undo
+    /// action.
+    pub fn get_mut(&mut self, index: usize) -> &mut D::Value {
+        &mut self.values[index]
+    }
+
+    /// Updates the element at the given index. The old value will saved (and perhaps restored) if
+    /// a snapshot is active.
+    pub fn set(&mut self, index: usize, new_elem: D::Value) {
+        let old_elem = mem::replace(&mut self.values[index], new_elem);
+        if self.in_snapshot() {
+            self.undo_log.push(SetElem(index, old_elem));
+        }
+    }
+
+    /// Updates all elements. Potentially more efficient -- but
+    /// otherwise equivalent to -- invoking `set` for each element.
+    pub fn set_all(&mut self, mut new_elems: impl FnMut(usize) -> D::Value) {
+        if !self.in_snapshot() {
+            for (index, slot) in self.values.iter_mut().enumerate() {
+                *slot = new_elems(index);
+            }
+        } else {
+            for i in 0..self.values.len() {
+                self.set(i, new_elems(i));
+            }
+        }
+    }
+
+    pub fn update<OP>(&mut self, index: usize, op: OP)
+    where
+        OP: FnOnce(&mut D::Value),
+        D::Value: Clone,
+    {
+        if self.in_snapshot() {
+            let old_elem = self.values[index].clone();
+            self.undo_log.push(SetElem(index, old_elem));
+        }
+        op(&mut self.values[index]);
+    }
+
+    pub fn start_snapshot(&mut self) -> Snapshot {
+        self.num_open_snapshots += 1;
+        Snapshot {
+            value_count: self.values.len(),
+            undo_len: self.undo_log.len(),
+        }
+    }
+
+    pub fn actions_since_snapshot(&self, snapshot: &Snapshot) -> &[UndoLog<D>] {
+        &self.undo_log[snapshot.undo_len..]
+    }
+
+    fn assert_open_snapshot(&self, snapshot: &Snapshot) {
+        // Failures here may indicate a failure to follow a stack discipline.
+        assert!(self.undo_log.len() >= snapshot.undo_len);
+        assert!(self.num_open_snapshots > 0);
+    }
+
+    pub fn rollback_to(&mut self, snapshot: Snapshot) {
+        debug!("rollback_to({})", snapshot.undo_len);
+
+        self.assert_open_snapshot(&snapshot);
+
+        while self.undo_log.len() > snapshot.undo_len {
+            match self.undo_log.pop().unwrap() {
+                NewElem(i) => {
+                    self.values.pop();
+                    assert!(self.values.len() == i);
+                }
+
+                SetElem(i, v) => {
+                    self.values[i] = v;
+                }
+
+                Other(u) => {
+                    D::reverse(&mut self.values, u);
+                }
+            }
+        }
+
+        self.num_open_snapshots -= 1;
+    }
+
+    /// Commits all changes since the last snapshot. Of course, they
+    /// can still be undone if there is a snapshot further out.
+    pub fn commit(&mut self, snapshot: Snapshot) {
+        debug!("commit({})", snapshot.undo_len);
+
+        self.assert_open_snapshot(&snapshot);
+
+        if self.num_open_snapshots == 1 {
+            // The root snapshot. It's safe to clear the undo log because
+            // there's no snapshot further out that we might need to roll back
+            // to.
+            assert!(snapshot.undo_len == 0);
+            self.undo_log.clear();
+        }
+
+        self.num_open_snapshots -= 1;
+    }
+}
+
+impl<D: SnapshotVecDelegate> ops::Deref for SnapshotVec<D> {
+    type Target = [D::Value];
+    fn deref(&self) -> &[D::Value] {
+        &*self.values
+    }
+}
+
+impl<D: SnapshotVecDelegate> ops::DerefMut for SnapshotVec<D> {
+    fn deref_mut(&mut self) -> &mut [D::Value] {
+        &mut *self.values
+    }
+}
+
+impl<D: SnapshotVecDelegate> ops::Index<usize> for SnapshotVec<D> {
+    type Output = D::Value;
+    fn index(&self, index: usize) -> &D::Value {
+        self.get(index)
+    }
+}
+
+impl<D: SnapshotVecDelegate> ops::IndexMut<usize> for SnapshotVec<D> {
+    fn index_mut(&mut self, index: usize) -> &mut D::Value {
+        self.get_mut(index)
+    }
+}
+
+impl<D: SnapshotVecDelegate> Extend<D::Value> for SnapshotVec<D> {
+    fn extend<T>(&mut self, iterable: T)
+    where
+        T: IntoIterator<Item = D::Value>,
+    {
+        let initial_len = self.values.len();
+        self.values.extend(iterable);
+        let final_len = self.values.len();
+
+        if self.in_snapshot() {
+            self.undo_log.extend((initial_len..final_len).map(|len| NewElem(len)));
+        }
+    }
+}
+
+impl<D: SnapshotVecDelegate> Clone for SnapshotVec<D>
+where
+    D::Value: Clone,
+    D::Undo: Clone,
+{
+    fn clone(&self) -> Self {
+        SnapshotVec {
+            values: self.values.clone(),
+            undo_log: self.undo_log.clone(),
+            num_open_snapshots: self.num_open_snapshots,
+        }
+    }
+}
+
+impl<D: SnapshotVecDelegate> Clone for UndoLog<D>
+where
+    D::Value: Clone,
+    D::Undo: Clone,
+{
+    fn clone(&self) -> Self {
+        match *self {
+            NewElem(i) => NewElem(i),
+            SetElem(i, ref v) => SetElem(i, v.clone()),
+            Other(ref u) => Other(u.clone()),
+        }
+    }
+}
+
+impl SnapshotVecDelegate for i32 {
+    type Value = i32;
+    type Undo = ();
+
+    fn reverse(_: &mut Vec<i32>, _: ()) {}
+}
+
+#[test]
+fn basic() {
+    let mut vec: SnapshotVec<i32> = SnapshotVec::default();
