Add read-only git provider

Implement submodule.Provider with the git CLI so submarine reads real
repositories: `submarine [dir]`, with `--mock` for the built-in data.

- Tree reads .gitmodules recursively; Status derives presence, checkout,
  worktree, staged pin, pushed state and pin -> ref resolution from local
  git commands only (no network).
- Concurrent Status calls are bounded by runtime.NumCPU().
- scripts/make-fixture.sh builds a superproject covering every state; the
  integration tests run against it.
- theme.Ago shows "never" for a zero time.
- Docs updated; decisions D14-D16 added.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

AGENTS.md

@@ -1,7 +1,7 @@
 # AGENTS.md
 
 submarine: a Go + Bubble Tea TUI for repositories with many git submodules.
-Currently a prototype on mock data.
+Currently a prototype with a read-only git provider and mock data.
 
 ## Read first
 
@@ -18,7 +18,9 @@ and decisions. Read the relevant document before changing behavior:
 ## Commands
 
 ```sh
-go run ./cmd/submarine        # run (mock data)
+go run ./cmd/submarine [dir]  # run on a repository (default: cwd)
+go run ./cmd/submarine --mock # run on mock data
+scripts/make-fixture.sh DIR   # build a test superproject in DIR
 go vet ./...
 go test ./...
 gofmt -l .                    # must print nothing
@@ -32,7 +34,8 @@ gofmt -l .                    # must print nothing
 - **Layering.** `internal/submodule` has no UI imports and decides what a
   state means (`Status.Issues()`). `internal/tui/theme` decides how it looks.
   Components never classify states or define colors themselves.
-- **Data access** only through `submodule.Provider`.
+- **Data access** only through `submodule.Provider`. The git provider is
+  read-only and must never use the network.
 - **Colors** only from `theme.Theme`, ANSI 16 colors for meaning (see
   `docs/ux-guide.md#colors`). Render segments on a background with a `base`
   style (`LevelStyle(level, base)`, `Delta(pin, base)`).

cmd/submarine/main.go

@@ -12,20 +12,37 @@ import (
 	zone "github.com/lrstanley/bubblezone"
 
 	"submarine/internal/submodule"
+	"submarine/internal/submodule/git"
 	"submarine/internal/submodule/mock"
 	"submarine/internal/tui"
 )
 
 func main() {
-	useMock := flag.Bool("mock", true, "use built-in mock data instead of a real repository")
+	useMock := flag.Bool("mock", false, "use built-in mock data instead of a real repository")
+	flag.Usage = func() {
+		fmt.Fprintln(flag.CommandLine.Output(), "usage: submarine [--mock] [dir]")
+		flag.PrintDefaults()
+	}
 	flag.Parse()
+	if flag.NArg() > 1 {
+		flag.Usage()
+		os.Exit(2)
+	}
 
 	var provider submodule.Provider
 	if *useMock {
 		provider = mock.New()
 	} else {
-		fmt.Fprintln(os.Stderr, "only --mock is supported in the prototype")
+		dir := "."
-		os.Exit(2)
+		if flag.NArg() == 1 {
+			dir = flag.Arg(0)
+		}
+		p, err := git.New(dir)
+		if err != nil {
+			fmt.Fprintln(os.Stderr, "submarine:", err)
+			os.Exit(2)
+		}
+		provider = p
 	}
 
 	ctx, cancel := context.WithCancel(context.Background())

docs/README.md

@@ -17,13 +17,14 @@ bare commit hash.
 
 ## Status
 
-**Prototype.** The UI is complete for browsing and runs on mock data
+**Prototype.** The UI is complete for browsing. It reads real repositories
-(`internal/submodule/mock`). There is no git backend yet, and actions
+through a read-only git provider (`internal/submodule/git`):
-(`u`, `b`, `f`, `/`, `!`) only show a "not implemented" message.
+`submarine [dir]`. `submarine --mock` shows the built-in mock data
+(`internal/submodule/mock`). Actions (`u`, `b`, `f`, `/`, `!`) only show a
+"not implemented" message.
 
 Next steps, roughly in order:
 
-1. Git provider implementing `submodule.Provider` (read-only).
+1. Actions: fetch, checkout pin (`submodule update`), bump pin to tip, init.
-2. Actions: fetch, checkout pin (`submodule update`), bump pin to tip, init.
+2. Filter (`/`) and "problems only" (`!`) views.
-3. Filter (`/`) and "problems only" (`!`) views.
+3. Refresh after actions and on demand.
-4. Refresh after actions and on demand.

docs/architecture.md

@@ -8,11 +8,15 @@ Go, [Bubble Tea](https://github.com/charmbracelet/bubbletea) (v1),
 ## Package layout
 
 ```
-cmd/submarine/main.go         wiring: choose provider (--mock), start the program
+cmd/submarine/main.go         wiring: choose provider (git or --mock), start the program
 internal/submodule/           domain model; NO UI imports
   model.go                    Repo, Submodule, Status, PinInfo, Commit, enums
   status.go                   Issues(), HiddenIssues(), Walk()
   provider.go                 Provider interface
+internal/submodule/git/       read-only provider using the git CLI
+  exec.go                     git runner (environment, errors)
+  parse.go                    pure parsers for git output
+  provider.go                 Tree, Status, pin → ref resolution
 internal/submodule/mock/      fixture provider with artificial latency
 internal/tui/                 Bubble Tea application
   app.go                      root model: layout, message routing, loading
@@ -52,6 +56,23 @@ type Provider interface {
 The mock provider (`internal/submodule/mock`) covers every state and sleeps
 100–800 ms per `Status` call to exercise the loading path.
 
