view rust/hg-core/src/ancestors.rs @ 49491:c6a1beba27e9

bisect: avoid copying ancestor list for non-merge commits

During a bisection, hg needs to compute a list of all ancestors for every
candidate commit. This is accomplished via a bottom-up traversal of the set of
candidates, during which each revision's ancestor list is populated using the
ancestor list of its parent(s). Previously, this involved copying the entire
list, which could be very long in if the bisection range was large.

To help improve this, we can observe that each candidate commit is visited
exactly once, at which point its ancestor list is copied into its children's
lists and then dropped. In the case of non-merge commits, a commit's ancestor
list consists exactly of its parent's list plus itself. This means that we can
trivially reuse the parent's existing list for one of its non-merge children,
which avoids copying entirely if that commit is the parent's only child. This
makes bisections over linear ranges of commits much faster.

During some informal testing in the large publicly-available `mozilla-central`
repository, this noticeably sped up bisections over large ranges of history:

Setup:
  $ cd mozilla-central
  $ hg bisect --reset
  $ hg bisect --good 0
  $ hg log -r tip -T '{rev}\n'
  628417

Test:
  $ time hg bisect --bad tip --noupdate

Before:
  real    3m35.927s
  user    3m35.553s
  sys     0m0.319s

After:
  real    1m41.142s
  user    1m40.810s
  sys     0m0.285s

// ancestors.rs
//
// Copyright 2018 Georges Racinet <[e-mail]>
//
// This software may be used and distributed according to the terms of the
// GNU General Public License version 2 or any later version.

//! Rust versions of generic DAG ancestors algorithms for Mercurial

use super::{Graph, GraphError, Revision, NULL_REVISION};
use crate::dagops;
use std::cmp::max;
use std::collections::{BinaryHeap, HashSet};

/// Iterator over the ancestors of a given list of revisions
/// This is a generic type, defined and implemented for any Graph, so that
/// it's easy to
///
/// - unit test in pure Rust
/// - bind to main Mercurial code, potentially in several ways and have these
///   bindings evolve over time
pub struct AncestorsIterator<G: Graph> {
    graph: G,
    visit: BinaryHeap<Revision>,
    seen: HashSet<Revision>,
    stoprev: Revision,
}

pub struct MissingAncestors<G: Graph> {
    graph: G,
    bases: HashSet<Revision>,
    max_base: Revision,
}

impl<G: Graph> AncestorsIterator<G> {
    /// Constructor.
    ///
    /// if `inclusive` is true, then the init revisions are emitted in
    /// particular, otherwise iteration starts from their parents.
    pub fn new(
        graph: G,
        initrevs: impl IntoIterator<Item = Revision>,
        stoprev: Revision,
        inclusive: bool,
    ) -> Result<Self, GraphError> {
        let filtered_initrevs = initrevs.into_iter().filter(|&r| r >= stoprev);
        if inclusive {
            let visit: BinaryHeap<Revision> = filtered_initrevs.collect();
            let seen = visit.iter().cloned().collect();
            return Ok(AncestorsIterator {
                visit,
                seen,
                stoprev,
                graph,
            });
        }
        let mut this = AncestorsIterator {
            visit: BinaryHeap::new(),
            seen: HashSet::new(),
            stoprev,
            graph,
        };
        this.seen.insert(NULL_REVISION);
        for rev in filtered_initrevs {
            for parent in this.graph.parents(rev)?.iter().cloned() {
                this.conditionally_push_rev(parent);
            }
        }
        Ok(this)
    }

    #[inline]
    fn conditionally_push_rev(&mut self, rev: Revision) {
        if self.stoprev <= rev && self.seen.insert(rev) {
            self.visit.push(rev);
        }
    }

    /// Consumes partially the iterator to tell if the given target
    /// revision
    /// is in the ancestors it emits.
    /// This is meant for iterators actually dedicated to that kind of
    /// purpose
    pub fn contains(&mut self, target: Revision) -> Result<bool, GraphError> {
        if self.seen.contains(&target) && target != NULL_REVISION {
            return Ok(true);
        }
        for item in self {
            let rev = item?;
            if rev == target {
                return Ok(true);
            }
            if rev < target {
                return Ok(false);
            }
        }
        Ok(false)
    }

    pub fn peek(&self) -> Option<Revision> {
        self.visit.peek().cloned()
    }