+    assert!(!vec.in_snapshot());
+    assert_eq!(vec.len(), 0);
+    vec.push(22);
+    vec.push(33);
+    assert_eq!(vec.len(), 2);
+    assert_eq!(*vec.get(0), 22);
+    assert_eq!(*vec.get(1), 33);
+    vec.set(1, 34);
+    assert_eq!(vec.len(), 2);
+    assert_eq!(*vec.get(0), 22);
+    assert_eq!(*vec.get(1), 34);
+
+    let snapshot = vec.start_snapshot();
+    assert!(vec.in_snapshot());
+
+    vec.push(44);
+    vec.push(55);
+    vec.set(1, 35);
+    assert_eq!(vec.len(), 4);
+    assert_eq!(*vec.get(0), 22);
+    assert_eq!(*vec.get(1), 35);
+    assert_eq!(*vec.get(2), 44);
+    assert_eq!(*vec.get(3), 55);
+
+    vec.rollback_to(snapshot);
+    assert!(!vec.in_snapshot());
+
+    assert_eq!(vec.len(), 2);
+    assert_eq!(*vec.get(0), 22);
+    assert_eq!(*vec.get(1), 34);
+}
+
+#[test]
+#[should_panic]
+fn out_of_order() {
+    let mut vec: SnapshotVec<i32> = SnapshotVec::default();
+    vec.push(22);
+    let snapshot1 = vec.start_snapshot();
+    vec.push(33);
+    let snapshot2 = vec.start_snapshot();
+    vec.push(44);
+    vec.rollback_to(snapshot1); // bogus, but accepted
+    vec.rollback_to(snapshot2); // asserts
+}
+
+#[test]
+fn nested_commit_then_rollback() {
+    let mut vec: SnapshotVec<i32> = SnapshotVec::default();
+    vec.push(22);
+    let snapshot1 = vec.start_snapshot();
+    let snapshot2 = vec.start_snapshot();
+    vec.set(0, 23);
+    vec.commit(snapshot2);
+    assert_eq!(*vec.get(0), 23);
+    vec.rollback_to(snapshot1);
+    assert_eq!(*vec.get(0), 22);
+}

src/ena/unify/backing_vec.rs

@@ -0,0 +1,220 @@
+#[cfg(feature = "persistent")]
+use dogged::DVec;
+use ena::snapshot_vec as sv;
+use std::ops::{self, Range};
+use std::marker::PhantomData;
+
+use super::{VarValue, UnifyKey, UnifyValue};
+
+#[allow(dead_code)] // rustc BUG
+#[allow(type_alias_bounds)]
+type Key<S: UnificationStore> = <S as UnificationStore>::Key;
+
+/// Largely internal trait implemented by the unification table
+/// backing store types. The most common such type is `InPlace`,
+/// which indicates a standard, mutable unification table.
+pub trait UnificationStore:
+    ops::Index<usize, Output = VarValue<Key<Self>>> + Clone + Default
+{
+    type Key: UnifyKey<Value = Self::Value>;
+    type Value: UnifyValue;
+    type Snapshot;
+
+    fn start_snapshot(&mut self) -> Self::Snapshot;
+
+    fn rollback_to(&mut self, snapshot: Self::Snapshot);
+
+    fn commit(&mut self, snapshot: Self::Snapshot);
+
+    fn values_since_snapshot(&self, snapshot: &Self::Snapshot) -> Range<usize>;
+
+    fn reset_unifications(
+        &mut self,
+        value: impl FnMut(u32) -> VarValue<Self::Key>,
+    );
+
+    fn len(&self) -> usize;
+
+    fn push(&mut self, value: VarValue<Self::Key>);
+
+    fn reserve(&mut self, num_new_values: usize);
+
+    fn update<F>(&mut self, index: usize, op: F)
+        where F: FnOnce(&mut VarValue<Self::Key>);
+
+    fn tag() -> &'static str {
+        Self::Key::tag()
+    }
+}
+
+/// Backing store for an in-place unification table.
+/// Not typically used directly.
+#[derive(Clone, Debug)]
+pub struct InPlace<K: UnifyKey> {
+    values: sv::SnapshotVec<Delegate<K>>
+}
+
+// HACK(eddyb) manual impl avoids `Default` bound on `K`.
+impl<K: UnifyKey> Default for InPlace<K> {
+    fn default() -> Self {
+        InPlace { values: sv::SnapshotVec::new() }
+    }
+}
+
+impl<K: UnifyKey> UnificationStore for InPlace<K> {
+    type Key = K;
+    type Value = K::Value;
+    type Snapshot = sv::Snapshot;
+
+    #[inline]
+    fn start_snapshot(&mut self) -> Self::Snapshot {
+        self.values.start_snapshot()
+    }
+
+    #[inline]
+    fn rollback_to(&mut self, snapshot: Self::Snapshot) {
+        self.values.rollback_to(snapshot);
+    }
+
+    #[inline]
+    fn commit(&mut self, snapshot: Self::Snapshot) {
+        self.values.commit(snapshot);
+    }
+
+    #[inline]
+    fn values_since_snapshot(&self, snapshot: &Self::Snapshot) -> Range<usize> {
+        snapshot.value_count..self.len()
+    }
+
+    #[inline]
+    fn reset_unifications(
+        &mut self,
+        mut value: impl FnMut(u32) -> VarValue<Self::Key>,
+    ) {
+        self.values.set_all(|i| value(i as u32));
+    }
+
+    fn len(&self) -> usize {
+        self.values.len()
+    }
+
+    #[inline]
+    fn push(&mut self, value: VarValue<Self::Key>) {
+        self.values.push(value);
+    }
+
+    #[inline]
+    fn reserve(&mut self, num_new_values: usize) {
+        self.values.reserve(num_new_values);
+    }
+
+    #[inline]
+    fn update<F>(&mut self, index: usize, op: F)
+        where F: FnOnce(&mut VarValue<Self::Key>)
+    {
+        self.values.update(index, op)
+    }
+}
+
+impl<K> ops::Index<usize> for InPlace<K>
+    where K: UnifyKey
+{
+    type Output = VarValue<K>;
+    fn index(&self, index: usize) -> &VarValue<K> {
+        &self.values[index]
+    }
+}
+
+#[derive(Copy, Clone, Debug)]
+struct Delegate<K>(PhantomData<K>);
+
+impl<K: UnifyKey> sv::SnapshotVecDelegate for Delegate<K> {
+    type Value = VarValue<K>;
+    type Undo = ();
+
+    fn reverse(_: &mut Vec<VarValue<K>>, _: ()) {}
+}
+
+#[cfg(feature = "persistent")]
+#[derive(Clone, Debug)]
+pub struct Persistent<K: UnifyKey> {
+    values: DVec<VarValue<K>>
+}
+
+// HACK(eddyb) manual impl avoids `Default` bound on `K`.
