Merge pull request #93 from infosiftr/gitfs-symlinks

Fix gitfs symlink handling

pkg/gitfs/fs.go

@@ -34,13 +34,15 @@ func CommitHash(repo *goGit.Repository, commit string) (fs.FS, error) {
 	if err != nil {
 		return nil, err
 	}
-	return gitFSFS{
-		gitFS: &gitFS{
+	f := &gitFS{
 		storer: repo.Storer,
 		tree:   tree,
 		name:   ".",
 		Mod:    CommitTime(gitCommit),
-		},
+	}
+	f.root = f
+	return gitFSFS{
+		gitFS: f,
 	}, nil
 }
 
@@ -54,6 +56,8 @@ type gitFSFS struct {
 // https://pkg.go.dev/io/fs#FileInfo
 // https://pkg.go.dev/io/fs#DirEntry
 type gitFS struct {
+	root *gitFS // used so we can rewind back to the root if we need to (see symlink handling code; should *only* be set in CommitHash / constructors)
+
 	storer goGitPlumbingStorer.EncodedObjectStorer
 	tree   *goGitPlumbingObject.Tree
 	entry  *goGitPlumbingObject.TreeEntry // might be nil ("." at the top-level of the repo)
@@ -172,7 +176,10 @@ func (f gitFS) statEntry(name string, entry *goGitPlumbingObject.TreeEntry, foll
 		if target, err := fi.resolveLink(); err != nil {
 			return nil, err
 		} else if target != "" {
-			return f.stat(target, followSymlinks)
+			// the value from resolveLink is relative to the root
+			return f.root.stat(target, followSymlinks)
+			// ideally this would "just" use "path.Rel" to make "target" relative to "f.name" instead, but "path.Rel" does not exist and only "filepath.Rel" does which would break this code on Windows, so instead we added a "root" pointer that we pass around forever that links us back to the root of our "Tree"
+			// we could technically solve this by judicious use of "../" (with enough "../" to catch all the "/" in "f.name"), but it seems simpler and more obvious (and less error prone) to just pass around a pointer to the root
 		}
 	}
 

pkg/gitfs/fs_test.go

@@ -8,6 +8,7 @@ import (
 	"github.com/docker-library/bashbrew/pkg/gitfs"
 
 	"github.com/go-git/go-git/v5"
+	goGitConfig "github.com/go-git/go-git/v5/config"
 	"github.com/go-git/go-git/v5/storage/memory"
 )
 
@@ -52,7 +53,7 @@ func TestCommitFS(t *testing.T) {
 	})
 }
 
-func TestSymlinkFS(t *testing.T) {
+func TestRootSymlinkFS(t *testing.T) {
 	// TODO instead of cloning a remote repository, synthesize a very simple Git repository right in the test here (benefit of the remote repository is that it's much larger, so fstest.TestFS has a lot more data to test against)
 	repo, err := git.Clone(memory.NewStorage(), nil, &git.CloneOptions{
 		URL:          "https://github.com/tianon/gosu.git", // just a repository with a known symlink (`.dockerignore` -> `.gitignore`)
@@ -93,3 +94,80 @@ func TestSymlinkFS(t *testing.T) {
 		}
 	})
 }
+
+func TestSubdirSymlinkFS(t *testing.T) {
+	// TODO instead of cloning a remote repository, synthesize a very simple Git repository right in the test here (benefit of the remote repository is that it's much larger, so fstest.TestFS has a lot more data to test against)
+	// Init + CreateRemoteAnonymous + Fetch because Clone doesn't support fetch-by-commit
+	repo, err := git.Init(memory.NewStorage(), nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	remote, err := repo.CreateRemoteAnonymous(&goGitConfig.RemoteConfig{
+		Name: "anonymous",
+		URLs: []string{"https://github.com/docker-library/busybox.git"}, // just a repository with a known symlink at a non-root level (`latest/musl/amd64/blobs/sha256/6e5e0f90c009d12db9478afe5656920e7bdd548e9fd8f50eab2be694102ae318` -> `../../image-config.json`)
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	commit := "668d52e6f0596e0fd0b1be1d8267c4b9240dc2b3"
+	err = remote.Fetch(&git.FetchOptions{
+		RefSpecs: []goGitConfig.RefSpec{goGitConfig.RefSpec(commit + ":FETCH_HEAD")},
+		Tags:     git.NoTags,
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	f, err := gitfs.CommitHash(repo, commit)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	t.Run("Open+ReadAll", func(t *testing.T) {
+		r, err := f.Open("latest/musl/amd64/blobs/sha256/6e5e0f90c009d12db9478afe5656920e7bdd548e9fd8f50eab2be694102ae318")
+		if err != nil {
+			t.Fatal(err)
+		}
+		defer func() {
+			if err := r.Close(); err != nil {
+				t.Fatal(err)
+			}
+		}()
+		b, err := io.ReadAll(r)
+		if err != nil {
+			t.Fatal(err)
+		}
+		expected := `{
+	"config": {
+		"Cmd": [
+			"sh"
+		]
+	},
+	"created": "2023-05-18T22:34:17Z",
+	"history": [
+		{
+			"created": "2023-05-18T22:34:17Z",
+			"created_by": "BusyBox 1.36.1 (musl), Alpine 3.19.1"
+		}
+	],
+	"rootfs": {
+		"type": "layers",
+		"diff_ids": [
+			"sha256:994bf8f4adc78c5c1e4a6b5e3b59ad57902b301e0e79255a3e95ea4b213a76bd"
+		]
+	},
+	"architecture": "amd64",
+	"os": "linux"
+}
+`
+		if string(b) != expected {
+			t.Fatalf("expected %q, got %q", expected, string(b))
+		}
+	})
+
+	// might as well run fstest again, now that we have a new filesystem tree 😅
+	t.Run("fstest.TestFS", func(t *testing.T) {
+		if err := fstest.TestFS(f, "latest/musl/amd64/blobs/sha256/6e5e0f90c009d12db9478afe5656920e7bdd548e9fd8f50eab2be694102ae318", "latest/musl/amd64/index.json"); err != nil {
+			t.Fatal(err)
+		}
+	})
+}