mirror of https://github.com/python/cpython
164 lines
5.7 KiB
Python
164 lines
5.7 KiB
Python
r"""Test correct treatment of various string literals by the parser.
|
|
|
|
There are four types of string literals:
|
|
|
|
'abc' -- normal str
|
|
r'abc' -- raw str
|
|
b'xyz' -- normal bytes
|
|
br'xyz' | rb'xyz' -- raw bytes
|
|
|
|
The difference between normal and raw strings is of course that in a
|
|
raw string, \ escapes (while still used to determine the end of the
|
|
literal) are not interpreted, so that r'\x00' contains four
|
|
characters: a backslash, an x, and two zeros; while '\x00' contains a
|
|
single character (code point zero).
|
|
|
|
The tricky thing is what should happen when non-ASCII bytes are used
|
|
inside literals. For bytes literals, this is considered illegal. But
|
|
for str literals, those bytes are supposed to be decoded using the
|
|
encoding declared for the file (UTF-8 by default).
|
|
|
|
We have to test this with various file encodings. We also test it with
|
|
exec()/eval(), which uses a different code path.
|
|
|
|
This file is really about correct treatment of encodings and
|
|
backslashes. It doesn't concern itself with issues like single
|
|
vs. double quotes or singly- vs. triply-quoted strings: that's dealt
|
|
with elsewhere (I assume).
|
|
"""
|
|
|
|
import os
|
|
import sys
|
|
import shutil
|
|
import tempfile
|
|
import unittest
|
|
import test.support
|
|
|
|
|
|
TEMPLATE = r"""# coding: %s
|
|
a = 'x'
|
|
assert ord(a) == 120
|
|
b = '\x01'
|
|
assert ord(b) == 1
|
|
c = r'\x01'
|
|
assert list(map(ord, c)) == [92, 120, 48, 49]
|
|
d = '\x81'
|
|
assert ord(d) == 0x81
|
|
e = r'\x81'
|
|
assert list(map(ord, e)) == [92, 120, 56, 49]
|
|
f = '\u1881'
|
|
assert ord(f) == 0x1881
|
|
g = r'\u1881'
|
|
assert list(map(ord, g)) == [92, 117, 49, 56, 56, 49]
|
|
"""
|
|
|
|
|
|
def byte(i):
|
|
return bytes([i])
|
|
|
|
|
|
class TestLiterals(unittest.TestCase):
|
|
|
|
def setUp(self):
|
|
self.save_path = sys.path[:]
|
|
self.tmpdir = tempfile.mkdtemp()
|
|
sys.path.insert(0, self.tmpdir)
|
|
|
|
def tearDown(self):
|
|
sys.path = self.save_path
|
|
shutil.rmtree(self.tmpdir, ignore_errors=True)
|
|
|
|
def test_template(self):
