crates/oxy-core/src/graph.rs at 5f661562b3a4 · andymccall/oxygen

# 5f661562b3a4 oxygen / crates / oxy-core / src / graph.rs
37.5 KiB · Rust · last changed The history of a range of lines: oxy log -L START,END:PATH 1 month ago source blame history raw
//! The per-file atom graph (spec §7.3), its materialization, and text conflict
//! detection (spec §9.1, §9.5).
//!
//! The graph is a pure function of the set of changes applied to it: inserts
//! add vertices and edges, deletes set a flag, and nothing is ever removed.

use crate::fasthash::{FastMap as HashMap, FastSet as HashSet};
use crate::hash::ChangeHash;
use crate::objects::AtomRange;
use std::collections::{BTreeMap, BTreeSet};

/// `(introducing change, index within that change for this node)`.
pub type AtomKey = (ChangeHash, u32);

#[derive(Clone, Copy, PartialEq, Eq, Hash, PartialOrd, Ord, Debug)]
pub enum Vertex {
    Start,
    Atom(AtomKey),
    End,
}

/// Transitive dependency test between changes; `depends_on(a, a)` is true.
pub trait DepOracle {
    fn depends_on(&self, a: &ChangeHash, b: &ChangeHash) -> bool;
}

/// No change depends on any other: every pair commutes.
pub struct NoDeps;
impl DepOracle for NoDeps {
    fn depends_on(&self, a: &ChangeHash, b: &ChangeHash) -> bool {
        a == b
    }
}

#[derive(Default, Clone, Debug)]
struct AtomInfo {
    /// `None` until the introducing change has been applied.
    content: Option<Vec<u8>>,
    deleted_by: BTreeSet<ChangeHash>,
}

#[derive(Clone, Debug)]
pub struct FileGraph {
    atoms: HashMap<AtomKey, AtomInfo>,
    /// Every vertex, with its direct successors.
    succ: HashMap<Vertex, Edges>,
}

/// A vertex's direct successors, sorted and without repeats. Nearly every
/// vertex has one, which is kept inline without an allocation.
#[derive(Clone, Debug, Default, PartialEq, Eq)]
enum Edges {
    #[default]
    None,
    One(Vertex),
    Many(Vec<Vertex>),
}

impl Edges {
    fn insert(&mut self, v: Vertex) {
        match self {
            Edges::None => *self = Edges::One(v),
            Edges::One(a) if *a == v => {}
            Edges::One(a) => {
                let a = *a;
                *self = Edges::Many(if a < v { vec![a, v] } else { vec![v, a] });
            }
            Edges::Many(list) => {
                if let Err(i) = list.binary_search(&v) {
                    list.insert(i, v);
                }
            }
        }
    }

    fn as_slice(&self) -> &[Vertex] {
        match self {
            Edges::None => &[],
            Edges::One(v) => std::slice::from_ref(v),
            Edges::Many(list) => list,
        }
    }

    fn iter(&self) -> std::slice::Iter<'_, Vertex> {
        self.as_slice().iter()
    }
}

impl Default for FileGraph {
    fn default() -> Self {
        Self::new()
    }
}

/// One piece of a materialized file.
#[derive(Clone, Debug, PartialEq, Eq)]
pub enum Segment {
    Lines(Vec<AtomKey>),
    /// Order conflict: alternatives that cannot be ordered against each other.
    Conflict(Vec<Vec<AtomKey>>),
    /// Delete/insert conflict: `inserted` was anchored to `deleted` by a change
    /// that commutes with the deletion.
    Zombie {
        deleted: Vec<AtomKey>,
        inserted: Vec<AtomKey>,
    },
}

#[derive(Clone, Debug, PartialEq, Eq, Default)]
pub struct Materialized {
    pub segments: Vec<Segment>,
}

impl Materialized {
    pub fn is_conflicted(&self) -> bool {
        self.segments
            .iter()
            .any(|s| !matches!(s, Segment::Lines(_)))
    }
    pub fn conflict_count(&self) -> usize {
        self.segments
            .iter()
            .filter(|s| !matches!(s, Segment::Lines(_)))
            .count()
    }
    /// Every live atom in materialized order, conflict sides concatenated.
    pub fn live_atoms(&self) -> Vec<AtomKey> {
        let mut out = Vec::new();
        for s in &self.segments {
            match s {
                Segment::Lines(a) => out.extend(a),
                Segment::Conflict(sides) => sides.iter().for_each(|x| out.extend(x)),
                Segment::Zombie { inserted, .. } => out.extend(inserted),
            }
        }
        out
    }
}

impl FileGraph {
    pub fn new() -> Self {
        let mut g = FileGraph {
            atoms: HashMap::default(),
            succ: HashMap::default(),
        };
        g.edge(Vertex::Start, Vertex::End);
        g
    }

    fn edge(&mut self, a: Vertex, b: Vertex) {
        self.succ.entry(a).or_default().insert(b);
        self.succ.entry(b).or_default();
    }