    /// Tell if the iterator is about an empty set
    ///
    /// The result does not depend whether the iterator has been consumed
    /// or not.
    /// This is mostly meant for iterators backing a lazy ancestors set
    pub fn is_empty(&self) -> bool {
        if self.visit.len() > 0 {
            return false;
        }
        if self.seen.len() > 1 {
            return false;
        }
        // at this point, the seen set is at most a singleton.
        // If not `self.inclusive`, it's still possible that it has only
        // the null revision
        self.seen.is_empty() || self.seen.contains(&NULL_REVISION)
    }
}

/// Main implementation for the iterator
///
/// The algorithm is the same as in `_lazyancestorsiter()` from `ancestors.py`
/// with a few non crucial differences:
///
/// - there's no filtering of invalid parent revisions. Actually, it should be
///   consistent and more efficient to filter them from the end caller.
/// - we don't have the optimization for adjacent revisions (i.e., the case
///   where `p1 == rev - 1`), because it amounts to update the first element of
///   the heap without sifting, which Rust's BinaryHeap doesn't let us do.
/// - we save a few pushes by comparing with `stoprev` before pushing
impl<G: Graph> Iterator for AncestorsIterator<G> {
    type Item = Result<Revision, GraphError>;

    fn next(&mut self) -> Option<Self::Item> {
        let current = match self.visit.peek() {
            None => {
                return None;
            }
            Some(c) => *c,
        };
        let [p1, p2] = match self.graph.parents(current) {
            Ok(ps) => ps,
            Err(e) => return Some(Err(e)),
        };
        if p1 < self.stoprev || !self.seen.insert(p1) {
            self.visit.pop();
        } else {
            *(self.visit.peek_mut().unwrap()) = p1;
        };

        self.conditionally_push_rev(p2);
        Some(Ok(current))
    }
}

impl<G: Graph> MissingAncestors<G> {
    pub fn new(graph: G, bases: impl IntoIterator<Item = Revision>) -> Self {
        let mut created = MissingAncestors {
            graph,
            bases: HashSet::new(),
            max_base: NULL_REVISION,
        };
        created.add_bases(bases);
        created
    }

    pub fn has_bases(&self) -> bool {
        !self.bases.is_empty()
    }

    /// Return a reference to current bases.
    ///
    /// This is useful in unit tests, but also setdiscovery.py does
    /// read the bases attribute of a ancestor.missingancestors instance.
    pub fn get_bases<'a>(&'a self) -> &'a HashSet<Revision> {
        &self.bases
    }

    /// Computes the relative heads of current bases.
    ///
    /// The object is still usable after this.
    pub fn bases_heads(&self) -> Result<HashSet<Revision>, GraphError> {
        dagops::heads(&self.graph, self.bases.iter())
    }

    /// Consumes the object and returns the relative heads of its bases.
    pub fn into_bases_heads(
        mut self,
    ) -> Result<HashSet<Revision>, GraphError> {
        dagops::retain_heads(&self.graph, &mut self.bases)?;
        Ok(self.bases)
    }

    /// Add some revisions to `self.bases`
    ///
    /// Takes care of keeping `self.max_base` up to date.
    pub fn add_bases(
        &mut self,
        new_bases: impl IntoIterator<Item = Revision>,
    ) {
        let mut max_base = self.max_base;
        self.bases.extend(
            new_bases
                .into_iter()
                .filter(|&rev| rev != NULL_REVISION)
                .map(|r| {
                    if r > max_base {
                        max_base = r;
                    }
                    r
                }),
        );
        self.max_base = max_base;
    }