+#[cfg(feature = "persistent")]
+impl<K: UnifyKey> Default for Persistent<K> {
+    fn default() -> Self {
+        Persistent { values: DVec::new() }
+    }
+}
+
+#[cfg(feature = "persistent")]
+impl<K: UnifyKey> UnificationStore for Persistent<K> {
+    type Key = K;
+    type Value = K::Value;
+    type Snapshot = Self;
+
+    #[inline]
+    fn start_snapshot(&mut self) -> Self::Snapshot {
+        self.clone()
+    }
+
+    #[inline]
+    fn rollback_to(&mut self, snapshot: Self::Snapshot) {
+        *self = snapshot;
+    }
+
+    #[inline]
+    fn commit(&mut self, _snapshot: Self::Snapshot) {}
+
+    #[inline]
+    fn values_since_snapshot(&self, snapshot: &Self::Snapshot) -> Range<usize> {
+        snapshot.len()..self.len()
+    }
+
+    #[inline]
+    fn reset_unifications(
+        &mut self,
+        mut value: impl FnMut(u32) -> VarValue<Self::Key>,
+    ) {
+        // Without extending dogged, there isn't obviously a more
+        // efficient way to do this. But it's pretty dumb. Maybe
+        // dogged needs a `map`.
+        for i in 0 .. self.values.len() {
+            self.values[i] = value(i as u32);
+        }
+    }
+
+    fn len(&self) -> usize {
+        self.values.len()
+    }
+
+    #[inline]
+    fn push(&mut self, value: VarValue<Self::Key>) {
+        self.values.push(value);
+    }
+
+    #[inline]
+    fn reserve(&mut self, _num_new_values: usize) {
+        // not obviously relevant to DVec.
+    }
+
+    #[inline]
+    fn update<F>(&mut self, index: usize, op: F)
+        where F: FnOnce(&mut VarValue<Self::Key>)
+    {
+        let p = &mut self.values[index];
+        op(p);
+    }
+}
+
+#[cfg(feature = "persistent")]
+impl<K> ops::Index<usize> for Persistent<K>
+    where K: UnifyKey
+{
+    type Output = VarValue<K>;
+    fn index(&self, index: usize) -> &VarValue<K> {
+        &self.values[index]
+    }
+}

src/ena/unify/mod.rs

@@ -0,0 +1,547 @@
+// Copyright 2012-2014 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.
+
+//! Union-find implementation. The main type is `UnificationTable`.
+//!
+//! You can define your own type for the *keys* in the table, but you
+//! must implement `UnifyKey` for that type. The assumption is that
+//! keys will be newtyped integers, hence we require that they
+//! implement `Copy`.
+//!
+//! Keys can have values associated with them. The assumption is that
+//! these values are cheaply cloneable (ideally, `Copy`), and some of
+//! the interfaces are oriented around that assumption. If you just
+//! want the classical "union-find" algorithm where you group things
+//! into sets, use the `Value` type of `()`.
+//!
+//! When you have keys with non-trivial values, you must also define
+//! how those values can be merged. As part of doing this, you can
+//! define the "error" type to return on error; if errors are not
+//! possible, use `NoError` (an uninstantiable struct). Using this
+//! type also unlocks various more ergonomic methods (e.g., `union()`
+//! in place of `unify_var_var()`).
+//!
+//! The best way to see how it is used is to read the `tests.rs` file;
+//! search for e.g. `UnitKey`.
+
+use std::marker;
+use std::fmt::Debug;
+use std::ops::Range;
+
+mod backing_vec;
+pub use self::backing_vec::{InPlace, UnificationStore};
+
+#[cfg(feature = "persistent")]
+pub use self::backing_vec::Persistent;
+
+
+#[cfg(test)]
+mod tests;
+
+/// This trait is implemented by any type that can serve as a type
+/// variable. We call such variables *unification keys*. For example,
+/// this trait is implemented by `IntVid`, which represents integral
+/// variables.
+///
+/// Each key type has an associated value type `V`. For example, for
+/// `IntVid`, this is `Option<IntVarValue>`, representing some
+/// (possibly not yet known) sort of integer.
+///
+/// Clients are expected to provide implementations of this trait; you
+/// can see some examples in the `test` module.
+pub trait UnifyKey: Copy + Clone + Debug + PartialEq {
+    type Value: UnifyValue;
+
+    fn index(&self) -> u32;
+
+    fn from_index(u: u32) -> Self;
+
+    fn tag() -> &'static str;
+
+    /// If true, then `self` should be preferred as root to `other`.
+    /// Note that we assume a consistent partial ordering, so
+    /// returning true implies that `other.prefer_as_root_to(self)`
+    /// would return false.  If there is no ordering between two keys
+    /// (i.e., `a.prefer_as_root_to(b)` and `b.prefer_as_root_to(a)`
+    /// both return false) then the rank will be used to determine the
+    /// root in an optimal way.
+    ///
+    /// NB. The only reason to implement this method is if you want to
+    /// control what value is returned from `find()`. In general, it
+    /// is better to let the unification table determine the root,
+    /// since overriding the rank can cause execution time to increase
+    /// dramatically.
+    #[allow(unused_variables)]
+    fn order_roots(
+        a: Self,
+        a_value: &Self::Value,
+        b: Self,
+        b_value: &Self::Value,
+    ) -> Option<(Self, Self)> {
+        None
+    }
+}
+
+/// Trait implemented for **values** associated with a unification
+/// key. This trait defines how to merge the values from two keys that
+/// are unioned together. This merging can be fallible. If you attempt
+/// to union two keys whose values cannot be merged, then the error is
+/// propagated up and the two keys are not unioned.
+///
+/// This crate provides implementations of `UnifyValue` for `()`
+/// (which is infallible) and `Option<T>` (where `T: UnifyValue`). The
+/// option implementation merges two sum-values using the `UnifyValue`
+/// implementation of `T`.
+///
+/// See also `EqUnifyValue`, which is a convenience trait for cases
+/// where the "merge" operation succeeds only if the two values are
+/// equal.
+pub trait UnifyValue: Clone + Debug {
+    /// Defines the type to return when merging of two values fails.
+    /// If merging is infallible, use the special struct `NoError`
+    /// found in this crate, which unlocks various more convenient
+    /// methods on the unification table.
+    type Error;
+
+    /// Given two values, produce a new value that combines them.
+    /// If that is not possible, produce an error.
