robustness against unicode errors on encoding detection

Cython/Compiler/Main.py

@@ -139,8 +139,10 @@ class Context:
 
     def parse(self, source_filename, type_names, pxd, full_module_name):
         # Parse the given source file and return a parse tree.
+        try:
             f = Utils.open_source_file(source_filename, "rU")
 
+            try:
                 if isinstance(source_filename, unicode):
                     name = source_filename
                 else:
@@ -149,15 +151,13 @@ class Context:
                         filename_encoding = sys.getdefaultencoding()
                     name = source_filename.decode(filename_encoding)
 
-        try:
-            try:
                     s = PyrexScanner(f, name, source_encoding = f.encoding,
                                      type_names = type_names, context = self)
                     tree = Parsing.p_module(s, pxd, full_module_name)
-            except UnicodeDecodeError, msg:
-                error((name, 0, 0), "Decoding error, missing or incorrect coding=<encoding-name> at top of source (%s)" % msg)
             finally:
                 f.close()
+        except UnicodeDecodeError, msg:
+            error((source_filename, 0, 0), "Decoding error, missing or incorrect coding=<encoding-name> at top of source (%s)" % msg)
         if Errors.num_errors > 0:
             raise CompileError
         return tree

Cython/Utils.py

@@ -41,12 +41,15 @@ def detect_file_encoding(source_filename):
     # PEPs 263 and 3120
     f = codecs.open(source_filename, "rU", encoding="UTF-8")
     try:
-        for line_no, line in enumerate(f):
-            encoding = _match_file_encoding(line)
+        chars = []
+        for i in range(2):
+            c = f.read(1)
+            while c and c != '\n':
+                chars.append(c)
+                c = f.read(1)
+            encoding = _match_file_encoding(u''.join(chars))
             if encoding:
                 return encoding.group(1)
-            if line_no == 1:
-                break
     finally:
         f.close()
     return "UTF-8"