+### Git provider
+
+`internal/submodule/git` shells out to `git -C <dir> …` (see D14).
+
+- Environment: `GIT_TERMINAL_PROMPT=0`, `GIT_OPTIONAL_LOCKS=0` (reads never
+  take `index.lock`), `LC_ALL=C`. No command touches the network.
+- `Tree` reads each repository's `.gitmodules` (`git config -f
+  .gitmodules`) and recurses into initialized submodules. It records, per
+  path, the parent repository and the `.gitmodules` branch for `Status`.
+  `Name` is the last element of `RelPath`.
+- `Status` runs about ten small git commands per submodule (see
+  [state-model.md](state-model.md#git-commands)). At most `runtime.NumCPU()`
+  `Status` calls run at once; the others wait on a semaphore and still
+  respect `ctx` (D16).
+- Superproject: branch from `symbolic-ref`, ahead/behind against
+  `@{upstream}`, "fetched" from the mtime of `FETCH_HEAD`.
+
 ## Loading flow
 
 1. `Init` runs `Tree()` → `treeLoadedMsg`.
@@ -91,14 +112,18 @@ go test ./...
 ```
 
 - `internal/submodule`: issue classification and hidden-issue aggregation.
+- `internal/submodule/git`: parser unit tests, and integration tests
+  against a fixture repository built by `scripts/make-fixture.sh` (skipped
+  if `git` is not installed).
 - `internal/tui/tree`: navigation rules. Tests need `zone.NewGlobal()` (see
   `TestMain`).
 
 Manual check in a headless terminal:
 
 ```sh
+scripts/make-fixture.sh /tmp/subfix   # superproject with every state
 go build -o /tmp/submarine ./cmd/submarine
-tmux new-session -d -s sub -x 130 -y 24 /tmp/submarine
+tmux new-session -d -s sub -x 130 -y 24 "/tmp/submarine /tmp/subfix/super"
 tmux send-keys -t sub Down Right
 tmux capture-pane -pt sub        # add -e to see colors
 tmux kill-session -t sub

docs/decisions.md

@@ -99,3 +99,32 @@ explicit action (`f`).
 
 **Decision.** Keep mouse support (click to select/toggle, wheel to move)
 using bubblezone markers per row.
+
+## D14 — Git provider shells out to the git CLI
+
+**Context.** The provider can use a Go library (go-git) or the `git` binary.
+**Decision.** Run `git -C <dir> …` with a fixed environment
+(`GIT_TERMINAL_PROMPT=0`, `GIT_OPTIONAL_LOCKS=0`, `LC_ALL=C`) and parse
+stable output formats (`--porcelain=v2`, custom `--format`).
+**Consequences.** Behaves exactly like the user's git (config, `gitdir:`
+files, worktrees, partial clones), no new dependency, fast on large repos.
+Requires `git` on `PATH`; one process per query.
+
+## D15 — Git provider semantics
+
+**Decision.** The pin is the **index** gitlink, because that is what the
+next commit records and what `submodule update` checks out; a difference to
+HEAD is `PinStaged`. Worktree state uses `--ignore-submodules=dirty`, so dirt
+in a nested submodule doesn't mark its parent (consistent with D7). A pin
+whose commit isn't present locally is a loading `error`, not `unpushed`,
+because without the commit nothing can be said about it.
+`--mock` is no longer the default; `submarine [dir]` reads a real repository.
+
+## D16 — Bounded concurrency inside the provider
+
+**Context.** The app starts one `Status` command per submodule at once
+(D11); with hundreds of submodules that would spawn thousands of git
+processes.
+**Decision.** The git provider limits concurrent `Status` calls to
+`runtime.NumCPU()` with a semaphore that respects `ctx`.
+**Consequences.** The UI stays unchanged; rows fill in steadily.

docs/requirements.md

@@ -88,6 +88,13 @@ commit hashes. It doesn't answer the questions users actually have:
 | `/` | Filter by name/path |
 | `!` | Show only submodules with issues |
 
+## Usage
+
+```sh
+submarine [dir]     # the repository containing dir (default: current directory)
+submarine --mock    # built-in mock data
+```
+
 ## Non-functional requirements
 
 - **N1** Startup must not block on the network. The tool works from local
@@ -100,5 +107,7 @@ commit hashes. It doesn't answer the questions users actually have:
 - **N5** Works without a Nerd Font. `SUBMARINE_ASCII=1` replaces the few
   remaining Unicode symbols with ASCII.
 - **N6** Works in narrow terminals (detail pane hides below 90 columns).
-- **N7** The UI is independent of the data source: replacing the mock
+- **N7** The UI is independent of the data source: the mock and the git
-  provider with a git provider requires no UI changes.
+  provider are interchangeable without UI changes.
+- **N8** The git provider is read-only and never uses the network; it uses
+  the `git` command line, so the user's git configuration applies.

docs/state-model.md

@@ -9,7 +9,7 @@ them.
 
 | Commit | Source |
 |---|---|
-| **Pinned** | The gitlink recorded in the superproject (HEAD or index) |
+| **Pinned** | The gitlink recorded in the superproject index (differs from HEAD → `PinStaged`) |
 | **Checked out** | The submodule's `HEAD` |
 | **Tip** | The tip of the ref the pin was resolved to, e.g. `origin/main` |
 
@@ -97,6 +97,38 @@ Distances: `git rev-list --left-right --count <pin>...<ref>` gives
 ahead of its ref can still be pushed (e.g. on another branch); only
 `Pushed == false` is dangerous.
 
+## Git commands
+
+How the git provider (`internal/submodule/git`) fills `Status`. All commands
+are local; `<pin>` is the index gitlink.
+
+| Field | Command (in the parent repo / in the submodule) |
+|---|---|
+| `Pinned`, `PinStaged` | parent: `ls-files -s -- <path>` vs `ls-tree HEAD -- <path>` |
+| `Presence` | directory missing → `DirMissing`; no `.git` in it → `NotInitialized` |
+| `Checkout` | `rev-parse HEAD`, `symbolic-ref -q HEAD` (no branch → detached) |
+| `Worktree` | `status --porcelain=v2 --ignore-submodules=dirty` |
+| `Pushed` | `for-each-ref --contains <pin> refs/remotes` is not empty |
+| `Pin.Ref` | rules above; the remote is `origin`, else the first remote |
+| `Ahead`, `Behind` | `rev-list --left-right --count <pin>...<ref>` |
+| `Between` | `log <pin>...<ref>`, at most 50 commits |
+
+Details:
+
+- Rule 1: `branch = .` means the superproject's current branch. The rule
+  applies only if `<remote>/<branch>` exists locally.
+- Rule 2 uses the checked-out branch's upstream (`<branch>@{upstream}`).
+- Rule 4 compares at most 16 containing branches (oldest tip first) and
+  picks the one with the fewest commits after the pin.
+- Rule 6 shows the short SHA; there is no tip and no distance.
+- `--ignore-submodules=dirty`: changes inside a nested submodule do not make
+  its parent `changed` (D7), but a moved nested pin does.
+- A missing submodule only gets a `Pinned` SHA; nested submodules of an
+  uninitialized submodule are unknown and not shown.
+- If the pinned commit doesn't exist in the submodule (not fetched), loading
+  fails with "pinned commit … not fetched" (shown as `error`) instead of
+  guessing `unpushed`.
+
 ## Distance notation
 
 | Display | Meaning |

docs/ux-guide.md

@@ -72,7 +72,9 @@ indent ▸ name [n]   status          ref             delta
 `path  branch ↑↓ · n submodules · <count> <issue>… · n pin staged · n loading · fetched <age>`
 
