Mailgun merges user-variables (metadata) into the webhook post data
interspersed with the actual event params. This can lead to ambiguity
interpreting post data.
To extract metadata from an event, Anymail had been attempting to avoid
that ambiguity by instead using X-Mailgun-Variables fields found in the
event's message-headers param. But message-headers isn't included in
some tracking events (opened, clicked, unsubscribed), resulting in
empty metadata for those events. (#76)
Also, conflicting metadata keys could confuse Anymail's Mailgun event
parsing, leading to unexpected values in the normalized event. (#77)
This commit:
* Cleans up Anymail's tracking webhook to be explicit about which
multi-value params it uses, avoiding conflicts with metadata keys.
Fixes#77.
* Extracts metadata from post params for opened, clicked and
unsubscribed events. All unknown event params are assumed to be
metadata. Fixes#76.
* Documents a few metadata key names where it's impossible (or likely
to be unreliable) for Anymail to extract metadata from the post data.
For reference, the order of params in the Mailgun's post data *appears*
to be (from live testing):
* For the timestamp, token and signature params, any user-variable with
the same name appears *before* the corresponding event data.
* For all other params, any user-variable with the same name as a
Mailgun event param appears *after* the Mailgun data.
Update internal-use ParsedEmail to be more like Python 3.6+
email.headerregistry.Address, and remove "internal use only"
recommendation.
(Prep for exposing inbound email headers in a convenient form.
Old names remain temporarily available for internal use;
should clean up at some point.)
SendGrid requires extra headers and metadata values be strings.
Anymail has always coerced int and float; this treats Python 2's
`long` integer type the same.
Fixes#74
* Change Anymail's test EmailBackend to collect sent messages in
django.core.mail.outbox, same as Django's own locmem EmailBackend.
(So Django's test runner will automatically clear accumulated mail
between test cases.)
* Rename EmailMessage `test_response` attr to `anymail_test_response`
to avoid conflicts, and record merged ESP send params in
new `anymail_send_params` attr.
* Add docs
Closes#36.
[RFC-5322 allows](https://tools.ietf.org/html/rfc5322#section-3.6.2)
multiple addresses in the From header.
Django's SMTP backend supports this, as a single comma-separated
string (*not* a list of strings like the recipient params):
from_email='one@example.com, two@example.com'
to=['one@example.com', 'two@example.com']
Both Mailgun and SparkPost support multiple From addresses
(and Postmark accepts them, though truncates to the first one
on their end). For compatibility with Django -- and because
Anymail attempts to support all ESP features -- Anymail now
allows multiple From addresses, too, for ESPs that support it.
Note: as a practical matter, deliverability with multiple
From addresses is pretty bad. (Google outright rejects them.)
This change also reworks Anymail's internal ParsedEmail object,
and approach to parsing addresses, for better consistency with
Django's SMTP backend and improved error messaging.
In particular, Django (and now Anymail) allows multiple email
addresses in a single recipient string:
to=['one@example.com', 'two@example.com, three@example.com']
len(to) == 2 # but there will be three recipients
Fixes#60
Issue a better error message if message.reply_to
is set to a single string.
(Would also like to do this for to, cc, and bcc,
but Django core EmailMessage.recipients is called
and stumbles over thoses cases before Anymail's
backend gets involved.)
Fixes#57
Track change to Mailgun's events API, which
no longer includes message recipients.
(Only affected check for successful send
in the integration tests; Anymail doesn't
use the events API outside test code.)
Fixes#58
Only real problem is in json serialization tests:
Python 3.6 [changed][1] the json serialization
error message to use the object's class name
rather than its repr. E.g.:
"Decimal('19.99') is not JSON serializable"
becomes:
"Object of type 'Decimal' is not JSON serializable"
Update tests that looked for specific serialization
error message to just look for the word "Decimal"
instead. (Works with all Python versions.)
[1]: https://bugs.python.org/issue26623
* **Future breaking change:**
Rename all Anymail backends to just `EmailBackend`,
matching Django's naming convention.
(E.g., switch to "anymail.backends.mailgun.EmailBackend"
rather than "anymail.backends.mailgun.MailgunBackend".)
The old names still work, but will issue a DeprecationWarning
and will be removed in some future release.
(Apologies for this change; the old naming convention was
a holdover from Djrill, and I wanted consistency with
other Django EmailBackends before hitting 1.0.)
Fixes#49.
Mandrill's webhook signature calculation uses the
*exact url* Mandrill is posting to. If HTTP basic
auth is also used, that auth is included in the url.
Anymail was using Django's request.build_absolute_uri,
which doesn't include HTTP basic auth. Anymail now
includes the auth in the calculation, if it was present
in the request.
This should eliminate the need to use the
ANYMAIL_MANDRILL_WEBHOOK_URL override,
if Django's SECURE_PROXY_SSL_HEADER and
USE_X_FORWARDED_HOST (and/or
USE_X_FORWARDED_PROTO) settings are correct
for your server.
(The calculated url is now also included in
the validation failure error message, to aid
debugging.)
Fixes#48
SendGrid: update to v3 send API
**SendGrid:** **[possibly-breaking]** Update SendGrid backend to newer Web API v3. This should be a transparent change for most projects. Exceptions: if you use SendGrid username/password auth, esp_extra with "x-smtpapi", or multiple Reply-To addresses, please review the [porting notes](http://anymail.readthedocs.io/en/latest/esps/sendgrid/#sendgrid-v3-upgrade).
