copy files off a card, verified by SHA256 (closes #5)

Store.Import copies every member of every file group and the images
and videos under PRIVATE/M4ROOT to dest/YYYY-MM-DD/<file name>. The day
comes from the new Object.ModTime. A file group stays in the folder of
its primary member.

Each copy is read back and compared by SHA256. A file with the same
hash already there is skipped, and anything else there is a conflict
and left alone. One failed file does not stop the others. A cancelled
context stops at once and removes the partial copy.

Object.Hash now shares its reading code with the import.

Model: opus-5-5

README.md

@@ -3,7 +3,7 @@
 ## Description
 
 `dcf` is a WTFPL-licensed Go library by [@sneak](https://sneak.berlin) that
-reads DCF memory cards from digital cameras.
+reads DCF memory cards from digital cameras and copies the files off them.
 
 DCF is the "Design rule for Camera File system". It is also JEITA specification
 number CP-3461. JEITA is the Japan Electronics and Information Technology
@@ -29,6 +29,7 @@ Then list the DCF stores on the mounted filesystems:
 package main
 
 import (
+	"context"
 	"fmt"
 	"log"
 
@@ -36,17 +37,53 @@ import (
 )
 
 func main() {
-	stores, err := dcf.GetDCFStores(0)
+	stores, err := dcf.GetStores(context.Background())
 	if err != nil {
 		log.Fatal(err)
 	}
 
-	for _, store := range *stores {
+	for _, store := range stores {
 		fmt.Println(store.String())
 	}
 }
 ```
 
+Call `dcf.GetStore` instead when you expect exactly one DCF store. It returns
+`dcf.ErrNoStore` when none is mounted, and `dcf.ErrMultipleStores` when more
+than one is.
+
+To copy the files off a card, call `Import` with a destination directory:
+
+```go
+package main
+
+import (
+	"context"
+	"fmt"
+	"log"
+
+	"sneak.berlin/go/dcf/pkg/dcf"
+)
+
+func main() {
+	ctx := context.Background()
+
+	store, err := dcf.GetStore(ctx)
+	if err != nil {
+		log.Fatal(err)
+	}
+
+	result, err := store.Import(ctx, "/path/to/photos")
+	if err != nil {
+		log.Fatal(err)
+	}
+
+	fmt.Println("copied:", result.Copied)
+	fmt.Println("skipped:", result.Skipped)
+	fmt.Println("conflicts:", result.Conflicts)
+}
+```
+
 To work on the library itself, you need docker and git:
 
 ```sh
@@ -105,26 +142,39 @@ overengineering and reusable code.
 
 All the code is in the `pkg/dcf` package. The repo has no commands.
 
-- `dcf.go` holds the public API. `GetDCFStores` finds the DCF stores and reads
+- `dcf.go` holds the public API for reading a card. `GetStores` finds the DCF
-  each one.
+  stores and reads each one. `GetStore` returns the one DCF store when exactly
+  one is mounted.
 - `helpers.go` finds the mount points that hold a DCF store.
 - `parse.go` reads one DCF store through an `fs.FS` rooted at its mount point.
+- `filegroup.go` sorts the files of one DCF directory into file groups.
+- `import.go` holds `Store.Import`, which copies files off a card.
 
 The main types are:
 
 - `Store` is one DCF store on one mounted filesystem.
-- `Directory` is one DCF directory. It has a directory number, a name and the
+- `Directory` is one DCF directory. It has a directory number, a name, the
-  objects inside it.
+  objects inside it and their file groups.
 - `Object` is one file in a store. A file in a DCF directory also has a
-  directory number and a file number.
+  directory number and a file number. `Object.ModTime` is its modification time.
+  `Object.Hash` reads the file through the `fs.FS` of its store. An `Object` you
+  build yourself has none, so its `Hash` returns an error.
+- `FileGroup` is one file group in a DCF directory.
 - `Image` and `Video` embed `Object`. The file extension decides which one a
   file becomes. A file with any other extension, such as a `THM` thumbnail, is
   neither.
+- `ImportResult` lists what `Store.Import` did with each file.
 
 A mounted filesystem is a DCF store when its root holds a `DCIM` or
 `PRIVATE/M4ROOT` directory. It needs only one of them. `dcf` never mounts a
 filesystem. Media anywhere else on the filesystem is not part of the store.
 
+`dcf` never follows a symlink.
+
+`GetStores` and `GetStore` take a `context.Context`. When it is cancelled, the
+read stops before the next directory. The returned error matches
+`context.Canceled`.
+
 `dcf` applies the naming rules of the DCF specification, CIPA DC-009-2010, under
 `DCIM`:
 
@@ -138,21 +188,87 @@ filesystem. Media anywhere else on the filesystem is not part of the store.
 - Directories that share a directory number are not DCF directories.
 - `JPG` files that share a file number in one directory are not DCF files.
 
-`Store.NonConforming` lists the path of every other file and directory under
+`Store.NonConforming` lists the path of everything else under `DCIM`. It
-`DCIM`. It includes everything inside a directory that breaks the rules. These
+includes everything inside a directory that breaks the rules. It also includes
-files have no DCF numbers, and they are not counted as images or videos.
+every symlink, whatever its name. These entries have no DCF numbers, and they
+are not counted as images or videos.
 
 The naming rules do not apply under `PRIVATE/M4ROOT`. The images and videos
 there are in `Store.Images` and `Store.Videos`, and their DCF numbers are 0.
+`dcf` skips any symlink there.
+
+A file group is made of the DCF files in one DCF directory that share a file
+number. The specification calls it a DCF object. A RAW file and its JPEG form a
+file group, and so do a video and its `THM` thumbnail.
+
+- The first four characters and the extension can differ within a group.
+- A DCF file that shares its number with no other file is a group of one.
+- Files in different DCF directories are never in the same group.
+- Only DCF files are grouped. Files in `Store.NonConforming` and under
+  `PRIVATE/M4ROOT` are in no group.
+
+`Directory.FileGroups` lists the groups in file number order. Each `FileGroup`
+has:
+
+- `DirectoryNumber` and `FileNumber`
+- `Members`, which is every file in the group
+- `Primary`, which is the member that stands for the whole group
+- `MembersCount`, which returns the number of members
+- `TotalSize`, which returns the total size of the members in bytes
+
+The primary member is the first of these that the group holds:
+
+1. a `JPG` file
+2. any other file in `Store.Images`
+3. a file in `Store.Videos`
+4. any other file except a `THM`
+5. a `THM`
+
+`Store.Images` and `Store.Videos` list each file on its own, whatever its group.
+
+`Store.Import` copies files off a card to a destination directory. It verifies
+each copy against the SHA256 hash of its source. It only reads the card, so it
+never deletes anything there.
+
+It copies:
+
+- every member of every file group
+- the images and videos under `PRIVATE/M4ROOT`
+
+It does not copy the files in `Store.NonConforming`.
+
+Each file goes to `YYYY-MM-DD/<file name>` under the destination directory. The
+date is the day of the file's modification time, in the computer's time zone.
+Every member of a file group goes to the day of its primary member, so a group
+never splits across two folders. Each copy keeps the modification time of its
+source. Only the user who ran the import can read the copies.
+
+`ImportResult` lists each destination path in one of three lists:
+
+- `Copied` holds the copies this import wrote and verified.
+- `Skipped` holds the paths that already held a file with the same SHA256 hash.
+- `Conflicts` holds the paths that already held anything else. `dcf` leaves that
+  alone and does not copy the file.
+
+A conflict is not an error, so a caller must check `Conflicts`. These are the
+usual causes:
+
+- Two different files have the same name and the same day. Two cards can reuse a
+  file number, and so can two DCF directories on one card. The first file is
+  copied, and the second is a conflict.
+- A partial file remains because the process died during a copy. Delete it and
+  import again.
+
+A copy that fails or does not match the hash of its source is removed. One file
+that fails does not stop the others. The returned error joins the failures, each
+naming its file. The result still lists what was done.
+
+When the context is cancelled, the import stops at once, even inside a file, and
+removes the partial copy. The returned error matches `context.Canceled`.
 
 ## TODO
 
-- Handle DCF file groups, where multiple objects share the same index number.
+No open items.
-  `dcf` does not handle these yet. If you need this, send me a link to a zip or
-  tar of a memory card that uses them, and I'll see what I can do.
-- Take a `context.Context` to cancel the walk of a store.
-- Return a plain slice from `GetDCFStores`, not a pointer to one.
-- Copy files off a card, and verify each copy against its SHA256 hash.
 
 ## License
 

pkg/dcf/dcf.go

@@ -2,22 +2,33 @@
 // Camera File system" specification (JEITA CP-3461). It finds the cards among
 // the mounted filesystems and parses the DCF directories and files on each one
 // into their directory numbers and file numbers. It lists the images and videos
-// on each card, and the names under DCIM that break the DCF naming rules.
+// on each card, and everything under DCIM that is not a DCF directory or file. It
+// groups the files in each DCF directory that share a file number. It copies the
+// files off a card, and verifies each copy against its SHA256 hash.
 package dcf
 
 import (
-	"crypto/sha256"
+	"context"
-	"encoding/hex"
+	"errors"
 	"fmt"
-	"io"
+	"io/fs"
 	"os"
 	"path/filepath"
 	"slices"
 	"strings"
+	"time"
 
 	"github.com/dustin/go-humanize"
 )
 
+var (
+	// ErrNoStore is the error from GetStore when no DCF store is mounted.
+	ErrNoStore = errors.New("no DCF store is mounted")
+	// ErrMultipleStores is the error from GetStore when more than one DCF store
+	// is mounted.
+	ErrMultipleStores = errors.New("more than one DCF store is mounted")
+)
+
 // Store represents a DCF store, which is a directory structure that contains
 // images and videos in the root of a mounted filesystem.
 type Store struct {
@@ -28,9 +39,11 @@ type Store struct {
 	// Directories, and those under PRIVATE/M4ROOT.
 	Images []*Image
 	Videos []*Video
-	// NonConforming holds the path, relative to RootDirectory, of every file
+	// NonConforming holds the path, relative to RootDirectory, of everything
-	// and directory under DCIM that breaks the DCF naming rules. It includes
+	// under DCIM that is not a DCF directory or a DCF file. It includes every
-	// everything inside a directory that breaks them.
+	// symlink, and everything inside a directory that breaks the DCF naming
+	// rules. A directory whose name is not valid UTF-8 is listed without what is
+	// inside it, because fs.FS refuses to open one.
 	NonConforming []string
 }
 
@@ -44,6 +57,22 @@ type Directory struct {
 	// Objects holds every DCF file in the directory, including those that are
 	// neither images nor videos.
 	Objects []*Object
+	// FileGroups holds the same files, grouped by file number, in file number
+	// order.
+	FileGroups []*FileGroup
+}
+
+// FileGroup represents one file group: the files in one DCF directory that share
+// a file number. The DCF specification calls this a DCF object. A file that
+// shares its number with no other file is a group of one.
+type FileGroup struct {
+	DirectoryNumber int
+	FileNumber      int
+	// Members holds every file in the group, including Primary, in name order.
+	// Two members can have the same extension.
+	Members []*Object
+	// Primary is the member that stands for the whole group.
+	Primary *Object
 }
 
 // Object represents a file in a DCF store. It is embedded in Image and Video
@@ -53,10 +82,15 @@ type Object struct {
 	Path         string
 	Size         int64
 	Extension    string
+	// ModTime is the modification time of the file.
+	ModTime time.Time
 	// DirectoryNumber and FileNumber are the DCF numbers of a file in a DCF
 	// directory. Both are 0 for a file under PRIVATE/M4ROOT.
 	DirectoryNumber int
 	FileNumber      int
+	// fsys is the fs.FS the store was read through. Hash reads the file through
+	// it.
+	fsys fs.FS
 }
 
 // Image represents an image file on a DCF store.
@@ -112,6 +146,21 @@ func (d *Store) TotalSize() int64 {
 	return d.TotalImageSize() + d.TotalVideoSize()
 }
 
+// MembersCount returns the number of files in the file group.
+func (g *FileGroup) MembersCount() int {
+	return len(g.Members)
+}
+
+// TotalSize returns the total size in bytes of the files in the file group.
+func (g *FileGroup) TotalSize() int64 {
+	totalSize := int64(0)
+	for _, member := range g.Members {
+		totalSize += member.Size
+	}
+
+	return totalSize
+}
+
 func (d *Store) String() string {
 	// A caller can set a negative Size, and uint64 would make that huge.
 	totalSize := uint64(max(d.TotalSize(), 0))
@@ -139,58 +188,88 @@ func (i *Image) String() string {
 	)
 }
 
