Merge pull request GitoxideLabs#1763 from GitoxideLabs/better-refspec-primitives

Remote reverse mapping

Author: Byron

gix-protocol/src/fetch/refmap/init.rs

@@ -110,7 +110,7 @@ impl RefMap {
         let num_explicit_specs = fetch_refspecs.len();
         let group = gix_refspec::MatchGroup::from_fetch_specs(all_refspecs.iter().map(gix_refspec::RefSpec::to_ref));
         let (res, fixes) = group
-            .match_remotes(remote_refs.iter().map(|r| {
+            .match_lhs(remote_refs.iter().map(|r| {
                 let (full_ref_name, target, object) = r.unpack();
                 gix_refspec::match_group::Item {
                     full_ref_name,

gix-refspec/src/match_group/mod.rs

@@ -3,7 +3,7 @@ use std::collections::BTreeSet;
 use crate::{parse::Operation, types::Mode, MatchGroup, RefSpecRef};
 
 pub(crate) mod types;
-pub use types::{Item, Mapping, Outcome, Source, SourceRef};
+pub use types::{match_lhs, match_rhs, Item, Mapping, Source, SourceRef};
 
 ///
 pub mod validate;
@@ -26,13 +26,20 @@ impl<'a> MatchGroup<'a> {
 }
 
 /// Matching
-impl<'a> MatchGroup<'a> {
+impl<'spec> MatchGroup<'spec> {
     /// Match all `items` against all *fetch* specs present in this group, returning deduplicated mappings from source to destination.
-    /// *Note that this method is correct only for specs*, even though it also *works for push-specs*.
+    /// `items` are expected to be references on the remote, which will be matched and mapped to obtain their local counterparts,
+    /// i.e. *left side of refspecs is mapped to their right side*.
+    /// *Note that this method is correct only for fetch-specs*, even though it also *works for push-specs*.
+    ///
+    /// Object names are never mapped and always returned as match.
     ///
     /// Note that negative matches are not part of the return value, so they are not observable but will be used to remove mappings.
     // TODO: figure out how to deal with push-specs, probably when push is being implemented.
-    pub fn match_remotes<'item>(self, mut items: impl Iterator<Item = Item<'item>> + Clone) -> Outcome<'a, 'item> {
+    pub fn match_lhs<'item>(
+        self,
+        mut items: impl Iterator<Item = Item<'item>> + Clone,
+    ) -> match_lhs::Outcome<'spec, 'item> {
         let mut out = Vec::new();
         let mut seen = BTreeSet::default();
         let mut push_unique = |mapping| {
@@ -67,16 +74,15 @@ impl<'a> MatchGroup<'a> {
                 continue;
             }
             for (item_index, item) in items.clone().enumerate() {
-                if let Some(matcher) = matcher {
-                    let (matched, rhs) = matcher.matches_lhs(item);
-                    if matched {
-                        push_unique(Mapping {
-                            item_index: Some(item_index),
-                            lhs: SourceRef::FullName(item.full_ref_name),
-                            rhs,
-                            spec_index,
-                        });
-                    }
+                let Some(matcher) = matcher else { continue };
+                let (matched, rhs) = matcher.matches_lhs(item);
+                if matched {
+                    push_unique(Mapping {
+                        item_index: Some(item_index),
+                        lhs: SourceRef::FullName(item.full_ref_name.into()),
+                        rhs,
+                        spec_index,
+                    });
                 }
             }
         }
@@ -88,12 +94,78 @@ impl<'a> MatchGroup<'a> {
                 .zip(self.specs.iter())
                 .filter_map(|(m, spec)| m.and_then(|m| (spec.mode == Mode::Negative).then_some(m)))
             {
-                out.retain(|m| match m.lhs {
+                out.retain(|m| match &m.lhs {
                     SourceRef::ObjectId(_) => true,
                     SourceRef::FullName(name) => {
                         !matcher
                             .matches_lhs(Item {
-                                full_ref_name: name,
+                                full_ref_name: name.as_ref(),
+                                target: &null_id,
+                                object: None,
+                            })
+                            .0
+                    }
+                });
+            }
+        }
+        match_lhs::Outcome {
+            group: self,
+            mappings: out,
+        }
+    }
+
+    /// Match all `items` against all *fetch* specs present in this group, returning deduplicated mappings from destination to source.
+    /// `items` are expected to be tracking references in the local clone, which will be matched and reverse-mapped to obtain their remote counterparts,
+    /// i.e. *right side of refspecs is mapped to their left side*.
+    /// *Note that this method is correct only for fetch-specs*, even though it also *works for push-specs*.
+    ///
+    /// Note that negative matches are not part of the return value, so they are not observable but will be used to remove mappings.
+    // Reverse-mapping is implemented here: https://github.com/git/git/blob/76cf4f61c87855ebf0784b88aaf737d6b09f504b/branch.c#L252
+    pub fn match_rhs<'item>(
+        self,
+        mut items: impl Iterator<Item = Item<'item>> + Clone,
+    ) -> match_rhs::Outcome<'spec, 'item> {
+        let mut out = Vec::<Mapping<'spec, 'item>>::new();
+        let mut seen = BTreeSet::default();
+        let mut push_unique = |mapping| {
+            if seen.insert(calculate_hash(&mapping)) {
+                out.push(mapping);
+            }
+        };
+        let mut matchers: Vec<Matcher<'_>> = self.specs.iter().copied().map(Matcher::from).collect();
+
+        let mut has_negation = false;
+        for (spec_index, (spec, matcher)) in self.specs.iter().zip(matchers.iter_mut()).enumerate() {
+            if spec.mode == Mode::Negative {
+                has_negation = true;
+                continue;
+            }
+            for (item_index, item) in items.clone().enumerate() {
+                let (matched, lhs) = matcher.matches_rhs(item);
+                if let Some(lhs) = lhs.filter(|_| matched) {
+                    push_unique(Mapping {
+                        item_index: Some(item_index),
+                        lhs: SourceRef::FullName(lhs),
+                        rhs: Some(item.full_ref_name.into()),
+                        spec_index,
+                    });
+                }
+            }
+        }
+
+        if let Some(hash_kind) = has_negation.then(|| items.next().map(|i| i.target.kind())).flatten() {
+            let null_id = hash_kind.null();
+            for matcher in matchers
+                .into_iter()
+                .zip(self.specs.iter())
+                .filter_map(|(m, spec)| (spec.mode == Mode::Negative).then_some(m))
+            {
+                out.retain(|m| match &m.lhs {
+                    SourceRef::ObjectId(_) => true,
+                    SourceRef::FullName(name) => {
+                        !matcher
+                            .matches_rhs(Item {
+                                full_ref_name: name.as_ref(),
                                 target: &null_id,
                                 object: None,
                             })
@@ -102,7 +174,7 @@ impl<'a> MatchGroup<'a> {
                 });
             }
         }
-        Outcome {
+        match_rhs::Outcome {
             group: self,
             mappings: out,
         }

