diff --git a/Cargo.lock b/Cargo.lock
index defeb6998..af806aa53 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -172,6 +172,16 @@ version = "1.0.7"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "3f9eec918d3f24069decb9af1554cad7c880e2da24a9afd88aca000531ab82c1"
 
+[[package]]
+name = "futf"
+version = "0.1.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7c9c1ce3fa9336301af935ab852c437817d14cd33690446569392e65170aac3b"
+dependencies = [
+ "mac",
+ "new_debug_unreachable",
+]
+
 [[package]]
 name = "getrandom"
 version = "0.1.14"
@@ -199,6 +209,7 @@ dependencies = [
  "anyhow",
  "ropey",
  "smallvec 1.4.0",
+ "tendril",
 ]
 
 [[package]]
@@ -232,6 +243,12 @@ dependencies = [
  "cfg-if",
 ]
 
+[[package]]
+name = "mac"
+version = "0.1.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c41e0c4fef86961ac6d6f8a82609f55f31b05e4fce149ac5710e439df7619ba4"
+
 [[package]]
 name = "maybe-uninit"
 version = "2.0.0"
@@ -250,6 +267,12 @@ version = "0.1.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "a64a92489e2744ce060c349162be1c5f33c6969234104dbd99ddb5feb08b8c15"
 
+[[package]]
+name = "new_debug_unreachable"
+version = "1.0.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e4a24736216ec316047a1fc4252e27dabb04218aa4a3f37c6e7ddbf1f9782b54"
+
 [[package]]
 name = "nom"
 version = "5.1.1"
@@ -458,10 +481,9 @@ checksum = "7fe5bd57d1d7414c6b5ed48563a2c855d995ff777729dcd91c369ec7fea395ae"
 [[package]]
 name = "ropey"
 version = "1.1.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ba326a8508a4add47e7b260333aa2d896213a5f3572fde11ed6e9130241b7f71"
+source = "git+https://github.com/cessen/ropey#083c34949274ef9800267e6bc64b76a45e401807"
 dependencies = [
- "smallvec 0.6.13",
+ "smallvec 1.4.0",
 ]
 
 [[package]]
@@ -554,6 +576,16 @@ dependencies = [
  "unicode-xid 0.2.0",
 ]
 
+[[package]]
+name = "tendril"
+version = "0.4.1"
+source = "git+https://github.com/servo/tendril#08f7f292ab82c00e9cf491b5918a76e53af92c8c"
+dependencies = [
+ "futf",
+ "mac",
+ "utf-8",
+]
+
 [[package]]
 name = "terminfo"
 version = "0.7.3"
@@ -639,6 +671,12 @@ version = "0.2.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "826e7639553986605ec5979c7dd957c7895e93eabed50ab2ffa7f6128a75097c"
 
+[[package]]
+name = "utf-8"
+version = "0.7.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "05e42f7c18b8f902290b009cde6d651262f956c98bc51bca4cd1d511c9cd85c7"
+
 [[package]]
 name = "utf8parse"
 version = "0.1.1"
diff --git a/helix-core/Cargo.toml b/helix-core/Cargo.toml
index f1f6264f4..fc6a1b530 100644
--- a/helix-core/Cargo.toml
+++ b/helix-core/Cargo.toml
@@ -7,7 +7,9 @@ edition = "2018"
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
 [dependencies]
-ropey = "1.1.0"
+# ropey = "1.1.0"
+ropey = { git = "https://github.com/cessen/ropey" }
 anyhow = "1.0.31"
 smallvec = "1.4.0"
+tendril = { git = "https://github.com/servo/tendril" }
 # slab = "0.4.2"
diff --git a/helix-core/src/lib.rs b/helix-core/src/lib.rs
index ceed961fc..7b5844012 100644
--- a/helix-core/src/lib.rs
+++ b/helix-core/src/lib.rs
@@ -10,4 +10,4 @@
 
 pub use state::State;
 