    /// Apply an `Insert` by `change` whose first new atom has index `first_index`.
    pub fn insert(
        &mut self,
        change: ChangeHash,
        first_index: u32,
        after: Vertex,
        before: Vertex,
        lines: &[Vec<u8>],
    ) {
        self.insert_owned(change, first_index, after, before, lines.to_vec());
    }

    /// [`FileGraph::insert`] taking the lines, so loading a state moves
    /// them in rather than copying. Each new atom's entry is made once, as
    /// the source of the next edge, and the chain's ends are ensured last.
    pub fn insert_owned(
        &mut self,
        change: ChangeHash,
        first_index: u32,
        after: Vertex,
        before: Vertex,
        lines: Vec<Vec<u8>>,
    ) {
        self.atoms.reserve(lines.len());
        self.succ.reserve(lines.len() + 1);
        let mut prev = after;
        for (i, line) in lines.into_iter().enumerate() {
            let key = (change, first_index + i as u32);
            self.atoms.entry(key).or_default().content = Some(line);
            let v = Vertex::Atom(key);
            self.succ.entry(prev).or_default().insert(v);
            prev = v;
        }
        self.succ.entry(prev).or_default().insert(before);
        self.succ.entry(before).or_default();
    }

    /// Apply a `Delete` of `range` by change `by`.
    pub fn delete(&mut self, by: ChangeHash, range: &AtomRange) {
        for i in range.from..range.to {
            self.atoms
                .entry((range.change, i))
                .or_default()
                .deleted_by
                .insert(by);
        }
    }

    pub fn is_live(&self, k: &AtomKey) -> bool {
        self.atoms
            .get(k)
            .is_some_and(|a| a.content.is_some() && a.deleted_by.is_empty())
    }

    pub fn content(&self, k: &AtomKey) -> Option<&[u8]> {
        self.atoms.get(k).and_then(|a| a.content.as_deref())
    }

    pub fn deleters(&self, k: &AtomKey) -> impl Iterator<Item = &ChangeHash> {
        self.atoms
            .get(k)
            .into_iter()
            .flat_map(|a| a.deleted_by.iter())
    }

    /// Number of atoms introduced by `change` in this graph (for index bookkeeping).
    pub fn atom_count(&self) -> usize {
        self.atoms.len()
    }

    /// The changes that inserted a live atom directly after `k`. A change
    /// that deletes `k` must depend on them (spec §8.1), or `is_ghost`
    /// reads the deletion as concurrent with those insertions.
    /// Every atom on some path from `first` to `last`, both included, live
    /// or deleted (spec §8): the atoms a range of lines is made of, and
    /// those that were deleted from between them. Empty when `last` does
    /// not follow `first`.
    pub fn between(&self, first: &AtomKey, last: &AtomKey) -> Vec<AtomKey> {
        let (from, to) = (Vertex::Atom(*first), Vertex::Atom(*last));
        let mut ahead: HashSet<Vertex> = HashSet::default();
        let mut stack = vec![from];
        while let Some(v) = stack.pop() {
            if ahead.insert(v) {
                stack.extend(
                    self.succ
                        .get(&v)
                        .map(|e| e.as_slice())
                        .unwrap_or(&[])
                        .iter()
                        .copied(),
                );
            }
        }
        if !ahead.contains(&to) {
            return vec![];
        }
        let mut pred: HashMap<Vertex, Vec<Vertex>> = HashMap::default();
        for (u, es) in &self.succ {
            if ahead.contains(u) {
                for w in es.iter() {
                    pred.entry(*w).or_default().push(*u);
                }
            }
        }
        let mut behind: HashSet<Vertex> = HashSet::default();
        let mut stack = vec![to];
        while let Some(v) = stack.pop() {
            if behind.insert(v) {
                stack.extend(pred.get(&v).into_iter().flatten().copied());
            }
        }
        let mut out: Vec<AtomKey> = ahead
            .intersection(&behind)
            .filter_map(|v| match v {
                Vertex::Atom(k) if self.atoms.contains_key(k) => Some(*k),
                _ => None,
            })
            .collect();
        out.sort();
        out
    }

