fix(ermitegit): keep a dangling ref out of ErrNotFound and trim names like git

commit: dedf508f11ea8c4e745bbbbc9fbf650af0fa9312
author: brice · 11d ago · 2026-10-11 00:39 UTC
parent: c38360a
signature: GPG VerifiedCD88F9E564E0102646FB6077A1A07C7C7C3E349D · arch-gpg Verified for the committer email

4 files · +50 -2

M internal/adapters/ermitegit/convert.go +13 -0


@@ -53,6 +53,12 @@ func listRefs(r *git.Repo) ([]domain.Ref, error) {
53	53		}
54	54		case errors.Is(err, git.ErrWrongType):
55	55		// A tag of a tree or a blob has no commit.
	56	+	case errors.Is(err, git.ErrNotFound):
	57	+	// The ref exists and its object does not: the repository is
	58	+	// damaged, which is not a missing repository or revision, so
	59	+	// the error must not match domain.ErrNotFound (gitcli fails
	60	+	// the same way). %v hides git.ErrNotFound from mapErr.
	61	+	return nil, fmt.Errorf("ref %s points to a missing object: %v", ref.Name, err)
56	62		default:
57	63		return nil, fmt.Errorf("follow %s: %w", ref.Name, err)
58	64		}

@@ -388,6 +394,9 @@ func convertCommit(c *git.Commit) (domain.Commit, error) {
388	394		}
389	395		end := [5]int{}
390	396		for i, f := range [...][]byte{c.Author.Name, c.Author.Email, c.Committer.Name, c.Committer.Email, c.Message} {
	397	+	if i == 0 \|\| i == 2 {
	398	+	f = trimNameRight(f)
	399	+	}
391	400		buf = append(buf, f...)
392	401		end[i] = len(buf)
393	402		}

@@ -419,6 +428,10 @@ func convertCommit(c *git.Commit) (domain.Commit, error) {
419	428		return out, nil
420	429		}
421	430
	431	+	// trimNameRight drops the trailing whitespace of a name the way git does (the
	432	+	// library trims spaces only).
	433	+	func trimNameRight(b []byte) []byte { return bytes.TrimRight(b, " \t\r\n") }
	434	+
422	435		func isUTF8Label(b []byte) bool {
423	436		return bytes.EqualFold(b, []byte("utf-8")) \|\| bytes.EqualFold(b, []byte("utf8"))
424	437		}


@@ -53,6 +53,12 @@ func listRefs(r *git.Repo) ([]domain.Ref, error) {
53	}	53		}
54	case errors.Is(err, git.ErrWrongType):	54		case errors.Is(err, git.ErrWrongType):
55	// A tag of a tree or a blob has no commit.	55		// A tag of a tree or a blob has no commit.
		56	+	case errors.Is(err, git.ErrNotFound):
		57	+	// The ref exists and its object does not: the repository is
		58	+	// damaged, which is not a missing repository or revision, so
		59	+	// the error must not match domain.ErrNotFound (gitcli fails
		60	+	// the same way). %v hides git.ErrNotFound from mapErr.
		61	+	return nil, fmt.Errorf("ref %s points to a missing object: %v", ref.Name, err)
56	default:	62		default:
57	return nil, fmt.Errorf("follow %s: %w", ref.Name, err)	63		return nil, fmt.Errorf("follow %s: %w", ref.Name, err)
58	}	64		}

@@ -388,6 +394,9 @@ func convertCommit(c *git.Commit) (domain.Commit, error) {
388	}	394		}
389	end := [5]int{}	395		end := [5]int{}
390	for i, f := range [...][]byte{c.Author.Name, c.Author.Email, c.Committer.Name, c.Committer.Email, c.Message} {	396		for i, f := range [...][]byte{c.Author.Name, c.Author.Email, c.Committer.Name, c.Committer.Email, c.Message} {
		397	+	if i == 0 \|\| i == 2 {
		398	+	f = trimNameRight(f)
		399	+	}
391	buf = append(buf, f...)	400		buf = append(buf, f...)
392	end[i] = len(buf)	401		end[i] = len(buf)
393	}	402		}

