Convert: Verify RAW previews and prefer djxl for JPEG XL #5673 #5693

internal/photoprism/convert_command.go

@@ -11,6 +11,7 @@ import (
 type ConvertCmd struct {
 	Cmd         *exec.Cmd
 	Orientation media.Orientation
+	VerifyImage bool
 }
 
 // String returns the conversion command as string e.g. for logging.
@@ -33,6 +34,13 @@ func (c *ConvertCmd) ResetOrientation() *ConvertCmd {
 	return c.WithOrientation(media.ResetOrientation)
 }
 
+// WithImageVerification marks the command's output for a decode check before acceptance,
+// so corrupt embedded previews are rejected and the loop tries the next command.
+func (c *ConvertCmd) WithImageVerification() *ConvertCmd {
+	c.VerifyImage = true
+	return c
+}
+
 // NewConvertCmd returns a new file converter command with default options.
 func NewConvertCmd(cmd *exec.Cmd) *ConvertCmd {
 	if cmd == nil {

internal/photoprism/convert_command_test.go

@@ -32,6 +32,14 @@ func TestNewConvertCmd(t *testing.T) {
 		assert.Equal(t, "/usr/bin/sips -Z 123 -s format jpeg --out file.jpeg file.heic", result.String())
 		assert.Equal(t, media.ResetOrientation, result.Orientation)
 	})
+	t.Run("WithImageVerification", func(t *testing.T) {
+		result := NewConvertCmd(
+			exec.Command("exiftool", "-q", "-q", "-b", "-PreviewImage", "file.cr3"),
+		)
+		assert.False(t, result.VerifyImage)
+		assert.Same(t, result, result.WithImageVerification())
+		assert.True(t, result.VerifyImage)
+	})
 }
 
 func TestNewConvertCmds(t *testing.T) {

internal/photoprism/convert_image.go

@@ -92,9 +92,10 @@ func (w *Convert) ToImage(f *MediaFile, force bool) (result *MediaFile, err erro
 
 	start := time.Now()
 
-	// PNG, GIF, BMP, TIFF, HEIC/HEIF, AVIF, and WebP can be handled natively, as can
+	// PNG, GIF, BMP, TIFF, HEIC/HEIF, AVIF, and WebP are handled natively. JPEG XL only when
-	// JPEG XL when the libvips runtime provides JPEG XL support.
+	// "djxl" is unavailable: some libvips builds report JPEG XL support but mis-render files
-	if f.IsImageOther() || (f.IsJpegXL() && thumb.JpegXLSupported() && w.conf.JpegXLEnabled()) {
+	// without erroring, so the reference "djxl" decoder is preferred when present.
+	if f.IsImageOther() || (f.IsJpegXL() && thumb.JpegXLSupported() && w.conf.JpegXLEnabled() && w.conf.JpegXLDecoderBin() == "") {
 		log.Infof("convert: converting %s to %s (%s)", clean.Log(filepath.Base(fileName)), clean.Log(filepath.Base(imageName)), f.FileType())
 
 		// Create PNG or JPEG image from source file.
@@ -181,19 +182,30 @@ func (w *Convert) ToImage(f *MediaFile, force bool) (result *MediaFile, err erro
 			log.Debugf("convert: %s (%s)", clean.Error(err), filepath.Base(cmd.Path))
 			continue
 		} else if fs.FileExistsNotEmpty(imageName) {
-			log.Infof("convert: %s created in %s (%s)", clean.Log(filepath.Base(imageName)), time.Since(start), filepath.Base(cmd.Path))
+			// The command wrote the target file directly (e.g. Darktable, RawTherapee).
-			fileOrientation = c.Orientation
-			break
 		} else if res := out.Bytes(); len(res) < 512 || !mimetype.Detect(res).Is(expectedMime) {
 			continue
 		} else if err = os.WriteFile(imageName, res, fs.ModeFile); err != nil {
 			log.Tracef("convert: %s (%s)", err, filepath.Base(cmd.Path))
 			continue
-		} else {
-			log.Infof("convert: %s created in %s (%s)", clean.Log(filepath.Base(imageName)), time.Since(start), filepath.Base(cmd.Path))
-			fileOrientation = c.Orientation
-			break
 		}
+
+		// Reject undecodable output (e.g. a truncated/bogus embedded RAW preview that passed
+		// the MIME sniff) so the loop tries the next converter instead of indexing a file
+		// whose thumbnails will fail.
+		if c.VerifyImage {
+			if err = thumb.Verify(imageName); err != nil {
+				log.Debugf("convert: discarding undecodable %s from %s (%s)", clean.Log(filepath.Base(imageName)), filepath.Base(cmd.Path), clean.Error(err))
+				if removeErr := os.Remove(imageName); removeErr != nil && !os.IsNotExist(removeErr) {
+					log.Tracef("convert: %s (%s)", removeErr, filepath.Base(cmd.Path))
+				}
+				continue
+			}
+		}
+
+		log.Infof("convert: %s created in %s (%s)", clean.Log(filepath.Base(imageName)), time.Since(start), filepath.Base(cmd.Path))
+		fileOrientation = c.Orientation
+		break
 	}
 
 	// Ok?