|
|
# Check that the template doesn't contain any non-printables
|
|
# except for \n.
|
|
for c in TEMPLATE:
|
|
assert c == '\n' or ' ' <= c <= '~', repr(c)
|
|
|
|
def test_eval_str_normal(self):
|
|
self.assertEqual(eval(""" 'x' """), 'x')
|
|
self.assertEqual(eval(r""" '\x01' """), chr(1))
|
|
self.assertEqual(eval(""" '\x01' """), chr(1))
|
|
self.assertEqual(eval(r""" '\x81' """), chr(0x81))
|
|
self.assertEqual(eval(""" '\x81' """), chr(0x81))
|
|
self.assertEqual(eval(r""" '\u1881' """), chr(0x1881))
|
|
self.assertEqual(eval(""" '\u1881' """), chr(0x1881))
|
|
|
|
def test_eval_str_raw(self):
|
|
self.assertEqual(eval(""" r'x' """), 'x')
|
|
self.assertEqual(eval(r""" r'\x01' """), '\\' + 'x01')
|
|
self.assertEqual(eval(""" r'\x01' """), chr(1))
|
|
self.assertEqual(eval(r""" r'\x81' """), '\\' + 'x81')
|
|
self.assertEqual(eval(""" r'\x81' """), chr(0x81))
|
|
self.assertEqual(eval(r""" r'\u1881' """), '\\' + 'u1881')
|
|
self.assertEqual(eval(""" r'\u1881' """), chr(0x1881))
|
|
|
|
def test_eval_bytes_normal(self):
|
|
self.assertEqual(eval(""" b'x' """), b'x')
|
|
self.assertEqual(eval(r""" b'\x01' """), byte(1))
|
|
self.assertEqual(eval(""" b'\x01' """), byte(1))
|
|
self.assertEqual(eval(r""" b'\x81' """), byte(0x81))
|
|
self.assertRaises(SyntaxError, eval, """ b'\x81' """)
|
|
self.assertEqual(eval(r""" b'\u1881' """), b'\\' + b'u1881')
|
|
self.assertRaises(SyntaxError, eval, """ b'\u1881' """)
|
|
|
|
def test_eval_bytes_raw(self):
|
|
self.assertEqual(eval(""" br'x' """), b'x')
|
|
self.assertEqual(eval(""" rb'x' """), b'x')
|
|
self.assertEqual(eval(r""" br'\x01' """), b'\\' + b'x01')
|
|
self.assertEqual(eval(r""" rb'\x01' """), b'\\' + b'x01')
|
|
self.assertEqual(eval(""" br'\x01' """), byte(1))
|
|
self.assertEqual(eval(""" rb'\x01' """), byte(1))
|
|
self.assertEqual(eval(r""" br'\x81' """), b"\\" + b"x81")
|
|
self.assertEqual(eval(r""" rb'\x81' """), b"\\" + b"x81")
|
|
self.assertRaises(SyntaxError, eval, """ br'\x81' """)
|
|
self.assertRaises(SyntaxError, eval, """ rb'\x81' """)
|
|
self.assertEqual(eval(r""" br'\u1881' """), b"\\" + b"u1881")
|
|
self.assertEqual(eval(r""" rb'\u1881' """), b"\\" + b"u1881")
|
|
self.assertRaises(SyntaxError, eval, """ br'\u1881' """)
|
|
self.assertRaises(SyntaxError, eval, """ rb'\u1881' """)
|
|
self.assertRaises(SyntaxError, eval, """ bb'' """)
|
|
self.assertRaises(SyntaxError, eval, """ rr'' """)
|
|
self.assertRaises(SyntaxError, eval, """ brr'' """)
|
|
self.assertRaises(SyntaxError, eval, """ bbr'' """)
|
|
self.assertRaises(SyntaxError, eval, """ rrb'' """)
|
|
self.assertRaises(SyntaxError, eval, """ rbb'' """)
|
|
|
|
def check_encoding(self, encoding, extra=""):
|
|
modname = "xx_" + encoding.replace("-", "_")
|
|
fn = os.path.join(self.tmpdir, modname + ".py")
|
|
f = open(fn, "w", encoding=encoding)
|
|
try:
|
|
f.write(TEMPLATE % encoding)
|
|
f.write(extra)
|
|
finally:
|
|
f.close()
|
|
__import__(modname)
|
|
del sys.modules[modname]
|
|
|
|
def test_file_utf_8(self):
|
|
extra = "z = '\u1234'; assert ord(z) == 0x1234\n"
|
|
self.check_encoding("utf-8", extra)
|
|
|
|
def test_file_utf_8_error(self):
|
|
extra = "b'\x80'\n"
|
|
self.assertRaises(SyntaxError, self.check_encoding, "utf-8", extra)
|
|
|
|
def test_file_utf8(self):
|
|
self.check_encoding("utf-8")
|
|
|
|
def test_file_iso_8859_1(self):
|
|
self.check_encoding("iso-8859-1")
|
|
|
|
def test_file_latin_1(self):
|
|
self.check_encoding("latin-1")
|
|
|
|
def test_file_latin9(self):
|
|
self.check_encoding("latin9")
|
|
|
|
|
|
def test_main():
|
|
test.support.run_unittest(__name__)
|
|
|
|
if __name__ == "__main__":
|
|
test_main()
|