@@ -419,6 +428,10 @@ func convertCommit(c *git.Commit) (domain.Commit, error) {
419	return out, nil	428		return out, nil
420	}	429		}
421		430
		431	+	// trimNameRight drops the trailing whitespace of a name the way git does (the
		432	+	// library trims spaces only).
		433	+	func trimNameRight(b []byte) []byte { return bytes.TrimRight(b, " \t\r\n") }
		434	+
422	func isUTF8Label(b []byte) bool {	435		func isUTF8Label(b []byte) bool {
423	return bytes.EqualFold(b, []byte("utf-8")) \|\| bytes.EqualFold(b, []byte("utf8"))	436		return bytes.EqualFold(b, []byte("utf-8")) \|\| bytes.EqualFold(b, []byte("utf8"))
424	}	437		}

M internal/readertest/compare.go +2 -2


@@ -55,7 +55,7 @@ func errClass(err error) string {
55	55		case errors.Is(err, context.Canceled):
56	56		return "canceled"
57	57		}
58		-	return "other: " + err.Error()
	58	+	return "other" // the messages of two readers differ
59	59		}
60	60
61	61		// Queries are the questions put to both readers about one repository.

@@ -116,7 +116,7 @@ func Derive(t testing.TB, dir string, b Budget) Queries {
116	116		var commits []string
117	117		{
118	118		// An empty repository has no commit: rev-list --all prints nothing.
119		-	cmd := exec.Command("git", "--git-dir="+dir, "rev-list", "--all")
	119	+	cmd := exec.Command("git", "--git-dir="+dir, "rev-list", "--all", "--missing=allow-any")
120	120		cmd.Env = []string{"PATH=" + os.Getenv("PATH"), "HOME=" + dir, "GIT_CONFIG_NOSYSTEM=1", "GIT_CONFIG_GLOBAL=/dev/null"}
121	121		out, err := cmd.Output()
122	122		require.NoError(t, err)


@@ -55,7 +55,7 @@ func errClass(err error) string {
55		case errors.Is(err, context.Canceled):	55		case errors.Is(err, context.Canceled):
56		return "canceled"	56		return "canceled"
57		}	57		}
58	-	return "other: " + err.Error()	58	+	return "other" // the messages of two readers differ
59		}	59		}
60			60
61		// Queries are the questions put to both readers about one repository.	61		// Queries are the questions put to both readers about one repository.

@@ -116,7 +116,7 @@ func Derive(t testing.TB, dir string, b Budget) Queries {
116		var commits []string	116		var commits []string
117		{	117		{
118		// An empty repository has no commit: rev-list --all prints nothing.	118		// An empty repository has no commit: rev-list --all prints nothing.
119	-	cmd := exec.Command("git", "--git-dir="+dir, "rev-list", "--all")	119	+	cmd := exec.Command("git", "--git-dir="+dir, "rev-list", "--all", "--missing=allow-any")
120		cmd.Env = []string{"PATH=" + os.Getenv("PATH"), "HOME=" + dir, "GIT_CONFIG_NOSYSTEM=1", "GIT_CONFIG_GLOBAL=/dev/null"}	120		cmd.Env = []string{"PATH=" + os.Getenv("PATH"), "HOME=" + dir, "GIT_CONFIG_NOSYSTEM=1", "GIT_CONFIG_GLOBAL=/dev/null"}
121		out, err := cmd.Output()	121		out, err := cmd.Output()
122		require.NoError(t, err)	122		require.NoError(t, err)

M internal/readertest/fixture.go +12 -0


@@ -70,6 +70,18 @@ func (f *Fixture) Git(args ...string) string {
70	70		return strings.TrimSpace(string(out))
71	71		}
72	72
	73	+	// WriteObject stores a raw object of the given type (commit, tag...) built by
	74	+	// hand, for what git itself refuses to write, and returns its id.
	75	+	func (f *Fixture) WriteObject(typ, content string) string {
	76	+	f.T.Helper()
	77	+	cmd := exec.Command("git", "hash-object", "-w", "-t", typ, "--stdin")
	78	+	cmd.Env = []string{"PATH=" + os.Getenv("PATH"), "HOME=" + f.Work, "GIT_CONFIG_NOSYSTEM=1", "GIT_CONFIG_GLOBAL=/dev/null", "GIT_DIR=" + f.Bare}
	79	+	cmd.Stdin = strings.NewReader(content)
	80	+	out, err := cmd.CombinedOutput()
	81	+	require.NoError(f.T, err, "git hash-object: %s", out)
	82	+	return strings.TrimSpace(string(out))
	83	+	}
	84	+
73	85		// Path returns the path of a file of the work tree.
74	86		func (f *Fixture) Path(p string) string { return filepath.Join(f.Work, filepath.FromSlash(p)) }
75	87