internal/photoprism/convert_image_jpeg.go

@@ -95,11 +95,11 @@ func (w *Convert) JpegConvertCmds(f *MediaFile, jpegName string, xmpName string)
 		if w.conf.ExifToolEnabled() {
 			result = append(result, NewConvertCmd(
 				// #nosec G204 -- arguments are built from validated config and file paths.
-				exec.Command(w.conf.ExifToolBin(), "-q", "-q", "-b", "-JpgFromRaw", f.FileName())),
+				exec.Command(w.conf.ExifToolBin(), "-q", "-q", "-b", "-JpgFromRaw", f.FileName())).WithImageVerification(),
 			)
 			result = append(result, NewConvertCmd(
 				// #nosec G204 -- arguments are built from validated config and file paths.
-				exec.Command(w.conf.ExifToolBin(), "-q", "-q", "-b", "-PreviewImage", f.FileName())),
+				exec.Command(w.conf.ExifToolBin(), "-q", "-q", "-b", "-PreviewImage", f.FileName())).WithImageVerification(),
 			)
 		}
 
@@ -121,7 +121,7 @@ func (w *Convert) JpegConvertCmds(f *MediaFile, jpegName string, xmpName string)
 		// Example: exiftool -b -PreviewImage -w IMG_4691.DNG.jpg IMG_4691.DNG
 		result = append(result, NewConvertCmd(
 			// #nosec G204 -- arguments are built from validated config and file paths.
-			exec.Command(w.conf.ExifToolBin(), "-q", "-q", "-b", "-PreviewImage", f.FileName())),
+			exec.Command(w.conf.ExifToolBin(), "-q", "-q", "-b", "-PreviewImage", f.FileName())).WithImageVerification(),
 		)
 	}
 
@@ -165,7 +165,7 @@ func (w *Convert) JpegConvertCmds(f *MediaFile, jpegName string, xmpName string)
 	if (f.IsTiff() || f.IsPsd()) && w.conf.ExifToolEnabled() {
 		result = append(result, NewConvertCmd(
 			// #nosec G204 -- arguments are built from validated config and file paths.
-			exec.Command(w.conf.ExifToolBin(), "-q", "-q", "-b", "-PhotoshopThumbnail", f.FileName())),
+			exec.Command(w.conf.ExifToolBin(), "-q", "-q", "-b", "-PhotoshopThumbnail", f.FileName())).WithImageVerification(),
 		)
 	}
 

internal/photoprism/convert_image_test.go

@@ -9,6 +9,7 @@ import (
 	"github.com/stretchr/testify/assert"
 
 	"github.com/photoprism/photoprism/internal/config"
+	"github.com/photoprism/photoprism/internal/thumb"
 	"github.com/photoprism/photoprism/pkg/fs"
 )
 
@@ -50,6 +51,38 @@ func TestConvert_ToImage(t *testing.T) {
 
 		_ = os.Remove(outputName)
 	})
+	t.Run("JpegXL", func(t *testing.T) {
+		if cnf.JpegXLDecoderBin() == "" {
+			t.Skip("djxl must be available for the JPEG XL conversion path")
+		}
+
+		fileName := filepath.Join(cnf.SamplesPath(), "dice.jxl")
+		outputName := filepath.Join(cnf.SidecarPath(), cnf.SamplesPath(), "dice.jxl.jpg")
+
+		_ = os.Remove(outputName)
+
+		mf, err := NewMediaFile(fileName)
+
+		if err != nil {
+			t.Fatal(err)
+		}
+
+		assert.True(t, mf.IsJpegXL())
+
+		img, err := convert.ToImage(mf, false)
+
+		if err != nil {
+			t.Fatal(err)
+		}
+
+		// With djxl available the native libvips path is skipped (some builds
+		// mis-render JPEG XL); the djxl-produced preview must be decodable.
+		assert.Equal(t, outputName, img.FileName())
+		assert.True(t, img.IsPreviewImage())
+		assert.NoError(t, thumb.Verify(img.FileName()))
+
+		_ = os.Remove(outputName)
+	})
 	t.Run("Raw", func(t *testing.T) {
 		jpegFilename := filepath.Join(cnf.ImportPath(), "fern_green.jpg")
 

internal/thumb/verify.go

@@ -0,0 +1,34 @@
+package thumb
+
+import (
+	"fmt"
+
+	"github.com/davidbyttow/govips/v2/vips"
+)
+
+// Verify reports an error if the image file cannot be decoded by the active rendering library.
+// Used to reject corrupt converter output (e.g. a truncated embedded RAW preview that passes a
+// MIME sniff but later fails libvips) so the conversion loop can try the next converter.
+func Verify(fileName string) error {
+	if fileName == "" {
+		return fmt.Errorf("verify: empty filename")
+	}
+
+	// Use the loader the thumbnailer will use so the check matches GenerateThumbnails.
+	if Library == LibVips {
+		VipsInit()
+
+		img, err := vips.LoadImageFromFile(fileName, VipsImportParams())
+		if err != nil {
+			return err
+		}
+
+		img.Close()
+
+		return nil
+	}
+
+	_, err := Open(fileName, 1)
+
+	return err
+}

internal/thumb/verify_test.go

@@ -0,0 +1,29 @@
+package thumb
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestVerify(t *testing.T) {
+	t.Run("ValidJPEG", func(t *testing.T) {
+		assert.NoError(t, Verify("testdata/example.jpg"))
+	})
+	t.Run("ValidPNG", func(t *testing.T) {
+		assert.NoError(t, Verify("testdata/example.png"))
+	})
+	t.Run("EmptyFilename", func(t *testing.T) {
+		assert.Error(t, Verify(""))
+	})
+	t.Run("NotAnImage", func(t *testing.T) {
+		name := filepath.Join(t.TempDir(), "garbage.jpg")
+		assert.NoError(t, os.WriteFile(name, []byte("this is not a valid jpeg payload, only plain text"), 0o644))
+		assert.Error(t, Verify(name))
+	})
+	t.Run("MissingFile", func(t *testing.T) {
+		assert.Error(t, Verify(filepath.Join(t.TempDir(), "does-not-exist.jpg")))
+	})
+}