    /// The changes that inserted or deleted any of `atoms`.
    pub fn touching(&self, atoms: &[AtomKey]) -> BTreeSet<ChangeHash> {
        let mut out = BTreeSet::new();
        for k in atoms {
            out.insert(k.0);
            if let Some(info) = self.atoms.get(k) {
                out.extend(info.deleted_by.iter().copied());
            }
        }
        out
    }

    pub fn live_successor_changes(&self, k: &AtomKey) -> Vec<ChangeHash> {
        self.succ
            .get(&Vertex::Atom(*k))
            .map(|ss| {
                ss.iter()
                    .filter_map(|s| match s {
                        Vertex::Atom(y) if self.is_live(y) => Some(y.0),
                        _ => None,
                    })
                    .collect()
            })
            .unwrap_or_default()
    }

    fn is_ghost(&self, k: &AtomKey, info: &AtomInfo, deps: &dyn DepOracle) -> bool {
        // A deleted atom is a ghost when a live direct successor was inserted
        // by a change that commutes with every deleter: neither depends on
        // the other (spec §9.1, delete/insert). If the inserter came after the
        // deletion, or the deleter saw the insertion, there is no conflict.
        self.succ.get(&Vertex::Atom(*k)).is_some_and(|ss| {
            ss.iter().any(|s| match s {
                Vertex::Atom(y) if self.is_live(y) => !info
                    .deleted_by
                    .iter()
                    .any(|d| deps.depends_on(d, &y.0) || deps.depends_on(&y.0, d)),
                _ => false,
            })
        })
    }

    /// Successors of `v` in the graph with non-kept vertices contracted away.
    fn reduced_succ(
        &self,
        v: Vertex,
        kept: &HashSet<Vertex>,
        memo: &mut HashMap<Vertex, BTreeSet<Vertex>>,
    ) -> BTreeSet<Vertex> {
        let mut out = BTreeSet::new();
        let mut stack: Vec<Vertex> = self
            .succ
            .get(&v)
            .map(|s| s.iter().copied().collect())
            .unwrap_or_default();
        let mut seen = HashSet::default();
        while let Some(s) = stack.pop() {
            if kept.contains(&s) {
                out.insert(s);
            } else if seen.insert(s) {
                if let Some(m) = memo.get(&s) {
                    out.extend(m.iter().copied());
                } else if let Some(next) = self.succ.get(&s) {
                    stack.extend(next.iter().copied());
                }
            }
        }
        out
    }

    /// Order the live atoms and locate conflicts.
    pub fn materialize(&self, deps: &dyn DepOracle) -> Materialized {
        // 1. Vertices that survive contraction: live atoms, ghosts, Start, End.
        let mut kept: HashSet<Vertex> = HashSet::default();
        let mut ghosts: HashSet<AtomKey> = HashSet::default();
        kept.insert(Vertex::Start);
        kept.insert(Vertex::End);
        for (k, info) in &self.atoms {
            if info.content.is_none() {
                continue;
            }
            if info.deleted_by.is_empty() {
                kept.insert(Vertex::Atom(*k));
            } else if self.is_ghost(k, info, deps) {
                kept.insert(Vertex::Atom(*k));
                ghosts.insert(*k);
            }
        }

        // 2. Reduced graph over kept vertices.
        let mut memo = HashMap::default();
        let mut rsucc: HashMap<Vertex, BTreeSet<Vertex>> = HashMap::default();
        let mut indeg: HashMap<Vertex, usize> = HashMap::default();
        for &u in &kept {
            indeg.entry(u).or_insert(0);
            let rs = self.reduced_succ(u, &kept, &mut memo);
            for &w in &rs {
                *indeg.entry(w).or_insert(0) += 1;
            }
            rsucc.insert(u, rs);
        }

        // 3. Kahn's algorithm with side labelling for conflict regions.
        let mut ready: BTreeSet<Vertex> = indeg
            .iter()
            .filter(|(_, &d)| d == 0)
            .map(|(v, _)| *v)
            .collect();
        let mut side: HashMap<Vertex, u32> = HashMap::default();
        let mut joins: HashSet<Vertex> = HashSet::default();
        let mut next_side = 0u32;
        let mut plain: Vec<Vertex> = Vec::new();
        let mut conflict: Option<BTreeMap<u32, Vec<Vertex>>> = None;
        let mut segments: Vec<Segment> = Vec::new();

        let flush_plain = |plain: &mut Vec<Vertex>, segments: &mut Vec<Segment>| {
            let atoms: Vec<AtomKey> = plain
                .iter()
                .filter_map(|v| {
                    if let Vertex::Atom(k) = v {
                        Some(*k)
                    } else {
                        None
                    }
                })
                .collect();
            plain.clear();
            if !atoms.is_empty() {
                segments.extend(split_zombies(atoms, &ghosts));
            }
        };
        let finish_conflict = |c: BTreeMap<u32, Vec<Vertex>>, segments: &mut Vec<Segment>| {
            let sides: Vec<Vec<AtomKey>> = c
                .into_values()
                .map(|vs| {
                    vs.into_iter()
                        .filter_map(|v| match v {
                            Vertex::Atom(k) if !ghosts.contains(&k) => Some(k),
                            _ => None,
                        })
                        .collect::<Vec<_>>()
                })
                .filter(|s| !s.is_empty())
                .collect();
            match sides.len() {
                0 => {}
                1 => segments.push(Segment::Lines(sides.into_iter().next().unwrap())),
                _ => segments.push(Segment::Conflict(sides)),
            }
        };

