[FIX] base: avoid bpo-35805 corrupting Message-Id

Python 3 before 3.8 has a bug that causes the email.policy classes to
incorrectly fold and RFC2047-encode "identification fields" in email
messages. This mainly applies to Message-Id, References, and In-Reply-To
fields.

We are impacted by this bug since odoo/odoo#35929 where we switched to
using the "modern" email.message API.

RFC2047 section 5 clearly states that those headers/fields are not to be
encoded, and that would violate RFC5322.

Further, such a folded Message-Id is considered non-RFC-conformant by
popular MTAs (GMail, Outlook), which will then generate *another*
Message-Id field, causing the original threading information to be lost.
Replies to such a modified message will reference the new, unknown
Message-Id, and won't be attached to the original thread.

The solution we adopt here is to monkey-patch the SMTP policies to
special-case those identification fields and deactivate the automatic
folding, until the bug is properly and fully fixed in the standard lib.

Some considerations taken into account for this patch:

- `email.policy.SMTP` is being monkey-patched globally to make sure we
  fix all possible places where Messages are being encoded/folded
- the fix is **not** made version-specific, considering that even in Python
  3.8 the official bugfix only applies to Message-Id, but still fails to
  protect other identification fields, like *References* and
  *In-Reply-To*. The author specifically noted that shortcoming [2].
  The fix wouldn't break anything on Python 3.8 anyway.
- the `noFoldPolicy` trick for preventing folding is done with no max
  line length at all. RFC5322, section 2.1.1 states [3] that the maximum
  length is 998 due to legacy implementations, but there is no provision
  to wrap identification fields that are longer than that. Wrapping at
  998 chars would corrupt the header anyway. We'll just count on the
  fact that we don't usually need 1k+ chars in those headers.

The invalid folding/encoding in action on Python 3.6 (in Python 3.8 only
the second header gets folded):

```py
>>> msg = email.message.EmailMessage(policy=email.policy.SMTP)
>>> msg['Message-Id'] = '<929227342217024.1596730490.324691772460938-example-30661-some.reference@test-123.example.com>'
>>> msg['In-Reply-To'] = '<92922734221723.1596730568.324691772460444-another-30661-parent.reference@test-123.example.com>'
>>> print(msg.as_string())
Message-Id: =?utf-8?q?=3C929227342217024=2E1596730490=2E324691772460938-exam?=
 =?utf-8?q?ple-30661-some=2Ereference=40test-123=2Eexample=2Ecom=3E?=
In-Reply-To: =?utf-8?q?=3C92922734221723=2E1596730568=2E324691772460444-anot?=
 =?utf-8?q?her-30661-parent=2Ereference=40test-123=2Eexample=2Ecom=3E?=

```

and the expected result after the fix:
```py
>>> msg = email.message.EmailMessage(policy=email.policy.SMTP)
>>> msg['Message-Id'] = '<929227342217024.1596730490.324691772460938-example-30661-some.reference@test-123.example.com>'
>>> msg['In-Reply-To'] = '<92922734221723.1596730568.324691772460444-another-30661-parent.reference@test-123.example.com>'
>>> print(msg.as_string())
Message-Id: <929227342217024.1596730490.324691772460938-example-30661-some.reference@test-123.example.com>
In-Reply-To: <92922734221723.1596730568.324691772460444-another-30661-parent.reference@test-123.example.com>

```

[1] bpo-35805: https://bugs.python.org/issue35805
[2] https://github.com/python/cpython/pull/13397#issuecomment-493618544
[3] https://tools.ietf.org/html/rfc5322#section-2.1.1

closes odoo/odoo#55656

X-original-commit: 02b78770147e2eda65a76d29c6f2fc3278581b22
Signed-off-by: Olivier Dony (odo) <odo@openerp.com>

1 files changed, 15 insertions, 0 deletions

diff --git a/odoo/addons/base/models/ir_mail_server.py b/odoo/addons/base/models/ir_mail_server.py
index 9c72e54fdbb..ad8f03ca637 100644
--- a/odoo/addons/base/models/ir_mail_server.py
+++ b/odoo/addons/base/models/ir_mail_server.py
@@ -33,6 +33,21 @@ def _print_debug(self, *args):
     _logger.debug(' '.join(str(a) for a in args))
 smtplib.SMTP._print_debug = _print_debug
 
+# Python 3: workaround for bpo-35805, only partially fixed in Python 3.8.
+RFC5322_IDENTIFICATION_HEADERS = {'message-id', 'in-reply-to', 'references', 'resent-msg-id'}
+_noFoldPolicy = email.policy.SMTP.clone(max_line_length=None)
+class IdentificationFieldsNoFoldPolicy(email.policy.EmailPolicy):
+    # Override _fold() to avoid folding identification fields, excluded by RFC2047 section 5
+    # These are particularly important to preserve, as MTAs will often rewrite non-conformant
+    # Message-ID headers, causing a loss of thread information (replies are lost)
+    def _fold(self, name, value, *args, **kwargs):
+        if name.lower() in RFC5322_IDENTIFICATION_HEADERS:
+            return _noFoldPolicy._fold(name, value, *args, **kwargs)
+        return super()._fold(name, value, *args, **kwargs)
+
+# Global monkey-patch for our preferred SMTP policy, preserving the non-default linesep
+email.policy.SMTP = IdentificationFieldsNoFoldPolicy(linesep=email.policy.SMTP.linesep)
+
 # Python 2: replace smtplib's stderr
 class WriteToLogger(object):
     def write(self, s):