    /// Remove all ancestors of self.bases from the revs set (in place)
    pub fn remove_ancestors_from(
        &mut self,
        revs: &mut HashSet<Revision>,
    ) -> Result<(), GraphError> {
        revs.retain(|r| !self.bases.contains(r));
        // the null revision is always an ancestor. Logically speaking
        // it's debatable in case bases is empty, but the Python
        // implementation always adds NULL_REVISION to bases, making it
        // unconditionnally true.
        revs.remove(&NULL_REVISION);
        if revs.is_empty() {
            return Ok(());
        }
        // anything in revs > start is definitely not an ancestor of bases
        // revs <= start need to be investigated
        if self.max_base == NULL_REVISION {
            return Ok(());
        }

        // whatever happens, we'll keep at least keepcount of them
        // knowing this gives us a earlier stop condition than
        // going all the way to the root
        let keepcount = revs.iter().filter(|r| **r > self.max_base).count();

        let mut curr = self.max_base;
        while curr != NULL_REVISION && revs.len() > keepcount {
            if self.bases.contains(&curr) {
                revs.remove(&curr);
                self.add_parents(curr)?;
            }
            curr -= 1;
        }
        Ok(())
    }

    /// Add the parents of `rev` to `self.bases`
    ///
    /// This has no effect on `self.max_base`
    #[inline]
    fn add_parents(&mut self, rev: Revision) -> Result<(), GraphError> {
        if rev == NULL_REVISION {
            return Ok(());
        }
        for p in self.graph.parents(rev)?.iter().cloned() {
            // No need to bother the set with inserting NULL_REVISION over and
            // over
            if p != NULL_REVISION {
                self.bases.insert(p);
            }
        }
        Ok(())
    }

    /// Return all the ancestors of revs that are not ancestors of self.bases
    ///
    /// This may include elements from revs.
    ///
    /// Equivalent to the revset (::revs - ::self.bases). Revs are returned in
    /// revision number order, which is a topological order.
    pub fn missing_ancestors(
        &mut self,
        revs: impl IntoIterator<Item = Revision>,
    ) -> Result<Vec<Revision>, GraphError> {
        // just for convenience and comparison with Python version
        let bases_visit = &mut self.bases;
        let mut revs: HashSet<Revision> = revs
            .into_iter()
            .filter(|r| !bases_visit.contains(r))
            .collect();
        let revs_visit = &mut revs;
        let mut both_visit: HashSet<Revision> =
            revs_visit.intersection(&bases_visit).cloned().collect();
        if revs_visit.is_empty() {
            return Ok(Vec::new());
        }
        let max_revs = revs_visit.iter().cloned().max().unwrap();
        let start = max(self.max_base, max_revs);

        // TODO heuristics for with_capacity()?
        let mut missing: Vec<Revision> = Vec::new();
        for curr in (0..=start).rev() {
            if revs_visit.is_empty() {
                break;
            }
            if both_visit.remove(&curr) {
                // curr's parents might have made it into revs_visit through
                // another path
                for p in self.graph.parents(curr)?.iter().cloned() {
                    if p == NULL_REVISION {
                        continue;
                    }
                    revs_visit.remove(&p);
                    bases_visit.insert(p);
                    both_visit.insert(p);
                }
            } else if revs_visit.remove(&curr) {
                missing.push(curr);
                for p in self.graph.parents(curr)?.iter().cloned() {
                    if p == NULL_REVISION {
                        continue;
                    }
                    if bases_visit.contains(&p) {
                        // p is already known to be an ancestor of revs_visit
                        revs_visit.remove(&p);
                        both_visit.insert(p);
                    } else if both_visit.contains(&p) {
                        // p should have been in bases_visit
                        revs_visit.remove(&p);
                        bases_visit.insert(p);
                    } else {
                        // visit later
                        revs_visit.insert(p);
                    }
                }
            } else if bases_visit.contains(&curr) {
                for p in self.graph.parents(curr)?.iter().cloned() {
                    if p == NULL_REVISION {
                        continue;
                    }
                    if revs_visit.remove(&p) || both_visit.contains(&p) {
                        // p is an ancestor of bases_visit, and is implicitly
                        // in revs_visit, which means p is ::revs & ::bases.
                        bases_visit.insert(p);
                        both_visit.insert(p);
                    } else {
                        bases_visit.insert(p);
                    }
                }
            }
        }
        missing.reverse();
        Ok(missing)
    }
}