-pub use transaction::{Change, Transaction};
+pub use transaction::{Change, ChangeSet, Transaction};
diff --git a/helix-core/src/selection.rs b/helix-core/src/selection.rs
index 24a8be467..03d5db0e3 100644
--- a/helix-core/src/selection.rs
+++ b/helix-core/src/selection.rs
@@ -83,7 +83,7 @@ pub fn extend(&self, from: usize, to: usize) -> Self {
 /// A selection consists of one or more selection ranges.
 pub struct Selection {
     // TODO: decide how many ranges to inline SmallVec<[Range; 1]>
-    ranges: Vec<Range>,
+    ranges: SmallVec<[Range; 1]>,
     primary_index: usize,
 }
 
@@ -100,7 +100,7 @@ pub fn as_single(self) -> Self {
             self
         } else {
             Self {
-                ranges: vec![self.ranges[self.primary_index]],
+                ranges: smallvec![self.ranges[self.primary_index]],
                 primary_index: 0,
             }
         }
@@ -112,18 +112,18 @@ pub fn as_single(self) -> Self {
     /// Constructs a selection holding a single range.
     pub fn single(anchor: usize, head: usize) -> Self {
         Self {
-            ranges: vec![Range { anchor, head }],
+            ranges: smallvec![Range { anchor, head }],
             primary_index: 0,
         }
     }
 
-    pub fn new(ranges: Vec<Range>, primary_index: usize) -> Self {
-        fn normalize(mut ranges: Vec<Range>, primary_index: usize) -> Selection {
+    pub fn new(ranges: SmallVec<[Range; 1]>, primary_index: usize) -> Self {
+        fn normalize(mut ranges: SmallVec<[Range; 1]>, primary_index: usize) -> Selection {
             let primary = ranges[primary_index];
             ranges.sort_unstable_by_key(|range| range.from());
             let mut primary_index = ranges.iter().position(|&range| range == primary).unwrap();
 
-            let mut result: Vec<Range> = Vec::new();
+            let mut result: SmallVec<[Range; 1]> = SmallVec::new();
 
             // TODO: we could do with one vec by removing elements as we mutate
 
@@ -174,7 +174,7 @@ mod test {
     #[test]
     fn test_create_normalizes_and_merges() {
         let sel = Selection::new(
-            vec![
+            smallvec![
                 Range::new(10, 12),
                 Range::new(6, 7),
                 Range::new(4, 5),
@@ -200,7 +200,7 @@ fn test_create_normalizes_and_merges() {
     #[test]
     fn test_create_merges_adjacent_points() {
         let sel = Selection::new(
-            vec![
+            smallvec![
                 Range::new(10, 12),
                 Range::new(12, 12),
                 Range::new(12, 12),
diff --git a/helix-core/src/transaction.rs b/helix-core/src/transaction.rs
index ecbe0c509..b9824bc2c 100644
--- a/helix-core/src/transaction.rs
+++ b/helix-core/src/transaction.rs
@@ -1,25 +1,293 @@
-pub struct Change {
-    from: usize,
-    to: usize,
-    insert: Option<String>,
+// pub struct Change {
+//     from: usize,
+//     to: usize,
+//     insert: Option<String>,
+// }
+
+// 40 bytes (8 + 24 + 8) -> strings are really big 24 as String, 16 as &str
+// pub struct Change {
+//     /// old extent
+//     old_extent: usize,
+//     /// inserted text, new extent equal to insert length
+//     insert: Option<String>,
+//     /// distance from the previous change
+//     distance: usize,
+// }
+
+use crate::{Buffer, Selection};
+
+use ropey::Rope;
+use tendril::StrTendril as Tendril;
+
+// TODO: divided into three different operations, I sort of like having just
+// Splice { extent, Option<text>, distance } better.
+// insert: Splice { extent: 0, text: Some("a"), distance: 2 }
+// delete: Splice { extent: 2, text: None, distance: 2 }
+// replace: Splice { extent: 2, text: Some("abc"), distance: 2 }
+// unchanged?: Splice { extent: 0, text: None, distance: 2 }
+// harder to compose though.
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub enum Change {
+    /// Move cursor by n characters.
+    Retain(usize),
+    /// Delete n characters.
+    Delete(usize),
+    /// Insert text at position.
+    Insert(Tendril),
 }
 
 impl Change {
-    pub fn new(from: usize, to: usize, insert: Option<String>) {
+    pub fn new(from: usize, to: usize, insert: Option<Tendril>) {
         // old_extent, new_extent, insert
     }
 }
 