        while !ready.is_empty() {
            if ready.len() >= 2 {
                if conflict.is_none() {
                    flush_plain(&mut plain, &mut segments);
                    conflict = Some(BTreeMap::new());
                }
                for r in &ready {
                    if !side.contains_key(r) {
                        side.insert(*r, next_side);
                        next_side += 1;
                    }
                }
            }
            let v = ready.pop_first().unwrap();
            let my_side = side.get(&v).copied();
            match (&mut conflict, my_side) {
                (Some(c), Some(s)) => c.entry(s).or_default().push(v),
                (Some(_), None) => {
                    let c = conflict.take().unwrap();
                    finish_conflict(c, &mut segments);
                    plain.push(v);
                }
                (None, _) => plain.push(v),
            }
            for &w in &rsucc[&v] {
                let d = indeg.get_mut(&w).unwrap();
                *d -= 1;
                if conflict.is_some() && !joins.contains(&w) {
                    match (my_side, side.get(&w).copied()) {
                        (Some(s), None) => {
                            side.insert(w, s);
                        }
                        (Some(s), Some(t)) if s != t => {
                            side.remove(&w);
                            joins.insert(w);
                        }
                        (None, Some(_)) => {
                            side.remove(&w);
                            joins.insert(w);
                        }
                        _ => {}
                    }
                }
                if *d == 0 {
                    ready.insert(w);
                }
            }
        }
        if let Some(c) = conflict.take() {
            finish_conflict(c, &mut segments);
        }
        flush_plain(&mut plain, &mut segments);
        Materialized { segments }
    }

    /// Render a materialization to bytes with conflict markers (spec §9.5).
    /// `label` gives the short description of a change for marker headers.
    pub fn render(&self, m: &Materialized, label: &dyn Fn(&ChangeHash) -> String) -> Vec<u8> {
        let total = m.conflict_count();
        let mut n = 0;
        let mut out: Vec<u8> = Vec::new();
        let push_atoms = |out: &mut Vec<u8>, atoms: &[AtomKey]| {
            for a in atoms {
                if let Some(c) = self.content(a) {
                    out.extend_from_slice(c);
                }
            }
        };
        let newline_if_needed = |out: &mut Vec<u8>| {
            if !out.is_empty() && *out.last().unwrap() != b'\n' {
                out.push(b'\n');
            }
        };
        for seg in &m.segments {
            match seg {
                Segment::Lines(a) => push_atoms(&mut out, a),
                Segment::Conflict(sides) => {
                    n += 1;
                    let k = sides.len();
                    for (i, s) in sides.iter().enumerate() {
                        newline_if_needed(&mut out);
                        let who = s.first().map(|a| label(&a.0)).unwrap_or_default();
                        let marker = if i == 0 { "<<<<<<<" } else { "=======" };
                        out.extend_from_slice(
                            format!("{marker} oxy conflict {n} of {total}, side {} of {k}: change {who}\n", i + 1).as_bytes(),
                        );
                        push_atoms(&mut out, s);
                    }
                    newline_if_needed(&mut out);
                    out.extend_from_slice(
                        format!(">>>>>>> oxy conflict {n} of {total} end\n").as_bytes(),
                    );
                }
                Segment::Zombie { deleted, inserted } => {
                    n += 1;
                    newline_if_needed(&mut out);
                    let deleter = deleted
                        .first()
                        .and_then(|a| self.deleters(a).next().copied())
                        .map(|d| label(&d))
                        .unwrap_or_default();
                    let inserter = inserted.first().map(|a| label(&a.0)).unwrap_or_default();
                    out.extend_from_slice(
                        format!("<<<<<<< oxy conflict {n} of {total}, side 1 of 2: deleted by change {deleter}\n-------\n").as_bytes(),
                    );
                    push_atoms(&mut out, deleted);
                    newline_if_needed(&mut out);
                    out.extend_from_slice(
                        format!(
                            "======= oxy conflict {n} of {total}, side 2 of 2: change {inserter}\n"
                        )
                        .as_bytes(),
                    );
                    push_atoms(&mut out, inserted);
                    newline_if_needed(&mut out);
                    out.extend_from_slice(
                        format!(">>>>>>> oxy conflict {n} of {total} end\n").as_bytes(),
                    );
                }
            }
        }
        out
    }
}

fn vertex_value(v: &Vertex) -> crate::cbor::Value {
    use crate::cbor::Value;
    match v {
        Vertex::Start => Value::Uint(0),
        Vertex::End => Value::Uint(1),
        Vertex::Atom((c, i)) => {
            Value::Array(vec![Value::Bytes(c.0.to_vec()), Value::Uint(*i as u64)])
        }
    }
}

