We seemed to have been doing too much of sharpening on the thumbnails.
The purpose of sharpening here was to just counter the softening
effects of a resize on an image but overdoing it is bad.
Value sharpen(0.5,0.2,true) seems to look good for achieving the
best results here on different displays as revealed in the manual
hit and trial based testing.
Thanks to @borisyankov for pointing out the issue and suggesting
the values.
For some webhook endpoints where the third-party API requires us to do
this, the user's API key might appear in error emails through
appearing in the `QUERY_STRING` parameter. Fix that by filtering any
actual content from those; what we usually need for debugging is just
what set of parameters were provided.
The APNS client libraries (especially the hyper.http20 one) were
determined via profiling to take significant time during the import
process, so we move them to be lazily imported in order to optimize
the overall Zulip import process. This save up to about 100ms in
import time.
These libraries are only used in certain Django processes inside
zulipchat.com, and so are unnecessary both in development as well as
for self-hosted Zulip servers.
This is a prepartory commit for the upcoming changes. It was meaningful
to extract this one out because this function is essentially a condition
check on whether a given url is one of the user_uploads or an external
one. Based on its value we decide whether a url must be thumbnailed or
not and thus this function will also be used in an upcoming commit
patching lib/thumbnail.py to do the same check before thumbnail url
generation.
We are basically adding a check for url's to be external (belonging
to some 3rd party web site hosting the image) or be one of the
user uploaded files. User uploaded files are served by a separate
endpoint which is /user_uploads/. Any other local url such as
/user_avatars/ or /static/ should never be sent to thumbor for
thumbnailing.
Not sending /user_avatars/ to thumbor for thumbnailing makes sense
because they are already properly thumbnailed and stored properly.
/static/ urls host very few images we use for demo and can be safely
be excluded from thumbnailing.
The Zulip API is to be used on both development and production
servers, and really we just need to talk about zuliprc files.
There's a similar issue for the JS docs, but we need to fix the
copy/paste issues with those as well.
We don't want really long urls to lead to truncated
keys, or we could theoretically have two different
urls get mixed up previews.
Also, this suppresses warnings about exceeding the
250 char limit.
Finally, this gives the key a proper prefix.
We use UserMessageLite to avoid Django overhead, and we
do updates in chunks of 10000. (The export may be broken
into several files already, but a reasonable chunking at
import time is good defense against running out of memory.)
Now that we allow multiple users to have registered the same token, we
need to configure calls to unregister tokens to only query the
targeted user_id.
We conveniently were already passing the `user_id` into the push
notification bouncer for the remove API, so no migration for older
Zulip servers is required.
If cordelia searches on pm-with:iago@zulip.com,cordelia@zulip.com,
we now properly treat that the same way as pm-with:iago@zulip.com.
Before this fix, the query would initially go through the
huddle code path. The symptom wasn't completely obvious, as
eventually a deeper function would return a recipient id
corresponding to a single PM with @iago@zulip.com, but we would
only get messages where iago was the recipient, and not any
messages where he was the sender to cordelia.
I put the helper function for this in zerver/lib/addressee, which
is somewhat speculative. Eventually, we'll want pm-with queries
to allow for user ids, and I imagine there will be some shared
logic with other Addressee code in terms of how we handle these
strings. The way we deal with lists of emails/users for various
endpoints is kind of haphazard in the current code, although
granted it's mostly just repeating the same simple patterns. It
would be nice for some of this code to converge a bit. This
affects new messages, typing indicators, search filters, etc.,
and some endpoints have strange legacy stuff like supporting
JSON-encoded lists, so it's not trivial to clean this up.
Tweaked by tabbott to add some additional tests.
For our bots that use GenericOutgoingWebhookService
(which are basically Zulip style bots), we now
include a "content-type" header of "application/json".
We accomplish this by having the service classes
implement their own custom method called
`send_data_to_server`. For the Slack-related
code, we just extracted code from `do_rest_call`,
and then for the Zulip-related code, we added
a `headers` parameter.
If we omit methods in subclasses, they're likely to
be caught by linters or unit tests, and even if they
aren't, raising NotImplementedError doesn't actually
prevent user problems.
I've been fighting these in refactoring, and it's
just been a bunch of busy work, plus comments are
highly likely to bitrot.
