mirror of https://github.com/python/cpython
50 lines
1.3 KiB
Python
50 lines
1.3 KiB
Python
"""Test the parser and generator are inverses.
|
|
|
|
Note that this is only strictly true if we are parsing RFC valid messages and
|
|
producing RFC valid messages.
|
|
"""
|
|
|
|
import io
|
|
import unittest
|
|
from email import policy, message_from_bytes
|
|
from email.generator import BytesGenerator
|
|
from test.test_email import TestEmailBase, parameterize
|
|
|
|
# This is like textwrap.dedent for bytes, except that it uses \r\n for the line
|
|
# separators on the rebuilt string.
|
|
def dedent(bstr):
|
|
lines = bstr.splitlines()
|
|
if not lines[0].strip():
|
|
raise ValueError("First line must contain text")
|
|
stripamt = len(lines[0]) - len(lines[0].lstrip())
|
|
return b'\r\n'.join(
|
|
[x[stripamt:] if len(x)>=stripamt else b''
|
|
for x in lines])
|
|
|
|
|
|
@parameterize
|
|
class TestInversion(TestEmailBase, unittest.TestCase):
|
|
|
|
def msg_as_input(self, msg):
|
|
m = message_from_bytes(msg, policy=policy.SMTP)
|
|
b = io.BytesIO()
|
|
g = BytesGenerator(b)
|
|
g.flatten(m)
|
|
self.assertEqual(b.getvalue(), msg)
|
|
|
|
# XXX: spaces are not preserved correctly here yet in the general case.
|
|
msg_params = {
|
|
'header_with_one_space_body': (dedent(b"""\
|
|
From: abc@xyz.com
|
|
X-Status:\x20
|
|
Subject: test
|
|
|
|
foo
|
|
"""),),
|
|
|
|
}
|
|
|
|
|
|
if __name__ == '__main__':
|
|
unittest.main()
|