don't mask encoding errors when decoding a string #6289

Lib/test/test_coding.py

@@ -21,6 +21,18 @@ class CodingTest(unittest.TestCase):
         fp.close()
         self.assertRaises(SyntaxError, compile, text, filename, 'exec')
 
+    def test_error_from_string(self):
+        # See http://bugs.python.org/issue6289
+        input = u"# coding: ascii\n\N{SNOWMAN}".encode('utf-8')
+        try:
+            compile(input, "<string>", "exec")
+        except SyntaxError as e:
+            expected = "'ascii' codec can't decode byte 0xe2 in position 16: " \
+                "ordinal not in range(128)"
+            self.assertTrue(str(e).startswith(expected))
+        else:
+            self.fail("didn't raise")
+
 def test_main():
     test.test_support.run_unittest(CodingTest)
 

Misc/NEWS

@@ -12,6 +12,8 @@ What's New in Python 2.7 alpha 1
 Core and Builtins
 -----------------
 
+- Issue #6289: Encoding errors from compile() were being masked.
+
 - When no module is given in a relative import, the module field of the
   ImportFrom AST node is now None instead of an empty string.
 

Parser/tokenizer.c

@@ -619,11 +619,8 @@ decode_str(const char *str, struct tok_state *tok)
 	if (tok->enc != NULL) {
 		assert(utf8 == NULL);
 		utf8 = translate_into_utf8(str, tok->enc);
-		if (utf8 == NULL) {
-			PyErr_Format(PyExc_SyntaxError,
-				"unknown encoding: %s", tok->enc);
+		if (utf8 == NULL)
 			return error_ret(tok);
-		}
 		str = PyString_AsString(utf8);
 	}
 #endif