Added experimental support for fixing Windows-1252 characters embedded in UTF-8 documents.

3 files changed, 239 insertions, 0 deletions

diff --git a/NEWS.txt b/NEWS.txt
index 47ffd28..a4d207e 100644
--- a/NEWS.txt
+++ b/NEWS.txt
@@ -1,3 +1,8 @@
+= 4.1.0 (unreleased) =
+
+* Added experimental support for fixing Windows-1252 characters
+  embedded in UTF-8 documents.
+
 = 4.0.5 (20120427) =
 
 * Added a new method, wrap(), which wraps an element in a tag. Renamed
diff --git a/bs4/dammit.py b/bs4/dammit.py
index ec62b99..a3301ee 100644
--- a/bs4/dammit.py
+++ b/bs4/dammit.py
@@ -587,3 +587,199 @@ class UnicodeDammit:
         b'\xfe' : 'b',
         b'\xff' : 'y',
         }
+
+    # A map used when removing rogue Windows-1252/ISO-8859-1
+    # characters in otherwise UTF-8 documents.
+    #
+    # Note that \x81, \x8d, \x8f, \x90, and \x9d are undefined in
+    # Windows-1252.
+    WINDOWS_1252_TO_UTF8 = {
+        0x80 : b'\xe2\x82\xac', # €
+        0x82 : b'\xe2\x80\x9a', # ‚
+        0x83 : b'\xc6\x92',     # ƒ
+        0x84 : b'\xe2\x80\x9e', # „
+        0x85 : b'\xe2\x80\xa6', # …
+        0x86 : b'\xe2\x80\xa0', # †
+        0x87 : b'\xe2\x80\xa1', # ‡
+        0x88 : b'\xcb\x86',     # ˆ
+        0x89 : b'\xe2\x80\xb0', # ‰
+        0x8a : b'\xc5\xa0',     # Š
+        0x8b : b'\xe2\x80\xb9', # ‹
+        0x8c : b'\xc5\x92',     # Œ
+        0x8e : b'\xc5\xbd',     # Ž
+        0x91 : b'\xe2\x80\x98', # ‘
+        0x92 : b'\xe2\x80\x99', # ’
+        0x93 : b'\xe2\x80\x9c', # “
+        0x94 : b'\xe2\x80\x9d', # ”
+        0x95 : b'\xe2\x80\xa2', # •
+        0x96 : b'\xe2\x80\x93', # –
+        0x97 : b'\xe2\x80\x94', # —
+        0x98 : b'\xcb\x9c',     # ˜
+        0x99 : b'\xe2\x84\xa2', # ™
+        0x9a : b'\xc5\xa1',     # š
+        0x9b : b'\xe2\x80\xba', # ›
+        0x9c : b'\xc5\x93',     # œ
+        0x9e : b'\xc5\xbe',     # ž
+        0x9f : b'\xc5\xb8',     # Ÿ
+        0xa0 : b'\xc2\xa0',     #  
+        0xa1 : b'\xc2\xa1',     # ¡
+        0xa2 : b'\xc2\xa2',     # ¢
+        0xa3 : b'\xc2\xa3',     # £
+        0xa4 : b'\xc2\xa4',     # ¤
+        0xa5 : b'\xc2\xa5',     # ¥
+        0xa6 : b'\xc2\xa6',     # ¦
+        0xa7 : b'\xc2\xa7',     # §
+        0xa8 : b'\xc2\xa8',     # ¨
+        0xa9 : b'\xc2\xa9',     # ©
+        0xaa : b'\xc2\xaa',     # ª
+        0xab : b'\xc2\xab',     # «
+        0xac : b'\xc2\xac',     # ¬
+        0xad : b'\xc2\xad',     # ­
+        0xae : b'\xc2\xae',     # ®
+        0xaf : b'\xc2\xaf',     # ¯
+        0xb0 : b'\xc2\xb0',     # °
+        0xb1 : b'\xc2\xb1',     # ±
+        0xb2 : b'\xc2\xb2',     # ²
+        0xb3 : b'\xc2\xb3',     # ³
+        0xb4 : b'\xc2\xb4',     # ´
+        0xb5 : b'\xc2\xb5',     # µ
+        0xb6 : b'\xc2\xb6',     # ¶
+        0xb7 : b'\xc2\xb7',     # ·
+        0xb8 : b'\xc2\xb8',     # ¸
+        0xb9 : b'\xc2\xb9',     # ¹
+        0xba : b'\xc2\xba',     # º
+        0xbb : b'\xc2\xbb',     # »
+        0xbc : b'\xc2\xbc',     # ¼
+        0xbd : b'\xc2\xbd',     # ½
+        0xbe : b'\xc2\xbe',     # ¾
+        0xbf : b'\xc2\xbf',     # ¿
+        0xc0 : b'\xc3\x80',     # À
+        0xc1 : b'\xc3\x81',     # Á
+        0xc2 : b'\xc3\x82',     # Â
+        0xc3 : b'\xc3\x83',     # Ã
+        0xc4 : b'\xc3\x84',     # Ä
+        0xc5 : b'\xc3\x85',     # Å
+        0xc6 : b'\xc3\x86',     # Æ
+        0xc7 : b'\xc3\x87',     # Ç
+        0xc8 : b'\xc3\x88',     # È
+        0xc9 : b'\xc3\x89',     # É
+        0xca : b'\xc3\x8a',     # Ê
+        0xcb : b'\xc3\x8b',     # Ë
+        0xcc : b'\xc3\x8c',     # Ì
+        0xcd : b'\xc3\x8d',     # Í
+        0xce : b'\xc3\x8e',     # Î
+        0xcf : b'\xc3\x8f',     # Ï
+        0xd0 : b'\xc3\x90',     # Ð
+        0xd1 : b'\xc3\x91',     # Ñ
+        0xd2 : b'\xc3\x92',     # Ò
+        0xd3 : b'\xc3\x93',     # Ó
+        0xd4 : b'\xc3\x94',     # Ô
+        0xd5 : b'\xc3\x95',     # Õ
+        0xd6 : b'\xc3\x96',     # Ö
+        0xd7 : b'\xc3\x97',     # ×
+        0xd8 : b'\xc3\x98',     # Ø
+        0xd9 : b'\xc3\x99',     # Ù
+        0xda : b'\xc3\x9a',     # Ú
+        0xdb : b'\xc3\x9b',     # Û
+        0xdc : b'\xc3\x9c',     # Ü
+        0xdd : b'\xc3\x9d',     # Ý