-pub struct Transaction {}
-
 // ChangeSpec = Change | ChangeSet | Vec<Change>
 // ChangeDesc as a ChangeSet without text: can't be applied, cheaper to store.
 // ChangeSet = ChangeDesc with Text
+#[derive(Debug)]
 pub struct ChangeSet {
     // basically Vec<ChangeDesc> where ChangeDesc = (current len, replacement len?)
     // (0, n>0) for insertion, (n>0, 0) for deletion, (>0, >0) for replacement
-    sections: Vec<(usize, isize)>,
+    // sections: Vec<(usize, isize)>,
+    changes: Vec<Change>,
+    /// The required document length. Will refuse to apply changes unless it matches.
+    len: usize,
 }
-//
+
+impl ChangeSet {
+    pub fn new(buf: &Buffer) -> Self {
+        let len = buf.contents.len_chars();
+        Self {
+            changes: vec![Change::Retain(len)],
+            len,
+        }
+    }
+
+    // TODO: from iter
+
+    /// Combine two changesets together.
+    /// In other words,  If `this` goes `docA` → `docB` and `other` represents `docB` → `docC`, the
+    /// returned value will represent the change `docA` → `docC`.
+    pub fn compose(self, other: ChangeSet) -> Result<Self, ()> {
+        if self.len != other.len {
+            // length mismatch
+            return Err(());
+        }
+
+        let len = self.changes.len();
+
+        let mut changes_a = self.changes.into_iter();
+        let mut changes_b = other.changes.into_iter();
+
+        let mut head_a = changes_a.next();
+        let mut head_b = changes_b.next();
+
+        let mut changes: Vec<Change> = Vec::with_capacity(len); // TODO: max(a, b), shrink_to_fit() afterwards
+
+        loop {
+            use std::cmp::Ordering;
+            use Change::*;
+            match (head_a, head_b) {
+                // we are done
+                (None, None) => {
+                    break;
+                }
+                // deletion in A
+                (Some(change @ Delete(..)), b) => {
+                    changes.push(change);
+                    head_a = changes_a.next();
+                    head_b = b;
+                }
+                // insertion in B
+                (a, Some(change @ Insert(..))) => {
+                    changes.push(change);
+                    head_a = a;
+                    head_b = changes_b.next();
+                }
+                (None, _) => return Err(()),
+                (_, None) => return Err(()),
+                (Some(Retain(i)), Some(Retain(j))) => match i.cmp(&j) {
+                    Ordering::Less => {
+                        changes.push(Retain(i));
+                        head_a = changes_a.next();
+                        head_b = Some(Retain(j - i));
+                    }
+                    Ordering::Equal => {
+                        changes.push(Retain(i));
+                        head_a = changes_a.next();
+                        head_b = changes_b.next();
+                    }
+                    Ordering::Greater => {
+                        changes.push(Retain(j));
+                        head_a = Some(Retain(i - j));
+                        head_b = changes_b.next();
+                    }
+                },
+                (Some(Insert(mut s)), Some(Delete(j))) => {
+                    let len = s.chars().count();
+                    match len.cmp(&j) {
+                        Ordering::Less => {
+                            head_a = changes_a.next();
+                            head_b = Some(Delete(j - len));
+                        }
+                        Ordering::Equal => {
+                            head_a = changes_a.next();
+                            head_b = changes_b.next();
+                        }
+                        Ordering::Greater => {
+                            // figure out the byte index of the truncated string end
+                            let (pos, _) = s.char_indices().nth(len - j).unwrap();
+                            // calculate the difference
+                            let to_drop = s.len() - pos;
+                            s.pop_back(to_drop as u32);
+                            head_a = Some(Insert(s));
+                            head_b = changes_b.next();
+                        }
+                    }
+                }
+                (Some(Insert(mut s)), Some(Retain(j))) => {
+                    let len = s.chars().count();
+                    match len.cmp(&j) {
+                        Ordering::Less => {
+                            changes.push(Insert(s));