 Issue counts are ordered by severity (unpushed, conflict, missing, off pin,
-changed), each in its level color. Zero counts are omitted.
+changed), each in its level color. Zero counts are omitted. `<age>` is
+`never` if the superproject has never been fetched; a detached superproject
+shows `(detached)` as its branch.
 
 ## Colors
 

internal/submodule/git/exec.go

@@ -0,0 +1,52 @@
+package git
+
+import (
+	"bytes"
+	"context"
+	"errors"
+	"fmt"
+	"os"
+	"os/exec"
+	"strings"
+)
+
+// gitEnv keeps git non-interactive, lock-free for reads and its output
+// parseable.
+var gitEnv = append(os.Environ(),
+	"GIT_TERMINAL_PROMPT=0",
+	"GIT_OPTIONAL_LOCKS=0",
+	"LC_ALL=C",
+)
+
+// run executes git in dir and returns stdout with the trailing newline
+// removed. The error includes git's stderr.
+func run(ctx context.Context, dir string, args ...string) (string, error) {
+	cmd := exec.CommandContext(ctx, "git", append([]string{"-C", dir}, args...)...)
+	cmd.Env = gitEnv
+	var stdout, stderr bytes.Buffer
+	cmd.Stdout, cmd.Stderr = &stdout, &stderr
+	if err := cmd.Run(); err != nil {
+		if ctx.Err() != nil {
+			return "", ctx.Err()
+		}
+		msg := strings.TrimSpace(stderr.String())
+		if msg == "" {
+			msg = err.Error()
+		}
+		return "", fmt.Errorf("git %s: %s", strings.Join(args, " "), msg)
+	}
+	return strings.TrimRight(stdout.String(), "\n"), nil
+}
+
+// try is run for commands whose failure only means "no answer", e.g.
+// "no upstream" or "ref does not exist". Context errors still propagate.
+func try(ctx context.Context, dir string, args ...string) (string, bool, error) {
+	out, err := run(ctx, dir, args...)
+	if err != nil {
+		if errors.Is(err, context.Canceled) || errors.Is(err, context.DeadlineExceeded) {
+			return "", false, err
+		}
+		return "", false, nil
+	}
+	return out, true, nil
+}

internal/submodule/git/parse.go

@@ -0,0 +1,125 @@
+package git
+
+import (
+	"fmt"
+	"strconv"
+	"strings"
+	"time"
+
+	"submarine/internal/submodule"
+)
+
+// gitmodule is one entry of a .gitmodules file.
+type gitmodule struct {
+	name, path, url, branch string
+}
+
+// parseGitmodules parses the output of
+// `git config -f .gitmodules --get-regexp '^submodule\..*\.(path|url|branch)$'`.
+// Entries keep the order of first appearance; entries without a path are
+// dropped.
+func parseGitmodules(out string) []gitmodule {
+	var order []string
+	byName := map[string]*gitmodule{}
+	for _, line := range strings.Split(out, "\n") {
+		key, value, _ := strings.Cut(line, " ")
+		key, ok := strings.CutPrefix(key, "submodule.")
+		if !ok {
+			continue
+		}
+		dot := strings.LastIndexByte(key, '.')
+		if dot < 0 {
+			continue
+		}
+		name, attr := key[:dot], key[dot+1:]
+		g := byName[name]
+		if g == nil {
+			g = &gitmodule{name: name}
+			byName[name] = g
+			order = append(order, name)
+		}
+		switch attr {
+		case "path":
+			g.path = value
+		case "url":
+			g.url = value
+		case "branch":
+			g.branch = value
+		}
+	}
+	var mods []gitmodule
+	for _, name := range order {
+		if g := byName[name]; g.path != "" {
+			mods = append(mods, *g)
+		}
+	}
+	return mods
+}
+
+// parseWorktree summarizes `git status --porcelain=v2` output. A conflict
+// outranks modifications, which outrank untracked files.
+func parseWorktree(out string) submodule.Worktree {
+	w := submodule.WorktreeClean
+	for _, line := range strings.Split(out, "\n") {
+		switch {
+		case strings.HasPrefix(line, "u "):
+			return submodule.WorktreeConflict
+		case strings.HasPrefix(line, "1 "), strings.HasPrefix(line, "2 "):
+			w = submodule.WorktreeModified
+		case strings.HasPrefix(line, "? ") && w == submodule.WorktreeClean:
+			w = submodule.WorktreeUntracked
+		}
+	}
+	return w
+}
+
+// parseLeftRight parses `git rev-list --left-right --count a...b`.
+func parseLeftRight(out string) (left, right int, err error) {
+	f := strings.Fields(out)
+	if len(f) != 2 {
+		return 0, 0, fmt.Errorf("unexpected rev-list output %q", out)
+	}
+	if left, err = strconv.Atoi(f[0]); err != nil {
+		return 0, 0, err
+	}
+	if right, err = strconv.Atoi(f[1]); err != nil {
+		return 0, 0, err
+	}
+	return left, right, nil
+}
+
+// commitFormat is the `git log --format` matching parseCommits: fields
+// separated by US, records terminated by RS.
+const commitFormat = "--format=%h%x1f%s%x1f%an%x1f%ct%x1e"
+
+// parseCommits parses `git log` output produced with commitFormat.
+func parseCommits(out string) []submodule.Commit {
+	var cs []submodule.Commit
+	for _, rec := range strings.Split(out, "\x1e") {
+		f := strings.Split(strings.TrimLeft(rec, "\n"), "\x1f")
+		if len(f) != 4 {
+			continue
+		}
+		c := submodule.Commit{SHA: f[0], Subject: f[1], Author: f[2]}
+		if sec, err := strconv.ParseInt(f[3], 10, 64); err == nil {
+			c.When = time.Unix(sec, 0)
+		}
+		cs = append(cs, c)
+	}
+	return cs
+}
+
+// parseGitlink returns the SHA of a gitlink (mode 160000) entry in
+// `git ls-files -s` or `git ls-tree` output, or "" if there is none.
+func parseGitlink(out string) string {
+	for _, line := range strings.Split(out, "\n") {
+		f := strings.Fields(line)
+		switch {
+		case len(f) >= 3 && f[0] == "160000" && f[1] == "commit": // ls-tree
+			return f[2]
+		case len(f) >= 2 && f[0] == "160000": // ls-files -s
+			return f[1]
+		}
+	}
+	return ""
+}