-// Hash returns the SHA256 hash of the file as a hex string.
+// Hash returns the SHA256 hash of the file as a hex string. It reads the file
+// through the fs.FS of its store. An Object built by hand has none, so Hash
+// returns an error that matches fs.ErrInvalid.
 func (d *Object) Hash() (string, error) {
-	return pathToSHA256(d.FullFilePath())
+	hash, err := d.hash(context.Background())
-}
-
-func pathToSHA256(path string) (string, error) {
-	file, err := os.Open(filepath.Clean(path))
 	if err != nil {
-		return "", err
+		return "", fmt.Errorf("hashing %s: %w", d.FullFilePath(), err)
 	}
 
-	defer func() { _ = file.Close() }()
+	return hash, nil
-
-	hash := sha256.New()
-
-	_, err = io.Copy(hash, file)
-	if err != nil {
-		return "", err
 }
 
-	return hex.EncodeToString(hash.Sum(nil)), nil
+// hash returns the SHA256 hash of the file as a hex string, and stops when ctx is
+// cancelled.
+func (d *Object) hash(ctx context.Context) (string, error) {
+	if d.fsys == nil {
+		return "", fs.ErrInvalid
 	}
 
-// GetDCFStores returns a list of DCF stores found on the system in the root of
+	return hashFile(ctx, d.fsys, d.Path)
-// any mounted filesystems. The integer argument is the number of DCF stores to
+}
-// return. If the argument is 0, all DCF stores are returned. It does not mount
+
-// filesystems or search outside of filesystem root directories. It does,
+// GetStores finds the DCF stores in the root of the mounted filesystems and reads
-// however, read the DCIM and PRIVATE/M4ROOT directories of each filesystem to
+// each one. It does not mount filesystems or search outside of filesystem root
-// populate the returned Stores.
+// directories. It reads the DCIM and PRIVATE/M4ROOT directories of each
-func GetDCFStores(requestedCount int) (*[]Store, error) {
+// filesystem to populate the returned Stores. It never follows a symlink. If a
-	dcfStorePaths, err := findDCFMountPoints(requestedCount)
+// card cannot be read, it returns an error that names the card's mount point.
+// When ctx is cancelled, the read stops before the next directory, and the error
+// matches ctx.Err().
+func GetStores(ctx context.Context) ([]*Store, error) {
+	dcfStorePaths, err := findDCFMountPoints(ctx)
 	if err != nil {
 		return nil, err
 	}
 
-	dcfStores := []Store{}
+	return parseStores(ctx, dcfStorePaths)
+}
 
-	for _, dcfStorePath := range dcfStorePaths {
+// GetStore returns the one DCF store when exactly one is mounted. It returns
-		dcfStore, err := parseStore(os.DirFS(dcfStorePath), dcfStorePath)
+// ErrNoStore when none is mounted, and ErrMultipleStores when more than one is.
+// It reads every card first, the way GetStores does, so a card that cannot be
+// read returns that error instead.
+func GetStore(ctx context.Context) (*Store, error) {
+	stores, err := GetStores(ctx)
 	if err != nil {
 		return nil, err
 	}
 
-		dcfStores = append(dcfStores, *dcfStore)
+	return onlyStore(stores)
 }
 
-	return &dcfStores, nil
+// onlyStore returns the store when stores holds exactly one.
+func onlyStore(stores []*Store) (*Store, error) {
+	switch len(stores) {
+	case 0:
+		return nil, ErrNoStore
+	case 1:
+		return stores[0], nil
+	default:
+		return nil, ErrMultipleStores
+	}
+}
+
+// parseStores reads the card at each of the given mount points.
+func parseStores(ctx context.Context, mountpoints []string) ([]*Store, error) {
+	dcfStores := []*Store{}
+
+	for _, mountpoint := range mountpoints {
+		dcfStore, err := parseStore(ctx, os.DirFS(mountpoint), mountpoint)
+		if err != nil {
+			return nil, err
+		}
+
+		dcfStores = append(dcfStores, dcfStore)
+	}
+
+	return dcfStores, nil
 }
 
 func isImageFile(path string) bool {
 	imageExtensions := []string{
-		".jpg", ".jpeg", ".png", ".gif", ".bmp", ".tiff", ".raw", ".arw",
+		".jpg", ".jpeg", ".png", ".gif", ".bmp", ".tif", ".tiff", ".raw", ".arw",
 	}
 
 	return slices.Contains(imageExtensions, strings.ToLower(filepath.Ext(path)))

pkg/dcf/dcf_internal_test.go

@@ -1,6 +1,144 @@
 package dcf
 
-import "testing"
+import (
+	"context"
+	"errors"
+	"io/fs"
+	"path/filepath"
+	"strings"
+	"testing"
+	"testing/fstest"
+)
+
+func TestParseStoresReadsEachMountPoint(t *testing.T) {
+	t.Parallel()
+
+	first := newCardOnDisk(t, "DCIM/100ABCDE/ABCD0001.JPG")
+	second := newCardOnDisk(t, "PRIVATE/M4ROOT/CLIP/C0001.MP4")
+
+	stores, err := parseStores(t.Context(), []string{first, second})
+	if err != nil {
+		t.Fatalf("parseStores: %v", err)
+	}
+
+	if len(stores) != 2 {
+		t.Fatalf("parseStores returned %d stores, want 2", len(stores))
+	}
+
+	cases := map[string]Object{
+		filepath.Join(first, "DCIM/100ABCDE/ABCD0001.JPG"):     stores[0].Images[0].Object,
+		filepath.Join(second, "PRIVATE/M4ROOT/CLIP/C0001.MP4"): stores[1].Videos[0].Object,
+	}
+
+	for want, object := range cases {
+		if got := object.FullFilePath(); got != want {
+			t.Errorf("FullFilePath() = %q, want %q", got, want)
+		}
+
+		got, err := object.Hash()
+		if err != nil {
+			t.Fatalf("Hash() returned error: %v", err)
+		}
+
+		// The published SHA-256 test vector for "abc".
+		wantHash := "ba7816bf8f01cfea414140de5dae2223b00361a396177a9cb410ff61f20015ad"
+		if got != wantHash {
+			t.Errorf("Hash() = %q, want %q", got, wantHash)
+		}
+	}
+}
+
+func TestParseStoresReturnsErrorNamingTheCard(t *testing.T) {
+	t.Parallel()
+
+	// Every lookup under a plain file fails.
+	mountpoint := filepath.Join(newCardOnDisk(t, "FILE"), "FILE")
+
+	_, err := parseStores(t.Context(), []string{mountpoint})
+	if err == nil || !strings.Contains(err.Error(), mountpoint) {
+		t.Fatalf("parseStores error = %v, want one naming %s", err, mountpoint)
+	}
+}
+
+func TestParseStoresReturnsErrorWhenCancelled(t *testing.T) {
+	t.Parallel()
+
+	ctx, cancel := context.WithCancel(t.Context())
+	cancel()
+
+	mountpoint := newCardOnDisk(t, "DCIM/100ABCDE/ABCD0001.JPG")
+
+	_, err := parseStores(ctx, []string{mountpoint})
+	if !errors.Is(err, context.Canceled) {
+		t.Fatalf("parseStores error = %v, want %v", err, context.Canceled)
+	}
+}
+
+func TestOnlyStore(t *testing.T) {
+	t.Parallel()
+
+	first := newMountPoint(t, dcimDirectory)
+	second := newMountPoint(t, m4RootDirectory)
+
+	stores, err := parseStores(t.Context(), []string{first, second})
+	if err != nil {
+		t.Fatalf("parseStores: %v", err)
+	}
+
+	_, err = onlyStore(stores[:0])
+	if !errors.Is(err, ErrNoStore) {
+		t.Errorf("onlyStore of no cards: error = %v, want %v", err, ErrNoStore)
+	}
+
+	store, err := onlyStore(stores[:1])
+	if err != nil {
+		t.Fatalf("onlyStore of one card: %v", err)
+	}
+
+	if store.RootDirectory != first {
+		t.Errorf("onlyStore returned the card at %s, want %s",
+			store.RootDirectory, first)
+	}
+
+	_, err = onlyStore(stores)
+	if !errors.Is(err, ErrMultipleStores) {
+		t.Errorf("onlyStore of two cards: error = %v, want %v",
+			err, ErrMultipleStores)
+	}
+}
+
+func TestObjectHashReadsThroughTheCard(t *testing.T) {
+	t.Parallel()
+
+	card := fstest.MapFS{"DCIM/100ABCDE/ABCD0001.JPG": {Data: []byte("abc")}}
+
+	got, err := mustParseStore(t, card).Images[0].Hash()
+	if err != nil {
+		t.Fatalf("Hash() returned error: %v", err)
+	}
+
+	// The published SHA-256 test vector for "abc".
+	want := "ba7816bf8f01cfea414140de5dae2223b00361a396177a9cb410ff61f20015ad"
+	if got != want {
+		t.Fatalf("Hash() = %q, want %q", got, want)
+	}
+}
+
+func TestObjectHashOfMissingFileReturnsError(t *testing.T) {
+	t.Parallel()
+
+	const filePath = "DCIM/100ABCDE/ABCD0001.JPG"
+
+	card := newCard(filePath)
+	image := mustParseStore(t, card).Images[0]
+
+	delete(card, filePath)
+
+	_, err := image.Hash()
+	if !errors.Is(err, fs.ErrNotExist) {
+		t.Fatalf("Hash() error = %v, want %v", err, fs.ErrNotExist)
+	}
+}
 
 func TestIsImageFile(t *testing.T) {
 	t.Parallel()
@@ -8,6 +146,8 @@ func TestIsImageFile(t *testing.T) {
 	cases := map[string]bool{
 		"DCIM/100MSDCF/DSC00001.jpg":    true,
 		"DCIM/100MSDCF/DSC00001.ARW":    true,
+		"DCIM/100MSDCF/DSC00001.TIF":    true,
+		"DCIM/100MSDCF/DSC00001.tif":    true,
 		"PRIVATE/M4ROOT/CLIP/C0001.MP4": false,
 		"DCIM/100MSDCF/notes.txt":       false,
 		"DCIM/100MSDCF/noextension":     false,

pkg/dcf/dcf_test.go

@@ -1,8 +1,9 @@
 package dcf_test
 
 import (
-	"os"
+	"context"
-	"path/filepath"
+	"errors"
+	"io/fs"
 	"testing"
 
 	"sneak.berlin/go/dcf/pkg/dcf"
@@ -125,37 +126,30 @@ func TestVideoString(t *testing.T) {
 	}
 }
 
-func TestObjectHashIsSHA256OfFile(t *testing.T) {
+func TestObjectHashOfObjectBuiltByHandReturnsError(t *testing.T) {
 	t.Parallel()
 
-	root := t.TempDir()
+	object := dcf.Object{DCFStoreRoot: cardRoot, Path: "a.jpg"}
-
-	err := os.WriteFile(filepath.Join(root, "a.jpg"), []byte("abc"), 0o600)
-	if err != nil {
-		t.Fatalf("WriteFile: %v", err)
-	}
-
-	object := dcf.Object{DCFStoreRoot: root, Path: "a.jpg"}
-
-	got, err := object.Hash()
-	if err != nil {
-		t.Fatalf("Hash() returned error: %v", err)
-	}
-
-	// The published SHA-256 test vector for "abc".
-	want := "ba7816bf8f01cfea414140de5dae2223b00361a396177a9cb410ff61f20015ad"
-	if got != want {
-		t.Fatalf("Hash() = %q, want %q", got, want)
-	}
-}
-
-func TestObjectHashOfMissingFileReturnsError(t *testing.T) {
-	t.Parallel()
-
-	object := dcf.Object{DCFStoreRoot: t.TempDir(), Path: "absent.jpg"}
 
