Fixing CI/CD pipeline

This commit is contained in:
Miroslav Štampar
2026-06-28 19:03:21 +02:00
parent 2297c81309
commit 4c869817d4
10 changed files with 41 additions and 22 deletions
+3 -2
View File
@@ -455,6 +455,7 @@ class TestParseRequestFileBurp(unittest.TestCase):
self._method = conf.method
self._headers = conf.headers
conf.scope = None
conf.method = None # avoid a leaked conf.method overriding the parsed verb
def tearDown(self):
conf.scope = self._scope
@@ -974,7 +975,7 @@ class TestGetPageWordSet(unittest.TestCase):
class TestNormalizeUnicode(unittest.TestCase):
def test_accents_stripped(self):
# normalizeUnicode collapses accented chars to their ASCII base
self.assertEqual(normalizeUnicode(u"éè"), "ee")
self.assertEqual(normalizeUnicode(u"\xe9\xe8"), "ee")
def test_plain_ascii_unchanged(self):
self.assertEqual(normalizeUnicode(u"abc123"), "abc123")
@@ -1342,7 +1343,7 @@ class TestCommonDecodeIntToUnicode(unittest.TestCase):
from lib.core.common import decodeIntToUnicode
set_dbms(DBMS.PGSQL)
# value > 255 on PGSQL takes the _unichr(value) branch
self.assertEqual(decodeIntToUnicode(0x2122), u"™")
self.assertEqual(decodeIntToUnicode(0x2122), u"\u2122")
class TestCommonDecodeDbmsHex(unittest.TestCase):