fn vertex_from(v: &crate::cbor::Value) -> Result<Vertex, crate::objects::ObjectError> {
    use crate::cbor::Value;
    use crate::objects::ObjectError;
    match v {
        Value::Uint(0) => Ok(Vertex::Start),
        Value::Uint(1) => Ok(Vertex::End),
        Value::Array(a) if a.len() == 2 => Ok(Vertex::Atom((
            crate::objects::oref(&a[0], "vertex")?,
            crate::objects::u32_(&a[1], "vertex")?,
        ))),
        _ => Err(ObjectError::Schema(
            "vertex: expected 0, 1, or [hash, index]".into(),
        )),
    }
}

impl FileGraph {
    /// Serialize for a checkpoint: `{0: atoms [[hash, index, content|null, [deleters]]], 1: edges [[from, to]]}`.
    pub fn to_value(&self) -> crate::cbor::Value {
        use crate::cbor::Value;
        let mut atoms: Vec<(&AtomKey, &AtomInfo)> = self.atoms.iter().collect();
        atoms.sort_by_key(|(k, _)| **k);
        let mut edges: Vec<(Vertex, Vertex)> = self
            .succ
            .iter()
            .flat_map(|(a, bs)| bs.iter().map(move |b| (*a, *b)))
            .collect();
        edges.sort();
        Value::map([
            (
                0,
                Value::Array(
                    atoms
                        .into_iter()
                        .map(|((c, i), info)| {
                            Value::Array(vec![
                                Value::Bytes(c.0.to_vec()),
                                Value::Uint(*i as u64),
                                info.content
                                    .as_ref()
                                    .map(|b| Value::Bytes(b.clone()))
                                    .unwrap_or(Value::Null),
                                Value::Array(
                                    info.deleted_by
                                        .iter()
                                        .map(|d| Value::Bytes(d.0.to_vec()))
                                        .collect(),
                                ),
                            ])
                        })
                        .collect(),
                ),
            ),
            (
                1,
                Value::Array(
                    edges
                        .iter()
                        .map(|(a, b)| Value::Array(vec![vertex_value(a), vertex_value(b)]))
                        .collect(),
                ),
            ),
        ])
    }

    pub fn from_value(v: &crate::cbor::Value) -> Result<FileGraph, crate::objects::ObjectError> {
        use crate::objects::{array, bytes, field, map, oref, u32_, ObjectError};
        let m = map(v, "file graph")?;
        let mut g = FileGraph {
            atoms: HashMap::default(),
            succ: HashMap::default(),
        };
        for a in array(field(m, 0, "atoms")?, "atoms")? {
            let e = array(a, "atom")?;
            if e.len() != 4 {
                return Err(ObjectError::Schema("atom: expected 4 fields".into()));
            }
            let key = (oref(&e[0], "atom change")?, u32_(&e[1], "atom index")?);
            let content = if e[2].is_null() {
                None
            } else {
                Some(bytes(&e[2], "atom content")?.to_vec())
            };
            let deleted_by = array(&e[3], "deleters")?
                .iter()
                .map(|d| oref(d, "deleter"))
                .collect::<Result<BTreeSet<_>, _>>()?;
            g.atoms.insert(
                key,
                AtomInfo {
                    content,
                    deleted_by,
                },
            );
        }
        for e in array(field(m, 1, "edges")?, "edges")? {
            let pair = array(e, "edge")?;
            if pair.len() != 2 {
                return Err(ObjectError::Schema("edge: expected [from, to]".into()));
            }
            g.edge(vertex_from(&pair[0])?, vertex_from(&pair[1])?);
        }
        Ok(g)
    }
}

/// Split a plain run into `Lines` and `Zombie` segments around ghost atoms.
fn split_zombies(atoms: Vec<AtomKey>, ghosts: &HashSet<AtomKey>) -> Vec<Segment> {
    let mut out = Vec::new();
    let mut lines: Vec<AtomKey> = Vec::new();
    let mut i = 0;
    while i < atoms.len() {
        if !ghosts.contains(&atoms[i]) {
            lines.push(atoms[i]);
            i += 1;
            continue;
        }
        if !lines.is_empty() {
            out.push(Segment::Lines(std::mem::take(&mut lines)));
        }
        let mut deleted = Vec::new();
        while i < atoms.len() && ghosts.contains(&atoms[i]) {
            deleted.push(atoms[i]);
            i += 1;
        }
        let mut inserted = Vec::new();
        if i < atoms.len() {
            let ch = atoms[i].0;
            while i < atoms.len() && atoms[i].0 == ch && !ghosts.contains(&atoms[i]) {
                inserted.push(atoms[i]);
                i += 1;
            }
        }
        out.push(Segment::Zombie { deleted, inserted });
    }
    if !lines.is_empty() {
        out.push(Segment::Lines(lines));
    }
    out
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::hash::ObjectRef;