 	_, err := object.Hash()
-	if err == nil {
+	if !errors.Is(err, fs.ErrInvalid) {
-		t.Fatal("Hash() of a missing file returned no error")
+		t.Fatalf("Hash() error = %v, want %v", err, fs.ErrInvalid)
+	}
+}
+
+func TestGetStoresAndGetStoreReturnCancellation(t *testing.T) {
+	t.Parallel()
+
+	ctx, cancel := context.WithCancel(t.Context())
+	cancel()
+
+	_, err := dcf.GetStores(ctx)
+	if !errors.Is(err, context.Canceled) {
+		t.Errorf("GetStores error = %v, want %v", err, context.Canceled)
+	}
+
+	_, err = dcf.GetStore(ctx)
+	if !errors.Is(err, context.Canceled) {
+		t.Errorf("GetStore error = %v, want %v", err, context.Canceled)
 	}
 }

pkg/dcf/filegroup.go

@@ -0,0 +1,67 @@
+package dcf
+
+import (
+	"cmp"
+	"path"
+	"slices"
+	"strings"
+)
+
+// newFileGroups sorts the objects of one DCF directory into file groups, in file
+// number order. Section 4.3.2.2 of the specification says the files in a DCF
+// directory that share a file number are one group. Their first four characters
+// and their extensions can differ.
+func newFileGroups(objects []*Object) []*FileGroup {
+	var groups []*FileGroup
+
+	groupWithNumber := map[int]*FileGroup{}
+
+	for _, object := range objects {
+		group, found := groupWithNumber[object.FileNumber]
+		if !found {
+			group = &FileGroup{
+				DirectoryNumber: object.DirectoryNumber,
+				FileNumber:      object.FileNumber,
+			}
+			groupWithNumber[object.FileNumber] = group
+			groups = append(groups, group)
+		}
+
+		group.Members = append(group.Members, object)
+	}
+
+	for _, group := range groups {
+		group.Primary = primaryMember(group.Members)
+	}
+
+	slices.SortFunc(groups, func(a, b *FileGroup) int {
+		return cmp.Compare(a.FileNumber, b.FileNumber)
+	})
+
+	return groups
+}
+
+// primaryMember returns the member that stands for the whole group. Section
+// 6.2.1.3 of the specification gives a DCF basic file priority over a DCF extended
+// image file. It says nothing about a DCF optional file, the other kind of JPG, so
+// ranking every JPG first is a choice. Section 4.6 makes a THM only the thumbnail
+// of another file, so it comes last. The specification leaves the rest open, so an
+// image comes before a video, and a video before any other file. Among equals, the
+// first in name order wins.
+func primaryMember(members []*Object) *Object {
+	choices := []func(string) bool{isJPG, isImageFile, isVideoFile, isNotTHM}
+
+	for _, choice := range choices {
+		for _, member := range members {
+			if choice(member.Path) {
+				return member
+			}
+		}
+	}
+
+	return members[0]
+}
+
+func isNotTHM(name string) bool {
+	return !strings.EqualFold(path.Ext(name), ".THM")
+}

pkg/dcf/filegroup_internal_test.go

@@ -0,0 +1,232 @@
+package dcf
+
+import (
+	"slices"
+	"testing"
+	"testing/fstest"
+)
+
+// mustFileGroups returns the file groups of directory, and stops the test unless
+// there are exactly count of them.
+func mustFileGroups(t *testing.T, directory *Directory, count int) []*FileGroup {
+	t.Helper()
+
+	if len(directory.FileGroups) != count {
+		t.Fatalf("%s has %d file groups, want %d",
+			directory.Name, len(directory.FileGroups), count)
+	}
+
+	return directory.FileGroups
+}
+
+// assertFileGroup checks the numbers, the primary member and the members of group.
+func assertFileGroup(
+	t *testing.T,
+	group *FileGroup,
+	directoryNumber, fileNumber int,
+	primary string,
+	members ...string,
+) {
+	t.Helper()
+
+	if group.DirectoryNumber != directoryNumber || group.FileNumber != fileNumber {
+		t.Errorf("group numbers = %d and %d, want %d and %d",
+			group.DirectoryNumber, group.FileNumber, directoryNumber, fileNumber)
+	}
+
+	if group.Primary.Path != primary {
+		t.Errorf("Primary = %q, want %q", group.Primary.Path, primary)
+	}
+
+	paths := make([]string, 0, len(group.Members))
+
+	for _, member := range group.Members {
+		paths = append(paths, member.Path)
+	}
+
+	assertPaths(t, "members", paths, members...)
+
+	if got := group.MembersCount(); got != len(members) {
+		t.Errorf("MembersCount() = %d, want %d", got, len(members))
+	}
+}
+
+// assertSizes checks the total size of group and the size of each member.
+func assertSizes(t *testing.T, group *FileGroup, total int64, sizes ...int64) {
+	t.Helper()
+
+	if got := group.TotalSize(); got != total {
+		t.Errorf("TotalSize() = %d, want %d", got, total)
+	}
+
+	got := make([]int64, 0, len(group.Members))
+
+	for _, member := range group.Members {
+		got = append(got, member.Size)
+	}
+
+	if !slices.Equal(got, sizes) {
+		t.Errorf("member sizes = %v, want %v", got, sizes)
+	}
+}
+
+// assertPrimaryMember checks which of the named members primaryMember picks.
+func assertPrimaryMember(t *testing.T, want string, names ...string) {
+	t.Helper()
+
+	members := make([]*Object, 0, len(names))
+
+	for _, name := range names {
+		members = append(members, &Object{Path: name})
+	}
+
+	if got := primaryMember(members).Path; got != want {
+		t.Errorf("primaryMember of %q = %q, want %q", names, got, want)
+	}
+}
+
+func TestParseStoreGroupsFilesSharingAFileNumber(t *testing.T) {
+	t.Parallel()
+
+	store := mustParseStore(t, fstest.MapFS{
+		"DCIM/100ABCDE/DSC00001.ARW": {Data: make([]byte, 2000)},
+		"DCIM/100ABCDE/DSC00001.JPG": {Data: make([]byte, 500)},
+		"DCIM/100ABCDE/MVI_0002.MP4": {Data: make([]byte, 9000)},
+		"DCIM/100ABCDE/MVI_0002.THM": {Data: make([]byte, 100)},
+		"DCIM/100ABCDE/DSC00003.JPG": {Data: make([]byte, 700)},
+		"DCIM/101ABCDE/DSC00001.JPG": {Data: make([]byte, 300)},
+	})
+
+	first := mustFileGroups(t, store.Directories[0], 3)
+	second := mustFileGroups(t, store.Directories[1], 1)
+
+	assertFileGroup(t, first[0], 100, 1, "DCIM/100ABCDE/DSC00001.JPG",
+		"DCIM/100ABCDE/DSC00001.ARW",
+		"DCIM/100ABCDE/DSC00001.JPG",
+	)
+	assertSizes(t, first[0], 2500, 2000, 500)
+
+	assertFileGroup(t, first[1], 100, 2, "DCIM/100ABCDE/MVI_0002.MP4",
+		"DCIM/100ABCDE/MVI_0002.MP4",
+		"DCIM/100ABCDE/MVI_0002.THM",
+	)
+	assertSizes(t, first[1], 9100, 9000, 100)
+
+	assertFileGroup(t, first[2], 100, 3, "DCIM/100ABCDE/DSC00003.JPG",
+		"DCIM/100ABCDE/DSC00003.JPG",
+	)
+	assertSizes(t, first[2], 700, 700)
+
+	assertFileGroup(t, second[0], 101, 1, "DCIM/101ABCDE/DSC00001.JPG",
+		"DCIM/101ABCDE/DSC00001.JPG",
+	)
+	assertSizes(t, second[0], 300, 300)
+
+	// The images, the videos and the totals still count each file on its own.
+	if got := store.ImagesCount(); got != 4 {
+		t.Errorf("ImagesCount() = %d, want 4", got)
+	}
+
+	if got := store.VideosCount(); got != 1 {
+		t.Errorf("VideosCount() = %d, want 1", got)
+	}
+
+	if got := store.TotalSize(); got != 12500 {
+		t.Errorf("TotalSize() = %d, want 12500", got)
+	}
+}
+
+func TestParseStoreGroupsFilesWhoseFirstFourCharactersDiffer(t *testing.T) {
+	t.Parallel()
+
+	store := mustParseStore(t, newCard(
+		"DCIM/100ABCDE/ABCD0004.TIF",
+		"DCIM/100ABCDE/THMB0004.THM",
+	))
+
+	groups := mustFileGroups(t, store.Directories[0], 1)
+
+	assertFileGroup(t, groups[0], 100, 4, "DCIM/100ABCDE/ABCD0004.TIF",
+		"DCIM/100ABCDE/ABCD0004.TIF",
+		"DCIM/100ABCDE/THMB0004.THM",
+	)
+}
+
+func TestParseStoreOrdersFileGroupsByFileNumber(t *testing.T) {
+	t.Parallel()
+
+	store := mustParseStore(t, newCard(
+		"DCIM/100ABCDE/ABCD0002.JPG",
+		"DCIM/100ABCDE/WXYZ0001.JPG",
+	))
+
+	groups := mustFileGroups(t, store.Directories[0], 2)
+
+	assertFileGroup(t, groups[0], 100, 1, "DCIM/100ABCDE/WXYZ0001.JPG",
+		"DCIM/100ABCDE/WXYZ0001.JPG",
+	)
+	assertFileGroup(t, groups[1], 100, 2, "DCIM/100ABCDE/ABCD0002.JPG",
+		"DCIM/100ABCDE/ABCD0002.JPG",
+	)
+}
+
+func TestParseStoreLeavesOtherFilesOutOfFileGroups(t *testing.T) {
+	t.Parallel()
+
+	store := mustParseStore(t, newCard(
+		"DCIM/100ABCDE/ABCD0004.ARW",
+		"DCIM/100ABCDE/ABCD0004.JPG",
+		"DCIM/100ABCDE/PQRS0004.JPG",
+		"DCIM/100ABCDE/README.TXT",
+		"DCIM/MISC/ABCD0005.JPG",
+		"PRIVATE/M4ROOT/CLIP/C0006.MP4",
+	))
+
+	if len(store.Directories) != 1 {
+		t.Fatalf("got %d directories, want 1", len(store.Directories))
+	}
+
+	groups := mustFileGroups(t, store.Directories[0], 1)
+
+	assertFileGroup(t, groups[0], 100, 4, "DCIM/100ABCDE/ABCD0004.ARW",
+		"DCIM/100ABCDE/ABCD0004.ARW",
+	)
+}
+
+func TestParseStoreGivesDirectoryWithoutDCFFilesNoFileGroups(t *testing.T) {
+	t.Parallel()
+
+	store := mustParseStore(t, newCard("DCIM/100ABCDE/README.TXT"))
+
+	mustFileGroups(t, store.Directories[0], 0)
+}
+
+func TestPrimaryMember(t *testing.T) {
+	t.Parallel()
+
+	// A JPG comes first, in either case.
+	assertPrimaryMember(t, "A0010001.JPG", "A0010001.ARW", "A0010001.JPG")
+	assertPrimaryMember(t, "A0010001.JPG", "A0010001.JPG", "A0010001.MOV")
+	assertPrimaryMember(t, "a0010001.jpg", "a0010001.arw", "a0010001.jpg")
+
+	// Then any other image.
+	assertPrimaryMember(t, "Z0010001.ARW", "A0010001.MP4", "Z0010001.ARW")
+	assertPrimaryMember(t, "Z0010001.ARW", "A0010001.WAV", "Z0010001.ARW")
+	assertPrimaryMember(t, "A0010001.TIF", "A0010001.MOV", "A0010001.TIF")
+	assertPrimaryMember(t, "Z0010001.TIF", "A0010001.WAV", "Z0010001.TIF")
+
+	// Then a video.
+	assertPrimaryMember(t, "A0010001.MP4", "A0010001.AAC", "A0010001.MP4")
+	assertPrimaryMember(t, "Z0010001.MP4", "A0010001.THM", "Z0010001.MP4")
+
+	// Then any other file except a THM, in either case.
+	assertPrimaryMember(t, "Z0010001.WAV", "A0010001.THM", "Z0010001.WAV")
+	assertPrimaryMember(t, "a0010001.wav", "a0010001.thm", "a0010001.wav")
+
+	// A THM is primary only when the group holds nothing else.
+	assertPrimaryMember(t, "A0010001.THM", "A0010001.THM")
+
+	// Among equals, the first in name order wins.
+	assertPrimaryMember(t, "A0010001.THM", "A0010001.THM", "Z0010001.THM")
+	assertPrimaryMember(t, "A0010001.WAV", "A0010001.WAV", "Z0010001.XML")
+}

pkg/dcf/helpers.go

@@ -1,21 +1,28 @@
 package dcf
 
 import (
+	"context"
 	"fmt"
 	"log/slog"
 	"os"
-	"path/filepath"
 	"slices"
 
 	"github.com/shirou/gopsutil/disk"
 )
 
-func findAllMountPoints() ([]string, error) {
+func findAllMountPoints(ctx context.Context) ([]string, error) {
 	mountpoints := []string{}
 
-	partitions, err := disk.Partitions(false) // physical devices only, so false
+	// disk.PartitionsWithContext does not check ctx on Linux.
+	err := ctx.Err()
 	if err != nil {
-		return nil, err
+		return nil, fmt.Errorf("listing the mounted filesystems: %w", err)
+	}
+
+	// physical devices only, so false
+	partitions, err := disk.PartitionsWithContext(ctx, false)
+	if err != nil {
+		return nil, fmt.Errorf("listing the mounted filesystems: %w", err)
 	}
 
 	for _, partition := range partitions {
@@ -27,21 +34,18 @@ func findAllMountPoints() ([]string, error) {
 	return mountpoints, nil
 }
 
-// findDCFMountPoints returns a list of strings of the paths of mountpoints
+// findDCFMountPoints returns the paths of the mountpoints that hold a DCF store.
-// that hold a DCF store. Its only argument is an integer of how many mountpoints
+func findDCFMountPoints(ctx context.Context) ([]string, error) {
-// to return. If the argument is 0, all mountpoints are returned.
+	mountpoints, err := findAllMountPoints(ctx)
-func findDCFMountPoints(requestedCount int) ([]string, error) {
-	mountpoints, err := findAllMountPoints()
 	if err != nil {
 		return nil, err
 	}
 