+    fn unify_values(value1: &Self, value2: &Self) -> Result<Self, Self::Error>;
+}
+
+/// A convenient helper for unification values which must be equal or
+/// else an error occurs. For example, if you are unifying types in a
+/// simple functional language, this may be appropriate, since (e.g.)
+/// you can't unify a type variable bound to `int` with one bound to
+/// `float` (but you can unify two type variables both bound to
+/// `int`).
+///
+/// Any type which implements `EqUnifyValue` automatially implements
+/// `UnifyValue`; if the two values are equal, merging is permitted.
+/// Otherwise, the error `(v1, v2)` is returned, where `v1` and `v2`
+/// are the two unequal values.
+pub trait EqUnifyValue: Eq + Clone + Debug {}
+
+impl<T: EqUnifyValue> UnifyValue for T {
+    type Error = (T, T);
+
+    fn unify_values(value1: &Self, value2: &Self) -> Result<Self, Self::Error> {
+        if value1 == value2 {
+            Ok(value1.clone())
+        } else {
+            Err((value1.clone(), value2.clone()))
+        }
+    }
+}
+
+/// A struct which can never be instantiated. Used
+/// for the error type for infallible cases.
+#[derive(Debug)]
+pub struct NoError {
+    _dummy: (),
+}
+
+/// Value of a unification key. We implement Tarjan's union-find
+/// algorithm: when two keys are unified, one of them is converted
+/// into a "redirect" pointing at the other. These redirects form a
+/// DAG: the roots of the DAG (nodes that are not redirected) are each
+/// associated with a value of type `V` and a rank. The rank is used
+/// to keep the DAG relatively balanced, which helps keep the running
+/// time of the algorithm under control. For more information, see
+/// <http://en.wikipedia.org/wiki/Disjoint-set_data_structure>.
+#[derive(PartialEq, Clone, Debug)]
+pub struct VarValue<K: UnifyKey> { // FIXME pub
+    parent: K, // if equal to self, this is a root
+    value: K::Value, // value assigned (only relevant to root)
+    rank: u32, // max depth (only relevant to root)
+}
+
+/// Table of unification keys and their values. You must define a key type K
+/// that implements the `UnifyKey` trait. Unification tables can be used in two-modes:
+///
+/// - in-place (`UnificationTable<InPlace<K>>` or `InPlaceUnificationTable<K>`):
+///   - This is the standard mutable mode, where the array is modified
+///     in place.
+///   - To do backtracking, you can employ the `snapshot` and `rollback_to`
+///     methods.
+/// - persistent (`UnificationTable<Persistent<K>>` or `PersistentUnificationTable<K>`):
+///   - In this mode, we use a persistent vector to store the data, so that
+///     cloning the table is an O(1) operation.
+///   - This implies that ordinary operations are quite a bit slower though.
+///   - Requires the `persistent` feature be selected in your Cargo.toml file.
+#[derive(Clone, Debug, Default)]
+pub struct UnificationTable<S: UnificationStore> {
+    /// Indicates the current value of each key.
+    values: S,
+}
+
+/// A unification table that uses an "in-place" vector.
+#[allow(type_alias_bounds)]
+pub type InPlaceUnificationTable<K: UnifyKey> = UnificationTable<InPlace<K>>;
+
+/// A unification table that uses a "persistent" vector.
+#[cfg(feature = "persistent")]
+#[allow(type_alias_bounds)]
+pub type PersistentUnificationTable<K: UnifyKey> = UnificationTable<Persistent<K>>;
+
+/// At any time, users may snapshot a unification table.  The changes
+/// made during the snapshot may either be *committed* or *rolled back*.
+pub struct Snapshot<S: UnificationStore> {
+    // Link snapshot to the unification store `S` of the table.
+    marker: marker::PhantomData<S>,
+    snapshot: S::Snapshot,
+}
+
+impl<K: UnifyKey> VarValue<K> {
+    fn new_var(key: K, value: K::Value) -> VarValue<K> {
+        VarValue::new(key, value, 0)
+    }
+
+    fn new(parent: K, value: K::Value, rank: u32) -> VarValue<K> {
+        VarValue {
+            parent: parent, // this is a root
+            value: value,
+            rank: rank,
+        }
+    }
+
+    fn redirect(&mut self, to: K) {
+        self.parent = to;
+    }
+
+    fn root(&mut self, rank: u32, value: K::Value) {
+        self.rank = rank;
+        self.value = value;
+    }
+
+    fn parent(&self, self_key: K) -> Option<K> {
+        self.if_not_self(self.parent, self_key)
+    }
+
+    fn if_not_self(&self, key: K, self_key: K) -> Option<K> {
+        if key == self_key {
+            None
+        } else {
+            Some(key)
+        }
+    }
+}
+
+// We can't use V:LatticeValue, much as I would like to,
+// because frequently the pattern is that V=Option<U> for some
+// other type parameter U, and we have no way to say
+// Option<U>:LatticeValue.
+
+impl<S: UnificationStore> UnificationTable<S> {
+    pub fn new() -> Self {
+        Self::default()
+    }
+
+    /// Starts a new snapshot. Each snapshot must be either
+    /// rolled back or committed in a "LIFO" (stack) order.
+    pub fn snapshot(&mut self) -> Snapshot<S> {
+        Snapshot {
+            marker: marker::PhantomData::<S>,
+            snapshot: self.values.start_snapshot(),
+        }
+    }
+
+    /// Reverses all changes since the last snapshot. Also
+    /// removes any keys that have been created since then.
+    pub fn rollback_to(&mut self, snapshot: Snapshot<S>) {
+        debug!("{}: rollback_to()", S::tag());
+        self.values.rollback_to(snapshot.snapshot);
+    }
+
+    /// Commits all changes since the last snapshot. Of course, they
+    /// can still be undone if there is a snapshot further out.
+    pub fn commit(&mut self, snapshot: Snapshot<S>) {
+        debug!("{}: commit()", S::tag());
+        self.values.commit(snapshot.snapshot);
+    }
+
+    /// Creates a fresh key with the given value.
+    pub fn new_key(&mut self, value: S::Value) -> S::Key {
+        let len = self.values.len();
+        let key: S::Key = UnifyKey::from_index(len as u32);
+        self.values.push(VarValue::new_var(key, value));
+        debug!("{}: created new key: {:?}", S::tag(), key);
+        key
+    }
+
+    /// Reserve memory for `num_new_keys` to be created. Does not
+    /// actually create the new keys; you must then invoke `new_key`.