+        0xde : b'\xc3\x9e',     # Þ
+        0xdf : b'\xc3\x9f',     # ß
+        0xe0 : b'\xc3\xa0',     # à
+        0xe1 : b'\xa1',     # á
+        0xe2 : b'\xc3\xa2',     # â
+        0xe3 : b'\xc3\xa3',     # ã
+        0xe4 : b'\xc3\xa4',     # ä
+        0xe5 : b'\xc3\xa5',     # å
+        0xe6 : b'\xc3\xa6',     # æ
+        0xe7 : b'\xc3\xa7',     # ç
+        0xe8 : b'\xc3\xa8',     # è
+        0xe9 : b'\xc3\xa9',     # é
+        0xea : b'\xc3\xaa',     # ê
+        0xeb : b'\xc3\xab',     # ë
+        0xec : b'\xc3\xac',     # ì
+        0xed : b'\xc3\xad',     # í
+        0xee : b'\xc3\xae',     # î
+        0xef : b'\xc3\xaf',     # ï
+        0xf0 : b'\xc3\xb0',     # ð
+        0xf1 : b'\xc3\xb1',     # ñ
+        0xf2 : b'\xc3\xb2',     # ò
+        0xf3 : b'\xc3\xb3',     # ó
+        0xf4 : b'\xc3\xb4',     # ô
+        0xf5 : b'\xc3\xb5',     # õ
+        0xf6 : b'\xc3\xb6',     # ö
+        0xf7 : b'\xc3\xb7',     # ÷
+        0xf8 : b'\xc3\xb8',     # ø
+        0xf9 : b'\xc3\xb9',     # ù
+        0xfa : b'\xc3\xba',     # ú
+        0xfb : b'\xc3\xbb',     # û
+        0xfc : b'\xc3\xbc',     # ü
+        0xfd : b'\xc3\xbd',     # ý
+        0xfe : b'\xc3\xbe',     # þ
+        }
+
+    MULTIBYTE_MARKERS_AND_SIZES = [
+        (0xc2, 0xdf, 2), # 2-byte characters start with a byte C2-DF
+        (0xe0, 0xef, 3), # 3-byte characters start with E0-EF
+        (0xf0, 0xf4, 4), # 4-byte characters start with F0-F4
+        ]
+
+    FIRST_MULTIBYTE_MARKER = MULTIBYTE_MARKERS_AND_SIZES[0][0]
+    LAST_MULTIBYTE_MARKER = MULTIBYTE_MARKERS_AND_SIZES[-1][1]
+
+    @classmethod
+    def fix_embedded_windows_1252(cls, in_bytes, actual_encoding="utf8"):
+        """Fix Windows-1252 characters embedded in some other encoding.
+
+        Also fixes embedded ISO-8859-1, which is a subset of Windows-1252.
+
+        Currently the only encoding supported is UTF-8.
+
+        The input must be a bytestring. If you've already converted
+        the document to Unicode, you're too late.
+
+        The output is a bytestring in which Windows-1252 characters
+        have been converted to their UTF-8 equivalents.
+        """
+        if actual_encoding.lower() not in ('utf8', 'utf-8'):
+            raise NotImplementedError(
+                "UTF-8 is the only currently supported encoding "
+                "for Windows-1252 removal.")
+
+        byte_chunks = []
+
+        chunk_start = 0
+        pos = 0
+        while pos < len(in_bytes):
+            byte = in_bytes[pos]
+            if not isinstance(byte, int):
+                # Python 2.x
+                byte = ord(byte)
+            if byte >= cls.FIRST_MULTIBYTE_MARKER and byte <= cls.LAST_MULTIBYTE_MARKER:
+                # This is the start of a UTF-8 multibyte character. Skip
+                # to the end.
+                for start, end, size in cls.MULTIBYTE_MARKERS_AND_SIZES:
+                    if byte >= start and byte <= end:
+                        pos += size
+                        break
+            elif byte >= 0x80 and byte in cls.WINDOWS_1252_TO_UTF8:
+                # We found a Windows-1252 character!
+                # Save the string up to this point as a chunk.
+                byte_chunks.append(in_bytes[chunk_start:pos])
+
+                # Now translate the Windows-1252 character into UTF-8
+                # and add it as another, one-byte chunk.
+                byte_chunks.append(cls.WINDOWS_1252_TO_UTF8[byte])
+                pos += 1
+                chunk_start = pos
+            else:
+                # Go on to the next character.
+                pos += 1
+        if chunk_start == 0:
+            # The string is unchanged.
+            return in_bytes
+        else:
+            # Store the final chunk.
+            byte_chunks.append(in_bytes[chunk_start:])
+        return b''.join(byte_chunks)
+
diff --git a/bs4/tests/test_soup.py b/bs4/tests/test_soup.py
index bb97e52..ef58521 100644
--- a/bs4/tests/test_soup.py
+++ b/bs4/tests/test_soup.py
@@ -286,6 +286,44 @@ class TestUnicodeDammit(unittest.TestCase):
         self.assertEqual(u"<a>áé</a>", dammit.unicode_markup)
         self.assertEqual("utf-16le", dammit.original_encoding)
 