-	return filterDCFMountPoints(mountpoints, requestedCount), nil
+	return filterDCFMountPoints(mountpoints), nil
 }
 
-// filterDCFMountPoints returns the first requestedCount of the mountpoints that
+// filterDCFMountPoints returns the mountpoints that hold a DCF store.
-// hold a DCF store. If requestedCount is 0, it returns all of them.
+func filterDCFMountPoints(mountpoints []string) []string {
-func filterDCFMountPoints(mountpoints []string, requestedCount int) []string {
 	filteredMountpoints := []string{}
 
 	for _, mountpoint := range mountpoints {
@@ -52,9 +56,6 @@ func filterDCFMountPoints(mountpoints []string, requestedCount int) []string {
 		slog.Debug(fmt.Sprintf("Keeping mountpoint %s\n", mountpoint))
 
 		filteredMountpoints = append(filteredMountpoints, mountpoint)
-		if (requestedCount > 0) && (len(filteredMountpoints) >= requestedCount) {
-			break
-		}
 	}
 
 	return filteredMountpoints
@@ -63,12 +64,12 @@ func filterDCFMountPoints(mountpoints []string, requestedCount int) []string {
 // isDCFMountPoint reports whether the root of mountpoint holds a DCIM or
 // PRIVATE/M4ROOT directory, which is how we detect DCF stores.
 func isDCFMountPoint(mountpoint string) bool {
-	for _, directory := range []string{dcimDirectory, m4RootDirectory} {
+	fsys := os.DirFS(mountpoint)
-		path := filepath.Join(mountpoint, directory)
 
-		info, err := os.Stat(path)
+	for _, directory := range []string{dcimDirectory, m4RootDirectory} {
-		if err == nil && info.IsDir() {
+		found, err := hasDirectory(fsys, directory)
-			slog.Debug(fmt.Sprintf("Found %s directory at %s\n", directory, path))
+		if err == nil && found {
+			slog.Debug(fmt.Sprintf("Found %s directory at %s\n", directory, mountpoint))
 
 			return true
 		}

pkg/dcf/helpers_internal_test.go

@@ -57,29 +57,31 @@ func TestIsDCFMountPointIgnoresPlainFile(t *testing.T) {
 	}
 }
 
-func TestFilterDCFMountPointsReturnsRequestedCount(t *testing.T) {
+func TestIsDCFMountPointIgnoresSymlink(t *testing.T) {
+	t.Parallel()
+
+	outside := newMountPoint(t, "M4ROOT")
+
+	for _, link := range []string{dcimDirectory, privateDirectory, m4RootDirectory} {
+		root := newMountPoint(t, filepath.Dir(link))
+		addSymlink(t, root, link, outside)
+
+		if isDCFMountPoint(root) {
+			t.Errorf("isDCFMountPoint counted a symlink at %s as a directory", link)
+		}
+	}
+}
+
+func TestFilterDCFMountPointsKeepsThoseHoldingAStore(t *testing.T) {
 	t.Parallel()
 
 	first := newMountPoint(t, "DCIM")
 	second := newMountPoint(t, "PRIVATE/M4ROOT")
-	third := newMountPoint(t, "DCIM")
 
-	// This one holds no DCF store, so it must not use up the requested count.
+	got := filterDCFMountPoints([]string{newMountPoint(t), first, second})
-	mountpoints := []string{newMountPoint(t), first, second, third}
 
-	cases := map[int][]string{
+	want := []string{first, second}
-		0: {first, second, third},
-		1: {first},
-		2: {first, second},
-		3: {first, second, third},
-		4: {first, second, third},
-	}
-
-	for requestedCount, want := range cases {
-		got := filterDCFMountPoints(mountpoints, requestedCount)
 	if !slices.Equal(got, want) {
-			t.Errorf("filterDCFMountPoints(%d) = %v, want %v",
+		t.Errorf("filterDCFMountPoints = %v, want %v", got, want)
-				requestedCount, got, want)
-		}
 	}
 }

pkg/dcf/import.go

@@ -0,0 +1,279 @@
+package dcf
+
+import (
+	"context"
+	"crypto/sha256"
+	"encoding/hex"
+	"errors"
+	"fmt"
+	"io"
+	"io/fs"
+	"os"
+	"path"
+	"path/filepath"
+	"time"
+)
+
+// The modes of what Import creates. Only the user who ran it can read the copies.
+const (
+	importDirectoryMode = 0o750
+	importFileMode      = 0o600
+)
+
+var errHashMismatch = errors.New("the SHA256 hash of the copy differs from the source")
+
+// ImportResult lists what Store.Import did with each file, by destination path.
+type ImportResult struct {
+	// Copied holds the copies this import wrote and verified.
+	Copied []string
+	// Skipped holds the paths that already held a file with the same SHA256 hash.
+	Skipped []string
+	// Conflicts holds the paths that already held anything else. Import left it
+	// alone and did not copy the file.
+	Conflicts []string
+}
+
+// importFile is one file for Import to copy, and the path to copy it to.
+type importFile struct {
+	source *Object
+	target string
+}
+
+// Import copies the files of the store to dest, and verifies each copy against
+// the SHA256 hash of its source. It copies every member of every file group, and
+// the images and videos under PRIVATE/M4ROOT. It only reads the card.
+//
+// Each file goes to dest/YYYY-MM-DD/<file name>, from its modification time. Every
+// member of a file group goes to the day of its primary member, so a group never
+// splits across two folders.
+//
+// A file with the same hash already at the path is skipped. Anything else there is
+// a conflict, which is not an error. Import never overwrites anything.
+//
+// One file that fails does not stop the others. The returned error joins the
+// failures, each naming its file, and the result still lists what was done. When
+// ctx is cancelled, the import stops at once and removes the partial copy, and the
+// error matches ctx.Err().
+func (d *Store) Import(ctx context.Context, dest string) (ImportResult, error) {
+	var (
+		result   ImportResult
+		failures []error
+	)
+
+	for _, file := range d.importFiles(dest) {
+		err := ctx.Err()
+		if err != nil {
+			failures = append(failures, fmt.Errorf("import stopped: %w", err))
+
+			break
+		}
+
+		err = copyFile(ctx, file, &result)
+		if err != nil {
+			failures = append(failures,
+				fmt.Errorf("importing %s: %w", file.source.FullFilePath(), err))
+		}
+	}
+
+	return result, errors.Join(failures...)
+}
+
+// importFiles lists the files of the store that Import copies to dest.
+func (d *Store) importFiles(dest string) []importFile {
+	var files []importFile
+
+	add := func(source *Object, day time.Time) {
+		name := path.Base(source.Path)
+
+		files = append(files, importFile{
+			source: source,
+			target: filepath.Join(dest, day.Format(time.DateOnly), name),
+		})
+	}
+
+	for _, directory := range d.Directories {
+		for _, group := range directory.FileGroups {
+			for _, member := range group.Members {
+				add(member, group.Primary.ModTime)
+			}
+		}
+	}
+
+	// The files under PRIVATE/M4ROOT are the ones with no DCF numbers.
+	for _, image := range d.Images {
+		if image.DirectoryNumber == 0 {
+			add(&image.Object, image.ModTime)
+		}
+	}
+
+	for _, video := range d.Videos {
+		if video.DirectoryNumber == 0 {
+			add(&video.Object, video.ModTime)
+		}
+	}
+
+	return files
+}
+
+// copyFile copies one file, and adds its target to the list in result that says
+// what happened. It adds the target to no list when it returns an error.
+func copyFile(ctx context.Context, file importFile, result *ImportResult) error {
+	sourceHash, err := file.source.hash(ctx)
+	if err != nil {
+		return err
+	}
+
+	err = os.MkdirAll(filepath.Dir(file.target), importDirectoryMode)
+	if err != nil {
+		return err
+	}
+
+	// O_EXCL fails on anything already at the path, even a symlink, so nothing
+	// there can be overwritten.
+	target, err := os.OpenFile(
+		file.target, os.O_WRONLY|os.O_CREATE|os.O_EXCL, importFileMode,
+	)
+	if errors.Is(err, fs.ErrExist) {
+		return checkExisting(ctx, file.target, sourceHash, result)
+	}
+
+	if err != nil {
+		return err
+	}
+
+	err = writeCopy(ctx, file, target, sourceHash)
+	if err != nil {
+		return errors.Join(err, os.Remove(file.target))
+	}
+
+	result.Copied = append(result.Copied, file.target)
+
+	return nil
+}
+
+// checkExisting adds target to the skipped files if it is a file with the given
+// hash, and to the conflicts if it is anything else.
+func checkExisting(
+	ctx context.Context,
+	target, sourceHash string,
+	result *ImportResult,
+) error {
+	info, err := os.Lstat(target)
+	if err != nil {
+		return err
+	}
+
+	if !info.Mode().IsRegular() {
+		result.Conflicts = append(result.Conflicts, target)
+
+		return nil
+	}
+
+	targetHash, err := hashOnDisk(ctx, target)
+	if err != nil {
+		return err
+	}
+
+	if targetHash == sourceHash {
+		result.Skipped = append(result.Skipped, target)
+	} else {
+		result.Conflicts = append(result.Conflicts, target)
+	}
+
+	return nil
+}
+
+// writeCopy fills the new file target from the source, and closes it. It then
+// reads the written file back to compare its hash with the hash of the source.
+func writeCopy(
+	ctx context.Context,
+	file importFile,
+	target *os.File,
+	sourceHash string,
+) error {
+	err := writeAndClose(ctx, file.source, target)
+	if err != nil {
+		return err
+	}
+
+	targetHash, err := hashOnDisk(ctx, file.target)
+	if err != nil {
+		return err
+	}
+
+	if targetHash != sourceHash {
+		return errHashMismatch
+	}
+
+	// The zero time leaves the access time as it is.
+	return os.Chtimes(file.target, time.Time{}, file.source.ModTime)
+}
+
+// writeAndClose writes the bytes of source to target, syncs target and closes it.
+func writeAndClose(ctx context.Context, source *Object, target *os.File) error {
+	defer func() { _ = target.Close() }()
+
+	sourceFile, err := source.fsys.Open(source.Path)
+	if err != nil {
+		return err
+	}
+
+	defer func() { _ = sourceFile.Close() }()
+
+	_, err = io.Copy(target, contextReader(ctx, sourceFile))
+	if err != nil {
+		return err
+	}
+
+	err = target.Sync()
+	if err != nil {
+		return err
+	}
+
+	return target.Close()
+}
+
+// hashOnDisk returns the SHA256 hash of the file at filePath on the local
+// filesystem.
+func hashOnDisk(ctx context.Context, filePath string) (string, error) {
+	return hashFile(ctx, os.DirFS(filepath.Dir(filePath)), filepath.Base(filePath))
+}
+
+// hashFile returns the SHA256 hash of the named file as a hex string.
+func hashFile(ctx context.Context, fsys fs.FS, name string) (string, error) {
+	file, err := fsys.Open(name)
+	if err != nil {
+		return "", err
+	}
+
+	defer func() { _ = file.Close() }()
+
+	hash := sha256.New()
+
+	_, err = io.Copy(hash, contextReader(ctx, file))
+	if err != nil {
+		return "", err
+	}
+
+	return hex.EncodeToString(hash.Sum(nil)), nil
+}
+
+// readerFunc is a function that works as an io.Reader.
+type readerFunc func([]byte) (int, error)
+
+func (f readerFunc) Read(p []byte) (int, error) {
+	return f(p)
+}
+
+// contextReader returns a reader that fails with ctx.Err() once ctx is cancelled,
+// so a copy stops inside a file.
+func contextReader(ctx context.Context, reader io.Reader) io.Reader {
+	return readerFunc(func(p []byte) (int, error) {
+		err := ctx.Err()
+		if err != nil {
+			return 0, err
+		}
+
+		return reader.Read(p)
+	})
+}