+                            head_a = changes_a.next();
+                            head_b = Some(Retain(j - len));
+                        }
+                        Ordering::Equal => {
+                            changes.push(Insert(s));
+                            head_a = changes_a.next();
+                            head_b = changes_b.next();
+                        }
+                        Ordering::Greater => {
+                            // figure out the byte index of the truncated string end
+                            let (pos, _) = s.char_indices().nth(j).unwrap();
+                            // calculate the difference
+                            let to_drop = s.len() - pos;
+                            s.pop_back(to_drop as u32);
+                            head_a = Some(Insert(s));
+                            head_b = changes_b.next();
+                        }
+                    }
+                }
+                (Some(Retain(i)), Some(Delete(j))) => match i.cmp(&j) {
+                    Ordering::Less => {
+                        changes.push(Delete(i));
+                        head_a = changes_a.next();
+                        head_b = Some(Delete(j - i));
+                    }
+                    Ordering::Equal => {
+                        changes.push(Delete(j));
+                        head_a = changes_a.next();
+                        head_b = changes_b.next();
+                    }
+                    Ordering::Greater => {
+                        changes.push(Delete(j));
+                        head_a = Some(Retain(i - j));
+                        head_b = changes_b.next();
+                    }
+                },
+            };
+        }
+
+        Ok(Self {
+            len: self.len,
+            changes,
+        })
+    }
+
+    /// Given another change set starting in the same document, maps this
+    /// change set over the other, producing a new change set that can be
+    /// applied to the document produced by applying `other`. When
+    /// `before` is `true`, order changes as if `this` comes before
+    /// `other`, otherwise (the default) treat `other` as coming first.
+    ///
+    /// Given two changes `A` and `B`, `A.compose(B.map(A))` and
+    /// `B.compose(A.map(B, true))` will produce the same document. This
+    /// provides a basic form of [operational
+    /// transformation](https://en.wikipedia.org/wiki/Operational_transformation),
+    /// and can be used for collaborative editing.
+    pub fn map(self, other: Self) -> Self {
+        unimplemented!()
+    }
+
+    /// Returns a new changeset that reverts this one. Useful for `undo` implementation.
+    pub fn invert(self) -> Self {
+        unimplemented!()
+    }
+
+    pub fn apply(&self, text: &mut Rope) {
+        // TODO: validate text.chars() == self.len
+
+        let mut pos = 0;
+
+        for change in self.changes.iter() {
+            use Change::*;
+            match change {
+                Retain(n) => {
+                    pos += n;
+                }
+                Delete(n) => {
+                    text.remove(pos..pos + *n);
+                    // pos += n;
+                }
+                Insert(s) => {
+                    text.insert(pos, s);
+                    pos += s.len();
+                }
+            }
+        }
+    }
+
+    // iter over changes
+}
+
 // trait Transaction
 // trait StrictTransaction
+
+/// Transaction represents a single undoable unit of changes. Several changes can be grouped into
+/// a single transaction.
+pub struct Transaction {
+    /// Changes made to the buffer.
+    changes: ChangeSet,
+    /// When set, explicitly updates the selection.
+    selection: Option<Selection>,
+    // effects, annotations
+    // scroll_into_view
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    fn composition() {
+        use Change::*;
+
+        let a = ChangeSet {
+            changes: vec![
+                Retain(5),
+                Insert("!".into()),
+                Retain(1),
+                Delete(2),
+                Insert("ab".into()),
+            ],
+            len: 7,
+        };
+
+        let b = ChangeSet {
+            changes: vec![Delete(5), Insert("world".into()), Retain(4)],
+            len: 7,
+        };
+
+        let mut text = Rope::from("hello xz");
+
+        // should probably return cloned text
+        a.compose(b).unwrap().apply(&mut text);
+
+        unimplemented!("{:?}", text);
+    }
+
+    #[test]
+    fn map() {}
+}