+    pub fn reserve(&mut self, num_new_keys: usize) {
+        self.values.reserve(num_new_keys);
+    }
+
+    /// Clears all unifications that have been performed, resetting to
+    /// the initial state. The values of each variable are given by
+    /// the closure.
+    pub fn reset_unifications(
+        &mut self,
+        mut value: impl FnMut(S::Key) -> S::Value,
+    ) {
+        self.values.reset_unifications(|i| {
+            let key = UnifyKey::from_index(i as u32);
+            let value = value(key);
+            VarValue::new_var(key, value)
+        });
+    }
+
+    /// Returns the number of keys created so far.
+    pub fn len(&self) -> usize {
+        self.values.len()
+    }
+
+    /// Returns the keys of all variables created since the `snapshot`.
+    pub fn vars_since_snapshot(
+        &self,
+        snapshot: &Snapshot<S>,
+    ) -> Range<S::Key> {
+        let range = self.values.values_since_snapshot(&snapshot.snapshot);
+        S::Key::from_index(range.start as u32)..S::Key::from_index(range.end as u32)
+    }
+
+    /// Obtains the current value for a particular key.
+    /// Not for end-users; they can use `probe_value`.
+    fn value(&self, key: S::Key) -> &VarValue<S::Key> {
+        &self.values[key.index() as usize]
+    }
+
+    /// Find the root node for `vid`. This uses the standard
+    /// union-find algorithm with path compression:
+    /// <http://en.wikipedia.org/wiki/Disjoint-set_data_structure>.
+    ///
+    /// NB. This is a building-block operation and you would probably
+    /// prefer to call `probe` below.
+    fn get_root_key(&mut self, vid: S::Key) -> S::Key {
+        let redirect = {
+            match self.value(vid).parent(vid) {
+                None => return vid,
+                Some(redirect) => redirect,
+            }
+        };
+
+        let root_key: S::Key = self.get_root_key(redirect);
+        if root_key != redirect {
+            // Path compression
+            self.update_value(vid, |value| value.parent = root_key);
+        }
+
+        root_key
+    }
+
+    fn update_value<OP>(&mut self, key: S::Key, op: OP)
+    where
+        OP: FnOnce(&mut VarValue<S::Key>),
+    {
+        self.values.update(key.index() as usize, op);
+        debug!("Updated variable {:?} to {:?}", key, self.value(key));
+    }
+
+    /// Either redirects `node_a` to `node_b` or vice versa, depending
+    /// on the relative rank. The value associated with the new root
+    /// will be `new_value`.
+    ///
+    /// NB: This is the "union" operation of "union-find". It is
+    /// really more of a building block. If the values associated with
+    /// your key are non-trivial, you would probably prefer to call
+    /// `unify_var_var` below.
+    fn unify_roots(&mut self, key_a: S::Key, key_b: S::Key, new_value: S::Value) {
+        debug!("unify(key_a={:?}, key_b={:?})", key_a, key_b);
+
+        let rank_a = self.value(key_a).rank;
+        let rank_b = self.value(key_b).rank;
+        if let Some((new_root, redirected)) =
+            S::Key::order_roots(
+                key_a,
+                &self.value(key_a).value,
+                key_b,
+                &self.value(key_b).value,
+            ) {
+            // compute the new rank for the new root that they chose;
+            // this may not be the optimal choice.
+            let new_rank = if new_root == key_a {
+                debug_assert!(redirected == key_b);
+                if rank_a > rank_b {
+                    rank_a
+                } else {
+                    rank_b + 1
+                }
+            } else {
+                debug_assert!(new_root == key_b);
+                debug_assert!(redirected == key_a);
+                if rank_b > rank_a {
+                    rank_b
+                } else {
+                    rank_a + 1
+                }
+            };
+            self.redirect_root(new_rank, redirected, new_root, new_value);
+        } else if rank_a > rank_b {
+            // a has greater rank, so a should become b's parent,
+            // i.e., b should redirect to a.
+            self.redirect_root(rank_a, key_b, key_a, new_value);
+        } else if rank_a < rank_b {
+            // b has greater rank, so a should redirect to b.
+            self.redirect_root(rank_b, key_a, key_b, new_value);
+        } else {
+            // If equal, redirect one to the other and increment the
+            // other's rank.
+            self.redirect_root(rank_a + 1, key_a, key_b, new_value);
+        }
+    }
+
+    /// Internal method to redirect `old_root_key` (which is currently
+    /// a root) to a child of `new_root_key` (which will remain a
+    /// root). The rank and value of `new_root_key` will be updated to
+    /// `new_rank` and `new_value` respectively.
+    fn redirect_root(
+        &mut self,
+        new_rank: u32,
+        old_root_key: S::Key,
+        new_root_key: S::Key,
+        new_value: S::Value,
+    ) {
+        self.update_value(old_root_key, |old_root_value| {
+            old_root_value.redirect(new_root_key);
+        });
+        self.update_value(new_root_key, |new_root_value| {
+            new_root_value.root(new_rank, new_value);
+        });
+    }
+}
+
+/// ////////////////////////////////////////////////////////////////////////
+/// Public API
+
+impl<'tcx, S, K, V> UnificationTable<S>
+where
+    S: UnificationStore<Key = K, Value = V>,
+    K: UnifyKey<Value = V>,
+    V: UnifyValue,
+{
+    /// Unions two keys without the possibility of failure; only
+    /// applicable when unify values use `NoError` as their error
+    /// type.
+    pub fn union<K1, K2>(&mut self, a_id: K1, b_id: K2)
+    where
+        K1: Into<K>,
+        K2: Into<K>,
+        V: UnifyValue<Error = NoError>,
+    {
+        self.unify_var_var(a_id, b_id).unwrap();
+    }
+
+    /// Unions a key and a value without the possibility of failure;
+    /// only applicable when unify values use `NoError` as their error
+    /// type.
+    pub fn union_value<K1>(&mut self, id: K1, value: V)
+    where
+        K1: Into<K>,
+        V: UnifyValue<Error = NoError>,
+    {
+        self.unify_var_value(id, value).unwrap();
+    }
+
+    /// Given two keys, indicates whether they have been unioned together.
+    pub fn unioned<K1, K2>(&mut self, a_id: K1, b_id: K2) -> bool
+    where
+        K1: Into<K>,
+        K2: Into<K>,
+    {
+        self.find(a_id) == self.find(b_id)
+    }
+
+    /// Given a key, returns the (current) root key.