pkg/dcf/import_internal_test.go

@@ -0,0 +1,521 @@
+package dcf
+
+import (
+	"context"
+	"errors"
+	"io/fs"
+	"maps"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+	"testing/fstest"
+	"time"
+)
+
+// testDay is the folder that a file from cardFile goes to. The other constants
+// are the paths and contents of the files on most cards in these tests.
+const (
+	testDay = "2026-10-03"
+
+	firstJPG      = "DCIM/100ABCDE/DSC00011.JPG"
+	firstJPGData  = "first jpeg"
+	secondJPG     = "DCIM/100ABCDE/DSC00012.JPG"
+	secondJPGData = "second jpeg"
+	testClip      = "PRIVATE/M4ROOT/CLIP/C0011.MP4"
+	testClipData  = "clip"
+)
+
+// cardFile returns a file on a card, modified at noon on testDay.
+func cardFile(contents string) *fstest.MapFile {
+	return &fstest.MapFile{
+		Data:    []byte(contents),
+		ModTime: time.Date(2026, 10, 3, 12, 0, 0, 0, time.UTC),
+	}
+}
+
+// inTestDay returns the paths of the named files in the testDay folder of dest.
+func inTestDay(dest string, names ...string) []string {
+	paths := make([]string, 0, len(names))
+
+	for _, name := range names {
+		paths = append(paths, filepath.Join(dest, testDay, name))
+	}
+
+	return paths
+}
+
+// assertDest checks that dest holds exactly the given files, as a map from each
+// file's path under dest to its contents.
+func assertDest(t *testing.T, dest string, want map[string]string) {
+	t.Helper()
+
+	fsys := os.DirFS(dest)
+	got := map[string]string{}
+
+	err := fs.WalkDir(fsys, ".",
+		func(name string, entry fs.DirEntry, err error) error {
+			if err != nil || !entry.Type().IsRegular() {
+				return err
+			}
+
+			data, err := fs.ReadFile(fsys, name)
+			got[name] = string(data)
+
+			return err
+		})
+	if err != nil {
+		t.Fatalf("reading %s: %v", dest, err)
+	}
+
+	if !maps.Equal(got, want) {
+		t.Errorf("destination holds %q, want %q", got, want)
+	}
+}
+
+func assertEmptyDirectory(t *testing.T, directory string) {
+	t.Helper()
+
+	entries, err := os.ReadDir(directory)
+	if err != nil {
+		t.Fatalf("ReadDir: %v", err)
+	}
+
+	if len(entries) != 0 {
+		t.Errorf("%s holds %v, want nothing", directory, entries)
+	}
+}
+
+// addDestFile writes a file at the named path under dest.
+func addDestFile(t *testing.T, dest, name, contents string) {
+	t.Helper()
+
+	fullPath := filepath.Join(dest, name)
+
+	err := os.MkdirAll(filepath.Dir(fullPath), 0o750)
+	if err != nil {
+		t.Fatalf("MkdirAll: %v", err)
+	}
+
+	err = os.WriteFile(fullPath, []byte(contents), 0o600)
+	if err != nil {
+		t.Fatalf("WriteFile: %v", err)
+	}
+}
+
+func TestImportCopiesEveryFileAndKeepsItsModificationTime(t *testing.T) {
+	t.Parallel()
+
+	dest := t.TempDir()
+
+	store := mustParseStore(t, fstest.MapFS{
+		"DCIM/100ABCDE/DSC00011.ARW":         cardFile("raw"),
+		firstJPG:                             cardFile(firstJPGData),
+		"DCIM/101ABCDE/MVI_0012.MP4":         cardFile("video"),
+		"DCIM/101ABCDE/MVI_0012.THM":         cardFile("thumbnail"),
+		"DCIM/MISC/DSC00013.JPG":             cardFile("not conforming"),
+		testClip:                             cardFile(testClipData),
+		"PRIVATE/M4ROOT/CLIP/C0011M01.XML":   cardFile("neither image nor video"),
+		"PRIVATE/M4ROOT/THMBNL/C0011T01.JPG": cardFile("clip thumbnail"),
+	})
+
+	result, err := store.Import(t.Context(), dest)
+	if err != nil {
+		t.Fatalf("Import: %v", err)
+	}
+
+	assertPaths(t, "Copied", result.Copied, inTestDay(dest,
+		"DSC00011.ARW",
+		"DSC00011.JPG",
+		"MVI_0012.MP4",
+		"MVI_0012.THM",
+		"C0011T01.JPG",
+		"C0011.MP4",
+	)...)
+	assertPaths(t, "Skipped", result.Skipped)
+	assertPaths(t, "Conflicts", result.Conflicts)
+
+	assertDest(t, dest, map[string]string{
+		testDay + "/DSC00011.ARW": "raw",
+		testDay + "/DSC00011.JPG": firstJPGData,
+		testDay + "/MVI_0012.MP4": "video",
+		testDay + "/MVI_0012.THM": "thumbnail",
+		testDay + "/C0011.MP4":    testClipData,
+		testDay + "/C0011T01.JPG": "clip thumbnail",
+	})
+
+	for _, copied := range result.Copied {
+		info, err := os.Stat(copied)
+		if err != nil {
+			t.Fatalf("Stat: %v", err)
+		}
+
+		if want := cardFile("").ModTime; !info.ModTime().Equal(want) {
+			t.Errorf("%s modified at %v, want %v", copied, info.ModTime(), want)
+		}
+	}
+}
+
+func TestImportTwiceSkipsEveryFileTheSecondTime(t *testing.T) {
+	t.Parallel()
+
+	dest := t.TempDir()
+
+	store := mustParseStore(t, fstest.MapFS{
+		firstJPG: cardFile(firstJPGData),
+		testClip: cardFile(testClipData),
+	})
+
+	_, err := store.Import(t.Context(), dest)
+	if err != nil {
+		t.Fatalf("first Import: %v", err)
+	}
+
+	result, err := store.Import(t.Context(), dest)
+	if err != nil {
+		t.Fatalf("second Import: %v", err)
+	}
+
+	assertPaths(t, "Copied", result.Copied)
+	assertPaths(t, "Skipped", result.Skipped,
+		inTestDay(dest, "DSC00011.JPG", "C0011.MP4")...)
+	assertPaths(t, "Conflicts", result.Conflicts)
+
+	assertDest(t, dest, map[string]string{
+		testDay + "/DSC00011.JPG": firstJPGData,
+		testDay + "/C0011.MP4":    testClipData,
+	})
+}
+
+func TestImportLeavesADifferentFileAloneAndReportsAConflict(t *testing.T) {
+	t.Parallel()
+
+	dest := t.TempDir()
+	addDestFile(t, dest, testDay+"/DSC00011.JPG", "different")
+
+	store := mustParseStore(t, fstest.MapFS{
+		firstJPG:  cardFile(firstJPGData),
+		secondJPG: cardFile(secondJPGData),
+	})
+
+	result, err := store.Import(t.Context(), dest)
+	if err != nil {
+		t.Fatalf("Import: %v", err)
+	}
+
+	assertPaths(t, "Copied", result.Copied, inTestDay(dest, "DSC00012.JPG")...)
+	assertPaths(t, "Skipped", result.Skipped)
+	assertPaths(t, "Conflicts", result.Conflicts, inTestDay(dest, "DSC00011.JPG")...)
+
+	assertDest(t, dest, map[string]string{
+		testDay + "/DSC00011.JPG": "different",
+		testDay + "/DSC00012.JPG": secondJPGData,
+	})
+}
+
+func TestImportReportsAConflictBetweenTwoFilesWithOneName(t *testing.T) {
+	t.Parallel()
+
+	dest := t.TempDir()
+
+	store := mustParseStore(t, fstest.MapFS{
+		firstJPG:                     cardFile(firstJPGData),
+		"DCIM/101ABCDE/DSC00011.JPG": cardFile(secondJPGData),
+	})
+
+	result, err := store.Import(t.Context(), dest)
+	if err != nil {
+		t.Fatalf("Import: %v", err)
+	}
+
+	assertPaths(t, "Copied", result.Copied, inTestDay(dest, "DSC00011.JPG")...)
+	assertPaths(t, "Conflicts", result.Conflicts, inTestDay(dest, "DSC00011.JPG")...)
+	assertDest(t, dest, map[string]string{testDay + "/DSC00011.JPG": firstJPGData})
+}
+
+func TestImportReportsAConflictAtADirectoryOrASymlink(t *testing.T) {
+	t.Parallel()
+
+	dest := t.TempDir()
+
+	// The symlink points at a file with the same bytes, which must not count.
+	outside := newCardOnDisk(t, "OUTSIDE.JPG")
+
+	err := os.MkdirAll(filepath.Join(dest, testDay, "DSC00011.JPG"), 0o750)
+	if err != nil {
+		t.Fatalf("MkdirAll: %v", err)
+	}
+
+	addSymlink(t, dest, testDay+"/DSC00012.JPG", filepath.Join(outside, "OUTSIDE.JPG"))
+
+	store := mustParseStore(t, fstest.MapFS{
+		firstJPG:  cardFile("abc"),
+		secondJPG: cardFile("abc"),
+	})
+
+	result, err := store.Import(t.Context(), dest)
+	if err != nil {
+		t.Fatalf("Import: %v", err)
+	}
+
+	assertPaths(t, "Copied", result.Copied)
+	assertPaths(t, "Skipped", result.Skipped)
+	assertPaths(t, "Conflicts", result.Conflicts,
+		inTestDay(dest, "DSC00011.JPG", "DSC00012.JPG")...)
+}
+
+func TestImportPutsAFileGroupInTheFolderOfItsPrimaryMember(t *testing.T) {
+	t.Parallel()
+
+	dest := t.TempDir()
+
+	beforeMidnight := time.Date(2026, 10, 3, 23, 59, 59, 0, time.UTC)
+	afterMidnight := time.Date(2026, 10, 4, 0, 0, 1, 0, time.UTC)
+
+	store := mustParseStore(t, fstest.MapFS{
+		"DCIM/100ABCDE/DSC00011.ARW": {Data: []byte("raw"), ModTime: afterMidnight},
+		firstJPG:                     {Data: []byte("jpeg"), ModTime: beforeMidnight},
+		secondJPG:                    {Data: []byte("next"), ModTime: afterMidnight},
+	})
+
+	_, err := store.Import(t.Context(), dest)
+	if err != nil {
+		t.Fatalf("Import: %v", err)
+	}
+
+	assertDest(t, dest, map[string]string{
+		"2026-10-03/DSC00011.ARW": "raw",
+		"2026-10-03/DSC00011.JPG": "jpeg",
+		"2026-10-04/DSC00012.JPG": "next",
+	})
+
+	// The copy keeps its own modification time, whichever folder it is in.
+	info, err := os.Stat(filepath.Join(dest, "2026-10-03/DSC00011.ARW"))
+	if err != nil {
+		t.Fatalf("Stat: %v", err)
+	}
+
+	if !info.ModTime().Equal(afterMidnight) {
+		t.Errorf("modified at %v, want %v", info.ModTime(), afterMidnight)
+	}
+}
+
+// changingFS is a card whose file at name holds other bytes from its second read
+// on. It counts the reads in opens.
+type changingFS struct {
+	fstest.MapFS
+
+	name  string
+	opens *int
+}
+
+func (f changingFS) Open(name string) (fs.File, error) {
+	if name == f.name {
+		*f.opens++
+		if *f.opens > 1 {
+			return fstest.MapFS{name: cardFile("other bytes")}.Open(name)
+		}
+	}
+
+	return f.MapFS.Open(name)
+}
+
+func TestImportRemovesACopyThatDoesNotMatchItsSource(t *testing.T) {
+	t.Parallel()
+
+	dest := t.TempDir()
+	opens := 0
+
+	store := mustParseStore(t, changingFS{
+		MapFS: fstest.MapFS{
+			firstJPG:  cardFile(firstJPGData),
+			secondJPG: cardFile(secondJPGData),
+		},
+		name:  firstJPG,
+		opens: &opens,
+	})
+
+	result, err := store.Import(t.Context(), dest)
+	if !errors.Is(err, errHashMismatch) {
+		t.Fatalf("Import error = %v, want %v", err, errHashMismatch)
+	}
+
+	if !strings.Contains(err.Error(), firstJPG) {
+		t.Errorf("Import error = %v, want one naming %s", err, firstJPG)
+	}
+
+	assertPaths(t, "Copied", result.Copied, inTestDay(dest, "DSC00012.JPG")...)
+	assertDest(t, dest, map[string]string{testDay + "/DSC00012.JPG": secondJPGData})
+}
+
+func TestImportCopiesTheOtherFilesWhenOneCannotBeRead(t *testing.T) {
+	t.Parallel()
+
+	dest := t.TempDir()
+
+	store := mustParseStore(t, failingFS{
+		MapFS: fstest.MapFS{
+			firstJPG:  cardFile(firstJPGData),
+			secondJPG: cardFile(secondJPGData),
+			testClip:  cardFile(testClipData),
+		},
+		call: "Open",
+		name: firstJPG,
+	})
+
+	result, err := store.Import(t.Context(), dest)
+	if !errors.Is(err, fs.ErrPermission) {
+		t.Fatalf("Import error = %v, want %v", err, fs.ErrPermission)
+	}
+
+	if !strings.Contains(err.Error(), firstJPG) {
+		t.Errorf("Import error = %v, want one naming %s", err, firstJPG)
+	}
+
+	assertPaths(t, "Copied", result.Copied,
+		inTestDay(dest, "DSC00012.JPG", "C0011.MP4")...)
+	assertDest(t, dest, map[string]string{
+		testDay + "/DSC00012.JPG": secondJPGData,
+		testDay + "/C0011.MP4":    testClipData,
+	})
+}
+
+func TestImportNamesEveryFileThatFails(t *testing.T) {
+	t.Parallel()
+
+	dest := t.TempDir()
+
+	// A plain file where the day folder belongs fails every file of that day.
+	addDestFile(t, dest, testDay, "not a folder")
+
+	store := mustParseStore(t, fstest.MapFS{
+		firstJPG:  cardFile(firstJPGData),
+		secondJPG: cardFile(secondJPGData),
+	})
+
+	result, err := store.Import(t.Context(), dest)
+	if err == nil {
+		t.Fatal("Import returned no error")
+	}
+
+	for _, name := range []string{firstJPG, secondJPG} {
+		if !strings.Contains(err.Error(), name) {
+			t.Errorf("Import error = %v, want one naming %s", err, name)
+		}
+	}
+
+	assertPaths(t, "Copied", result.Copied)
+}
+
+func TestImportOfObjectBuiltByHandReturnsError(t *testing.T) {
+	t.Parallel()
+
+	dest := t.TempDir()
+
+	store := &Store{Images: []*Image{{Object: Object{Path: "DSC00011.JPG"}}}}
+
+	_, err := store.Import(t.Context(), dest)
+	if !errors.Is(err, fs.ErrInvalid) {
+		t.Fatalf("Import error = %v, want %v", err, fs.ErrInvalid)
+	}
+
+	assertEmptyDirectory(t, dest)
+}
+
+// cancelingFile is an open file that cancels the import after each read.
+type cancelingFile struct {
+	fs.File
+
+	cancel context.CancelFunc
+}
+
+func (f cancelingFile) Read(p []byte) (int, error) {
+	n, err := f.File.Read(p)
+
+	f.cancel()
+
+	return n, err
+}
+
+// cancelingReadFS is a card that cancels the import during the read of the named
+// file, the time that file is opened for the openCount-th time. It counts the
+// reads in opens.
+type cancelingReadFS struct {
+	fstest.MapFS
+
+	name      string
+	openCount int
+	opens     *int
+	cancel    context.CancelFunc
+}
+
+func (f cancelingReadFS) Open(name string) (fs.File, error) {
+	file, err := f.MapFS.Open(name)
+	if err != nil || name != f.name {
+		return file, err
+	}
+
+	*f.opens++
+	if *f.opens != f.openCount {
+		return file, nil
+	}
+
+	return cancelingFile{File: file, cancel: f.cancel}, nil
+}
+
+func TestImportStopsInsideAFileWhenCancelledAndRemovesThePartialCopy(t *testing.T) {
+	t.Parallel()
+
+	// The first read of a file is hashing it, and the second is copying it.
+	for _, openCount := range []int{1, 2} {
+		ctx, cancel := context.WithCancel(t.Context())
+		dest := t.TempDir()
+		opens := 0
+
+		store := mustParseStore(t, cancelingReadFS{
+			MapFS: fstest.MapFS{
+				// Larger than one read, so the cancellation falls inside the file.
+				firstJPG:  {Data: make([]byte, 1<<20)},
+				secondJPG: cardFile(secondJPGData),
+			},
+			name:      firstJPG,
+			openCount: openCount,
+			opens:     &opens,
+			cancel:    cancel,
+		})
+
+		result, err := store.Import(ctx, dest)
+		if !errors.Is(err, context.Canceled) {
+			t.Errorf("cancelled in read %d: error = %v, want %v",
+				openCount, err, context.Canceled)
+		}
+
+		assertPaths(t, "Copied", result.Copied)
+		assertDest(t, dest, map[string]string{})
+
+		if opens != openCount {
+			t.Errorf("cancelled in read %d: file opened %d times", openCount, opens)
+		}
+
+		cancel()
+	}
+}
+
+func TestImportCopiesNothingWhenAlreadyCancelled(t *testing.T) {
+	t.Parallel()
+
+	ctx, cancel := context.WithCancel(t.Context())
+	cancel()
+
+	dest := t.TempDir()
+
+	store := mustParseStore(t, fstest.MapFS{firstJPG: cardFile(firstJPGData)})
+
+	result, err := store.Import(ctx, dest)
+	if !errors.Is(err, context.Canceled) {
+		t.Fatalf("Import error = %v, want %v", err, context.Canceled)
+	}
+
+	assertPaths(t, "Copied", result.Copied)
+	assertEmptyDirectory(t, dest)
+}

