We no longer use all the alert words for all the users in the
entire realm when we look for alert words in a newly sent/edited
message. Now we limit the search to only all the alert words
for all the users who will get UserMessage records. This will
hopefully make a big difference for big realms where most messages
are only sent to a small subset of users.
The bugdown parser no longer has a concept of which users need which
alert words, since it can't really do anything actionable with that info
from a rendering standpoint.
Instead, our calling code passes in a set of search words to the parser.
The parser returns the list of words it finds in the message.
Then the model method builds up the list of user ids that should be
flagged as having alert words in the message.
This refactoring is a little more involved than I'd like, but there are
still some circular dependency issues with rendering code, so I need to
pass in the rather complicated realm_alert_words data structure all the way
from the action through the model to the renderer.
This change shouldn't change the overall behavior of the system, except
that it does remove some duplicate regex checks that were occurring when
multiple users may have had the same alert word.
If you supplied an unrecognizable address to our email system,
or you had EMAIL_GATEWAY_PATTERN configured wrong,
the get_missed_message_token_from_address() used to crash
hard and cryptically with a traceback saying that you can't
call startswith() on a None object.
Now we throw a ZulipEmailForwardError exception. This will
still lead to a traceback, but it should be easier to diagnose
the problem.
In our email mirror, we have a special format for missed
message emails that uses a 32-bit randomly generated token
that we put into redis that is then prefixed with "mm" for
a total of 34 characters.
We had a bug where we would mis-classify emails like
mmcfoo@example.com as being these system-generated emails
that were part of the redis setup.
It's actually a little unclear how the bug in the library
function would have manifested from the user's point of view,
but it was definitely buggy code, and it's possibly related in
a subtle way to an error report we got from a customer where
only one of their users, who happened to have a name like
mmcfoo, was having problems with the mirror.
It appears that the assertRaisesRegexp approach we had before didn't
work properly on some systems, likely due to a bad interact with a
i18n (we haven't definitively determined the cause).
We now raise an exception in bugdown.do_convert() if rendering
fails, to avoid silent failures, and then calling code can convert
the exception to a JsonableError.
We now have 100% coverage on views/push_notifications.py, modulo
some dead code which will be addressed in the next commit.
There were some existing tests in text_external.py, but that
module is really intended for tests that hit external services.
The view is a really simple API that updates a DB table, and the
new test code focuses on error handling and idempotency as well
as the happy path.
Apparently, in urllib.parse, one need to extract the query string from
the rest of the URL before parsing the query string, otherwise the
very first query parameter will have rest of the URL in its name.
This results in a nondeterministic failure that happens 1/N of the
time, where N is the number of fields marshalled from a dictionary
into the query string.
Steve Howell
Rishi Gupta
Tomasz Kolek
sunnypalace
Tim Abbott
umkay