+    pub fn find<K1>(&mut self, id: K1) -> K
+    where
+        K1: Into<K>,
+    {
+        let id = id.into();
+        self.get_root_key(id)
+    }
+
+    /// Unions together two variables, merging their values. If
+    /// merging the values fails, the error is propagated and this
+    /// method has no effect.
+    pub fn unify_var_var<K1, K2>(&mut self, a_id: K1, b_id: K2) -> Result<(), V::Error>
+    where
+        K1: Into<K>,
+        K2: Into<K>,
+    {
+        let a_id = a_id.into();
+        let b_id = b_id.into();
+
+        let root_a = self.get_root_key(a_id);
+        let root_b = self.get_root_key(b_id);
+
+        if root_a == root_b {
+            return Ok(());
+        }
+
+        let combined = V::unify_values(&self.value(root_a).value, &self.value(root_b).value)?;
+
+        Ok(self.unify_roots(root_a, root_b, combined))
+    }
+
+    /// Sets the value of the key `a_id` to `b`, attempting to merge
+    /// with the previous value.
+    pub fn unify_var_value<K1>(&mut self, a_id: K1, b: V) -> Result<(), V::Error>
+    where
+        K1: Into<K>,
+    {
+        let a_id = a_id.into();
+        let root_a = self.get_root_key(a_id);
+        let value = V::unify_values(&self.value(root_a).value, &b)?;
+        self.update_value(root_a, |node| node.value = value);
+        Ok(())
+    }
+
+    /// Returns the current value for the given key. If the key has
+    /// been union'd, this will give the value from the current root.
+    pub fn probe_value<K1>(&mut self, id: K1) -> V
+    where
+        K1: Into<K>,
+    {
+        let id = id.into();
+        let id = self.get_root_key(id);
+        self.value(id).value.clone()
+    }
+}
+
+
+///////////////////////////////////////////////////////////////////////////
+
+impl UnifyValue for () {
+    type Error = NoError;
+
+    fn unify_values(_: &(), _: &()) -> Result<(), NoError> {
+        Ok(())
+    }
+}
+
+impl<V: UnifyValue> UnifyValue for Option<V> {
+    type Error = V::Error;
+
+    fn unify_values(a: &Option<V>, b: &Option<V>) -> Result<Self, V::Error> {
+        match (a, b) {
+            (&None, &None) => Ok(None),
+            (&Some(ref v), &None) |
+            (&None, &Some(ref v)) => Ok(Some(v.clone())),
+            (&Some(ref a), &Some(ref b)) => {
+                match V::unify_values(a, b) {
+                    Ok(v) => Ok(Some(v)),
+                    Err(err) => Err(err),
+                }
+            }
+        }
+    }
+}

src/ena/unify/tests.rs

@@ -0,0 +1,476 @@
+// Copyright 2015 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.
+
+// Naming the benchmarks using uppercase letters helps them sort
+// better.
+#![allow(non_snake_case)]
+
+#[cfg(feature = "bench")]
+extern crate test;
+#[cfg(feature = "bench")]
+use self::test::Bencher;
+use std::cmp;
+use ena::unify::{NoError, InPlace, InPlaceUnificationTable, UnifyKey, EqUnifyValue, UnifyValue};
+use ena::unify::{UnificationStore, UnificationTable};
+#[cfg(feature = "persistent")]
+use ena::unify::Persistent;
+
+#[derive(Copy, Clone, Debug, Hash, PartialEq, Eq)]
+struct UnitKey(u32);
+
+impl UnifyKey for UnitKey {
+    type Value = ();
+    fn index(&self) -> u32 {
+        self.0
+    }
+    fn from_index(u: u32) -> UnitKey {
+        UnitKey(u)
+    }
+    fn tag() -> &'static str {
+        "UnitKey"
+    }
+}
+
+macro_rules! all_modes {
+    ($name:ident for $t:ty => $body:tt) => {
+        fn test_body<$name: UnificationStore<Key = $t, Value = <$t as UnifyKey>::Value>>() {
+            $body
+        }
+
+        test_body::<InPlace<$t>>();
+
+        #[cfg(feature = "persistent")]
+        test_body::<Persistent<$t>>();
+    }
+}
+
+#[test]
+fn basic() {
+    all_modes! {
+        S for UnitKey => {
+            let mut ut: UnificationTable<S> = UnificationTable::new();
+            let k1 = ut.new_key(());
+            let k2 = ut.new_key(());
+            assert_eq!(ut.unioned(k1, k2), false);
+            ut.union(k1, k2);
+            assert_eq!(ut.unioned(k1, k2), true);
+        }
+    }
+}
+
+#[test]
+fn big_array() {
+    all_modes! {
+        S for UnitKey => {
+            let mut ut: UnificationTable<S> = UnificationTable::new();
+            let mut keys = Vec::new();
+            const MAX: usize = 1 << 15;
+
+            for _ in 0..MAX {
+                keys.push(ut.new_key(()));
+            }
+
+            for i in 1..MAX {
+                let l = keys[i - 1];
+                let r = keys[i];
+                ut.union(l, r);
+            }
+
+            for i in 0..MAX {
+                assert!(ut.unioned(keys[0], keys[i]));
+            }
+        }
+    }
+}
+
+#[cfg(feature = "bench")]
+fn big_array_bench_generic<S: UnificationStore<Key=UnitKey, Value=()>>(b: &mut Bencher) {
+    let mut ut: UnificationTable<S> = UnificationTable::new();
+    let mut keys = Vec::new();
+    const MAX: usize = 1 << 15;
+
+    for _ in 0..MAX {
+        keys.push(ut.new_key(()));
+    }
+
+    b.iter(|| {
+        for i in 1..MAX {
+            let l = keys[i - 1];
+            let r = keys[i];
+            ut.union(l, r);
+        }
+
+        for i in 0..MAX {
+            assert!(ut.unioned(keys[0], keys[i]));
+        }
+    })
+}
+
+#[cfg(feature = "bench")]
+#[bench]
+fn big_array_bench_InPlace(b: &mut Bencher) {
+    big_array_bench_generic::<InPlace<UnitKey>>(b);
+}
+
+#[cfg(all(feature = "bench", feature = "persistent"))]