pkg/dcf/parse.go

@@ -1,13 +1,14 @@
 package dcf
 
 import (
+	"context"
 	"errors"
+	"fmt"
 	"io/fs"
 	"path"
 	"regexp"
 	"strconv"
 	"strings"
-	"syscall"
 )
 
 // The two directories at the root of a card that hold its files.
@@ -26,43 +27,38 @@ var (
 
 // parseStore reads one card from fsys, which is rooted at the card's mount
 // point. The root argument is the path of that mount point. Only DCIM and
-// PRIVATE/M4ROOT are read. A card can lack either one.
+// PRIVATE/M4ROOT are read. A card can lack either one. A symlink is never
-func parseStore(fsys fs.FS, root string) (*Store, error) {
+// followed. An error names the mount point, because fsys reports only paths
+// inside it. When ctx is cancelled, the read stops before the next directory.
+func parseStore(ctx context.Context, fsys fs.FS, root string) (*Store, error) {
 	store := &Store{RootDirectory: root}
 
-	hasDCIM, err := hasDirectory(fsys, dcimDirectory)
+	err := parseDCIM(ctx, fsys, store)
-	if err != nil {
+	if err == nil {
-		return nil, err
+		err = parseM4Root(ctx, fsys, store)
 	}
 
-	if hasDCIM {
-		err = parseDCIM(fsys, store)
 	if err != nil {
-			return nil, err
+		return nil, fmt.Errorf("reading the card at %s: %w", root, err)
-		}
-	}
-
-	hasM4Root, err := hasDirectory(fsys, m4RootDirectory)
-	if err != nil {
-		return nil, err
-	}
-
-	if hasM4Root {
-		err = parseM4Root(fsys, store)
-		if err != nil {
-			return nil, err
-		}
 	}
 
 	return store, nil
 }
 
-// hasDirectory reports whether the card holds the named directory. Only a
+// hasDirectory reports whether the card holds the named directory. A plain file
-// missing directory counts as false: any other failure to look is an error. A
+// or a symlink at any step of the path counts as missing. Any other failure to
-// plain file named PRIVATE makes a lookup of PRIVATE/M4ROOT fail with ENOTDIR.
+// look is an error.
 func hasDirectory(fsys fs.FS, name string) (bool, error) {
-	info, err := fs.Stat(fsys, name)
+	parent := path.Dir(name)
-	if errors.Is(err, fs.ErrNotExist) || errors.Is(err, syscall.ENOTDIR) {
+	if parent != "." {
+		found, err := hasDirectory(fsys, parent)
+		if err != nil || !found {
+			return false, err
+		}
+	}
+
+	info, err := fs.Lstat(fsys, name)
+	if errors.Is(err, fs.ErrNotExist) {
 		return false, nil
 	}
 
@@ -73,11 +69,26 @@ func hasDirectory(fsys fs.FS, name string) (bool, error) {
 	return info.IsDir(), nil
 }
 
+// readDir returns the entries of the named directory, unless ctx is cancelled.
+func readDir(ctx context.Context, fsys fs.FS, name string) ([]fs.DirEntry, error) {
+	err := ctx.Err()
+	if err != nil {
+		return nil, err
+	}
+
+	return fs.ReadDir(fsys, name)
+}
+
 // parseDCIM adds the DCF directories under DCIM to store. Everything else there
 // is not conforming. That includes directories that share a directory number,
 // which section 7.1.2 of the specification treats as non-DCF directories.
-func parseDCIM(fsys fs.FS, store *Store) error {
+func parseDCIM(ctx context.Context, fsys fs.FS, store *Store) error {
-	entries, err := fs.ReadDir(fsys, dcimDirectory)
+	found, err := hasDirectory(fsys, dcimDirectory)
+	if err != nil || !found {
+		return err
+	}
+
+	entries, err := readDir(ctx, fsys, dcimDirectory)
 	if err != nil {
 		return err
 	}
@@ -94,10 +105,9 @@ func parseDCIM(fsys fs.FS, store *Store) error {
 		number := parseNumber(dcfDirectoryName, entry.Name())
 
 		if entry.IsDir() && number != 0 && directoriesWithNumber[number] == 1 {
-			err = parseDirectory(fsys, store, entry.Name(), number)
+			err = parseDirectory(ctx, fsys, store, entry.Name(), number)
 		} else {
-			entryPath := path.Join(dcimDirectory, entry.Name())
+			err = addNonConforming(ctx, fsys, store, dcimDirectory, entry)
-			err = addNonConforming(fsys, store, entryPath)
 		}
 
 		if err != nil {
@@ -109,13 +119,20 @@ func parseDCIM(fsys fs.FS, store *Store) error {
 }
 
 // parseDirectory adds one DCF directory and the DCF files directly inside it to
-// store. Everything else there is not conforming. That includes JPG files that
+// store. A DCF file is a regular file, so a symlink with a DCF name is not one.
-// share a file number, which section 7.2.2 of the specification says must not
+// Everything else there is not conforming. That includes JPG files that share a
-// show DCF numbers. Other files that share a file number do conform.
+// file number, which section 7.2.2 of the specification says must not show DCF
-func parseDirectory(fsys fs.FS, store *Store, name string, number int) error {
+// numbers. Other files that share a file number do conform.
+func parseDirectory(
+	ctx context.Context,
+	fsys fs.FS,
+	store *Store,
+	name string,
+	number int,
+) error {
 	directoryPath := path.Join(dcimDirectory, name)
 
-	entries, err := fs.ReadDir(fsys, directoryPath)
+	entries, err := readDir(ctx, fsys, directoryPath)
 	if err != nil {
 		return err
 	}
@@ -123,7 +140,7 @@ func parseDirectory(fsys fs.FS, store *Store, name string, number int) error {
 	jpgsWithNumber := map[int]int{}
 
 	for _, entry := range entries {
-		if !entry.IsDir() && isJPG(entry.Name()) {
+		if entry.Type().IsRegular() && isJPG(entry.Name()) {
 			jpgsWithNumber[parseNumber(dcfFileName, entry.Name())]++
 		}
 	}
@@ -131,12 +148,11 @@ func parseDirectory(fsys fs.FS, store *Store, name string, number int) error {
 	directory := &Directory{Number: number, Name: name}
 
 	for _, entry := range entries {
-		entryPath := path.Join(directoryPath, entry.Name())
 		fileNumber := parseNumber(dcfFileName, entry.Name())
 		sharedJPG := isJPG(entry.Name()) && jpgsWithNumber[fileNumber] > 1
 
-		if entry.IsDir() || fileNumber == 0 || sharedJPG {
+		if !entry.Type().IsRegular() || fileNumber == 0 || sharedJPG {
-			err = addNonConforming(fsys, store, entryPath)
+			err = addNonConforming(ctx, fsys, store, directoryPath, entry)
 			if err != nil {
 				return err
 			}
@@ -144,7 +160,9 @@ func parseDirectory(fsys fs.FS, store *Store, name string, number int) error {
 			continue
 		}
 
-		object, err := newObject(store, entryPath, entry)
+		entryPath := path.Join(directoryPath, entry.Name())
+
+		object, err := newObject(fsys, store, entryPath, entry)
 		if err != nil {
 			return err
 		}
@@ -156,25 +174,44 @@ func parseDirectory(fsys fs.FS, store *Store, name string, number int) error {
 		addMedia(store, object)
 	}
 
+	directory.FileGroups = newFileGroups(directory.Objects)
 	store.Directories = append(store.Directories, directory)
 
 	return nil
 }
 
 // parseM4Root adds the images and videos under PRIVATE/M4ROOT to store. The DCF
-// naming rules do not apply there, so these files have no DCF numbers.
+// naming rules do not apply there, so these files have no DCF numbers. It skips
-func parseM4Root(fsys fs.FS, store *Store) error {
+// a symlink. It also skips a name that is not valid UTF-8, because fs.FS refuses
+// to open one.
+func parseM4Root(ctx context.Context, fsys fs.FS, store *Store) error {
+	found, err := hasDirectory(fsys, m4RootDirectory)
+	if err != nil || !found {
+		return err
+	}
+
 	return fs.WalkDir(fsys, m4RootDirectory,
 		func(filePath string, entry fs.DirEntry, err error) error {
 			if err != nil {
 				return err
 			}
 
+			validName := fs.ValidPath(filePath)
+
+			if entry.IsDir() && !validName {
+				return fs.SkipDir
+			}
+
+			// fs.WalkDir reads a directory right after this call for it.
 			if entry.IsDir() {
+				return ctx.Err()
+			}
+
+			if !entry.Type().IsRegular() || !validName {
 				return nil
 			}
 
