Use merge tree to improve conflict checking performance when possible

modules/git/commit_test.go

@@ -4,7 +4,6 @@
 package git
 
 import (
-	"os"
 	"path/filepath"
 	"strings"
 	"testing"
@@ -339,18 +338,3 @@ func TestGetCommitFileStatusMerges(t *testing.T) {
 	assert.Equal(t, expected.Removed, commitFileStatus.Removed)
 	assert.Equal(t, expected.Modified, commitFileStatus.Modified)
 }
-
-func Test_GetCommitBranchStart(t *testing.T) {
-	bareRepo1Path := filepath.Join(testReposDir, "repo1_bare")
-	repo, err := OpenRepository(t.Context(), bareRepo1Path)
-	assert.NoError(t, err)
-	defer repo.Close()
-	commit, err := repo.GetBranchCommit("branch1")
-	assert.NoError(t, err)
-	assert.Equal(t, "2839944139e0de9737a044f78b0e4b40d989a9e3", commit.ID.String())
-
-	startCommitID, err := repo.GetCommitBranchStart(os.Environ(), "branch1", commit.ID.String())
-	assert.NoError(t, err)
-	assert.NotEmpty(t, startCommitID)
-	assert.Equal(t, "95bb4d39648ee7e325106df01a621c530863a653", startCommitID)
-}

modules/git/diff.go

@@ -7,6 +7,7 @@ import (
 	"bufio"
 	"bytes"
 	"context"
+	"errors"
 	"fmt"
 	"io"
 	"os"
@@ -288,20 +289,18 @@ func CutDiffAroundLine(originalDiff io.Reader, line int64, old bool, numbersOfLi
 }
 
 // GetAffectedFiles returns the affected files between two commits
-func GetAffectedFiles(repo *Repository, branchName, oldCommitID, newCommitID string, env []string) ([]string, error) {
-	if oldCommitID == emptySha1ObjectID.String() || oldCommitID == emptySha256ObjectID.String() {
-		startCommitID, err := repo.GetCommitBranchStart(env, branchName, newCommitID)
-		if err != nil {
-			return nil, err
-		}
-		if startCommitID == "" {
-			return nil, fmt.Errorf("cannot find the start commit of %s", newCommitID)
-		}
-		oldCommitID = startCommitID
+func GetAffectedFiles(ctx context.Context, repoPath, oldCommitID, newCommitID string, env []string) ([]string, error) {
+	if oldCommitID == emptySha1ObjectID.String() {
+		oldCommitID = emptySha1ObjectID.Type().EmptyTree().String()
+	} else if oldCommitID == emptySha256ObjectID.String() {
+		oldCommitID = emptySha256ObjectID.Type().EmptyTree().String()
+	} else if oldCommitID == "" {
+		return nil, errors.New("oldCommitID is empty")
 	}
+
 	stdoutReader, stdoutWriter, err := os.Pipe()
 	if err != nil {
-		log.Error("Unable to create os.Pipe for %s", repo.Path)
+		log.Error("Unable to create os.Pipe for %s", repoPath)
 		return nil, err
 	}
 	defer func() {
@@ -314,7 +313,7 @@ func GetAffectedFiles(repo *Repository, branchName, oldCommitID, newCommitID str
 	// Run `git diff --name-only` to get the names of the changed files
 	err = gitcmd.NewCommand("diff", "--name-only").AddDynamicArguments(oldCommitID, newCommitID).
 		WithEnv(env).
-		WithDir(repo.Path).
+		WithDir(repoPath).
 		WithStdout(stdoutWriter).
 		WithPipelineFunc(func(ctx context.Context, cancel context.CancelFunc) error {
 			// Close the writer end of the pipe to begin processing
@@ -334,9 +333,9 @@ func GetAffectedFiles(repo *Repository, branchName, oldCommitID, newCommitID str
 			}
 			return scanner.Err()
 		}).
-		Run(repo.Ctx)
+		Run(ctx)
 	if err != nil {
-		log.Error("Unable to get affected files for commits from %s to %s in %s: %v", oldCommitID, newCommitID, repo.Path, err)
+		log.Error("Unable to get affected files for commits from %s to %s in %s: %v", oldCommitID, newCommitID, repoPath, err)
 	}
 
 	return affectedFiles, err

modules/git/git.go

@@ -31,6 +31,7 @@ type Features struct {
 	SupportHashSha256      bool           // >= 2.42, SHA-256 repositories no longer an ‘experimental curiosity’
 	SupportedObjectFormats []ObjectFormat // sha1, sha256
 	SupportCheckAttrOnBare bool           // >= 2.40
+	SupportGitMergeTree    bool           // >= 2.38
 }
 
 var defaultFeatures *Features
@@ -75,6 +76,7 @@ func loadGitVersionFeatures() (*Features, error) {
 		features.SupportedObjectFormats = append(features.SupportedObjectFormats, Sha256ObjectFormat)
 	}
 	features.SupportCheckAttrOnBare = features.CheckVersionAtLeast("2.40")
+	features.SupportGitMergeTree = features.CheckVersionAtLeast("2.38")
 	return features, nil
 }
 

modules/git/gitcmd/command.go

@@ -456,6 +456,17 @@ func IsErrorExitCode(err error, code int) bool {
 	return false
 }
 