Closes#28
In BasePayload, ensure any Django ugettext_lazy
(or similar) are converted to real strings before
handing off to ESP code. This resolves problems where
calling code expects it can use lazy strings "anywhere",
but non-Django code (requests, ESP packages) don't
always handle them correctly.
* Add utils helpers for lazy objects (is_lazy, force_non_lazy*)
* Add lazy object handling to utils.Attachment
* Add lazy object handling converters to BasePayload attr
processing where appropriate. (This ends up varying by
the expected attribute type.)
Fixes#34.
Anymail was requiring Mandrill's webhook authentication key for the initial webhook url validation request from Mandrill, but Mandrill doesn't issue the key until that validation request succeeds.
* Defer complaining about missing Mandrill webhook key until actual event post.
* Document the double-deploy process required to set up Mandrill webhooks.
Fixes#46.
Add support for Postmark's recently-released [delivery tracking webhook] to Anymail's normailized status event handling. The existing Anymail tracking webhook URL can be copied to "Delivery webhook" in your Postmark outbound server settings.
Closes#45.
A message's `from_email` and each address in its `to`, `cc`, and `bcc` lists must contain exactly one email address. Previous code would silently ignore additional addresses, leading to unusual behavior. Now, raises new `AnymailInvalidAddress` exception.
Example: `from_email='Widgets, Inc. <widgets@example.com>'` is invalid: it needs double-quotes around the "Widgets, Inc." display-name portion. In earlier versions, this probably would have sent the message from something like "From: Widgets <@localhost>". Now, it will raise an exception.
**Potentially-breaking change:** If your code is using an unquoted display-name containing a comma in an email address, it will now raise an error. In earlier versions, this may have appeared to succeed, but was almost certainly not doing what you intended.
Fixes#44.
Compatibility with Python 2.7 versions older than 2.7.7
* Use Django's constant_time_compare method
* Include sparkpost in test requirements
* Don't use non-public `EnvironmentVarGuard` in tests
Fixes#41
Drop `from __future__ import unicode_literals`;
it was there for Python 3.2 compatibility (which
Anymail doesn't support). Ensures tests use normal
strs in Python 2.x.
Allow custom MAILGUN_SENDER_DOMAIN in Anymail
settings. (Replaces need to use global esp_extra.)
Improve docs to cover cases where this is needed.
(esp_extra sender_domain is still supported for
overriding individual messages.)
Fixes#26.
When using a stored template, SparkPost disallows
subject, text, and html. Django's EmailMessage default
empty strings are enough to provoke "Both content
object and template_id are specified" from SparkPost,
so remove them (if empty) when using stored templates.
Update docs and tests; add integration test for template_id.
Fixes#24
To conserve our ESP test accounts' send quotas, don't run
the live API integration tests 13 times in every Travis run.
Instead, just run them twice, on a representative set
of Python/Django combinations:
* Once on Python 2.7 (currently with Django 1.8)
* Once on Python 3.x (currently 3.5 with Django 1.9)
(Prep for running weekly tests on Travis cron.)
The *non*-integration tests still run on all combos.
* Introduce RUN_LIVE_TESTS environment var to control
whether live API integration test cases should run.
Default True, except in Travis-CI runs default False.
* Enable RUN_LIVE_TESTS in .travis.yml matrix for the
Python/Django combos listed above.
python-sparkpost generates a transmissions.send
payload which is now considered invalid by the API,
if you try to use both `cc` (or `bcc`) and the
`recipients` dict structure required for merge_data.
[Anymail had been generating that recipients dict
structure in all cases, for simplicity. Sometime
between 2016-06-07 and 2016-06-22, SparkPost
began rejecting that if it appeared in the `header_to`
constructed by python-sparkpost.]
Convert readthedocs links for their .org -> .io migration for hosted projects
As per [their blog post of the 27th April](https://blog.readthedocs.com/securing-subdomains/) ‘Securing subdomains’:
> Starting today, Read the Docs will start hosting projects from subdomains on the domain readthedocs.io, instead of on readthedocs.org. This change addresses some security concerns around site cookies while hosting user generated data on the same domain as our dashboard.
Test Plan: Manually visited all the links I’ve modified.
Catch and re-raise requests.RequestException in
AnymailRequestsBackend.post_to_esp.
* AnymailRequestsAPIError is needed for proper
fail_silently handling.
* Retain original requests exception type, to avoid
breaking existing code that might look for specific
requests exceptions.
Closes#16
* csrf_exempt must be applied to View.dispatch,
not View.post.
* In base WebhookTestCase, enable Django test Client
enforce_csrf_checks. (Test Client by default disables
CSRF protection.)
Closes#19
* Restructure runtests.py as suggested in
https://docs.djangoproject.com/en/1.9/topics/testing/advanced/#using-the-django-test-runner-to-test-reusable-applications
* Load version-specific Django settings modules
(rather than trying to make a single settings.configure()
work with all supported versions).
* Use `django-admin startproject` defaults for all
settings modules. (Tests compatibility with typical
apps, middleware, and other settings.)
* Set up tests-specific url config; switch to literal
urls in test cases. (Eliminates url `reverse` from
tests.)
* Make runtests.py executable
Closes#18
Also includes:
* Change AnymailTestMixin.assertDoesNotWarn
to filter specific warning classes.
* Look specifically for AnymailInsecureWebhookWarning
in WebhookBasicAuthTestsMixin.test_warns_if_no_auth
(because we don't care *in that test case* about
DeprecatedInDjango10 warnings).
