import python_strict_unicode from sys import version_info test_bytes = 'hello \x01world\x99' BYTES = 'BYTES' test_unicode = u'h\udce9llo w\u00f6rld' # Python < 2.6 rejects the b prefix for byte string literals as a SyntaxError, # so instead create Python3 bytes objects by encoding unicode strings as # latin-1, which maps code points 0-255 directly to the corresponding bytes. if version_info[0] >= 3: test_bytes = test_bytes.encode('latin-1') BYTES = BYTES.encode('latin-1') # Test that byte string inputs and outputs work as expected bdbl = python_strict_unicode.double_str(test_bytes) if bdbl != test_bytes + test_bytes: raise RuntimeError("Failed to double string") if type(bdbl) != type(BYTES): raise RuntimeError("Wrong type output for string") bout = python_strict_unicode.same_str(test_bytes) if bout != test_bytes: raise RuntimeError("Failed to copy char*") if type(bout) != type(BYTES): raise RuntimeError("Wrong type output for char*") # Test that unicode string inputs and outputs work as expected udbl = python_strict_unicode.double_wstr(test_unicode) if udbl != test_unicode + test_unicode: raise RuntimeError("Failed to double wide string") if type(udbl) != type(u''): raise RuntimeError("Wrong type output for wide string") uout = python_strict_unicode.same_wstr(test_unicode) if uout != test_unicode: raise RuntimeError("Failed to copy wchar_t*") if type(uout) != type(u''): raise RuntimeError("Wrong type output for wchar_t*") # Test that overloading is handled properly bovr = python_strict_unicode.overload(test_bytes) if bovr != BYTES: raise RuntimeError("Failed to return bytes from overload") if type(bovr) != type(BYTES): raise RuntimeError("Wrong type output from overload") uovr = python_strict_unicode.overload(test_unicode) if uovr != u'UNICODE': raise RuntimeError("Failed to return unicode from overload") if type(uovr) != type(u''): raise RuntimeERror("Wrong type output from overload") # Test that bytes aren't accepted as wide strings and unicode isn't accepted as narrow strings try: python_strict_unicode.double_str(test_unicode) error = 1 except TypeError: error = 0 if error: raise RuntimeError("Unicode accepted for string") try: python_strict_unicode.same_str(test_unicode) error = 1 except TypeError: error = 0 if error: raise RuntimeError("Unicode accepted for char*") try: python_strict_unicode.double_wstr(test_bytes) error = 1 except TypeError: error = 0 if error: raise RuntimeError("Bytes accepted for wstring") try: python_strict_unicode.same_wstr(test_bytes) error = 1 except TypeError: error = 0 if error: raise RuntimeError("Bytes accepted for wchar_t*")