+func ExitCode(err error) (int, bool) {
+	if err == nil {
+		return 0, true
+	}
+	var exitError *exec.ExitError
+	if errors.As(err, &exitError) {
+		return exitError.ExitCode(), true
+	}
+	return 0, false
+}
+
 // RunStdString runs the command and returns stdout/stderr as string. and store stderr to returned error (err combined with stderr).
 func (c *Command) RunStdString(ctx context.Context) (stdout, stderr string, runErr RunStdError) {
 	stdoutBytes, stderrBytes, runErr := c.WithParentCallerInfo().runStdBytes(ctx)

modules/git/repo_commit.go

@@ -580,34 +580,3 @@ func (repo *Repository) AddLastCommitCache(cacheKey, fullName, sha string) error
 	}
 	return nil
 }
-
-// GetCommitBranchStart returns the commit where the branch diverged
-func (repo *Repository) GetCommitBranchStart(env []string, branch, endCommitID string) (string, error) {
-	cmd := gitcmd.NewCommand("log", prettyLogFormat)
-	cmd.AddDynamicArguments(endCommitID)
-
-	stdout, _, runErr := cmd.WithDir(repo.Path).
-		WithEnv(env).
-		RunStdBytes(repo.Ctx)
-	if runErr != nil {
-		return "", runErr
-	}
-
-	parts := bytes.SplitSeq(bytes.TrimSpace(stdout), []byte{'\n'})
-
-	// check the commits one by one until we find a commit contained by another branch
-	// and we think this commit is the divergence point
-	for commitID := range parts {
-		branches, err := repo.getBranches(env, string(commitID), 2)
-		if err != nil {
-			return "", err
-		}
-		for _, b := range branches {
-			if b != branch {
-				return string(commitID), nil
-			}
-		}
-	}
-
-	return "", nil
-}

modules/gitrepo/fetch.go

@@ -0,0 +1,17 @@
+// Copyright 2025 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package gitrepo
+
+import (
+	"context"
+
+	"code.gitea.io/gitea/modules/git/gitcmd"
+)
+
+func FetchRemoteCommit(ctx context.Context, repo, remoteRepo Repository, commitID string) error {
+	_, err := RunCmdString(ctx, repo, gitcmd.NewCommand("fetch", "--no-tags").
+		AddDynamicArguments(repoPath(remoteRepo)).
+		AddDynamicArguments(commitID))
+	return err
+}

modules/gitrepo/merge.go

@@ -0,0 +1,74 @@
+// Copyright 2025 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package gitrepo
+
+import (
+	"bytes"
+	"context"
+	"errors"
+	"fmt"
+	"strings"
+
+	"code.gitea.io/gitea/modules/git"
+	"code.gitea.io/gitea/modules/git/gitcmd"
+)
+
+func MergeBase(ctx context.Context, repo Repository, commit1, commit2 string) (string, error) {
+	mergeBase, err := RunCmdString(ctx, repo, gitcmd.NewCommand("merge-base", "--").
+		AddDynamicArguments(commit1, commit2))
+	if err != nil {
+		return "", fmt.Errorf("get merge-base of %s and %s failed: %w", commit1, commit2, err)
+	}
+	return strings.TrimSpace(mergeBase), nil
+}
+
+// parseMergeTreeOutput parses the output of git merge-tree --write-tree -z --name-only --no-messages
+// For a successful merge, the output is a simply one line <OID of toplevel tree>NUL
+// Whereas for a conflicted merge, the output is:
+// <OID of toplevel tree>NUL
+// <Conflicted file name 1>NUL
+// <Conflicted file name 2>NUL
+// ...
+// ref: https://git-scm.com/docs/git-merge-tree/2.38.0#OUTPUT
+func parseMergeTreeOutput(output string) (string, []string, error) {
+	fields := strings.Split(strings.TrimSuffix(output, "\x00"), "\x00")
+	if len(fields) == 0 {
+		return "", nil, errors.New("unexpected empty output")
+	}
+	if len(fields) == 1 {
+		return strings.TrimSpace(fields[0]), nil, nil
+	}
+	return strings.TrimSpace(fields[0]), fields[1:], nil
+}
+
+// MergeTree performs a merge between two commits (baseRef and headRef) with an optional merge base.
+// It returns the resulting tree hash, a list of conflicted files (if any), and an error if the operation fails.
+// If there are no conflicts, the list of conflicted files will be nil.
+func MergeTree(ctx context.Context, repo Repository, baseRef, headRef, mergeBase string) (string, bool, []string, error) {
+	cmd := gitcmd.NewCommand("merge-tree", "--write-tree", "-z", "--name-only", "--no-messages")
+	if git.DefaultFeatures().CheckVersionAtLeast("2.40") && mergeBase != "" {
+		cmd.AddOptionFormat("--merge-base=%s", mergeBase)
+	}
+
+	stdout := &bytes.Buffer{}
+	gitErr := RunCmd(ctx, repo, cmd.AddDynamicArguments(baseRef, headRef).WithStdout(stdout))
+	exitCode, ok := gitcmd.ExitCode(gitErr)
+	if !ok {
+		return "", false, nil, fmt.Errorf("run merge-tree failed: %w", gitErr)
+	}
+
+	switch exitCode {
+	case 0, 1:
+		treeID, conflictedFiles, err := parseMergeTreeOutput(stdout.String())
+		if err != nil {
+			return "", false, nil, fmt.Errorf("parse merge-tree output failed: %w", err)
+		}
+		// For a successful, non-conflicted merge, the exit status is 0. When the merge has conflicts, the exit status is 1.
+		// A merge can have conflicts without having individual files conflict
+		// https://git-scm.com/docs/git-merge-tree/2.38.0#_mistakes_to_avoid
+		return treeID, exitCode == 1, conflictedFiles, nil
+	default:
+		return "", false, nil, fmt.Errorf("run merge-tree exit abnormally: %w", gitErr)
+	}
+}

