Mailgun: Fix lost attachments with non-ASCII filenames.

Workaround requests/requests#4652 (urllib3/urllib3#303), where
uploaded files in multipart/form-data are improperly given RFC 2231
encoded filenames. That format is not accepted by Mailgun's API (and is
prohibited by RFC 7578), resulting in the attachments being silently
dropped.

Fix is to patch up the multipart/form-data before posting to remove
the RFC 2231 encoding.

Fixes #125

anymail/backends/mailgun.py

@@ -1,4 +1,7 @@
 from datetime import datetime
+from email.utils import encode_rfc2231
+
+from requests import Request
 
 from ..exceptions import AnymailRequestsAPIError, AnymailError
 from ..message import AnymailRecipientStatus
@@ -78,6 +81,35 @@ class MailgunPayload(RequestsPayload):
                                backend=self.backend, email_message=self.message, payload=self)
         return "%s/messages" % self.sender_domain
 
+    def get_request_params(self, api_url):
+        params = super(MailgunPayload, self).get_request_params(api_url)
+        non_ascii_filenames = [filename
+                               for (field, (filename, content, mimetype)) in params["files"]
+                               if filename is not None and not isascii(filename)]
+        if non_ascii_filenames:
+            # Workaround https://github.com/requests/requests/issues/4652:
+            # Mailgun expects RFC 7578 compliant multipart/form-data, and is confused
+            # by Requests/urllib3's improper use of RFC 2231 encoded filename parameters
+            # ("filename*=utf-8''...") in Content-Disposition headers.
+            # The workaround is to pre-generate the (non-compliant) form-data body, and
+            # replace 'filename*={RFC 2231 encoded}' with 'filename="{UTF-8 bytes}"'.
+            # Replace _only_ the filenames that will be problems (not all "filename*=...")
+            # to minimize potential side effects--e.g., in attached messages that might
+            # have their own attachments with (correctly) RFC 2231 encoded filenames.
+            prepared = Request(**params).prepare()
+            form_data = prepared.body  # bytes
+            for filename in non_ascii_filenames:  # text
+                rfc2231_filename = encode_rfc2231(  # wants a str (text in PY3, bytes in PY2)
+                    filename if isinstance(filename, str) else filename.encode("utf-8"),
+                    charset="utf-8")
+                form_data = form_data.replace(
+                    b'filename*=%s' % rfc2231_filename.encode("utf-8"),
+                    b'filename="%s"' % filename.encode("utf-8"))
+            params["data"] = form_data
+            params["headers"]["Content-Type"] = prepared.headers["Content-Type"]  # multipart/form-data; boundary=...
+            params["files"] = None  # these are now in the form_data body
+        return params
+
     def serialize_data(self):
         self.populate_recipient_variables()
         return self.data
@@ -113,6 +145,7 @@ class MailgunPayload(RequestsPayload):
     def init_payload(self):
         self.data = {}   # {field: [multiple, values]}
         self.files = []  # [(field, multiple), (field, values)]
+        self.headers = {}
 
     def set_from_email_list(self, emails):
         # Mailgun supports multiple From email addresses
@@ -204,3 +237,15 @@ class MailgunPayload(RequestsPayload):
         # Allow override of sender_domain via esp_extra
         # (but pop it out of params to send to Mailgun)
         self.sender_domain = self.data.pop("sender_domain", self.sender_domain)
+
+
+def isascii(s):
+    """Returns True if str s is entirely ASCII characters.
+
+    (Compare to Python 3.7 `str.isascii()`.)
+    """
+    try:
+        s.encode("ascii")
+    except UnicodeEncodeError:
+        return False
+    return True