-			object, err := newObject(store, filePath, entry)
+			object, err := newObject(fsys, store, filePath, entry)
 			if err != nil {
 				return err
 			}
@@ -185,23 +222,46 @@ func parseM4Root(fsys fs.FS, store *Store) error {
 		})
 }
 
-// addNonConforming lists the named file or directory on store as not
+// addNonConforming lists one entry of the directory dir on store as not
-// conforming, along with everything inside it.
+// conforming. If the entry is a directory, it lists everything inside it too. It
-func addNonConforming(fsys fs.FS, store *Store, name string) error {
+// does not follow a symlink. It does not read a directory whose name is not
-	return fs.WalkDir(fsys, name,
+// valid UTF-8, because fs.FS refuses to open one.
-		func(entryPath string, _ fs.DirEntry, err error) error {
+func addNonConforming(
+	ctx context.Context,
+	fsys fs.FS,
+	store *Store,
+	dir string,
+	entry fs.DirEntry,
+) error {
+	entryPath := path.Join(dir, entry.Name())
+	store.NonConforming = append(store.NonConforming, entryPath)
+
+	if !entry.IsDir() || !fs.ValidPath(entryPath) {
+		return nil
+	}
+
+	entries, err := readDir(ctx, fsys, entryPath)
 	if err != nil {
 		return err
 	}
 
-			store.NonConforming = append(store.NonConforming, entryPath)
+	for _, inner := range entries {
+		err = addNonConforming(ctx, fsys, store, entryPath, inner)
+		if err != nil {
+			return err
+		}
+	}
 
 	return nil
-		})
 }
 
 // newObject returns the Object for the file at filePath, with no DCF numbers.
-func newObject(store *Store, filePath string, entry fs.DirEntry) (Object, error) {
+func newObject(
+	fsys fs.FS,
+	store *Store,
+	filePath string,
+	entry fs.DirEntry,
+) (Object, error) {
 	info, err := entry.Info()
 	if err != nil {
 		return Object{}, err
@@ -212,6 +272,8 @@ func newObject(store *Store, filePath string, entry fs.DirEntry) (Object, error)
 		Path:         filePath,
 		Size:         info.Size(),
 		Extension:    path.Ext(filePath),
+		ModTime:      info.ModTime(),
+		fsys:         fsys,
 	}, nil
 }
 

pkg/dcf/parse_internal_test.go

@@ -1,9 +1,11 @@
 package dcf
 
 import (
+	"context"
 	"errors"
 	"io/fs"
 	"os"
+	"path"
 	"path/filepath"
 	"reflect"
 	"slices"
@@ -11,7 +13,10 @@ import (
 	"testing/fstest"
 )
 
-const testRoot = "/mnt/card"
+const (
+	testRoot         = "/mnt/card"
+	privateDirectory = "PRIVATE"
+)
 
 // newCard returns a card holding an empty file at each of the given paths.
 func newCard(paths ...string) fstest.MapFS {
@@ -24,10 +29,44 @@ func newCard(paths ...string) fstest.MapFS {
 	return card
 }
 
+// newCardOnDisk returns a temporary directory holding the three bytes "abc" in a
+// file at each of the given paths. It is for tests that fstest.MapFS cannot do.
+func newCardOnDisk(t *testing.T, paths ...string) string {
+	t.Helper()
+
+	root := t.TempDir()
+
+	for _, filePath := range paths {
+		fullPath := filepath.Join(root, filePath)
+
+		err := os.MkdirAll(filepath.Dir(fullPath), 0o750)
+		if err != nil {
+			t.Fatalf("MkdirAll: %v", err)
+		}
+
+		err = os.WriteFile(fullPath, []byte("abc"), 0o600)
+		if err != nil {
+			t.Fatalf("WriteFile: %v", err)
+		}
+	}
+
+	return root
+}
+
+// addSymlink creates a symlink to target at the named path under root.
+func addSymlink(t *testing.T, root, name, target string) {
+	t.Helper()
+
+	err := os.Symlink(target, filepath.Join(root, name))
+	if err != nil {
+		t.Fatalf("Symlink: %v", err)
+	}
+}
+
 func mustParseStore(t *testing.T, fsys fs.FS) *Store {
 	t.Helper()
 
-	store, err := parseStore(fsys, testRoot)
+	store, err := parseStore(t.Context(), fsys, testRoot)
 	if err != nil {
 		t.Fatalf("parseStore: %v", err)
 	}
@@ -75,6 +114,16 @@ func directoryNumbers(store *Store) []int {
 	return numbers
 }
 
+// wantDirectory returns the Directory that parseStore should build for objects.
+func wantDirectory(number int, name string, objects ...*Object) *Directory {
+	return &Directory{
+		Number:     number,
+		Name:       name,
+		Objects:    objects,
+		FileGroups: newFileGroups(objects),
+	}
+}
+
 func assertPaths(t *testing.T, what string, got []string, want ...string) {
 	t.Helper()
 
@@ -109,6 +158,7 @@ func TestParseStorePutsEachFileWhereItBelongs(t *testing.T) {
 		Extension:       ".ARW",
 		DirectoryNumber: 100,
 		FileNumber:      1,
+		fsys:            card,
 	}
 	mp4 := Object{
 		DCFStoreRoot:    testRoot,
@@ -117,6 +167,7 @@ func TestParseStorePutsEachFileWhereItBelongs(t *testing.T) {
 		Extension:       ".MP4",
 		DirectoryNumber: 101,
 		FileNumber:      2,
+		fsys:            card,
 	}
 	thm := Object{
 		DCFStoreRoot:    testRoot,
@@ -125,17 +176,19 @@ func TestParseStorePutsEachFileWhereItBelongs(t *testing.T) {
 		Extension:       ".THM",
 		DirectoryNumber: 101,
 		FileNumber:      2,
+		fsys:            card,
 	}
 	clip := Object{
 		DCFStoreRoot: testRoot,
 		Path:         "PRIVATE/M4ROOT/CLIP/C0002.MP4",
 		Size:         400,
 		Extension:    ".MP4",
+		fsys:         card,
 	}
 
 	wantDirectories := []*Directory{
-		{Number: 100, Name: "100MSDCF", Objects: []*Object{&arw}},
+		wantDirectory(100, "100MSDCF", &arw),
-		{Number: 101, Name: "101CANON", Objects: []*Object{&mp4, &thm}},
+		wantDirectory(101, "101CANON", &mp4, &thm),
 	}
 	if !reflect.DeepEqual(store.Directories, wantDirectories) {
 		t.Errorf("Directories = %+v, want %+v", store.Directories, wantDirectories)
@@ -216,6 +269,8 @@ func TestParseNumberOfDirectoryName(t *testing.T) {
 		"100ABC-E":    0,
 		"100ABCD":     0,
 		"100ABCDEF":   0,
+		"1000ABCDE":   0,
+		"X100ABCDE":   0,
 		"TEMP":        0,
 	}
 
@@ -239,12 +294,14 @@ func TestParseNumberOfFileName(t *testing.T) {
 		"ABCD0000.JPG":    0,
 		"A0004.JPG":       0,
 		"ABCDE0005.JPG":   0,
+		"ABCD10000.JPG":   0,
 		"abcdefg0006.JPG": 0,
 		"+-@]0007.JPG":    0,
 		"ABCD00~1.JPG":    0,
 		"ABCDEFGH.JPG":    0,
 		"ABCD0001":        0,
 		"ABCD0001.":       0,
+		"ABCD0001.JP":     0,
 		"ABCD0001.JPEG":   0,
 		"ABCD0001.J-G":    0,
 		"._DSC0001.JPG":   0,
@@ -261,19 +318,18 @@ func TestParseNumberOfFileName(t *testing.T) {
 func TestParseStoreTreatsLowerCaseNamesAsDCFNames(t *testing.T) {
 	t.Parallel()
 
-	store := mustParseStore(t, newCard("DCIM/101abcde/abcd0002.jpg"))
+	card := newCard("DCIM/101abcde/abcd0002.jpg")
 
-	want := []*Directory{{
+	store := mustParseStore(t, card)
-		Number: 101,
+
-		Name:   "101abcde",
+	want := []*Directory{wantDirectory(101, "101abcde", &Object{
-		Objects: []*Object{{
 		DCFStoreRoot:    testRoot,
 		Path:            "DCIM/101abcde/abcd0002.jpg",
 		Extension:       ".jpg",
 		DirectoryNumber: 101,
 		FileNumber:      2,
-		}},
+		fsys:            card,
-	}}
+	})}
 	if !reflect.DeepEqual(store.Directories, want) {
 		t.Errorf("Directories = %+v, want %+v", store.Directories, want)
 	}
@@ -372,6 +428,49 @@ func TestParseStoreKeepsOtherFilesSharingAFileNumber(t *testing.T) {
 	assertPaths(t, "NonConforming", store.NonConforming)
 }
 
+func TestParseStoreKeepsARWSharingANumberWithTwoJPGs(t *testing.T) {
+	t.Parallel()
+
+	store := mustParseStore(t, newCard(
+		"DCIM/100ABCDE/ABCD0004.ARW",
+		"DCIM/100ABCDE/ABCD0004.JPG",
+		"DCIM/100ABCDE/PQRS0004.JPG",
+	))
+
+	assertPaths(t, "objects", objectPaths(store.Directories[0]),
+		"DCIM/100ABCDE/ABCD0004.ARW",
+	)
+	assertPaths(t, "NonConforming", store.NonConforming,
+		"DCIM/100ABCDE/ABCD0004.JPG",
+		"DCIM/100ABCDE/PQRS0004.JPG",
+	)
+}
+
+func TestParseStoreDoesNotCountDirectoryAsSharingAFileNumber(t *testing.T) {
+	t.Parallel()
+
+	store := mustParseStore(t, newCard(
+		"DCIM/100ABCDE/ABCD0003.JPG",
+		"DCIM/100ABCDE/PQRS0003.JPG/README.TXT",
+	))
+
+	assertPaths(t, "objects", objectPaths(store.Directories[0]),
+		"DCIM/100ABCDE/ABCD0003.JPG",
+	)
+	assertPaths(t, "NonConforming", store.NonConforming,
+		"DCIM/100ABCDE/PQRS0003.JPG",
+		"DCIM/100ABCDE/PQRS0003.JPG/README.TXT",
+	)
+}
+
+func TestParseStoreDoesNotCountDirectoryNamedLikeAVideo(t *testing.T) {
+	t.Parallel()
+
+	store := mustParseStore(t, newCard("PRIVATE/M4ROOT/CLIP.MP4/C0001.XML"))
+
+	assertPaths(t, "Videos", videoPaths(store))
+}
+
 func TestParseStoreRejectsSubdirectoryOfDCFDirectory(t *testing.T) {
 	t.Parallel()
 
@@ -441,24 +540,7 @@ func TestParseStoreTreatsPlainFilesAsMissingDirectories(t *testing.T) {
 func TestParseStoreReadsRealFilesystemWithPlainFileNamedPrivate(t *testing.T) {
 	t.Parallel()
 
-	root := t.TempDir()
+	root := newCardOnDisk(t, "DCIM/100ABCDE/ABCD0001.JPG", "PRIVATE")
-
-	err := os.MkdirAll(filepath.Join(root, "DCIM", "100ABCDE"), 0o750)
-	if err != nil {
-		t.Fatalf("MkdirAll: %v", err)
-	}
-
-	jpg := filepath.Join(root, "DCIM", "100ABCDE", "ABCD0001.JPG")
-
-	err = os.WriteFile(jpg, []byte("abc"), 0o600)
-	if err != nil {
-		t.Fatalf("WriteFile: %v", err)
-	}
-
-	err = os.WriteFile(filepath.Join(root, "PRIVATE"), nil, 0o600)
-	if err != nil {
-		t.Fatalf("WriteFile: %v", err)
-	}
 
 	store := mustParseStore(t, os.DirFS(root))
 
@@ -469,18 +551,365 @@ func TestParseStoreReadsRealFilesystemWithPlainFileNamedPrivate(t *testing.T) {
 	}
 }
 
-// unreadableFS fails every read the way a failing card does.
+// A real filesystem can hold a name with the byte 0xff, which is not valid UTF-8.
-type unreadableFS struct{}
+// fs.FS refuses to open such a name.
-
+func TestParseStoreListsNamesThatFSRefuses(t *testing.T) {
-func (unreadableFS) Open(string) (fs.File, error) {
-	return nil, fs.ErrPermission
-}
-
-func TestParseStoreReturnsReadErrors(t *testing.T) {
 	t.Parallel()
 
