Merge pull request #818 from dolda2000/itxt

Added support for encoding and decoding iTXt chunks.

PIL/PngImagePlugin.py

@@ -147,6 +147,17 @@ class ChunkStream:
 
         return cids
 
+# --------------------------------------------------------------------
+# Subclass of string to allow iTXt chunks to look like strings while
+# keeping their extra information
+
+class iTXt(str):
+    @staticmethod
+    def __new__(cls, text, lang, tkey):
+        self = str.__new__(cls, text)
+        self.lang = lang
+        self.tkey = tkey
+        return self
 
 # --------------------------------------------------------------------
 # PNG chunk container (for use with save(pnginfo=))
@@ -159,14 +170,36 @@ class PngInfo:
     def add(self, cid, data):
         self.chunks.append((cid, data))
 
+    def add_itxt(self, key, value, lang="", tkey="", zip=False):
+        if not isinstance(key, bytes):
+            key = key.encode("latin-1", "strict")
+        if not isinstance(value, bytes):
+            value = value.encode("utf-8", "strict")
+        if not isinstance(lang, bytes):
+            lang = lang.encode("utf-8", "strict")
+        if not isinstance(tkey, bytes):
+            tkey = tkey.encode("utf-8", "strict")
+
+        if zip:
+            import zlib
+            self.add(b"iTXt", key + b"\0\x01\0" + lang + b"\0" + tkey + b"\0" + zlib.compress(value))
+        else:
+            self.add(b"iTXt", key + b"\0\0\0" + lang + b"\0" + tkey + b"\0" + value)
+
     def add_text(self, key, value, zip=0):
+        if isinstance(value, iTXt):
+            return self.add_itxt(key, value, value.lang, value.tkey, bool(zip))
+
         # The tEXt chunk stores latin-1 text
+        if not isinstance(value, bytes):
+            try:
+                value = value.encode('latin-1', 'strict')
+            except UnicodeError:
+                return self.add_itxt(key, value, zip=bool(zip))
+
         if not isinstance(key, bytes):
             key = key.encode('latin-1', 'strict')
 
-        if not isinstance(value, bytes):
-            value = value.encode('latin-1', 'replace')
-
         if zip:
             import zlib
             self.add(b"zTXt", key + b"\0\0" + zlib.compress(value))
@@ -329,6 +362,43 @@ class PngStream(ChunkStream):
             self.im_info[k] = self.im_text[k] = v
         return s
 
+    def chunk_iTXt(self, pos, length):
+
+        # international text
+        r = s = ImageFile._safe_read(self.fp, length)
+        try:
+            k, r = r.split(b"\0", 1)
+        except ValueError:
+            return s
+        if len(r) < 2:
+            return s
+        cf, cm, r = i8(r[0]), i8(r[1]), r[2:]
+        try:
+            lang, tk, v = r.split(b"\0", 2)
+        except ValueError:
+            return s
+        if cf != 0:
+            if cm == 0:
+                import zlib
+                try:
+                    v = zlib.decompress(v)
+                except zlib.error:
+                    return s
+            else:
+                return s
+        if bytes is not str:
+            try:
+                k = k.decode("latin-1", "strict")
+                lang = lang.decode("utf-8", "strict")
+                tk = tk.decode("utf-8", "strict")
+                v = v.decode("utf-8", "strict")
+            except UnicodeError:
+                return s
+
+        self.im_info[k] = self.im_text[k] = iTXt(v, lang, tk)
+        
+        return s
+
 # --------------------------------------------------------------------
 # PNG reader
 

Tests/test_file_png.py

@@ -129,6 +129,39 @@ class TestFilePng(PillowTestCase):
             HEAD + chunk(b'zTXt', b'spam\0\0' + zlib.compress(b'egg')) + TAIL)
         self.assertEqual(im.info,  {'spam': 'egg'})
 