+#[bench]
+fn big_array_bench_Persistent(b: &mut Bencher) {
+    big_array_bench_generic::<Persistent<UnitKey>>(b);
+}
+
+#[cfg(feature = "bench")]
+fn big_array_bench_in_snapshot_generic<S: UnificationStore<Key=UnitKey, Value=()>>(b: &mut Bencher) {
+    let mut ut: UnificationTable<S> = UnificationTable::new();
+    let mut keys = Vec::new();
+    const MAX: usize = 1 << 15;
+
+    for _ in 0..MAX {
+        keys.push(ut.new_key(()));
+    }
+
+    b.iter(|| {
+        let snapshot = ut.snapshot();
+
+        for i in 1..MAX {
+            let l = keys[i - 1];
+            let r = keys[i];
+            ut.union(l, r);
+        }
+
+        for i in 0..MAX {
+            assert!(ut.unioned(keys[0], keys[i]));
+        }
+
+        ut.rollback_to(snapshot);
+    })
+}
+
+#[cfg(feature = "bench")]
+#[bench]
+fn big_array_bench_in_snapshot_InPlace(b: &mut Bencher) {
+    big_array_bench_in_snapshot_generic::<InPlace<UnitKey>>(b);
+}
+
+#[cfg(all(feature = "bench", feature = "persistent"))]
+#[bench]
+fn big_array_bench_in_snapshot_Persistent(b: &mut Bencher) {
+    big_array_bench_in_snapshot_generic::<Persistent<UnitKey>>(b);
+}
+
+#[cfg(feature = "bench")]
+fn big_array_bench_clone_generic<S: UnificationStore<Key=UnitKey, Value=()>>(b: &mut Bencher) {
+    let mut ut: UnificationTable<S> = UnificationTable::new();
+    let mut keys = Vec::new();
+    const MAX: usize = 1 << 15;
+
+    for _ in 0..MAX {
+        keys.push(ut.new_key(()));
+    }
+
+    b.iter(|| {
+        let saved_table = ut.clone();
+
+        for i in 1..MAX {
+            let l = keys[i - 1];
+            let r = keys[i];
+            ut.union(l, r);
+        }
+
+        for i in 0..MAX {
+            assert!(ut.unioned(keys[0], keys[i]));
+        }
+
+        ut = saved_table;
+    })
+}
+
+#[cfg(feature = "bench")]
+#[bench]
+fn big_array_bench_clone_InPlace(b: &mut Bencher) {
+    big_array_bench_clone_generic::<InPlace<UnitKey>>(b);
+}
+
+#[cfg(all(feature = "bench", feature = "persistent"))]
+#[bench]
+fn big_array_bench_clone_Persistent(b: &mut Bencher) {
+    big_array_bench_clone_generic::<Persistent<UnitKey>>(b);
+}
+
+#[test]
+fn even_odd() {
+    all_modes! {
+        S for UnitKey => {
+            let mut ut: UnificationTable<S> = UnificationTable::new();
+            let mut keys = Vec::new();
+            const MAX: usize = 1 << 10;
+
+            for i in 0..MAX {
+                let key = ut.new_key(());
+                keys.push(key);
+
+                if i >= 2 {
+                    ut.union(key, keys[i - 2]);
+                }
+            }
+
+            for i in 1..MAX {
+                assert!(!ut.unioned(keys[i - 1], keys[i]));
+            }
+
+            for i in 2..MAX {
+                assert!(ut.unioned(keys[i - 2], keys[i]));
+            }
+        }
+    }
+}
+
+#[derive(Copy, Clone, Debug, Hash, PartialEq, Eq)]
+struct IntKey(u32);
+
+impl UnifyKey for IntKey {
+    type Value = Option<i32>;
+    fn index(&self) -> u32 {
+        self.0
+    }
+    fn from_index(u: u32) -> IntKey {
+        IntKey(u)
+    }
+    fn tag() -> &'static str {
+        "IntKey"
+    }
+}
+
+impl EqUnifyValue for i32 {}
+
+#[test]
+fn unify_same_int_twice() {
+    all_modes! {
+        S for IntKey => {
+            let mut ut: UnificationTable<S> = UnificationTable::new();
+            let k1 = ut.new_key(None);
+            let k2 = ut.new_key(None);
+            assert!(ut.unify_var_value(k1, Some(22)).is_ok());
+            assert!(ut.unify_var_value(k2, Some(22)).is_ok());
+            assert!(ut.unify_var_var(k1, k2).is_ok());
+            assert_eq!(ut.probe_value(k1), Some(22));
+        }
+    }
+}
+
+#[test]
+fn unify_vars_then_int_indirect() {
+    all_modes! {
+        S for IntKey => {
+            let mut ut: UnificationTable<S> = UnificationTable::new();
+            let k1 = ut.new_key(None);
+            let k2 = ut.new_key(None);
+            assert!(ut.unify_var_var(k1, k2).is_ok());
+            assert!(ut.unify_var_value(k1, Some(22)).is_ok());
+            assert_eq!(ut.probe_value(k2), Some(22));
+        }
+    }
+}
+
+#[test]
+fn unify_vars_different_ints_1() {
+    all_modes! {
+        S for IntKey => {
+            let mut ut: UnificationTable<S> = UnificationTable::new();
+            let k1 = ut.new_key(None);
+            let k2 = ut.new_key(None);
+            assert!(ut.unify_var_var(k1, k2).is_ok());
+            assert!(ut.unify_var_value(k1, Some(22)).is_ok());
+            assert!(ut.unify_var_value(k2, Some(23)).is_err());
+        }
+    }
+}
+
+#[test]
+fn unify_vars_different_ints_2() {
+    all_modes! {
+        S for IntKey => {
+            let mut ut: UnificationTable<S> = UnificationTable::new();
+            let k1 = ut.new_key(None);
+            let k2 = ut.new_key(None);
+            assert!(ut.unify_var_var(k2, k1).is_ok());
+            assert!(ut.unify_var_value(k1, Some(22)).is_ok());
+            assert!(ut.unify_var_value(k2, Some(23)).is_err());
+        }
+    }
+}
+
+#[test]
+fn unify_distinct_ints_then_vars() {
+    all_modes! {
+        S for IntKey => {
+            let mut ut: UnificationTable<S> = UnificationTable::new();
+            let k1 = ut.new_key(None);
+            let k2 = ut.new_key(None);
+            assert!(ut.unify_var_value(k1, Some(22)).is_ok());
+            assert!(ut.unify_var_value(k2, Some(23)).is_ok());