-	_, err := parseStore(unreadableFS{}, testRoot)
+	root := newCardOnDisk(t,
+		"DCIM/100ABCDE/ABCD0001.JPG",
+		"DCIM/100ABCDE/A\xffDIR/INNER",
+		"DCIM/100ABCDE/A\xffFILE",
+		"DCIM/A\xffDIR/INNER",
+		"DCIM/A\xffFILE",
+		"DCIM/MISC/A\xffDIR/INNER",
+		"DCIM/MISC/A\xffFILE",
+	)
+
+	store := mustParseStore(t, os.DirFS(root))
+
+	assertPaths(t, "Images", imagePaths(store), "DCIM/100ABCDE/ABCD0001.JPG")
+	assertPaths(t, "NonConforming", store.NonConforming,
+		"DCIM/100ABCDE/A\xffDIR",
+		"DCIM/100ABCDE/A\xffFILE",
+		"DCIM/A\xffDIR",
+		"DCIM/A\xffFILE",
+		"DCIM/MISC",
+		"DCIM/MISC/A\xffDIR",
+		"DCIM/MISC/A\xffFILE",
+	)
+}
+
+func TestParseStoreListsDanglingSymlinks(t *testing.T) {
+	t.Parallel()
+
+	root := newCardOnDisk(t, "DCIM/100ABCDE/ABCD0001.JPG", "DCIM/MISC/README.TXT")
+
+	addSymlink(t, root, "DCIM/100ABCDE/ABCD0002.JPG", "absent")
+	addSymlink(t, root, "DCIM/100ABCDE/LINK", "absent")
+	addSymlink(t, root, "DCIM/101ABCDE", "absent")
+	addSymlink(t, root, "DCIM/LINK", "absent")
+	addSymlink(t, root, "DCIM/MISC/LINK", "absent")
+
+	store := mustParseStore(t, os.DirFS(root))
+
+	if got, want := directoryNumbers(store), []int{100}; !slices.Equal(got, want) {
+		t.Errorf("directory numbers = %v, want %v", got, want)
+	}
+
+	assertPaths(t, "Images", imagePaths(store), "DCIM/100ABCDE/ABCD0001.JPG")
+	assertPaths(t, "NonConforming", store.NonConforming,
+		"DCIM/100ABCDE/ABCD0002.JPG",
+		"DCIM/100ABCDE/LINK",
+		"DCIM/101ABCDE",
+		"DCIM/LINK",
+		"DCIM/MISC",
+		"DCIM/MISC/LINK",
+		"DCIM/MISC/README.TXT",
+	)
+}
+
+func TestParseStoreDoesNotFollowSymlinksUnderDCIM(t *testing.T) {
+	t.Parallel()
+
+	outside := newCardOnDisk(t, "OUTSIDE.JPG", "DIRECTORY/ABCD0009.JPG")
+	outsideFile := filepath.Join(outside, "OUTSIDE.JPG")
+	outsideDirectory := filepath.Join(outside, "DIRECTORY")
+
+	root := newCardOnDisk(t,
+		"DCIM/100ABCDE/ABCD0001.JPG",
+		"DCIM/102ABCDE/ABCD0002.JPG",
+		"DCIM/MISC/README.TXT",
+	)
+
+	// These two have DCF file names. The second has the number of a real JPG,
+	// which must not make that JPG shared.
+	addSymlink(t, root, "DCIM/100ABCDE/ABCD0003.JPG", outsideFile)
+	addSymlink(t, root, "DCIM/100ABCDE/PQRS0001.JPG", outsideFile)
+
+	// These two have DCF directory names. The second has the number of a real
+	// directory, which must not make that directory shared.
+	addSymlink(t, root, "DCIM/101ABCDE", outsideDirectory)
+	addSymlink(t, root, "DCIM/102PQRST", outsideDirectory)
+
+	addSymlink(t, root, "DCIM/LINK", outsideDirectory)
+	addSymlink(t, root, "DCIM/MISC/LINK", outsideDirectory)
+
+	store := mustParseStore(t, os.DirFS(root))
+
+	if got, want := directoryNumbers(store), []int{100, 102}; !slices.Equal(got, want) {
+		t.Errorf("directory numbers = %v, want %v", got, want)
+	}
+
+	assertPaths(t, "Images", imagePaths(store),
+		"DCIM/100ABCDE/ABCD0001.JPG",
+		"DCIM/102ABCDE/ABCD0002.JPG",
+	)
+
+	if got := store.TotalImageSize(); got != 6 {
+		t.Errorf("TotalImageSize() = %d, want 6", got)
+	}
+
+	assertPaths(t, "NonConforming", store.NonConforming,
+		"DCIM/100ABCDE/ABCD0003.JPG",
+		"DCIM/100ABCDE/PQRS0001.JPG",
+		"DCIM/101ABCDE",
+		"DCIM/102PQRST",
+		"DCIM/LINK",
+		"DCIM/MISC",
+		"DCIM/MISC/LINK",
+		"DCIM/MISC/README.TXT",
+	)
+}
+
+func TestParseStoreSkipsSymlinksAndNamesThatFSRefusesUnderM4Root(t *testing.T) {
+	t.Parallel()
+
+	outside := newCardOnDisk(t, "OUTSIDE.MP4")
+
+	// The names with the byte 0xff sort before CLIP, so a walk that stops at one
+	// of them does not reach it.
+	root := newCardOnDisk(t,
+		"PRIVATE/M4ROOT/A\xffDIR/C0002.MP4",
+		"PRIVATE/M4ROOT/A\xffFILE.MP4",
+		"PRIVATE/M4ROOT/CLIP/C0001.MP4",
+	)
+
+	addSymlink(t, root, "PRIVATE/M4ROOT/CLIP/C0003.MP4",
+		filepath.Join(outside, "OUTSIDE.MP4"))
+	addSymlink(t, root, "PRIVATE/M4ROOT/CLIP/C0004.MP4", "absent")
+	addSymlink(t, root, "PRIVATE/M4ROOT/LINK", outside)
+
+	store := mustParseStore(t, os.DirFS(root))
+
+	assertPaths(t, "Videos", videoPaths(store), "PRIVATE/M4ROOT/CLIP/C0001.MP4")
+
+	if got := store.TotalVideoSize(); got != 3 {
+		t.Errorf("TotalVideoSize() = %d, want 3", got)
+	}
+}
+
+func TestParseStoreTreatsSymlinksAsMissingDirectories(t *testing.T) {
+	t.Parallel()
+
+	outside := newCardOnDisk(t,
+		"100ABCDE/ABCD0001.JPG",
+		"M4ROOT/C0001.MP4",
+		"C0002.MP4",
+	)
+
+	cases := []struct{ link, target string }{
+		{dcimDirectory, outside},
+		{dcimDirectory, dcimDirectory},
+		{privateDirectory, outside},
+		{privateDirectory, privateDirectory},
+		{m4RootDirectory, outside},
+	}
+
+	for _, testCase := range cases {
+		root := newMountPoint(t, filepath.Dir(testCase.link))
+		addSymlink(t, root, testCase.link, testCase.target)
+
+		store := mustParseStore(t, os.DirFS(root))
+
+		if !reflect.DeepEqual(store, &Store{RootDirectory: testRoot}) {
+			t.Errorf("symlink at %s to %s read as %+v, want an empty store",
+				testCase.link, testCase.target, store)
+		}
+	}
+}
+
+// failingFS is a card on which one call fails for one path, the way a failing
+// card does. The call is Open, Lstat, ReadDir or Info.
+type failingFS struct {
+	fstest.MapFS
+
+	call string
+	name string
+}
+
+func (f failingFS) Open(name string) (fs.File, error) {
+	err := f.failure("Open", name)
+	if err != nil {
+		return nil, err
+	}
+
+	return f.MapFS.Open(name)
+}
+
+func (f failingFS) Lstat(name string) (fs.FileInfo, error) {
+	err := f.failure("Lstat", name)
+	if err != nil {
+		return nil, err
+	}
+
+	return f.MapFS.Lstat(name)
+}
+
+func (f failingFS) ReadDir(name string) ([]fs.DirEntry, error) {
+	err := f.failure("ReadDir", name)
+	if err != nil {
+		return nil, err
+	}
+
+	entries, err := f.MapFS.ReadDir(name)
+	if err != nil {
+		return nil, err
+	}
+
+	for i, entry := range entries {
+		err = f.failure("Info", path.Join(name, entry.Name()))
+		if err != nil {
+			entries[i] = failingEntry{DirEntry: entry, err: err}
+		}
+	}
+
+	return entries, nil
+}
+
+func (f failingFS) failure(call, name string) error {
+	if call == f.call && name == f.name {
+		return &fs.PathError{Op: call, Path: name, Err: fs.ErrPermission}
+	}
+
+	return nil
+}
+
+// failingEntry is a directory entry whose Info always returns err.
+type failingEntry struct {
+	fs.DirEntry
+
+	err error
+}
+
+func (e failingEntry) Info() (fs.FileInfo, error) {
+	return nil, e.err
+}
+
+func TestParseStoreReturnsReadErrorsNamingTheCard(t *testing.T) {
+	t.Parallel()
+
+	card := newCard(
+		"DCIM/100ABCDE/ABCD0001.JPG",
+		"DCIM/100ABCDE/SUB/README.TXT",
+		"DCIM/MISC/SUB/README.TXT",
+		"PRIVATE/M4ROOT/CLIP/C0001.MP4",
+	)
+
+	cases := map[string][]string{
+		"Lstat": {dcimDirectory, privateDirectory, m4RootDirectory},
+		"ReadDir": {
+			dcimDirectory,
+			"DCIM/100ABCDE",
+			"DCIM/100ABCDE/SUB",
+			"DCIM/MISC",
+			"DCIM/MISC/SUB",
+			m4RootDirectory,
+			"PRIVATE/M4ROOT/CLIP",
+		},
+		"Info": {"DCIM/100ABCDE/ABCD0001.JPG", "PRIVATE/M4ROOT/CLIP/C0001.MP4"},
+	}
+
+	for call, names := range cases {
+		for _, name := range names {
+			fsys := failingFS{MapFS: card, call: call, name: name}
+
+			_, err := parseStore(t.Context(), fsys, testRoot)
 			if !errors.Is(err, fs.ErrPermission) {
-		t.Fatalf("parseStore error = %v, want %v", err, fs.ErrPermission)
+				t.Errorf("%s of %s: error = %v, want %v",
+					call, name, err, fs.ErrPermission)
+
+				continue
+			}
+
+			want := "reading the card at " + testRoot + ": " +
+				call + " " + name + ": permission denied"
+			if err.Error() != want {
+				t.Errorf("error = %q, want %q", err.Error(), want)
+			}
+		}
+	}
+}
+
+// cancelingFS is a card that cancels the read when the read reaches the named
+// directory. It records every directory read in read.
+type cancelingFS struct {
+	fstest.MapFS
+
+	name   string
+	cancel context.CancelFunc
+	read   *[]string
+}
+
+func (f cancelingFS) ReadDir(name string) ([]fs.DirEntry, error) {
+	*f.read = append(*f.read, name)
+
+	if name == f.name {
+		f.cancel()
+	}
+
+	return f.MapFS.ReadDir(name)
+}
+
+func TestParseStoreStopsBeforeTheNextDirectoryWhenCancelled(t *testing.T) {
+	t.Parallel()
+
+	card := newCard(
+		"DCIM/100ABCDE/SUB/README.TXT",
+		"DCIM/OTHER/SUB/README.TXT",
+		"PRIVATE/M4ROOT/CLIP/SUB/C0001.MP4",
+	)
+
+	// Each of these directories holds another directory, which must not be read.
+	names := []string{
+		dcimDirectory,
+		"DCIM/100ABCDE",
+		"DCIM/OTHER",
+		m4RootDirectory,
+		"PRIVATE/M4ROOT/CLIP",
+	}
+
+	for _, name := range names {
+		ctx, cancel := context.WithCancel(t.Context())
+
+		var read []string
+
+		fsys := cancelingFS{MapFS: card, name: name, cancel: cancel, read: &read}
+
+		_, err := parseStore(ctx, fsys, testRoot)
+		if !errors.Is(err, context.Canceled) {
+			t.Errorf("cancelled at %s: error = %v, want %v",
+				name, err, context.Canceled)
+		}
+
+		if read[len(read)-1] != name {
+			t.Errorf("cancelled at %s: read %q", name, read)
+		}
+	}
+}
+
+func TestParseStoreReadsNoDirectoryWhenAlreadyCancelled(t *testing.T) {
+	t.Parallel()
+
+	ctx, cancel := context.WithCancel(t.Context())
+	cancel()
+
+	cards := []fstest.MapFS{
+		newCard("DCIM/README.TXT"),
+		newCard("PRIVATE/M4ROOT/C0001.MP4"),
+	}
+
+	for _, card := range cards {
+		var read []string
+
+		// The context is already cancelled, so this needs no name to cancel at.
+		fsys := cancelingFS{MapFS: card, read: &read}
+
+		_, err := parseStore(ctx, fsys, testRoot)
+		if !errors.Is(err, context.Canceled) {
+			t.Errorf("error = %v, want %v", err, context.Canceled)
+		}
+
+		assertPaths(t, "directories read", read)
 	}
 }