Merge pull request #1618 from GitoxideLabs/merge

octopus-merge (part 5: tree-merge-ORT three-way)

Author: Byron

.github/FUNDING.yml

@@ -1 +1,2 @@
 github: byron
+open_collective: gitoxide

Cargo.lock

@@ -338,14 +338,21 @@ Check out the [performance discussion][gix-diff-performance] as well.
 
 ### gix-merge
 
-* [x] three-way merge analysis of **blobs** with choice of how to resolve conflicts
+* [x] three-way content-merge analysis of **blobs** with choice of how to resolve conflicts
+    - [x] respect git attributes and drivers.
     - [ ] choose how to resolve conflicts on the data-structure
-    - [ ] produce a new blob based on data-structure containing possible resolutions
+    - [ ] more efficient handling of paths with `merge=binary` attributes (do not load them into memory)
+    - [x] produce a new blob based on data-structure containing possible resolutions
         - [x] `merge` style
         - [x] `diff3` style
         - [x] `zdiff` style
+    - [ ] various newlines-related options during the merge (see https://git-scm.com/docs/git-merge#Documentation/git-merge.txt-ignore-space-change).
     - [ ] a way to control inter-hunk merging based on proximity (maybe via `gix-diff` feature which could use the same)
-* [ ] diff-heuristics match Git perfectly
+* [x] **tree**-diff-heuristics match Git for its test-cases
+    - [ ] a way to generate an index with stages 
+        - *currently the data it provides won't generate index entries, and possibly can't be used for it yet*
+    - [ ] submodule merges (*right now they count as conflicts if they differ*)
+* [x] **commits** - with handling of multiple merge bases by recursive merge-base merge
 * [x] API documentation
     * [ ] Examples
 

crate-status.md

@@ -9,7 +9,8 @@ use anyhow::{anyhow, Result};
 use gix::{
     hash::ObjectId,
     object, objs, odb,
-    odb::{loose, pack, Write},
+    odb::{loose, pack},
+    prelude::Write,
     NestedProgress,
 };
 
@@ -96,8 +97,8 @@ enum OutputWriter {
     Sink(odb::Sink),
 }
 
-impl gix::odb::Write for OutputWriter {
-    fn write_buf(&self, kind: object::Kind, from: &[u8]) -> Result<ObjectId, gix::odb::write::Error> {
+impl gix::objs::Write for OutputWriter {
+    fn write_buf(&self, kind: object::Kind, from: &[u8]) -> Result<ObjectId, gix::objs::write::Error> {
         match self {
             OutputWriter::Loose(db) => db.write_buf(kind, from),
             OutputWriter::Sink(db) => db.write_buf(kind, from),
@@ -109,7 +110,7 @@ impl gix::odb::Write for OutputWriter {
         kind: object::Kind,
         size: u64,
         from: &mut dyn Read,
-    ) -> Result<ObjectId, gix::odb::write::Error> {
+    ) -> Result<ObjectId, gix::objs::write::Error> {
         match self {
             OutputWriter::Loose(db) => db.write_stream(kind, size, from),
             OutputWriter::Sink(db) => db.write_stream(kind, size, from),

gitoxide-core/src/pack/explode.rs

@@ -1,5 +1,6 @@
 use gix::bstr::{BString, ByteSlice};
 use gix::objs::tree::EntryMode;
+use gix::odb::store::RefreshMode;
 use gix::prelude::ObjectIdExt;
 
 pub fn tree(
@@ -9,6 +10,7 @@ pub fn tree(
     new_treeish: BString,
 ) -> anyhow::Result<()> {
     repo.object_cache_size_if_unset(repo.compute_object_cache_size_for_tree_diffs(&**repo.index_or_empty()?));
+    repo.objects.refresh = RefreshMode::Never;
 
     let old_tree_id = repo.rev_parse_single(old_treeish.as_bstr())?;
     let new_tree_id = repo.rev_parse_single(new_treeish.as_bstr())?;

gitoxide-core/src/repository/diff.rs

@@ -1,5 +1,5 @@
 use crate::OutputFormat;
-use anyhow::{bail, Context};
+use anyhow::{anyhow, bail, Context};
 use gix::bstr::BString;
 use gix::bstr::ByteSlice;
 use gix::merge::blob::builtin_driver::binary;
@@ -83,8 +83,11 @@ pub fn file(
         other: Some(theirs.as_bstr()),
     };
     let mut buf = repo.empty_reusable_buffer();
-    let (pick, resolution) = platform.merge(&mut buf, labels, repo.command_context()?)?;
-    let buf = platform.buffer_by_pick(pick).unwrap_or(&buf);
+    let (pick, resolution) = platform.merge(&mut buf, labels, &repo.command_context()?)?;
+    let buf = platform
+        .buffer_by_pick(pick)
+        .map_err(|_| anyhow!("Participating object was too large"))?
+        .unwrap_or(&buf);
     out.write_all(buf)?;
 
     if resolution == Resolution::Conflict {

gitoxide-core/src/repository/merge.rs renamed to gitoxide-core/src/repository/merge/file.rs