internal/submodule/git/parse_test.go

@@ -0,0 +1,85 @@
+package git
+
+import (
+	"slices"
+	"testing"
+
+	"submarine/internal/submodule"
+)
+
+func TestParseGitmodules(t *testing.T) {
+	out := "submodule.src/core.path src/core\n" +
+		"submodule.src/core.url ../core.git\n" +
+		"submodule.lib.v2.path libs/v2\n" +
+		"submodule.lib.v2.branch develop\n" +
+		"submodule.lib.v2.url git@example.com:lib.git\n" +
+		"submodule.nopath.url x\n"
+	want := []gitmodule{
+		{name: "src/core", path: "src/core", url: "../core.git"},
+		{name: "lib.v2", path: "libs/v2", url: "git@example.com:lib.git", branch: "develop"},
+	}
+	if got := parseGitmodules(out); !slices.Equal(got, want) {
+		t.Errorf("parseGitmodules() = %+v, want %+v", got, want)
+	}
+	if got := parseGitmodules(""); got != nil {
+		t.Errorf("parseGitmodules(\"\") = %+v, want nil", got)
+	}
+}
+
+func TestParseWorktree(t *testing.T) {
+	tests := []struct {
+		out  string
+		want submodule.Worktree
+	}{
+		{"", submodule.WorktreeClean},
+		{"? new.txt", submodule.WorktreeUntracked},
+		{"? new.txt\n1 .M N... 100644 100644 100644 a b README", submodule.WorktreeModified},
+		{"2 R. N... 100644 100644 100644 a b R100 new\told", submodule.WorktreeModified},
+		{"1 .M N... 100644 100644 100644 a b README\nu UU N... 1 2 3 4 a b c README", submodule.WorktreeConflict},
+	}
+	for _, tt := range tests {
+		if got := parseWorktree(tt.out); got != tt.want {
+			t.Errorf("parseWorktree(%q) = %v, want %v", tt.out, got, tt.want)
+		}
+	}
+}
+
+func TestParseLeftRight(t *testing.T) {
+	if l, r, err := parseLeftRight("3\t1"); err != nil || l != 3 || r != 1 {
+		t.Errorf("parseLeftRight = %d, %d, %v; want 3, 1, nil", l, r, err)
+	}
+	if _, _, err := parseLeftRight("3"); err == nil {
+		t.Error("parseLeftRight(\"3\") should fail")
+	}
+}
+
+func TestParseCommits(t *testing.T) {
+	out := "abc1234\x1fFix it\x1fAda\x1f1700000000\x1e\n" +
+		"def5678\x1fAdd\x1fBob\x1f1700000100\x1e\n" +
+		"malformed\x1e\n"
+	got := parseCommits(out)
+	want := []string{"abc1234", "def5678"}
+	if len(got) != len(want) {
+		t.Fatalf("parseCommits() = %+v, want SHAs %v", got, want)
+	}
+	for i, c := range got {
+		if c.SHA != want[i] {
+			t.Errorf("commit %d: SHA %q, want %q", i, c.SHA, want[i])
+		}
+	}
+	if got[0].Subject != "Fix it" || got[0].Author != "Ada" || got[0].When.Unix() != 1700000000 {
+		t.Errorf("parseCommits()[0] = %+v", got[0])
+	}
+}
+
+func TestParseGitlink(t *testing.T) {
+	if got := parseGitlink("160000 0123abc 0\tsrc/core"); got != "0123abc" {
+		t.Errorf("ls-files: got %q", got)
+	}
+	if got := parseGitlink("160000 commit 0123abc\tsrc/core"); got != "0123abc" {
+		t.Errorf("ls-tree: got %q", got)
+	}
+	if got := parseGitlink("100644 0123abc 0\tREADME"); got != "" {
+		t.Errorf("blob: got %q", got)
+	}
+}