+    def test_bad_itxt(self):
+
+        im = load(HEAD + chunk(b'iTXt') + TAIL)
+        self.assertEqual(im.info, {})
+
+        im = load(HEAD + chunk(b'iTXt', b'spam') + TAIL)
+        self.assertEqual(im.info, {})
+
+        im = load(HEAD + chunk(b'iTXt', b'spam\0') + TAIL)
+        self.assertEqual(im.info, {})
+
+        im = load(HEAD + chunk(b'iTXt', b'spam\0\x02') + TAIL)
+        self.assertEqual(im.info, {})
+
+        im = load(HEAD + chunk(b'iTXt', b'spam\0\0\0foo\0') + TAIL)
+        self.assertEqual(im.info, {})
+
+        im = load(HEAD + chunk(b'iTXt', b'spam\0\0\0en\0Spam\0egg') + TAIL)
+        self.assertEqual(im.info, {"spam": "egg"})
+        self.assertEqual(im.info["spam"].lang, "en")
+        self.assertEqual(im.info["spam"].tkey, "Spam")
+
+        im = load(HEAD + chunk(b'iTXt', b'spam\0\1\0en\0Spam\0' + zlib.compress(b"egg")[:1]) + TAIL)
+        self.assertEqual(im.info, {})
+
+        im = load(HEAD + chunk(b'iTXt', b'spam\0\1\1en\0Spam\0' + zlib.compress(b"egg")) + TAIL)
+        self.assertEqual(im.info, {})
+
+        im = load(HEAD + chunk(b'iTXt', b'spam\0\1\0en\0Spam\0' + zlib.compress(b"egg")) + TAIL)
+        self.assertEqual(im.info, {"spam": "egg"})
+        self.assertEqual(im.info["spam"].lang, "en")
+        self.assertEqual(im.info["spam"].tkey, "Spam")
+
     def test_interlace(self):
 
         file = "Tests/images/pil123p.png"
@@ -232,6 +265,50 @@ class TestFilePng(PillowTestCase):
         self.assertEqual(im.info, {'TXT': 'VALUE', 'ZIP': 'VALUE'})
         self.assertEqual(im.text, {'TXT': 'VALUE', 'ZIP': 'VALUE'})
 
+    def test_roundtrip_itxt(self):
+        # Check iTXt roundtripping
+
+        im = Image.new("RGB", (32, 32))
+        info = PngImagePlugin.PngInfo()
+        info.add_itxt("spam", "Eggs", "en", "Spam")
+        info.add_text("eggs", PngImagePlugin.iTXt("Spam", "en", "Eggs"), zip=True)
+
+        im = roundtrip(im, pnginfo=info)
+        self.assertEqual(im.info, {"spam": "Eggs", "eggs": "Spam"})
+        self.assertEqual(im.text, {"spam": "Eggs", "eggs": "Spam"})
+        self.assertEqual(im.text["spam"].lang, "en")
+        self.assertEqual(im.text["spam"].tkey, "Spam")
+        self.assertEqual(im.text["eggs"].lang, "en")
+        self.assertEqual(im.text["eggs"].tkey, "Eggs")
+
+    def test_nonunicode_text(self):
+        # Check so that non-Unicode text is saved as a tEXt rather than iTXt
+
+        im = Image.new("RGB", (32, 32))
+        info = PngImagePlugin.PngInfo()
+        info.add_text("Text", "Ascii")
+        im = roundtrip(im, pnginfo=info)
+        self.assertEqual(type(im.info["Text"]), str)
+
+    def test_unicode_text(self):
+        # Check preservation of non-ASCII characters on Python3
+        # This cannot really be meaningfully tested on Python2,
+        # since it didn't preserve charsets to begin with.
+
+        def rt_text(value):
+            im = Image.new("RGB", (32, 32))
+            info = PngImagePlugin.PngInfo()
+            info.add_text("Text", value)
+            im = roundtrip(im, pnginfo=info)
+            self.assertEqual(im.info, {"Text": value})
+
+        if str is not bytes:
+            rt_text(" Aa" + chr(0xa0) + chr(0xc4) + chr(0xff)) # Latin1
+            rt_text(chr(0x400) + chr(0x472) + chr(0x4ff))      # Cyrillic
+            rt_text(chr(0x4e00) + chr(0x66f0) +                # CJK
+                    chr(0x9fba) + chr(0x3042) + chr(0xac00))
+            rt_text("A" + chr(0xc4) + chr(0x472) + chr(0x3042)) # Combined
+
     def test_scary(self):
         # Check reading of evil PNG file.  For information, see:
         # http://scary.beasts.org/security/CESA-2004-001.txt