modules/gitrepo/merge_test.go

@@ -0,0 +1,26 @@
+// Copyright 2025 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package gitrepo
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func Test_parseMergeTreeOutput(t *testing.T) {
+	conflictedOutput := "837480c2773160381cbe6bcce90f7732789b5856\x00options/locale/locale_en-US.ini\x00services/webhook/webhook_test.go\x00"
+	treeID, conflictedFiles, err := parseMergeTreeOutput(conflictedOutput)
+	assert.NoError(t, err)
+	assert.Equal(t, "837480c2773160381cbe6bcce90f7732789b5856", treeID)
+	assert.Len(t, conflictedFiles, 2)
+	assert.Equal(t, "options/locale/locale_en-US.ini", conflictedFiles[0])
+	assert.Equal(t, "services/webhook/webhook_test.go", conflictedFiles[1])
+
+	nonConflictedOutput := "837480c2773160381cbe6bcce90f7732789b5856\x00"
+	treeID, conflictedFiles, err = parseMergeTreeOutput(nonConflictedOutput)
+	assert.NoError(t, err)
+	assert.Equal(t, "837480c2773160381cbe6bcce90f7732789b5856", treeID)
+	assert.Empty(t, conflictedFiles)
+}

routers/private/hook_pre_receive.go

@@ -242,7 +242,7 @@ func preReceiveBranch(ctx *preReceiveContext, oldCommitID, newCommitID string, r
 
 	globs := protectBranch.GetProtectedFilePatterns()
 	if len(globs) > 0 {
-		_, err := pull_service.CheckFileProtection(gitRepo, branchName, oldCommitID, newCommitID, globs, 1, ctx.env)
+		_, err := pull_service.CheckFileProtection(ctx, repo.RepoPath(), oldCommitID, newCommitID, globs, 1, ctx.env)
 		if err != nil {
 			if !pull_service.IsErrFilePathProtected(err) {
 				log.Error("Unable to check file protection for commits from %s to %s in %-v: %v", oldCommitID, newCommitID, repo, err)
@@ -300,7 +300,7 @@ func preReceiveBranch(ctx *preReceiveContext, oldCommitID, newCommitID string, r
 			// Allow commits that only touch unprotected files
 			globs := protectBranch.GetUnprotectedFilePatterns()
 			if len(globs) > 0 {
-				unprotectedFilesOnly, err := pull_service.CheckUnprotectedFiles(gitRepo, branchName, oldCommitID, newCommitID, globs, ctx.env)
+				unprotectedFilesOnly, err := pull_service.CheckUnprotectedFiles(ctx, repo, oldCommitID, newCommitID, globs, ctx.env)
 				if err != nil {
 					log.Error("Unable to check file protection for commits from %s to %s in %-v: %v", oldCommitID, newCommitID, repo, err)
 					ctx.JSON(http.StatusInternalServerError, private.Response{

services/pull/check.go

@@ -437,7 +437,7 @@ func checkPullRequestMergeable(id int64) {
 		return
 	}
 
-	if err := testPullRequestBranchMergeable(pr); err != nil {
+	if err := checkPullRequestMergeableAndUpdateStatus(ctx, pr); err != nil {
 		log.Error("testPullRequestTmpRepoBranchMergeable[%-v]: %v", pr, err)
 		pr.Status = issues_model.PullRequestStatusError
 		if err := pr.UpdateCols(ctx, "status"); err != nil {