internal/submodule/git/provider.go

@@ -0,0 +1,375 @@
+// Package git provides a read-only submodule.Provider backed by the git
+// command line. It never touches the network.
+package git
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"io/fs"
+	"os"
+	"path"
+	"path/filepath"
+	"runtime"
+	"strings"
+	"sync"
+
+	"submarine/internal/submodule"
+)
+
+// maxContainsCandidates limits how many remote branches containing the pin
+// are compared when resolving the pin by RefFromContains.
+const maxContainsCandidates = 16
+
+// maxBetween limits the commits listed between pin and tip.
+const maxBetween = 50
+
+// Provider reads a superproject and its submodules with git.
+type Provider struct {
+	root string        // absolute superproject root
+	sem  chan struct{} // bounds concurrent Status calls
+
+	mu   sync.RWMutex
+	mods map[string]info // by Submodule.Path; filled by Tree
+}
+
+// info is what Status needs about a submodule beyond its path.
+type info struct {
+	superDir string // absolute path of the parent repository
+	dir      string // absolute path of the submodule
+	relPath  string // relative to superDir, slash-separated
+	branch   string // branch from .gitmodules, may be "" or "."
+}
+
+// New returns a provider for the repository containing dir.
+func New(dir string) (*Provider, error) {
+	root, err := run(context.Background(), dir, "rev-parse", "--show-toplevel")
+	if err != nil {
+		return nil, fmt.Errorf("%s is not a git repository", dir)
+	}
+	return &Provider{
+		root: root,
+		sem:  make(chan struct{}, runtime.NumCPU()),
+	}, nil
+}
+
+func (p *Provider) Tree(ctx context.Context) (*submodule.Repo, error) {
+	repo := &submodule.Repo{Path: abbrevHome(p.root)}
+	if b, ok, err := try(ctx, p.root, "symbolic-ref", "--short", "-q", "HEAD"); err != nil {
+		return nil, err
+	} else if ok {
+		repo.Branch = b
+	} else {
+		repo.Branch = "(detached)"
+	}
+	if out, ok, err := try(ctx, p.root, "rev-list", "--left-right", "--count", "HEAD...@{upstream}"); err != nil {
+		return nil, err
+	} else if ok {
+		repo.Ahead, repo.Behind, _ = parseLeftRight(out)
+	}
+	if dir, ok, err := try(ctx, p.root, "rev-parse", "--path-format=absolute", "--git-common-dir"); err != nil {
+		return nil, err
+	} else if ok {
+		if fi, err := os.Stat(filepath.Join(dir, "FETCH_HEAD")); err == nil {
+			repo.FetchedAt = fi.ModTime()
+		}
+	}
+
+	mods := map[string]info{}
+	subs, err := p.build(ctx, p.root, nil, mods)
+	if err != nil {
+		return nil, err
+	}
+	repo.Submodules = subs
+
+	p.mu.Lock()
+	p.mods = mods
+	p.mu.Unlock()
+	return repo, nil
+}
+
+// build reads the .gitmodules of the repository at dir and recurses into
+// initialized submodules.
+func (p *Provider) build(ctx context.Context, dir string, parent *submodule.Submodule, mods map[string]info) ([]*submodule.Submodule, error) {
+	if _, err := os.Stat(filepath.Join(dir, ".gitmodules")); err != nil {
+		return nil, nil
+	}
+	out, _, err := try(ctx, dir, "config", "-f", ".gitmodules", "--get-regexp", `^submodule\..*\.(path|url|branch)$`)
+	if err != nil {
+		return nil, err
+	}
+	var subs []*submodule.Submodule
+	for _, g := range parseGitmodules(out) {
+		rel := path.Clean(g.path)
+		full := rel
+		if parent != nil {
+			full = path.Join(parent.Path, rel)
+		}
+		s := &submodule.Submodule{
+			Name:    path.Base(rel),
+			Path:    full,
+			RelPath: rel,
+			URL:     g.url,
+			Parent:  parent,
+		}
+		subDir := filepath.Join(dir, filepath.FromSlash(rel))
+		mods[full] = info{superDir: dir, dir: subDir, relPath: rel, branch: g.branch}
+		if initialized(subDir) {
+			if s.Children, err = p.build(ctx, subDir, s, mods); err != nil {
+				return nil, err
+			}
+		}
+		subs = append(subs, s)
+	}
+	return subs, nil
+}
+
+func (p *Provider) Status(ctx context.Context, path string) (*submodule.Status, error) {
+	select {
+	case p.sem <- struct{}{}:
+		defer func() { <-p.sem }()
+	case <-ctx.Done():
+		return nil, ctx.Err()
+	}
+	p.mu.RLock()
+	in, ok := p.mods[path]
+	p.mu.RUnlock()
+	if !ok {
+		return nil, fmt.Errorf("unknown submodule %q", path)
+	}
+	return status(ctx, in)
+}
+
+func status(ctx context.Context, in info) (*submodule.Status, error) {
+	// pin: index gitlink, compared with HEAD
+	out, err := run(ctx, in.superDir, "ls-files", "-s", "--", in.relPath)
+	if err != nil {
+		return nil, err
+	}
+	pin := parseGitlink(out)
+	if pin == "" {
+		return nil, errors.New("no gitlink in the superproject index")
+	}
+	out, _, err = try(ctx, in.superDir, "ls-tree", "HEAD", "--", in.relPath)
+	if err != nil {
+		return nil, err
+	}
+	st := &submodule.Status{PinStaged: parseGitlink(out) != pin}
+
+	// presence
+	if _, err := os.Stat(in.dir); errors.Is(err, fs.ErrNotExist) {
+		st.Presence = submodule.DirMissing
+	} else if !initialized(in.dir) {
+		st.Presence = submodule.NotInitialized
+	}
+	if st.Presence != submodule.PresenceOK {
+		st.Pinned = &submodule.Commit{SHA: pin[:min(7, len(pin))]}
+		return st, nil
+	}
+
+	// checkout and worktree
+	head, err := run(ctx, in.dir, "rev-parse", "HEAD")
+	if err != nil {
+		return nil, err
+	}
+	branch, onBranch, err := try(ctx, in.dir, "symbolic-ref", "--short", "-q", "HEAD")
+	if err != nil {
+		return nil, err
+	}
+	switch {
+	case head != pin:
+		st.Checkout = submodule.CheckoutDiffers
+	case !onBranch:
+		st.Checkout = submodule.CheckoutDetached
+	}
+	out, err = run(ctx, in.dir, "status", "--porcelain=v2", "--ignore-submodules=dirty")