+    def test_fix_embedded_windows_1252(self):
+        # Here's a UTF8 document.
+        utf8 = (u"\N{SNOWMAN}" * 3).encode("utf8")
+
+        # Here's a Windows-1252 document.
+        windows_1252 = (
+            u"\N{LEFT DOUBLE QUOTATION MARK}Hi, I like Windows!"
+            u"\N{RIGHT DOUBLE QUOTATION MARK}").encode("windows_1252")
+
+        # Through some unholy alchemy, they've been stuck together.
+        doc = utf8 + windows_1252 + utf8
+
+        # The document can't be turned into UTF-8:
+        self.assertRaises(UnicodeDecodeError, doc.decode, "utf8")
+
+        # Unicode, Dammit thinks the whole document is Windows-1252,
+        # and decodes it into "☃☃☃“Hi, I like Windows!”☃☃☃"
+
+        # But if we run it through fix_embedded_windows_1252, it's fixed:
+
+        fixed = UnicodeDammit.fix_embedded_windows_1252(doc)
+        self.assertEqual(
+            u"☃☃☃“Hi, I like Windows!”☃☃☃", fixed.decode("utf8"))
+
+    def test_fix_embedded_windows_1252_ignores_multibyte_characters(self):
+        # Each of these characters has a UTF-8 representation ending
+        # in \x93. \x93 is a smart quote if interpreted as
+        # Windows-1252. But our code knows to skip over multibyte
+        # UTF-8 characters, so they'll survive the process unscathed.
+        for tricky_unicode_char in (
+            u"\N{LATIN SMALL LIGATURE OE}", # 2-byte char '\xc5\x93'
+            u"\N{LATIN SUBSCRIPT SMALL LETTER X}", # 3-byte char '\xe2\x82\x93'
+            u"\xf0\x90\x90\x93", # This is a CJK character, not sure which one.
+            ):
+            input = tricky_unicode_char.encode("utf8")
+            self.assertTrue(input.endswith(b'\x93'))
+            output = UnicodeDammit.fix_embedded_windows_1252(input)
+            self.assertEqual(output, input)
 
 class TestNamedspacedAttribute(SoupTest):