mirror of
https://github.com/CloverHackyColor/CloverBootloader.git
synced 2024-12-28 17:08:18 +01:00
b1264ef1e3
Signed-off-by: Sergey Isakov <isakov-sl@bk.ru>
176 lines
5.6 KiB
Python
Executable File
176 lines
5.6 KiB
Python
Executable File
## @file
|
|
# Unit tests for AutoGen.UniClassObject
|
|
#
|
|
# Copyright (c) 2015, Intel Corporation. All rights reserved.<BR>
|
|
#
|
|
# SPDX-License-Identifier: BSD-2-Clause-Patent
|
|
#
|
|
|
|
##
|
|
# Import Modules
|
|
#
|
|
import os
|
|
import unittest
|
|
|
|
import codecs
|
|
|
|
import TestTools
|
|
|
|
from Common.Misc import PathClass
|
|
import AutoGen.UniClassObject as BtUni
|
|
|
|
from Common import EdkLogger
|
|
EdkLogger.InitializeForUnitTest()
|
|
|
|
class Tests(TestTools.BaseToolsTest):
|
|
|
|
SampleData = u'''
|
|
#langdef en-US "English"
|
|
#string STR_A #language en-US "STR_A for en-US"
|
|
'''
|
|
|
|
def EncodeToFile(self, encoding, string=None):
|
|
if string is None:
|
|
string = self.SampleData
|
|
if encoding is not None:
|
|
data = codecs.encode(string, encoding)
|
|
else:
|
|
data = string
|
|
path = 'input.uni'
|
|
self.WriteTmpFile(path, data)
|
|
return PathClass(self.GetTmpFilePath(path))
|
|
|
|
def ErrorFailure(self, error, encoding, shouldPass):
|
|
msg = error + ' should '
|
|
if shouldPass:
|
|
msg += 'not '
|
|
msg += 'be generated for '
|
|
msg += '%s data in a .uni file' % encoding
|
|
self.fail(msg)
|
|
|
|
def UnicodeErrorFailure(self, encoding, shouldPass):
|
|
self.ErrorFailure('UnicodeError', encoding, shouldPass)
|
|
|
|
def EdkErrorFailure(self, encoding, shouldPass):
|
|
self.ErrorFailure('EdkLogger.FatalError', encoding, shouldPass)
|
|
|
|
def CheckFile(self, encoding, shouldPass, string=None):
|
|
path = self.EncodeToFile(encoding, string)
|
|
try:
|
|
BtUni.UniFileClassObject([path])
|
|
if shouldPass:
|
|
return
|
|
except UnicodeError:
|
|
if not shouldPass:
|
|
return
|
|
else:
|
|
self.UnicodeErrorFailure(encoding, shouldPass)
|
|
except EdkLogger.FatalError:
|
|
if not shouldPass:
|
|
return
|
|
else:
|
|
self.EdkErrorFailure(encoding, shouldPass)
|
|
except Exception:
|
|
pass
|
|
|
|
self.EdkErrorFailure(encoding, shouldPass)
|
|
|
|
def testUtf16InUniFile(self):
|
|
self.CheckFile('utf_16', shouldPass=True)
|
|
|
|
def testSupplementaryPlaneUnicodeCharInUtf16File(self):
|
|
#
|
|
# Supplementary Plane characters can exist in UTF-16 files,
|
|
# but they are not valid UCS-2 characters.
|
|
#
|
|
# This test makes sure that BaseTools rejects these characters
|
|
# if seen in a .uni file.
|
|
#
|
|
data = u'''
|
|
#langdef en-US "English"
|
|
#string STR_A #language en-US "CodePoint (\U00010300) > 0xFFFF"
|
|
'''
|
|
|
|
self.CheckFile('utf_16', shouldPass=False, string=data)
|
|
|
|
def testSurrogatePairUnicodeCharInUtf16File(self):
|
|
#
|
|
# Surrogate Pair code points are used in UTF-16 files to
|
|
# encode the Supplementary Plane characters. But, a Surrogate
|
|
# Pair code point which is not followed by another Surrogate
|
|
# Pair code point might be interpreted as a single code point
|
|
# with the Surrogate Pair code point.
|
|
#
|
|
# This test makes sure that BaseTools rejects these characters
|
|
# if seen in a .uni file.
|
|
#
|
|
data = codecs.BOM_UTF16_LE + b'//\x01\xd8 '
|
|
|
|
self.CheckFile(encoding=None, shouldPass=False, string=data)
|
|
|
|
def testValidUtf8File(self):
|
|
self.CheckFile(encoding='utf_8', shouldPass=True)
|
|
|
|
def testValidUtf8FileWithBom(self):
|
|
#
|
|
# Same test as testValidUtf8File, but add the UTF-8 BOM
|
|
#
|
|
data = codecs.BOM_UTF8 + codecs.encode(self.SampleData, 'utf_8')
|
|
|
|
self.CheckFile(encoding=None, shouldPass=True, string=data)
|
|
|
|
def test32bitUnicodeCharInUtf8File(self):
|
|
data = u'''
|
|
#langdef en-US "English"
|
|
#string STR_A #language en-US "CodePoint (\U00010300) > 0xFFFF"
|
|
'''
|
|
|
|
self.CheckFile('utf_16', shouldPass=False, string=data)
|
|
|
|
def test32bitUnicodeCharInUtf8File(self):
|
|
data = u'''
|
|
#langdef en-US "English"
|
|
#string STR_A #language en-US "CodePoint (\U00010300) > 0xFFFF"
|
|
'''
|
|
|
|
self.CheckFile('utf_8', shouldPass=False, string=data)
|
|
|
|
def test32bitUnicodeCharInUtf8Comment(self):
|
|
data = u'''
|
|
// Even in comments, we reject non-UCS-2 chars: \U00010300
|
|
#langdef en-US "English"
|
|
#string STR_A #language en-US "A"
|
|
'''
|
|
|
|
self.CheckFile('utf_8', shouldPass=False, string=data)
|
|
|
|
def testSurrogatePairUnicodeCharInUtf8File(self):
|
|
#
|
|
# Surrogate Pair code points are used in UTF-16 files to
|
|
# encode the Supplementary Plane characters. In UTF-8, it is
|
|
# trivial to encode these code points, but they are not valid
|
|
# code points for characters, since they are reserved for the
|
|
# UTF-16 Surrogate Pairs.
|
|
#
|
|
# This test makes sure that BaseTools rejects these characters
|
|
# if seen in a .uni file.
|
|
#
|
|
data = b'\xed\xa0\x81'
|
|
|
|
self.CheckFile(encoding=None, shouldPass=False, string=data)
|
|
|
|
def testSurrogatePairUnicodeCharInUtf8FileWithBom(self):
|
|
#
|
|
# Same test as testSurrogatePairUnicodeCharInUtf8File, but add
|
|
# the UTF-8 BOM
|
|
#
|
|
data = codecs.BOM_UTF8 + b'\xed\xa0\x81'
|
|
|
|
self.CheckFile(encoding=None, shouldPass=False, string=data)
|
|
|
|
TheTestSuite = TestTools.MakeTheTestSuite(locals())
|
|
|
|
if __name__ == '__main__':
|
|
allTests = TheTestSuite()
|
|
unittest.TextTestRunner().run(allTests)
|