    fn h(n: u8) -> ChangeHash {
        ObjectRef([n; 32])
    }
    fn lines(v: &[&str]) -> Vec<Vec<u8>> {
        v.iter().map(|s| s.as_bytes().to_vec()).collect()
    }
    fn text(g: &FileGraph, m: &Materialized) -> String {
        String::from_utf8(g.render(m, &|c| c.short(4))).unwrap()
    }

    struct Deps(HashMap<ChangeHash, Vec<ChangeHash>>);
    impl DepOracle for Deps {
        fn depends_on(&self, a: &ChangeHash, b: &ChangeHash) -> bool {
            if a == b {
                return true;
            }
            let mut stack = vec![*a];
            let mut seen = HashSet::default();
            while let Some(x) = stack.pop() {
                if let Some(ds) = self.0.get(&x) {
                    for d in ds {
                        if d == b {
                            return true;
                        }
                        if seen.insert(*d) {
                            stack.push(*d);
                        }
                    }
                }
            }
            false
        }
    }

    #[test]
    fn sequential_inserts_and_deletes() {
        let mut g = FileGraph::new();
        g.insert(
            h(1),
            0,
            Vertex::Start,
            Vertex::End,
            &lines(&["a\n", "b\n", "c\n"]),
        );
        let m = g.materialize(&NoDeps);
        assert!(!m.is_conflicted());
        assert_eq!(text(&g, &m), "a\nb\nc\n");

        // Change 2 depends on 1, inserts between b and c and deletes a.
        g.insert(
            h(2),
            0,
            Vertex::Atom((h(1), 1)),
            Vertex::Atom((h(1), 2)),
            &lines(&["x\n"]),
        );
        g.delete(
            h(2),
            &AtomRange {
                change: h(1),
                from: 0,
                to: 1,
            },
        );
        let deps = Deps([(h(2), vec![h(1)])].into_iter().collect());
        let m = g.materialize(&deps);
        assert!(!m.is_conflicted());
        assert_eq!(text(&g, &m), "b\nx\nc\n");
        assert_eq!(m.live_atoms(), vec![(h(1), 1), (h(2), 0), (h(1), 2)]);
    }

    #[test]
    fn concurrent_inserts_at_same_anchor_conflict() {
        let mut g = FileGraph::new();
        g.insert(h(1), 0, Vertex::Start, Vertex::End, &lines(&["a\n", "b\n"]));
        g.insert(
            h(2),
            0,
            Vertex::Atom((h(1), 0)),
            Vertex::Atom((h(1), 1)),
            &lines(&["from-2\n"]),
        );
        g.insert(
            h(3),
            0,
            Vertex::Atom((h(1), 0)),
            Vertex::Atom((h(1), 1)),
            &lines(&["from-3\n", "more-3\n"]),
        );
        let deps = Deps(
            [(h(2), vec![h(1)]), (h(3), vec![h(1)])]
                .into_iter()
                .collect(),
        );
        let m = g.materialize(&deps);
        assert!(m.is_conflicted());
        assert_eq!(m.conflict_count(), 1);
        assert_eq!(
            m.segments,
            vec![
                Segment::Lines(vec![(h(1), 0)]),
                Segment::Conflict(vec![vec![(h(2), 0)], vec![(h(3), 0), (h(3), 1)]]),
                Segment::Lines(vec![(h(1), 1)]),
            ]
        );
        let t = text(&g, &m);
        assert!(t.starts_with("a\n<<<<<<< oxy conflict 1 of 1, side 1 of 2: change 0202\nfrom-2\n======= oxy conflict 1 of 1, side 2 of 2: change 0303\nfrom-3\nmore-3\n>>>>>>> oxy conflict 1 of 1 end\nb\n"), "{t}");
    }

    #[test]
    fn concurrent_inserts_at_different_anchors_do_not_conflict() {
        let mut g = FileGraph::new();
        g.insert(
            h(1),
            0,
            Vertex::Start,
            Vertex::End,
            &lines(&["a\n", "b\n", "c\n"]),
        );
        g.insert(
            h(2),
            0,
            Vertex::Start,
            Vertex::Atom((h(1), 0)),
            &lines(&["top\n"]),
        );
        g.insert(
            h(3),
            0,
            Vertex::Atom((h(1), 2)),
            Vertex::End,
            &lines(&["bottom\n"]),
        );
        let deps = Deps(
            [(h(2), vec![h(1)]), (h(3), vec![h(1)])]
                .into_iter()
                .collect(),
        );
        let m = g.materialize(&deps);
        assert!(!m.is_conflicted());
        assert_eq!(text(&g, &m), "top\na\nb\nc\nbottom\n");
    }