#[cfg(test)]
mod tests {

    use super::*;
    use crate::testing::{SampleGraph, VecGraph};
    use std::iter::FromIterator;

    fn list_ancestors<G: Graph>(
        graph: G,
        initrevs: Vec<Revision>,
        stoprev: Revision,
        inclusive: bool,
    ) -> Vec<Revision> {
        AncestorsIterator::new(graph, initrevs, stoprev, inclusive)
            .unwrap()
            .map(|res| res.unwrap())
            .collect()
    }

    #[test]
    /// Same tests as test-ancestor.py, without membership
    /// (see also test-ancestor.py.out)
    fn test_list_ancestor() {
        assert_eq!(list_ancestors(SampleGraph, vec![], 0, false), vec![]);
        assert_eq!(
            list_ancestors(SampleGraph, vec![11, 13], 0, false),
            vec![8, 7, 4, 3, 2, 1, 0]
        );
        assert_eq!(
            list_ancestors(SampleGraph, vec![1, 3], 0, false),
            vec![1, 0]
        );
        assert_eq!(
            list_ancestors(SampleGraph, vec![11, 13], 0, true),
            vec![13, 11, 8, 7, 4, 3, 2, 1, 0]
        );
        assert_eq!(
            list_ancestors(SampleGraph, vec![11, 13], 6, false),
            vec![8, 7]
        );
        assert_eq!(
            list_ancestors(SampleGraph, vec![11, 13], 6, true),
            vec![13, 11, 8, 7]
        );
        assert_eq!(
            list_ancestors(SampleGraph, vec![11, 13], 11, true),
            vec![13, 11]
        );
        assert_eq!(
            list_ancestors(SampleGraph, vec![11, 13], 12, true),
            vec![13]
        );
        assert_eq!(
            list_ancestors(SampleGraph, vec![10, 1], 0, true),
            vec![10, 5, 4, 2, 1, 0]
        );
    }

    #[test]
    /// Corner case that's not directly in test-ancestors.py, but
    /// that happens quite often, as demonstrated by running the whole
    /// suite.
    /// For instance, run tests/test-obsolete-checkheads.t
    fn test_nullrev_input() {
        let mut iter =
            AncestorsIterator::new(SampleGraph, vec![-1], 0, false).unwrap();
        assert_eq!(iter.next(), None)
    }

    #[test]
    fn test_contains() {
        let mut lazy =
            AncestorsIterator::new(SampleGraph, vec![10, 1], 0, true).unwrap();
        assert!(lazy.contains(1).unwrap());
        assert!(!lazy.contains(3).unwrap());

        let mut lazy =
            AncestorsIterator::new(SampleGraph, vec![0], 0, false).unwrap();
        assert!(!lazy.contains(NULL_REVISION).unwrap());
    }

    #[test]
    fn test_peek() {
        let mut iter =
            AncestorsIterator::new(SampleGraph, vec![10], 0, true).unwrap();
        // peek() gives us the next value
        assert_eq!(iter.peek(), Some(10));
        // but it's not been consumed
        assert_eq!(iter.next(), Some(Ok(10)));
        // and iteration resumes normally
        assert_eq!(iter.next(), Some(Ok(5)));

        // let's drain the iterator to test peek() at the end
        while iter.next().is_some() {}
        assert_eq!(iter.peek(), None);
    }

    #[test]
    fn test_empty() {
        let mut iter =
            AncestorsIterator::new(SampleGraph, vec![10], 0, true).unwrap();
        assert!(!iter.is_empty());
        while iter.next().is_some() {}
        assert!(!iter.is_empty());

        let iter =
            AncestorsIterator::new(SampleGraph, vec![], 0, true).unwrap();
        assert!(iter.is_empty());

        // case where iter.seen == {NULL_REVISION}
        let iter =
            AncestorsIterator::new(SampleGraph, vec![0], 0, false).unwrap();
        assert!(iter.is_empty());
    }