@@ -70,6 +70,18 @@ func (f *Fixture) Git(args ...string) string {
70	return strings.TrimSpace(string(out))	70		return strings.TrimSpace(string(out))
71	}	71		}
72		72
		73	+	// WriteObject stores a raw object of the given type (commit, tag...) built by
		74	+	// hand, for what git itself refuses to write, and returns its id.
		75	+	func (f *Fixture) WriteObject(typ, content string) string {
		76	+	f.T.Helper()
		77	+	cmd := exec.Command("git", "hash-object", "-w", "-t", typ, "--stdin")
		78	+	cmd.Env = []string{"PATH=" + os.Getenv("PATH"), "HOME=" + f.Work, "GIT_CONFIG_NOSYSTEM=1", "GIT_CONFIG_GLOBAL=/dev/null", "GIT_DIR=" + f.Bare}
		79	+	cmd.Stdin = strings.NewReader(content)
		80	+	out, err := cmd.CombinedOutput()
		81	+	require.NoError(f.T, err, "git hash-object: %s", out)
		82	+	return strings.TrimSpace(string(out))
		83	+	}
		84	+
73	// Path returns the path of a file of the work tree.	85		// Path returns the path of a file of the work tree.
74	func (f *Fixture) Path(p string) string { return filepath.Join(f.Work, filepath.FromSlash(p)) }	86		func (f *Fixture) Path(p string) string { return filepath.Join(f.Work, filepath.FromSlash(p)) }
75		87

M internal/readertest/fixtures.go +23 -0


@@ -4,10 +4,14 @@ package readertest
4	4
5	5		import (
6	6		"fmt"
	7	+	"os"
	8	+	"path/filepath"
7	9		"strings"
8	10		"testing"
9	11		"time"
10	12
	13	+	"github.com/stretchr/testify/require"
	14	+
11	15		"bnema.dev/ermite/internal/domain"
12	16		)
13	17

@@ -57,6 +61,7 @@ func Cases() []Case {
57	61		{"large blobs", buildLarge},
58	62		{"names and messages", buildNames},
59	63		{"refs", buildRefs},
	64	+	{"dangling ref", buildDangling},
60	65		{"submodule", buildSubmodule},
61	66		{"wide directory", buildWide},
62	67		{"detached HEAD", func(t testing.T, data string) Fixture {

@@ -194,6 +199,11 @@ func buildNames(t testing.T, data string) Fixture {
194	199		f.Git("-c", "user.name=Zoë <Ünï>", "commit", "--quiet", "--author=Zoë Ünï <zoe@example.org>", "-m", strings.Repeat("long ", 400))
195	200		f.Write("héllo wörld.txt", "x\n").Write("tab\tname.txt", "t\n").Write("-dash", "d\n")
196	201		f.Commit("names")
	202	+	// Git trims the trailing whitespace of a name when it writes a commit but
	203	+	// shows one that has some as it is read: the readers must agree on it.
	204	+	head, tree := f.Git("rev-parse", "HEAD"), f.Git("rev-parse", "HEAD^{tree}")
	205	+	f.Git("update-ref", "refs/heads/main", f.WriteObject("commit", "tree "+tree+"\nparent "+head+"\n"+
	206	+	"author Ada\t <a@b> 1767268860 +0000\ncommitter Grace \t\r <g@b> 1767268860 +0000\n\nraw names\n"))
197	207		f.Maintain()
198	208		return f
199	209		}

@@ -231,6 +241,19 @@ func buildRefs(t testing.T, data string) Fixture {
231	241		return f
232	242		}
233	243
	244	+	// buildDangling has a loose branch and a loose tag whose objects do not exist,
	245	+	// next to sound refs. git lists them, and fails when it must follow them.
	246	+	func buildDangling(t testing.T, data string) Fixture {
	247	+	f := NewFixture(t, data, "dangling")
	248	+	f.Write("a.txt", "1\n")
	249	+	f.Commit("one")
	250	+	ghost := strings.Repeat("0123456789", 4)
	251	+	for _, ref := range []string{"refs/heads/ghost", "refs/tags/ghost"} {
	252	+	require.NoError(t, os.WriteFile(filepath.Join(f.Bare, filepath.FromSlash(ref)), []byte(ghost+"\n"), 0o644))
	253	+	}
	254	+	return f
	255	+	}
	256	+
234	257		// buildSubmodule has a gitlink entry, a symlink and an executable.
235	258		func buildSubmodule(t testing.T, data string) Fixture {
236	259		f := NewFixture(t, data, "submod")