@@ -0,0 +1,5 @@
+mod file;
+pub use file::file;
+
+pub mod tree;
+pub use tree::function::tree;

gitoxide-core/src/repository/merge/mod.rs

@@ -0,0 +1,112 @@
+use crate::OutputFormat;
+
+pub struct Options {
+    pub format: OutputFormat,
+    pub resolve_content_merge: Option<gix::merge::blob::builtin_driver::text::Conflict>,
+    pub in_memory: bool,
+}
+
+pub(super) mod function {
+
+    use crate::OutputFormat;
+    use anyhow::{anyhow, bail, Context};
+    use gix::bstr::BString;
+    use gix::bstr::ByteSlice;
+    use gix::merge::blob::builtin_driver::binary;
+    use gix::merge::blob::builtin_driver::text::Conflict;
+    use gix::merge::tree::UnresolvedConflict;
+    use gix::prelude::Write;
+
+    use super::Options;
+
+    #[allow(clippy::too_many_arguments)]
+    pub fn tree(
+        mut repo: gix::Repository,
+        out: &mut dyn std::io::Write,
+        err: &mut dyn std::io::Write,
+        base: BString,
+        ours: BString,
+        theirs: BString,
+        Options {
+            format,
+            resolve_content_merge,
+            in_memory,
+        }: Options,
+    ) -> anyhow::Result<()> {
+        if format != OutputFormat::Human {
+            bail!("JSON output isn't implemented yet");
+        }
+        repo.object_cache_size_if_unset(repo.compute_object_cache_size_for_tree_diffs(&**repo.index_or_empty()?));
+        if in_memory {
+            repo.objects.enable_object_memory();
+        }
+        let (base_ref, base_id) = refname_and_tree(&repo, base)?;
+        let (ours_ref, ours_id) = refname_and_tree(&repo, ours)?;
+        let (theirs_ref, theirs_id) = refname_and_tree(&repo, theirs)?;
+
+        let mut options = repo.tree_merge_options()?;
+        if let Some(resolve) = resolve_content_merge {
+            options.blob_merge.text.conflict = resolve;
+            options.blob_merge.resolve_binary_with = match resolve {
+                Conflict::Keep { .. } => None,
+                Conflict::ResolveWithOurs => Some(binary::ResolveWith::Ours),
+                Conflict::ResolveWithTheirs => Some(binary::ResolveWith::Theirs),
+                Conflict::ResolveWithUnion => None,
+            };
+        }
+
+        let base_id_str = base_id.to_string();
+        let ours_id_str = ours_id.to_string();
+        let theirs_id_str = theirs_id.to_string();
+        let labels = gix::merge::blob::builtin_driver::text::Labels {
+            ancestor: base_ref
+                .as_ref()
+                .map_or(base_id_str.as_str().into(), |n| n.as_bstr())
+                .into(),
+            current: ours_ref
+                .as_ref()
+                .map_or(ours_id_str.as_str().into(), |n| n.as_bstr())
+                .into(),
+            other: theirs_ref
+                .as_ref()
+                .map_or(theirs_id_str.as_str().into(), |n| n.as_bstr())
+                .into(),
+        };
+        let mut res = repo.merge_trees(base_id, ours_id, theirs_id, labels, options)?;
+        {
+            let _span = gix::trace::detail!("Writing merged tree");
+            let mut written = 0;
+            let tree_id = res
+                .tree
+                .write(|tree| {
+                    written += 1;
+                    repo.write(tree)
+                })
+                .map_err(|err| anyhow!("{err}"))?;
+            writeln!(out, "{tree_id} (wrote {written} trees)")?;
+        }
+
+        if !res.conflicts.is_empty() {
+            writeln!(err, "{} possibly resolved conflicts", res.conflicts.len())?;
+        }
+        if res.has_unresolved_conflicts(UnresolvedConflict::Renames) {
+            bail!("Tree conflicted")
+        }
+        Ok(())
+    }
+
+    fn refname_and_tree(
+        repo: &gix::Repository,
+        revspec: BString,
+    ) -> anyhow::Result<(Option<BString>, gix::hash::ObjectId)> {
+        let spec = repo.rev_parse(revspec.as_bstr())?;
+        let tree_id = spec
+            .single()
+            .context("Expected revspec to expand to a single rev only")?
+            .object()?
+            .peel_to_tree()?
+            .id;
+        let refname = spec.first_reference().map(|r| r.name.shorten().as_bstr().to_owned());
+        Ok((refname, tree_id))
+    }
+}

gitoxide-core/src/repository/merge/tree.rs

@@ -1,4 +1,6 @@
+use crate::tree::visit::ChangeId;
 use crate::Rewrites;
+use std::collections::BTreeSet;
 
 /// Types related to the rename tracker for renames, rewrites and copies.
 pub mod tracker;
@@ -12,6 +14,8 @@ pub struct Tracker<T> {
     path_backing: Vec<u8>,
     /// How to track copies and/or rewrites.
     rewrites: Rewrites,
+    /// Previously emitted relation ids of rewrite pairs, with `(deleted source, added destination)`.
+    child_renames: BTreeSet<(ChangeId, ChangeId)>,
 }
 
 /// Determine in which set of files to search for copies.