gix-refspec/src/match_group/types.rs

@@ -1,6 +1,6 @@
 use std::borrow::Cow;
 
-use bstr::{BStr, BString};
+use bstr::BStr;
 use gix_hash::oid;
 
 use crate::RefSpecRef;
@@ -12,15 +12,38 @@ pub struct MatchGroup<'a> {
     pub specs: Vec<RefSpecRef<'a>>,
 }
 
-/// The outcome of any matching operation of a [`MatchGroup`].
 ///
-/// It's used to validate and process the contained [mappings][Mapping].
-#[derive(Debug, Clone)]
-pub struct Outcome<'spec, 'item> {
-    /// The match group that produced this outcome.
-    pub group: MatchGroup<'spec>,
-    /// The mappings derived from matching [items][Item].
-    pub mappings: Vec<Mapping<'item, 'spec>>,
+pub mod match_lhs {
+    use crate::match_group::Mapping;
+    use crate::MatchGroup;
+
+    /// The outcome of any matching operation of a [`MatchGroup`].
+    ///
+    /// It's used to validate and process the contained [mappings](Mapping).
+    #[derive(Debug, Clone)]
+    pub struct Outcome<'spec, 'item> {
+        /// The match group that produced this outcome.
+        pub group: MatchGroup<'spec>,
+        /// The mappings derived from matching [items](crate::match_group::Item).
+        pub mappings: Vec<Mapping<'item, 'spec>>,
+    }
+}
+
+///
+pub mod match_rhs {
+    use crate::match_group::Mapping;
+    use crate::MatchGroup;
+
+    /// The outcome of any matching operation of a [`MatchGroup`].
+    ///
+    /// It's used to validate and process the contained [mappings](Mapping).
+    #[derive(Debug, Clone)]
+    pub struct Outcome<'spec, 'item> {
+        /// The match group that produced this outcome.
+        pub group: MatchGroup<'spec>,
+        /// The mappings derived from matching [items](crate::match_group::Item).
+        pub mappings: Vec<Mapping<'spec, 'item>>,
+    }
 }
 
 /// An item to match, input to various matching operations.
@@ -34,13 +57,13 @@ pub struct Item<'a> {
     pub object: Option<&'a oid>,
 }
 