+	if err != nil {
+		return nil, err
+	}
+	st.Worktree = parseWorktree(out)
+
+	// the pin must exist locally to say anything about it
+	if _, ok, err := try(ctx, in.dir, "cat-file", "-e", pin+"^{commit}"); err != nil {
+		return nil, err
+	} else if !ok {
+		return nil, fmt.Errorf("pinned commit %.7s not fetched", pin)
+	}
+
+	// remote branches containing the pin, oldest tip first
+	out, err = run(ctx, in.dir, "for-each-ref", "--contains", pin, "--sort=committerdate", "--format=%(refname)", "refs/remotes")
+	if err != nil {
+		return nil, err
+	}
+	var containing []string
+	for _, r := range strings.Fields(out) {
+		if !strings.HasSuffix(r, "/HEAD") {
+			containing = append(containing, strings.TrimPrefix(r, "refs/remotes/"))
+		}
+	}
+	st.Pin.Pushed = len(containing) > 0
+
+	ref, source, err := resolve(ctx, in, pin, branch, containing)
+	if err != nil {
+		return nil, err
+	}
+	st.Pin.Ref, st.Pin.Source = ref, source
+
+	// commits
+	if st.Pinned, err = commit(ctx, in.dir, pin); err != nil {
+		return nil, err
+	}
+	st.CheckedOut = st.Pinned
+	if head != pin {
+		if st.CheckedOut, err = commit(ctx, in.dir, head); err != nil {
+			return nil, err
+		}
+	}
+	if source == submodule.RefFromSHA {
+		st.Pin.Ref = st.Pinned.SHA
+		return st, nil
+	}
+	if st.Tip, err = commit(ctx, in.dir, ref); err != nil {
+		return nil, err
+	}
+	out, err = run(ctx, in.dir, "rev-list", "--left-right", "--count", pin+"..."+ref)
+	if err != nil {
+		return nil, err
+	}
+	if st.Pin.Ahead, st.Pin.Behind, err = parseLeftRight(out); err != nil {
+		return nil, err
+	}
+	if st.Pin.Ahead+st.Pin.Behind > 0 {
+		out, err = run(ctx, in.dir, "log", commitFormat, fmt.Sprintf("--max-count=%d", maxBetween), pin+"..."+ref)
+		if err != nil {
+			return nil, err
+		}
+		st.Between = parseCommits(out)
+	}
+	return st, nil
+}
+
+// resolve finds the ref that describes the pin, following the order in
+// docs/state-model.md. containing lists the remote branches containing the
+// pin, oldest tip first.
+func resolve(ctx context.Context, in info, pin, branch string, containing []string) (string, submodule.RefSource, error) {
+	remote, err := remoteName(ctx, in.dir)
+	if err != nil {
+		return "", 0, err
+	}
+
+	// 1. branch from .gitmodules
+	if b := in.branch; b != "" && remote != "" {
+		if b == "." {
+			b, _, err = try(ctx, in.superDir, "symbolic-ref", "--short", "-q", "HEAD")
+			if err != nil {
+				return "", 0, err
+			}
+		}
+		if b != "" {
+			if _, ok, err := try(ctx, in.dir, "rev-parse", "--verify", "-q", "refs/remotes/"+remote+"/"+b); err != nil {
+				return "", 0, err
+			} else if ok {
+				return remote + "/" + b, submodule.RefFromGitmodules, nil
+			}
+		}
+	}
+
+	// 2. upstream of the checked-out branch
+	if branch != "" {
+		if up, ok, err := try(ctx, in.dir, "rev-parse", "--abbrev-ref", "--symbolic-full-name", branch+"@{upstream}"); err != nil {
+			return "", 0, err
+		} else if ok && up != "" {
+			return up, submodule.RefFromCurrentBranch, nil
+		}
+	}
+
+	// 3. remote default branch, if it contains the pin
+	if remote != "" {
+		if def, ok, err := try(ctx, in.dir, "symbolic-ref", "--short", "-q", "refs/remotes/"+remote+"/HEAD"); err != nil {
+			return "", 0, err
+		} else if ok {
+			if _, ok, err := try(ctx, in.dir, "merge-base", "--is-ancestor", pin, def); err != nil {
+				return "", 0, err
+			} else if ok {
+				return def, submodule.RefFromDefaultBranch, nil
+			}
+		}
+	}
+
+	// 4. remote branch containing the pin with the closest tip
+	best, bestN := "", -1
+	for _, r := range containing[:min(len(containing), maxContainsCandidates)] {
+		out, err := run(ctx, in.dir, "rev-list", "--count", pin+".."+r)
+		if err != nil {
+			return "", 0, err
+		}
+		var n int
+		if _, err := fmt.Sscan(out, &n); err == nil && (bestN < 0 || n < bestN) {
+			best, bestN = r, n
+		}
+	}
+	if best != "" {
+		return best, submodule.RefFromContains, nil
+	}
+
+	// 5. nearest tag
+	if tag, ok, err := try(ctx, in.dir, "describe", "--tags", "--abbrev=0", pin); err != nil {
+		return "", 0, err
+	} else if ok && tag != "" {
+		return tag, submodule.RefFromTag, nil
+	}
+
+	// 6. the commit itself
+	return "", submodule.RefFromSHA, nil
+}
+
+// remoteName returns "origin" if it exists, else the first remote, else "".
+func remoteName(ctx context.Context, dir string) (string, error) {
+	out, err := run(ctx, dir, "remote")
+	if err != nil {
+		return "", err
+	}
+	remotes := strings.Fields(out)
+	for _, r := range remotes {
+		if r == "origin" {
+			return r, nil
+		}
+	}
+	if len(remotes) > 0 {
+		return remotes[0], nil
+	}
+	return "", nil
+}
+
+func commit(ctx context.Context, dir, rev string) (*submodule.Commit, error) {
+	out, err := run(ctx, dir, "log", "-1", commitFormat, rev, "--")
+	if err != nil {
+		return nil, err
+	}
+	cs := parseCommits(out)
+	if len(cs) == 0 {
+		return nil, fmt.Errorf("no commit %s", rev)
+	}
+	return &cs[0], nil
+}
+
+// initialized reports whether dir is a checked-out repository (.git file or
+// directory present).
+func initialized(dir string) bool {
+	_, err := os.Stat(filepath.Join(dir, ".git"))
+	return err == nil
+}
+
+func abbrevHome(p string) string {
+	home, err := os.UserHomeDir()
+	if err != nil || home == "" {
+		return p
+	}
+	if p == home {
+		return "~"
+	}
+	if rest, ok := strings.CutPrefix(p, home+string(filepath.Separator)); ok {
+		return "~/" + filepath.ToSlash(rest)
+	}
+	return p
+}