+            assert!(ut.unify_var_var(k2, k1).is_err());
+        }
+    }
+}
+
+#[test]
+fn unify_root_value_1() {
+    all_modes! {
+        S for IntKey => {
+            let mut ut: UnificationTable<S> = UnificationTable::new();
+            let k1 = ut.new_key(None);
+            let k2 = ut.new_key(None);
+            let k3 = ut.new_key(None);
+            assert!(ut.unify_var_value(k1, Some(22)).is_ok());
+            assert!(ut.unify_var_var(k1, k2).is_ok());
+            assert!(ut.unify_var_value(k3, Some(23)).is_ok());
+            assert!(ut.unify_var_var(k1, k3).is_err());
+        }
+    }
+}
+
+#[test]
+fn unify_root_value_2() {
+    all_modes! {
+        S for IntKey => {
+            let mut ut: UnificationTable<S> = UnificationTable::new();
+            let k1 = ut.new_key(None);
+            let k2 = ut.new_key(None);
+            let k3 = ut.new_key(None);
+            assert!(ut.unify_var_value(k1, Some(22)).is_ok());
+            assert!(ut.unify_var_var(k2, k1).is_ok());
+            assert!(ut.unify_var_value(k3, Some(23)).is_ok());
+            assert!(ut.unify_var_var(k1, k3).is_err());
+        }
+    }
+}
+
+#[derive(Copy, Clone, Debug, Hash, PartialEq, Eq)]
+struct OrderedKey(u32);
+
+#[derive(Copy, Clone, Debug, Hash, PartialEq, Eq, PartialOrd, Ord)]
+struct OrderedRank(u32);
+
+impl UnifyKey for OrderedKey {
+    type Value = OrderedRank;
+    fn index(&self) -> u32 {
+        self.0
+    }
+    fn from_index(u: u32) -> OrderedKey {
+        OrderedKey(u)
+    }
+    fn tag() -> &'static str {
+        "OrderedKey"
+    }
+    fn order_roots(
+        a: OrderedKey,
+        a_rank: &OrderedRank,
+        b: OrderedKey,
+        b_rank: &OrderedRank,
+    ) -> Option<(OrderedKey, OrderedKey)> {
+        println!("{:?} vs {:?}", a_rank, b_rank);
+        if a_rank > b_rank {
+            Some((a, b))
+        } else if b_rank > a_rank {
+            Some((b, a))
+        } else {
+            None
+        }
+    }
+}
+
+impl UnifyValue for OrderedRank {
+    type Error = NoError;
+
+    fn unify_values(value1: &Self, value2: &Self) -> Result<Self, NoError> {
+        Ok(OrderedRank(cmp::max(value1.0, value2.0)))
+    }
+}
+
+#[test]
+fn ordered_key() {
+    all_modes! {
+        S for OrderedKey => {
+            let mut ut: UnificationTable<S> = UnificationTable::new();
+
+            let k0_1 = ut.new_key(OrderedRank(0));
+            let k0_2 = ut.new_key(OrderedRank(0));
+            let k0_3 = ut.new_key(OrderedRank(0));
+            let k0_4 = ut.new_key(OrderedRank(0));
+
+            ut.union(k0_1, k0_2); // rank of one of those will now be 1
+            ut.union(k0_3, k0_4); // rank of new root also 1
+            ut.union(k0_1, k0_3); // rank of new root now 2
+
+            let k0_5 = ut.new_key(OrderedRank(0));
+            let k0_6 = ut.new_key(OrderedRank(0));
+            ut.union(k0_5, k0_6); // rank of new root now 1
+
+            ut.union(k0_1, k0_5); // new root rank 2, should not be k0_5 or k0_6
+            assert!(vec![k0_1, k0_2, k0_3, k0_4].contains(&ut.find(k0_1)));
+        }
+    }
+}
+
+#[test]
+fn ordered_key_k1() {
+    all_modes! {
+        S for UnitKey => {
+            let mut ut: InPlaceUnificationTable<OrderedKey> = UnificationTable::new();
+
+            let k0_1 = ut.new_key(OrderedRank(0));
+            let k0_2 = ut.new_key(OrderedRank(0));
+            let k0_3 = ut.new_key(OrderedRank(0));
+            let k0_4 = ut.new_key(OrderedRank(0));
+
+            ut.union(k0_1, k0_2); // rank of one of those will now be 1
+            ut.union(k0_3, k0_4); // rank of new root also 1
+            ut.union(k0_1, k0_3); // rank of new root now 2
+
+            let k1_5 = ut.new_key(OrderedRank(1));
+            let k1_6 = ut.new_key(OrderedRank(1));
+            ut.union(k1_5, k1_6); // rank of new root now 1
+
+            ut.union(k0_1, k1_5); // even though k1 has lower rank, it wins
+            assert!(
+                vec![k1_5, k1_6].contains(&ut.find(k0_1)),
+                "unexpected choice for root: {:?}",
+                ut.find(k0_1)
+            );
+        }
+    }
+}
+
+/// Test that we *can* clone.
+#[test]
+fn clone_table() {
+    all_modes! {
+        S for IntKey => {
+            let mut ut: UnificationTable<S> = UnificationTable::new();
+            let k1 = ut.new_key(None);
+            let k2 = ut.new_key(None);
+            let k3 = ut.new_key(None);
+            assert!(ut.unify_var_value(k1, Some(22)).is_ok());
+            assert!(ut.unify_var_value(k2, Some(22)).is_ok());
+            assert!(ut.unify_var_var(k1, k2).is_ok());
+            assert_eq!(ut.probe_value(k3), None);
+
+            let mut ut1 = ut.clone();
+            assert_eq!(ut1.probe_value(k1), Some(22));
+            assert_eq!(ut1.probe_value(k3), None);
+
+            assert!(ut.unify_var_value(k3, Some(44)).is_ok());
+
+            assert_eq!(ut1.probe_value(k1), Some(22));
+            assert_eq!(ut1.probe_value(k3), None);
+            assert_eq!(ut.probe_value(k3), Some(44));
+        }
+    }
+}

src/lib.rs

@@ -17,11 +17,12 @@ pub mod solve;
 pub mod unify;
 pub mod infer;
 pub mod pretty_print_types;
+pub mod ena;
 
 extern crate im_rc;
 extern crate fraction;
 extern crate num;
 extern crate fxhash;
-extern crate ena;
 
 #[macro_use] extern crate combine;
+#[macro_use] extern crate log;