feat: first basic implementation of merge_base().

Author: Byron

gix-revision/src/merge_base.rs

@@ -1,6 +1,6 @@
 bitflags::bitflags! {
     /// The flags used in the graph for finding [merge bases](crate::merge_base()).
-    #[derive(Debug, Default, Copy, Clone)]
+    #[derive(Debug, Default, Copy, Clone, Eq, PartialEq)]
     pub struct Flags: u8 {
         /// The commit belongs to the graph reachable by the first commit
         const COMMIT1 = 1 << 0;
@@ -18,6 +18,8 @@ bitflags::bitflags! {
 #[derive(Debug, thiserror::Error)]
 #[allow(missing_docs)]
 pub enum Error {
+    #[error(transparent)]
+    IterParents(#[from] gix_revwalk::graph::commit::iter_parents::Error),
     #[error("A commit could not be found")]
     FindExistingCommit(#[from] gix_object::find::existing_iter::Error),
     #[error("A commit could not be decoded during traversal")]
@@ -51,6 +53,24 @@ pub(crate) mod function {
             return Ok(Some(vec![first]));
         }
 
+        let mut bases = paint_down_to_common(first, others, graph)?;
+        // TODO(ST): remove redundant
+        Ok(if bases.is_empty() {
+            None
+        } else {
+            let mut out = Vec::with_capacity(bases.len());
+            while let Some((_info, commit_id)) = bases.pop() {
+                out.push(commit_id);
+            }
+            Some(out)
+        })
+    }
+
+    fn paint_down_to_common<'name>(
+        first: ObjectId,
+        others: &[ObjectId],
+        graph: &mut Graph<'_, Flags>,
+    ) -> Result<PriorityQueue<GenThenTime, ObjectId>, Error> {
         let mut queue = PriorityQueue::<GenThenTime, ObjectId>::new();
         graph.insert_data(first, |commit| -> Result<_, Error> {
             queue.insert(commit.try_into()?, first);
@@ -63,10 +83,47 @@ pub(crate) mod function {
                 Ok(Flags::COMMIT2)
             })?;
         }
-        Ok(None)
+
+        let mut out = PriorityQueue::new();
+        while queue
+            .iter_unordered()
+            .any(|id| graph.get(id).map_or(false, |data| !data.contains(Flags::STALE)))
+        {
+            let (info, commit_id) = queue.pop().expect("we have non-stale");
+            let flags_mut = graph.get_mut(&commit_id).expect("everything queued is in graph");
+            let mut flags_without_result = *flags_mut & (Flags::COMMIT1 | Flags::COMMIT2 | Flags::STALE);
+            if flags_without_result == (Flags::COMMIT1 | Flags::COMMIT2) {
+                if !flags_mut.contains(Flags::RESULT) {
+                    *flags_mut |= Flags::RESULT;
+                    out.insert(info, commit_id);
+                }
+                flags_without_result |= Flags::STALE;
+            }
+
+            graph.insert_parents_with_lookup(&commit_id, &mut |parent_id, parent, ex_flags| -> Result<_, Error> {
+                let queue_info = match ex_flags {
+                    Some(ex_flags) => {
+                        if (*ex_flags & flags_without_result) != flags_without_result {
+                            *ex_flags |= flags_without_result;
+                            Some(GenThenTime::try_from(parent)?)
+                        } else {
+                            None
+                        }
+                    }
+                    None => Some(GenThenTime::try_from(parent)?),
+                };
+                if let Some(info) = queue_info {
+                    queue.insert(info, parent_id);
+                }
+                Ok(flags_without_result)
+            })?;
+        }
+
+        Ok(out)
     }
 
     // TODO(ST): Should this type be used for `describe` as well?
+    #[derive(Debug)]
     struct GenThenTime {
         /// Note that the special [`GENERATION_NUMBER_INFINITY`](gix_commitgraph::GENERATION_NUMBER_INFINITY) is used to indicate
         /// that no commitgraph is avaialble.

gix-revision/tests/fixtures/make_merge_base_repos.sh

@@ -45,6 +45,9 @@ mkcommit 100 DB
   baseline DA DA DB
 } > 1_disjoint.baseline
 
+# A graph that is purposefully using times that can't be trusted, i.e. the root E
+# has a higher time than its future commits, so that it would be preferred
+# unless if there was an additional pruning step to deal with this case.
 # E---D---C---B---A
 # \"-_         \   \
 #  \  `---------G   \

gix-revwalk/src/graph/mod.rs

@@ -90,11 +90,26 @@ impl<'find, T> Graph<'find, T> {
         self.map.get_mut(id)
     }
 
-    /// Insert `id` into the graph and associate it with `value`, returning the previous value associated with it if it existed.
+    /// Insert `id` into the graph and associate it with `value`, returning the previous value associated with `id` if it existed.
     pub fn insert(&mut self, id: gix_hash::ObjectId, value: T) -> Option<T> {
         self.map.insert(id, value)
     }
 
+    /// Insert `id` into the graph and associate it with the value returned by `make_data`,
+    /// and returning the previous value associated with `id` if it existed.
+    /// Fail if `id` doesn't exist in the object database.
+    pub fn insert_data<E>(
+        &mut self,
+        id: gix_hash::ObjectId,
+        mut make_data: impl FnMut(LazyCommit<'_>) -> Result<T, E>,
+    ) -> Result<Option<T>, E>
+    where
+        E: From<gix_object::find::existing_iter::Error>,
+    {
+        let value = make_data(self.lookup(&id).map_err(E::from)?)?;
+        Ok(self.map.insert(id, value))
+    }
+
     /// Remove all data from the graph to start over.
     pub fn clear(&mut self) {
         self.map.clear();