@@ -4,10 +4,14 @@ package readertest
4		4
5	import (	5		import (
6	"fmt"	6		"fmt"
		7	+	"os"
		8	+	"path/filepath"
7	"strings"	9		"strings"
8	"testing"	10		"testing"
9	"time"	11		"time"
10		12
		13	+	"github.com/stretchr/testify/require"
		14	+
11	"bnema.dev/ermite/internal/domain"	15		"bnema.dev/ermite/internal/domain"
12	)	16		)
13		17

@@ -57,6 +61,7 @@ func Cases() []Case {
57	{"large blobs", buildLarge},	61		{"large blobs", buildLarge},
58	{"names and messages", buildNames},	62		{"names and messages", buildNames},
59	{"refs", buildRefs},	63		{"refs", buildRefs},
		64	+	{"dangling ref", buildDangling},
60	{"submodule", buildSubmodule},	65		{"submodule", buildSubmodule},
61	{"wide directory", buildWide},	66		{"wide directory", buildWide},
62	{"detached HEAD", func(t testing.T, data string) Fixture {	67		{"detached HEAD", func(t testing.T, data string) Fixture {

@@ -194,6 +199,11 @@ func buildNames(t testing.T, data string) Fixture {
194	f.Git("-c", "user.name=Zoë <Ünï>", "commit", "--quiet", "--author=Zoë Ünï <zoe@example.org>", "-m", strings.Repeat("long ", 400))	199		f.Git("-c", "user.name=Zoë <Ünï>", "commit", "--quiet", "--author=Zoë Ünï <zoe@example.org>", "-m", strings.Repeat("long ", 400))
195	f.Write("héllo wörld.txt", "x\n").Write("tab\tname.txt", "t\n").Write("-dash", "d\n")	200		f.Write("héllo wörld.txt", "x\n").Write("tab\tname.txt", "t\n").Write("-dash", "d\n")
196	f.Commit("names")	201		f.Commit("names")
		202	+	// Git trims the trailing whitespace of a name when it writes a commit but
		203	+	// shows one that has some as it is read: the readers must agree on it.
		204	+	head, tree := f.Git("rev-parse", "HEAD"), f.Git("rev-parse", "HEAD^{tree}")
		205	+	f.Git("update-ref", "refs/heads/main", f.WriteObject("commit", "tree "+tree+"\nparent "+head+"\n"+
		206	+	"author Ada\t <a@b> 1767268860 +0000\ncommitter Grace \t\r <g@b> 1767268860 +0000\n\nraw names\n"))
197	f.Maintain()	207		f.Maintain()
198	return f	208		return f
199	}	209		}

@@ -231,6 +241,19 @@ func buildRefs(t testing.T, data string) Fixture {
231	return f	241		return f
232	}	242		}
233		243
		244	+	// buildDangling has a loose branch and a loose tag whose objects do not exist,
		245	+	// next to sound refs. git lists them, and fails when it must follow them.
		246	+	func buildDangling(t testing.T, data string) Fixture {
		247	+	f := NewFixture(t, data, "dangling")
		248	+	f.Write("a.txt", "1\n")
		249	+	f.Commit("one")
		250	+	ghost := strings.Repeat("0123456789", 4)
		251	+	for _, ref := range []string{"refs/heads/ghost", "refs/tags/ghost"} {
		252	+	require.NoError(t, os.WriteFile(filepath.Join(f.Bare, filepath.FromSlash(ref)), []byte(ghost+"\n"), 0o644))
		253	+	}
		254	+	return f
		255	+	}
		256	+
234	// buildSubmodule has a gitlink entry, a symlink and an executable.	257		// buildSubmodule has a gitlink entry, a symlink and an executable.
235	func buildSubmodule(t testing.T, data string) Fixture {	258		func buildSubmodule(t testing.T, data string) Fixture {
236	f := NewFixture(t, data, "submod")	259		f := NewFixture(t, data, "submod")