    /// A corrupted Graph, supporting error handling tests
    #[derive(Clone, Debug)]
    struct Corrupted;

    impl Graph for Corrupted {
        fn parents(&self, rev: Revision) -> Result<[Revision; 2], GraphError> {
            match rev {
                1 => Ok([0, -1]),
                r => Err(GraphError::ParentOutOfRange(r)),
            }
        }
    }

    #[test]
    fn test_initrev_out_of_range() {
        // inclusive=false looks up initrev's parents right away
        match AncestorsIterator::new(SampleGraph, vec![25], 0, false) {
            Ok(_) => panic!("Should have been ParentOutOfRange"),
            Err(e) => assert_eq!(e, GraphError::ParentOutOfRange(25)),
        }
    }

    #[test]
    fn test_next_out_of_range() {
        // inclusive=false looks up initrev's parents right away
        let mut iter =
            AncestorsIterator::new(Corrupted, vec![1], 0, false).unwrap();
        assert_eq!(iter.next(), Some(Err(GraphError::ParentOutOfRange(0))));
    }

    #[test]
    /// Test constructor, add/get bases and heads
    fn test_missing_bases() -> Result<(), GraphError> {
        let mut missing_ancestors =
            MissingAncestors::new(SampleGraph, [5, 3, 1, 3].iter().cloned());
        let mut as_vec: Vec<Revision> =
            missing_ancestors.get_bases().iter().cloned().collect();
        as_vec.sort();
        assert_eq!(as_vec, [1, 3, 5]);
        assert_eq!(missing_ancestors.max_base, 5);

        missing_ancestors.add_bases([3, 7, 8].iter().cloned());
        as_vec = missing_ancestors.get_bases().iter().cloned().collect();
        as_vec.sort();
        assert_eq!(as_vec, [1, 3, 5, 7, 8]);
        assert_eq!(missing_ancestors.max_base, 8);

        as_vec = missing_ancestors.bases_heads()?.iter().cloned().collect();
        as_vec.sort();
        assert_eq!(as_vec, [3, 5, 7, 8]);
        Ok(())
    }

    fn assert_missing_remove(
        bases: &[Revision],
        revs: &[Revision],
        expected: &[Revision],
    ) {
        let mut missing_ancestors =
            MissingAncestors::new(SampleGraph, bases.iter().cloned());
        let mut revset: HashSet<Revision> = revs.iter().cloned().collect();
        missing_ancestors
            .remove_ancestors_from(&mut revset)
            .unwrap();
        let mut as_vec: Vec<Revision> = revset.into_iter().collect();
        as_vec.sort();
        assert_eq!(as_vec.as_slice(), expected);
    }

    #[test]
    fn test_missing_remove() {
        assert_missing_remove(
            &[1, 2, 3, 4, 7],
            Vec::from_iter(1..10).as_slice(),
            &[5, 6, 8, 9],
        );
        assert_missing_remove(&[10], &[11, 12, 13, 14], &[11, 12, 13, 14]);
        assert_missing_remove(&[7], &[1, 2, 3, 4, 5], &[3, 5]);
    }

    fn assert_missing_ancestors(
        bases: &[Revision],
        revs: &[Revision],
        expected: &[Revision],
    ) {
        let mut missing_ancestors =
            MissingAncestors::new(SampleGraph, bases.iter().cloned());
        let missing = missing_ancestors
            .missing_ancestors(revs.iter().cloned())
            .unwrap();
        assert_eq!(missing.as_slice(), expected);
    }

    #[test]
    fn test_missing_ancestors() {
        // examples taken from test-ancestors.py by having it run
        // on the same graph (both naive and fast Python algs)
        assert_missing_ancestors(&[10], &[11], &[3, 7, 11]);
        assert_missing_ancestors(&[11], &[10], &[5, 10]);
        assert_missing_ancestors(&[7], &[9, 11], &[3, 6, 9, 11]);
    }

