Allow invalid encoding in EXIF UserComment field

However, emit a warning in this case.

Author: fkleon

gen/testdata_exiftool/images/invalid-encoding-usercomment.jpg.json

@@ -0,0 +1,31 @@
+[{
+  "SourceFile": "../testdata/images/invalid-encoding-usercomment.jpg",
+  "ExifTool": {
+    "ExifToolVersion": 12.76,
+    "Warning": "Invalid EXIF text encoding for UserComment"
+  },
+  "File": {
+    "FileName": "invalid-encoding-usercomment.jpg",
+    "Directory": "../testdata/images",
+    "FileSize": 381,
+    "FilePermissions": 100644,
+    "FileType": "JPEG",
+    "FileTypeExtension": "JPG",
+    "MIMEType": "image/jpeg",
+    "ExifByteOrder": "MM",
+    "ImageWidth": 1,
+    "ImageHeight": 1,
+    "EncodingProcess": 0,
+    "BitsPerSample": 8,
+    "ColorComponents": 1
+  },
+  "JFIF": {
+    "JFIFVersion": "1 1",
+    "ResolutionUnit": 0,
+    "XResolution": 1,
+    "YResolution": 1
+  },
+  "EXIF": {
+    "UserComment": "UserComment"
+  }
+}]

helpers.go

@@ -296,6 +296,12 @@ func (c vc) convertUserComment(ctx valueConverterContext, v any) any {
 	// UserComment tag is identified based on an ID code in a fixed 8-byte area at the start of the tag data area.
 	b, ok := typeAssert[[]byte](ctx, v)
 	if !ok {
+		// Handle plain string user comment (which is against spec; but commonly done)
+		// Exiftool prints a warning but returns the string as-is.
+		// See https://github.com/exiftool/exiftool/blob/13.27/lib/Image/ExifTool/Exif.pm#L5483
+		if text, ok := typeAssert[string](ctx, v); ok {
+			return text
+		}
 		return ""
 	}
 	if len(b) < 8 {

imagemeta_test.go

@@ -173,6 +173,37 @@ func TestDecodeCorrupt(t *testing.T) {
 	}
 }
 
+func TestDecodeUserCommentWithInvalidEncoding(t *testing.T) {
+	c := qt.New(t)
+
+	img, err := os.Open(filepath.Join("testdata", "images", "invalid-encoding-usercomment.jpg"))
+	c.Assert(err, qt.IsNil)
+	defer img.Close()
+
+	var userComment any
+	handleTag := func(ti imagemeta.TagInfo) error {
+		if ti.Tag == "UserComment" {
+			userComment = ti.Value
+		}
+		return nil
+	}
+
+	var warning error
+	warnf := func(format string, args ...any) {
+		warning = fmt.Errorf(format, args...)
+	}
+
+	err = imagemeta.Decode(imagemeta.Options{R: img, ImageFormat: imagemeta.JPEG, HandleTag: handleTag, Sources: imagemeta.EXIF, Warnf: warnf})
+	c.Assert(err, qt.IsNil)
+
+	// Expect user comment to be decoded
+	c.Assert(userComment, qt.IsNotNil)
+	c.Assert(userComment, eq, "UserComment")
+
+	// But with a warning about incorrect type
+	c.Assert(warning, qt.ErrorMatches, "UserComment: expected \\[\\]uint8, got string")
+}
+
 func TestDecodeCustomXMPHandler(t *testing.T) {
 	c := qt.New(t)
 
@@ -888,7 +919,8 @@ func withTestDataFile(t testing.TB, fn func(path string, info os.FileInfo, err e
 }
 
 var goldenSkip = map[string]bool{
-	"goexif/geodegrees_as_string.jpg": true, // The file has many EXIF errors. I think we do a better job than exiftools, but there are some differences.
+	"goexif/geodegrees_as_string.jpg":  true, // The file has many EXIF errors. I think we do a better job than exiftools, but there are some differences.
+	"invalid-encoding-usercomment.jpg": true, // The file has an EXIF error that produces a warning in imagemeta. It's tested separately.
 }
 
 var isSpaceDelimitedFloatRe = regexp.MustCompile(`^(\d+\.\d+) (\d+\.\d+)`)