-#[derive(Debug, Copy, Clone, PartialEq, Eq, Hash)]
-/// The source (or left-hand) side of a mapping, which references its name.
+#[derive(Debug, Clone, PartialEq, Eq, Hash)]
+/// The source (or left-hand) side of a mapping.
 pub enum SourceRef<'a> {
     /// A full reference name, which is expected to be valid.
     ///
     /// Validity, however, is not enforced here.
-    FullName(&'a BStr),
+    FullName(Cow<'a, BStr>),
     /// The name of an object that is expected to exist on the remote side.
     /// Note that it might not be advertised by the remote but part of the object graph,
     /// and thus gets sent in the pack. The server is expected to fail unless the desired
@@ -49,38 +72,27 @@ pub enum SourceRef<'a> {
 }
 
 impl SourceRef<'_> {
-    /// Create a fully owned instance from this one.
-    pub fn to_owned(&self) -> Source {
+    /// Create a fully owned instance by consuming this one.
+    pub fn into_owned(self) -> Source {
         match self {
-            SourceRef::ObjectId(id) => Source::ObjectId(*id),
-            SourceRef::FullName(name) => Source::FullName((*name).to_owned()),
+            SourceRef::ObjectId(id) => Source::ObjectId(id),
+            SourceRef::FullName(name) => Source::FullName(name.into_owned().into()),
         }
     }
 }
 
-#[derive(Debug, Clone, PartialEq, Eq, Hash)]
-/// The source (or left-hand) side of a mapping, which owns its name.
-pub enum Source {
-    /// A full reference name, which is expected to be valid.
-    ///
-    /// Validity, however, is not enforced here.
-    FullName(BString),
-    /// The name of an object that is expected to exist on the remote side.
-    /// Note that it might not be advertised by the remote but part of the object graph,
-    /// and thus gets sent in the pack. The server is expected to fail unless the desired
-    /// object is present but at some time it is merely a request by the user.
-    ObjectId(gix_hash::ObjectId),
-}
-
-impl std::fmt::Display for Source {
+impl std::fmt::Display for SourceRef<'_> {
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
         match self {
-            Source::FullName(name) => name.fmt(f),
-            Source::ObjectId(id) => id.fmt(f),
+            SourceRef::FullName(name) => name.fmt(f),
+            SourceRef::ObjectId(id) => id.fmt(f),
         }
     }
 }
 
+/// The source (or left-hand) side of a mapping, which owns its name.
+pub type Source = SourceRef<'static>;
+
 /// A mapping from a remote to a local refs for fetches or local to remote refs for pushes.
 ///
 /// Mappings are like edges in a graph, initially without any constraints.

gix-refspec/src/match_group/util.rs

@@ -12,8 +12,10 @@ pub struct Matcher<'a> {
 }
 
 impl<'a> Matcher<'a> {
-    /// Match `item` against this spec and return `(true, Some<rhs>)` to gain the other side of the match as configured, or `(true, None)`
-    /// if there was no `rhs` but the `item` matched. Lastly, return `(false, None)` if `item` didn't match at all.
+    /// Match the lefthand-side `item` against this spec and return `(true, Some<rhs>)` to gain the other,
+    /// transformed righthand-side of the match as configured by the refspec.
+    /// Or return `(true, None)` if there was no `rhs` but the `item` matched.
+    /// Lastly, return `(false, None)` if `item` didn't match at all.
     ///
     /// This may involve resolving a glob with an allocation, as the destination is built using the matching portion of a glob.
     pub fn matches_lhs(&self, item: Item<'_>) -> (bool, Option<Cow<'a, BStr>>) {
@@ -23,6 +25,20 @@ impl<'a> Matcher<'a> {
             (None, _) => (false, None),
         }
     }
+
+    /// Match the righthand-side `item` against this spec and return `(true, Some<lhs>)` to gain the other,
+    /// transformed lefthand-side of the match as configured by the refspec.
+    /// Or return `(true, None)` if there was no `lhs` but the `item` matched.
+    /// Lastly, return `(false, None)` if `item` didn't match at all.
+    ///
+    /// This may involve resolving a glob with an allocation, as the destination is built using the matching portion of a glob.
+    pub fn matches_rhs(&self, item: Item<'_>) -> (bool, Option<Cow<'a, BStr>>) {
+        match (self.lhs, self.rhs) {
+            (None, Some(rhs)) => (rhs.matches(item).is_match(), None),
+            (Some(lhs), Some(rhs)) => rhs.matches(item).into_match_outcome(lhs, item),
+            (_, None) => (false, None),
+        }
+    }
 }
 
 #[derive(Debug, Copy, Clone)]