internal/submodule/git/provider_test.go

@@ -0,0 +1,172 @@
+package git
+
+import (
+	"context"
+	"errors"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"runtime"
+	"slices"
+	"testing"
+
+	"submarine/internal/submodule"
+)
+
+// fixtureDir is the superproject built by scripts/make-fixture.sh, or ""
+// if git or bash is unavailable.
+var fixtureDir string
+
+func TestMain(m *testing.M) {
+	code := func() int {
+		tmp, err := os.MkdirTemp("", "submarine-fixture-")
+		if err != nil {
+			panic(err)
+		}
+		defer os.RemoveAll(tmp)
+		if _, err := exec.LookPath("git"); err == nil {
+			_, file, _, _ := runtime.Caller(0)
+			script := filepath.Join(filepath.Dir(file), "..", "..", "..", "scripts", "make-fixture.sh")
+			out, err := exec.Command("bash", script, filepath.Join(tmp, "fx")).CombinedOutput()
+			if err != nil {
+				panic("make-fixture.sh: " + err.Error() + "\n" + string(out))
+			}
+			fixtureDir = filepath.Join(tmp, "fx", "super")
+		}
+		return m.Run()
+	}()
+	os.Exit(code)
+}
+
+func loadFixture(t *testing.T) (*Provider, *submodule.Repo) {
+	t.Helper()
+	if fixtureDir == "" {
+		t.Skip("git not available")
+	}
+	p, err := New(fixtureDir)
+	if err != nil {
+		t.Fatal(err)
+	}
+	repo, err := p.Tree(context.Background())
+	if err != nil {
+		t.Fatal(err)
+	}
+	return p, repo
+}
+
+func TestTree(t *testing.T) {
+	_, repo := loadFixture(t)
+	if repo.Branch != "main" {
+		t.Errorf("Branch = %q, want main", repo.Branch)
+	}
+	if !repo.FetchedAt.IsZero() {
+		t.Errorf("FetchedAt = %v, want zero (superproject never fetched)", repo.FetchedAt)
+	}
+	var paths []string
+	submodule.Walk(repo.Submodules, func(s *submodule.Submodule) { paths = append(paths, s.Path) })
+	want := []string{
+		"src/core", "src/ui", "src/net", "libs/store", "libs/log", "libs/log/modules/nested",
+		"libs/feature", "libs/conflict", "libs/staged", "libs/missing", "vendor",
+	}
+	if !slices.Equal(paths, want) {
+		t.Errorf("paths = %v\nwant    %v", paths, want)
+	}
+	log := repo.Submodules[4]
+	nested := log.Children[0]
+	if nested.Parent != log || nested.Name != "nested" || nested.RelPath != "modules/nested" {
+		t.Errorf("nested = %+v", nested)
+	}
+}
+
+func TestStatus(t *testing.T) {
+	p, _ := loadFixture(t)
+	tests := []struct {
+		path     string
+		issues   []submodule.IssueKind
+		ref      string
+		source   submodule.RefSource
+		ahead    int
+		behind   int
+		staged   bool
+		checkout submodule.Checkout
+	}{
+		{path: "src/core", issues: []submodule.IssueKind{submodule.IssueOffPin}, ref: "origin/main", source: submodule.RefFromDefaultBranch, checkout: submodule.CheckoutDiffers},
+		{path: "src/ui", ref: "origin/main", source: submodule.RefFromCurrentBranch, behind: 2},
+		{path: "src/net", issues: []submodule.IssueKind{submodule.IssueUnpushed}, ref: "origin/main", source: submodule.RefFromCurrentBranch, ahead: 1},
+		{path: "libs/store", ref: "origin/develop", source: submodule.RefFromGitmodules},
+		{path: "libs/log", ref: "origin/main", source: submodule.RefFromCurrentBranch},
+		{path: "libs/log/modules/nested", issues: []submodule.IssueKind{submodule.IssueChanged}, ref: "origin/main", source: submodule.RefFromDefaultBranch, checkout: submodule.CheckoutDetached},
+		{path: "libs/feature", ref: "origin/feat/search", source: submodule.RefFromContains, behind: 1, checkout: submodule.CheckoutDetached},
+		{path: "libs/conflict", issues: []submodule.IssueKind{submodule.IssueConflict}, ref: "origin/main", source: submodule.RefFromCurrentBranch, staged: true},
+		{path: "libs/staged", ref: "origin/main", source: submodule.RefFromCurrentBranch, staged: true},
+		{path: "libs/missing", issues: []submodule.IssueKind{submodule.IssueMissing}},
+		{path: "vendor", issues: []submodule.IssueKind{submodule.IssueChanged}, ref: "origin/main", source: submodule.RefFromCurrentBranch, behind: 2},
+	}
+	for _, tt := range tests {
+		st, err := p.Status(context.Background(), tt.path)
+		if err != nil {
+			t.Errorf("%s: %v", tt.path, err)
+			continue
+		}
+		var kinds []submodule.IssueKind
+		for _, is := range st.Issues() {
+			kinds = append(kinds, is.Kind)
+		}
+		if !slices.Equal(kinds, tt.issues) {
+			t.Errorf("%s: issues %v, want %v", tt.path, kinds, tt.issues)
+		}
+		if st.Presence != submodule.PresenceOK {
+			continue
+		}
+		if st.Pin.Ref != tt.ref || st.Pin.Source != tt.source {
+			t.Errorf("%s: ref %q (%v), want %q (%v)", tt.path, st.Pin.Ref, st.Pin.Source, tt.ref, tt.source)
+		}
+		if st.Pin.Ahead != tt.ahead || st.Pin.Behind != tt.behind {
+			t.Errorf("%s: ↑%d↓%d, want ↑%d↓%d", tt.path, st.Pin.Ahead, st.Pin.Behind, tt.ahead, tt.behind)
+		}
+		if len(st.Between) != tt.ahead+tt.behind {
+			t.Errorf("%s: %d commits between, want %d", tt.path, len(st.Between), tt.ahead+tt.behind)
+		}
+		if st.PinStaged != tt.staged {
+			t.Errorf("%s: PinStaged %v, want %v", tt.path, st.PinStaged, tt.staged)
+		}
+		if st.Checkout != tt.checkout {
+			t.Errorf("%s: Checkout %v, want %v", tt.path, st.Checkout, tt.checkout)
+		}
+		if st.Pinned == nil || st.CheckedOut == nil || st.Tip == nil || st.Pinned.Author != "Ada Lovelace" {
+			t.Errorf("%s: commits %+v %+v %+v", tt.path, st.Pinned, st.CheckedOut, st.Tip)
+		}
+	}
+}
+
+func TestStatusMissingPresence(t *testing.T) {
+	p, _ := loadFixture(t)
+	st, err := p.Status(context.Background(), "libs/missing")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if st.Presence != submodule.NotInitialized || st.Pinned == nil || len(st.Pinned.SHA) != 7 {
+		t.Errorf("libs/missing: %+v", st)
+	}
+}
+
+func TestStatusErrors(t *testing.T) {
+	p, _ := loadFixture(t)
+	if _, err := p.Status(context.Background(), "nope"); err == nil {
+		t.Error("unknown path: want error")
+	}
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel()
+	if _, err := p.Status(ctx, "src/core"); !errors.Is(err, context.Canceled) {
+		t.Errorf("canceled: err = %v, want context.Canceled", err)
+	}
+}
+
+func TestNewNotARepo(t *testing.T) {
+	if _, err := exec.LookPath("git"); err != nil {
+		t.Skip("git not available")
+	}
+	if _, err := New(t.TempDir()); err == nil {
+		t.Error("New(non-repo): want error")
+	}
+}

internal/tui/theme/issues.go

@@ -55,8 +55,11 @@ func (t Theme) Delta(p submodule.PinInfo, base lipgloss.Style) string {
 	return s
 }
 