    /// An interesting case found by a random generator similar to
    /// the one in test-ancestor.py. An early version of Rust MissingAncestors
    /// failed this, yet none of the integration tests of the whole suite
    /// catched it.
    #[test]
    fn test_remove_ancestors_from_case1() {
        let graph: VecGraph = vec![
            [NULL_REVISION, NULL_REVISION],
            [0, NULL_REVISION],
            [1, 0],
            [2, 1],
            [3, NULL_REVISION],
            [4, NULL_REVISION],
            [5, 1],
            [2, NULL_REVISION],
            [7, NULL_REVISION],
            [8, NULL_REVISION],
            [9, NULL_REVISION],
            [10, 1],
            [3, NULL_REVISION],
            [12, NULL_REVISION],
            [13, NULL_REVISION],
            [14, NULL_REVISION],
            [4, NULL_REVISION],
            [16, NULL_REVISION],
            [17, NULL_REVISION],
            [18, NULL_REVISION],
            [19, 11],
            [20, NULL_REVISION],
            [21, NULL_REVISION],
            [22, NULL_REVISION],
            [23, NULL_REVISION],
            [2, NULL_REVISION],
            [3, NULL_REVISION],
            [26, 24],
            [27, NULL_REVISION],
            [28, NULL_REVISION],
            [12, NULL_REVISION],
            [1, NULL_REVISION],
            [1, 9],
            [32, NULL_REVISION],
            [33, NULL_REVISION],
            [34, 31],
            [35, NULL_REVISION],
            [36, 26],
            [37, NULL_REVISION],
            [38, NULL_REVISION],
            [39, NULL_REVISION],
            [40, NULL_REVISION],
            [41, NULL_REVISION],
            [42, 26],
            [0, NULL_REVISION],
            [44, NULL_REVISION],
            [45, 4],
            [40, NULL_REVISION],
            [47, NULL_REVISION],
            [36, 0],
            [49, NULL_REVISION],
            [NULL_REVISION, NULL_REVISION],
            [51, NULL_REVISION],
            [52, NULL_REVISION],
            [53, NULL_REVISION],
            [14, NULL_REVISION],
            [55, NULL_REVISION],
            [15, NULL_REVISION],
            [23, NULL_REVISION],
            [58, NULL_REVISION],
            [59, NULL_REVISION],
            [2, NULL_REVISION],
            [61, 59],
            [62, NULL_REVISION],
            [63, NULL_REVISION],
            [NULL_REVISION, NULL_REVISION],
            [65, NULL_REVISION],
            [66, NULL_REVISION],
            [67, NULL_REVISION],
            [68, NULL_REVISION],
            [37, 28],
            [69, 25],
            [71, NULL_REVISION],
            [72, NULL_REVISION],
            [50, 2],
            [74, NULL_REVISION],
            [12, NULL_REVISION],
            [18, NULL_REVISION],
            [77, NULL_REVISION],
            [78, NULL_REVISION],
            [79, NULL_REVISION],
            [43, 33],
            [81, NULL_REVISION],
            [82, NULL_REVISION],
            [83, NULL_REVISION],
            [84, 45],
            [85, NULL_REVISION],
            [86, NULL_REVISION],
            [NULL_REVISION, NULL_REVISION],
            [88, NULL_REVISION],
            [NULL_REVISION, NULL_REVISION],
            [76, 83],
            [44, NULL_REVISION],
            [92, NULL_REVISION],
            [93, NULL_REVISION],
            [9, NULL_REVISION],
            [95, 67],
            [96, NULL_REVISION],
            [97, NULL_REVISION],
            [NULL_REVISION, NULL_REVISION],
        ];
        let problem_rev = 28 as Revision;
        let problem_base = 70 as Revision;
        // making the problem obvious: problem_rev is a parent of problem_base
        assert_eq!(graph.parents(problem_base).unwrap()[1], problem_rev);

        let mut missing_ancestors: MissingAncestors<VecGraph> =
            MissingAncestors::new(
                graph,
                [60, 26, 70, 3, 96, 19, 98, 49, 97, 47, 1, 6]
                    .iter()
                    .cloned(),
            );
        assert!(missing_ancestors.bases.contains(&problem_base));

        let mut revs: HashSet<Revision> =
            [4, 12, 41, 28, 68, 38, 1, 30, 56, 44]
                .iter()
                .cloned()
                .collect();
        missing_ancestors.remove_ancestors_from(&mut revs).unwrap();
        assert!(!revs.contains(&problem_rev));
    }
}