This fixes a couple things:
* process_event() is a pretty vague name
* returning tuples should generally be avoided
* we were producing the same REST parameters in both
subclasses
* relative_url_path was always blank
* request_kwargs was always empty
Now process_event() is called build_bot_request(),
and it only returns request data,
not a tuple of `rest_operation` and `request_data`.
By no longer returning `rest_operation`, there are
fewer moving parts. We just have `do_rest_call` make
a POST call.
Before this change, we instantiated base_url into a superclass
of subclasses that returned base_url into a dictionary that
gets returned to our caller.
Now we just pull base_url out of service when we need to make
the REST call.
We move the JSON parsing step into the
higher level function: process_success_response().
In the unlikely event that we'll start integrating
with a solution that doesn't use JSON, we can deal
with that, and for now doing the parsing in one
place will help us make error reporting more
consistent.
In a subsequent commit we'll introduce better
error handling for malformed JSON.
The earlier code here, if it got a payload with
"response_string" as a key, would prefix the
corresponding value with "Success!". We just
want the bot to set its own content.
The code is reorganized here so that process_success()
always produces a value keyed by "content" from
incoming data, and then process_success_response()
doesn't do any fancy munging of the data.
Previously, Zulip did not correctly handle the case of a mobile device
being registered with a push device token being registered for
multiple accounts on the same server (which is a common case on
zulipchat.com). This was because our database `unique` and
`unique_together` indexes incorrectly enforced the token being unique
on a given server, rather than unique for a given user_id.
We fix this gap, and at the same time remove unnecessary (and
incorrectly racey) logic deleting and recreating the tokens in the
appropriate tables.
There's still an open mobile app bug causing repeated re-registrations
in a loop, but this should fix the fact that the relevant mobile bug
causes the server to 500.
Follow-up work that may be of value includes:
* Removing `ios_app_id`, which may not have much purpose.
* Renaming `last_updated` to `data_created`, since that's what it is now.
But none of those are critical to solving the actual bug here.
Fixes#8841.
We now allow outgoing webhooks to provide us a
"content" field, which is probably a more guessable
name than "response_string", particularly for folks
that use our other bot-related APIs. And we don't
modify content as we do response_string, i.e. no
"Success!" prefix.
If we're not too concerned about backward compatibility,
we can do a subsequent commit that makes "content"
and "response_string" true synonyms and get rid of
the "Success!" prefix, which was probably accidental
to begin with.
This commit starts by changing the third
argument of send_response_message to be a Dict
instead of a string, so that the data can be more
structured going forward.
That change makes the 2nd/3rd parameters both be
dicts, so to be defensive, I now have all the callers
pass in explicit keyword names. And then I rename
message to message_info, so that the callers have
more clear code.
And that changes the implementation inside of
send_response_message() a bit.
Sorry this commit is a bit coarse, but the intermediate
commits would have been kind of ugly, too.
At the end of the day, it's pretty simple:
bot_id: never changed
message_info: just renamed from message
response_data: is a Dict with the key of "content"
And the innards of send_response_message() are basically
simply dictionary lookups and function calls.
There's no reason to return a failure message in
process_success(), since it's implied to be part of
the success codepath. I didn't look at the full history
of how the strange API evolved, but the second element
of the tuple was clearly noise by the time I got here.
Neither of the subclasses ever set it, and none of the
consumers used it.
This two-line function wasn't really carrying its
weight, and it just made it harder to refactor the
overall codepath.
Eliminating the function forces us to mock at a slightly
deeper level, which is probably a good thing for what
the test intends to do. The deeper mock still verifies that
we're sending the message (good) without digging into
all the details of how we send it (good).
Note that we will still keep around the similarly named
`fail_with_message` helper, which is a lot more useful.
(The succeed/fail scenarios aren't really symmetric here.
For success, there are fewer codepaths that do more complex
things, whereas we have lots and lots of failure codepaths
that all do the same simple thing of replying with a canned
message.)
Before this change subclasses of OutgoingWebhookServiceInterface
would return a raw string as the first element of its return
tuple in process_success(). This is not a very flexible
design, as it prevents the bot from passing extra data like
`widget_content`.
It's also possible in the future that we'll want to let outgoing
bots reply directly to senders who mention them on streams, and
again the original design was overly constrained for that.
This commit does not actually change any functionality yet.
Aditya Bansal
Tim Abbott
Rishi Gupta
Yago González
Vishnu Ks
Steve Howell
Lyla Fischer
Pragati Agrawal