-// Ago formats a time relative to now, e.g. "5m ago".
+// Ago formats a time relative to now, e.g. "5m ago"; the zero time is "never".
 func Ago(t time.Time) string {
+	if t.IsZero() {
+		return "never"
+	}
 	d := time.Since(t)
 	switch {
 	case d < time.Minute:

scripts/make-fixture.sh

@@ -0,0 +1,126 @@
+#!/usr/bin/env bash
+# Builds a superproject with submodules in every state submarine shows.
+# Usage: scripts/make-fixture.sh <dir>   (dir must not exist)
+# The superproject is <dir>/super; bare "remotes" live in <dir>/remotes.
+set -euo pipefail
+
+[ $# -eq 1 ] || { echo "usage: $0 <dir>" >&2; exit 2; }
+[ ! -e "$1" ] || { echo "$1 already exists" >&2; exit 2; }
+mkdir -p "$1"
+root=$(cd "$1" && pwd)
+
+# Reproducible and isolated from the user's git config.
+export GIT_CONFIG_GLOBAL=/dev/null GIT_CONFIG_NOSYSTEM=1
+export GIT_CONFIG_COUNT=2
+export GIT_CONFIG_KEY_0=protocol.file.allow GIT_CONFIG_VALUE_0=always
+export GIT_CONFIG_KEY_1=init.defaultBranch GIT_CONFIG_VALUE_1=main
+export GIT_AUTHOR_NAME="Ada Lovelace" GIT_AUTHOR_EMAIL=ada@example.com
+export GIT_COMMITTER_NAME="Ada Lovelace" GIT_COMMITTER_EMAIL=ada@example.com
+
+g() { git -C "$@" >/dev/null 2>&1; }
+
+# commit <repo> <file> <message>
+commit() {
+	echo "$3" >>"$1/$2"
+	g "$1" add "$2"
+	g "$1" commit -m "$3"
+}
+
+# remote <name>: bare repo with two commits on main, pushed from a work clone
+remote() {
+	git init -q --bare "$root/remotes/$1.git"
+	git clone -q "$root/remotes/$1.git" "$root/work/$1" 2>/dev/null
+	commit "$root/work/$1" README "Initial $1"
+	commit "$root/work/$1" README "Add docs to $1"
+	g "$root/work/$1" push origin main
+}
+
+# upstream <name> <message>: a new commit on the remote's main
+upstream() {
+	commit "$root/work/$1" README "$2"
+	g "$root/work/$1" push origin main
+}
+
+mkdir -p "$root/remotes" "$root/work"
+for r in core ui net store log feature nested; do remote "$r"; done
+
+# "feature" has a branch that main doesn't contain
+g "$root/work/feature" switch -c feat/search
+commit "$root/work/feature" search "Add search"
+commit "$root/work/feature" search "Tune search"
+g "$root/work/feature" push origin feat/search
+g "$root/work/feature" switch main
+
+# "store" has a develop branch for the .gitmodules branch rule
+g "$root/work/store" switch -c develop
+commit "$root/work/store" dev "Start develop"
+g "$root/work/store" push origin develop
+g "$root/work/store" switch main
+
+# "nested" becomes a submodule of "log"
+g "$root/work/log" submodule add "$root/remotes/nested.git" modules/nested
+g "$root/work/log" commit -m "Add nested"
+g "$root/work/log" push origin main
+
+super="$root/super"
+git init -q "$super"
+commit "$super" README "Initial superproject"
+add() { g "$super" submodule add "${@:3}" "$root/remotes/$1.git" "$2"; }
+add core src/core
+add ui src/ui
+add net src/net
+add store libs/store -b develop
+add log libs/log
+add feature libs/feature
+add core libs/conflict
+add ui libs/staged
+add net libs/missing
+add ui vendor
+g "$super" submodule update --init --recursive
+g "$super" commit -m "Add submodules"
+
+# src/ui: remote moved on, fetched -> pin behind (↓2)
+upstream ui "Fix layout"
+upstream ui "Add theme"
+g "$super/src/ui" fetch
+g "$super/vendor" fetch
+g "$super/libs/staged" fetch
+
+# src/net: local commit pinned but never pushed -> unpushed
+commit "$super/src/net" NOTES "Local experiment"
+g "$super" add src/net
+g "$super" commit -m "Pin net experiment"
+
+# libs/feature: pinned inside feat/search only -> ref from contains (↓1)
+g "$super/libs/feature" fetch
+g "$super/libs/feature" checkout origin/feat/search~1
+g "$super" add libs/feature
+g "$super" commit -m "Pin feature search"
+
+# src/core: checkout moved away from the pin -> off pin
+g "$super/src/core" checkout HEAD~1
+
+# libs/log/modules/nested: uncommitted change, hidden under a collapsed parent
+echo dirty >>"$super/libs/log/modules/nested/README"
+
+# vendor: untracked file
+echo scratch >"$super/vendor/scratch.txt"
+
+# libs/staged: new pin staged in the superproject, not committed
+g "$super/libs/staged" merge --ff-only origin/main
+g "$super" add libs/staged
+
+# libs/conflict: merge conflict in the worktree
+g "$super/libs/conflict" switch -c left
+commit "$super/libs/conflict" README "Left change"
+g "$super/libs/conflict" switch main
+commit "$super/libs/conflict" README "Right change"
+g "$super/libs/conflict" push origin main
+git -C "$super/libs/conflict" merge left >/dev/null 2>&1 || true
+# keep the pin on the checkout so only the conflict shows
+g "$super" add libs/conflict
+
+# libs/missing: deinitialized -> missing
+g "$super" submodule deinit -f libs/missing
+
+echo "$super"