    #[test]
    fn concurrent_deletes_of_the_same_line_agree() {
        let mut g = FileGraph::new();
        g.insert(h(1), 0, Vertex::Start, Vertex::End, &lines(&["a\n", "b\n"]));
        g.delete(
            h(2),
            &AtomRange {
                change: h(1),
                from: 0,
                to: 1,
            },
        );
        g.delete(
            h(3),
            &AtomRange {
                change: h(1),
                from: 0,
                to: 1,
            },
        );
        let deps = Deps(
            [(h(2), vec![h(1)]), (h(3), vec![h(1)])]
                .into_iter()
                .collect(),
        );
        let m = g.materialize(&deps);
        assert!(!m.is_conflicted());
        assert_eq!(text(&g, &m), "b\n");
    }

    #[test]
    fn delete_insert_conflict_is_a_zombie() {
        let mut g = FileGraph::new();
        g.insert(h(1), 0, Vertex::Start, Vertex::End, &lines(&["a\n", "b\n"]));
        // 2 deletes a; 3 (concurrent) inserts after a.
        g.delete(
            h(2),
            &AtomRange {
                change: h(1),
                from: 0,
                to: 1,
            },
        );
        g.insert(
            h(3),
            0,
            Vertex::Atom((h(1), 0)),
            Vertex::Atom((h(1), 1)),
            &lines(&["after-a\n"]),
        );
        let deps = Deps(
            [(h(2), vec![h(1)]), (h(3), vec![h(1)])]
                .into_iter()
                .collect(),
        );
        let m = g.materialize(&deps);
        assert!(m.is_conflicted());
        assert_eq!(
            m.segments,
            vec![
                Segment::Zombie {
                    deleted: vec![(h(1), 0)],
                    inserted: vec![(h(3), 0)]
                },
                Segment::Lines(vec![(h(1), 1)]),
            ]
        );
        let t = text(&g, &m);
        assert!(t.contains("-------\na\n======="), "{t}");
        assert!(t.contains("after-a\n>>>>>>>"), "{t}");

        // If the deleter saw the insert (depends on 3), there is no conflict.
        let deps = Deps(
            [(h(2), vec![h(1), h(3)]), (h(3), vec![h(1)])]
                .into_iter()
                .collect(),
        );
        let m = g.materialize(&deps);
        assert!(!m.is_conflicted());
        assert_eq!(text(&g, &m), "after-a\nb\n");
    }

    #[test]
    fn insert_after_a_line_deleted_by_a_dependency_is_not_a_conflict() {
        // 1 inserts a,b; 2 deletes b (depends on 1); 3 depends on 2 yet anchors
        // after the dead b (a rewritten change may do this): no conflict.
        let mut g = FileGraph::new();
        g.insert(h(1), 0, Vertex::Start, Vertex::End, &lines(&["a\n", "b\n"]));
        g.delete(
            h(2),
            &AtomRange {
                change: h(1),
                from: 1,
                to: 2,
            },
        );
        g.insert(
            h(3),
            0,
            Vertex::Atom((h(1), 1)),
            Vertex::End,
            &lines(&["c\n"]),
        );
        let deps = Deps(
            [(h(2), vec![h(1)]), (h(3), vec![h(2)])]
                .into_iter()
                .collect(),
        );
        let m = g.materialize(&deps);
        assert!(!m.is_conflicted(), "{:?}", m.segments);
        assert_eq!(text(&g, &m), "a\nc\n");
        // A change that deletes a line it introduced itself, with a dependent
        // anchoring after it, is likewise fine.
        let mut g = FileGraph::new();
        g.insert(h(1), 0, Vertex::Start, Vertex::End, &lines(&["x\n"]));
        g.delete(
            h(1),
            &AtomRange {
                change: h(1),
                from: 0,
                to: 1,
            },
        );
        g.insert(
            h(2),
            0,
            Vertex::Atom((h(1), 0)),
            Vertex::End,
            &lines(&["y\n"]),
        );
        let deps = Deps([(h(2), vec![h(1)])].into_iter().collect());
        let m = g.materialize(&deps);
        assert!(!m.is_conflicted());
        assert_eq!(text(&g, &m), "y\n");
    }

    #[test]
    fn resolution_collapses_a_conflict() {
        let mut g = FileGraph::new();
        g.insert(h(1), 0, Vertex::Start, Vertex::End, &lines(&["a\n"]));
        g.insert(
            h(2),
            0,
            Vertex::Start,
            Vertex::Atom((h(1), 0)),
            &lines(&["x\n"]),
        );
        g.insert(
            h(3),
            0,
            Vertex::Start,
            Vertex::Atom((h(1), 0)),
            &lines(&["y\n"]),
        );
        let mut deps: HashMap<_, _> = [(h(2), vec![h(1)]), (h(3), vec![h(1)])]
            .into_iter()
            .collect();
        assert!(g.materialize(&Deps(deps.clone())).is_conflicted());
        // Resolution keeps y, drops x.
        g.delete(
            h(4),
            &AtomRange {
                change: h(2),
                from: 0,
                to: 1,
            },
        );
        deps.insert(h(4), vec![h(2), h(3)]);
        let m = g.materialize(&Deps(deps));
        assert!(!m.is_conflicted());
        assert_eq!(text(&g, &m), "y\na\n");
    }

    #[test]
    fn empty_and_unterminated_files() {
        let g = FileGraph::new();
        let m = g.materialize(&NoDeps);
        assert_eq!(m.segments, vec![]);
        assert_eq!(text(&g, &m), "");
        let mut g = FileGraph::new();
        g.insert(h(1), 0, Vertex::Start, Vertex::End, &lines(&["no newline"]));
        assert_eq!(text(&g, &g.materialize(&NoDeps)), "no newline");
    }

    #[test]
    fn serialization_round_trip() {
        let mut g = FileGraph::new();
        g.insert(h(1), 0, Vertex::Start, Vertex::End, &lines(&["a\n", "b\n"]));
        g.insert(
            h(2),
            0,
            Vertex::Atom((h(1), 0)),
            Vertex::Atom((h(1), 1)),
            &lines(&["x\n"]),
        );
        g.delete(
            h(3),
            &AtomRange {
                change: h(1),
                from: 1,
                to: 2,
            },
        );
        let v = g.to_value();
        let bytes = crate::cbor::encode(&v);
        let back = FileGraph::from_value(&crate::cbor::decode(&bytes).unwrap()).unwrap();
        let deps = Deps(
            [(h(2), vec![h(1)]), (h(3), vec![h(1)])]
                .into_iter()
                .collect(),
        );
        assert_eq!(back.materialize(&deps), g.materialize(&deps));
        assert_eq!(text(&back, &back.materialize(&deps)), "a\nx\n");
        assert_eq!(
            crate::cbor::encode(&back.to_value()),
            bytes,
            "deterministic"
        );
    }

    /// Spec §7.3: the graph is a function of the set of changes, not their order.
    #[test]
    fn application_order_is_irrelevant() {
        use rand::seq::SliceRandom;
        type Op = (
            ChangeHash,
            Vec<(u32, Vertex, Vertex, Vec<Vec<u8>>)>,
            Vec<AtomRange>,
        );
        let ops: Vec<Op> = vec![
            (
                h(1),
                vec![(0, Vertex::Start, Vertex::End, lines(&["a\n", "b\n", "c\n"]))],
                vec![],
            ),
            (
                h(2),
                vec![(
                    0,
                    Vertex::Atom((h(1), 0)),
                    Vertex::Atom((h(1), 1)),
                    lines(&["p\n"]),
                )],
                vec![],
            ),
            (
                h(3),
                vec![(
                    0,
                    Vertex::Atom((h(1), 0)),
                    Vertex::Atom((h(1), 1)),
                    lines(&["q\n", "r\n"]),
                )],
                vec![AtomRange {
                    change: h(1),
                    from: 2,
                    to: 3,
                }],
            ),
            (
                h(4),
                vec![(
                    0,
                    Vertex::Atom((h(3), 1)),
                    Vertex::Atom((h(1), 1)),
                    lines(&["s\n"]),
                )],
                vec![AtomRange {
                    change: h(1),
                    from: 1,
                    to: 2,
                }],
            ),
            (
                h(5),
                vec![(0, Vertex::Atom((h(1), 2)), Vertex::End, lines(&["z\n"]))],
                vec![],
            ),
        ];
        let deps = Deps(
            [
                (h(2), vec![h(1)]),
                (h(3), vec![h(1)]),
                (h(4), vec![h(3), h(1)]),
                (h(5), vec![h(1)]),
            ]
            .into_iter()
            .collect(),
        );
        let build = |order: &[usize]| {
            let mut g = FileGraph::new();
            for &i in order {
                let (c, ins, dels) = &ops[i];
                for (idx, a, b, l) in ins {
                    g.insert(*c, *idx, *a, *b, l);
                }
                for d in dels {
                    g.delete(*c, d);
                }
            }
            let m = g.materialize(&deps);
            (m.clone(), text(&g, &m))
        };
        let reference = build(&[0, 1, 2, 3, 4]);
        assert!(reference.0.is_conflicted());
        let mut rng = rand::thread_rng();
        let mut order: Vec<usize> = (0..ops.len()).collect();
        for _ in 0..50 {
            order.shuffle(&mut rng);
            assert_eq!(build(&order), reference, "order {order:?}");
        }
    }
}
`/`	Search
`g` `c`	Code
`g` `h`	History
`g` `b`	Bookmarks
`g` `t`	Trust
`y`	Permalink to this state (files and trees)
`?`	This sheet
Keyboard
