This was added by commit 7f174213ed, and
appears to have been designed for responses that are *successful* but
falsy. Logically, these should not implicitly represent a failure to
be retried if it were.
Note from tabbott: The background is that this idempotent retry loop
was a hacky workaround for a bug we never understood but saw daily in
production. Especially during server restarts / client reloads,
something would result in 200 responses with no data being seen by the
frontend, despite the Django server not having received/processed the
request. Fortunately, this strange failure mode appears to have
stopped happening in late 2019, so we can delete this hack.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
Commit 9aa5082d63 (#20673) incorrectly
changed the name of the error callback passed to channel.get. This
prevented reporting of errors while moving a topic. Fix it.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
Whether we sent a resolve topic notification or not may be useful in
the caller. It was originally intended to be used in #21712, but may
only be relevant for future logging.
Part of #21712.
This will have no real effect in most situations. However, a user
moves a topic to another stream while also adding/removing the
resolved-topic checkmark from the topic name, then the "This topic was
resolved" notificaiton will now appear just before the "This topic was
moved" notification rather than just after.
This is likely slightly less confusing to users, since the topic
having been moved from somewhere else is likely the most salient fact
to a reader.
We expect to change things to not send both notifications in an
upcoming commit.
This refactoring helps with #21712.
When we detect that multiple messages are selected, copying will copy the
entirety of each message that is selected. [more conversation on that
here](https://chat.zulip.org/#narrow/stream/6-frontend/topic/triple.20click.20.2B.20copy.20paste.20bug/near/1398292).
This commit fixes a bug where we would select the entirety of a message
following a selection even when none of that last message's text was
selected (only some HTML element in the messages's container).
Fixes#18706.
A message selection that goes off the end of the last message in a topic
could still be a selection of only one message. We shouldn't assume that
it's multiple messages, and so there shouldn't be an assignment of
`skip_same_td_check = true`.
The table_name property was only ever undefined for the
special all_messages_list object.
In 6f764ce4b3, we downgraded that object
to only have a MessageListData; as a result, we now never construct a
MessageList or MessageListView without `table_name` set correctly.
`current` should ideally be `undefined` here to reflect the
correct narrow state but to make sure background updates of
message_list go smoothly, we set it to home.
Our seat count calculation is different for guest user than normal users
(a number of initial guests are free, and additional marginal guests are
worth 1/5 of a seat) - so these checks we apply when a user is being
invited or signing up need to know whether it's a guest or non-guest
being added.
This is a simple generalization of get_latest_seat_count and is useful
for calculating "what will be the realm's license count if this
number of (guest) users is added?" without duplicating any of the math
logic. Will be used in the next commits.
Our billing FAQ says:
"For an organization with N other users, 5*N guest users are included at
no extra charge. After that, you will be charged at 1/5 of your regular
per-user pricing for each additional guest.".
It wasn't quite intuitive to me that
max(non_guests, math.ceil(guests / 5)) achieves that pricing, so it's
worth mentioning in a comment that it does and that that's why that
formula is used.
This commit reuses set_delete_own_message_policy_dropdown to enable
or disable the time limit settings based on change in
delete_own_message_policy dropdown.
This is a prep commit for deduplicating code and using this same
function set_delete_own_message_policy_dropdown for disabling or
enabling the limit setting while changing policy dropdown.
For multi word stream names, searching for particular phrases, like
the entirety of any word in it, caused no space to show before and/or
after that word. For example, searching for `core` resulted in
`**core**team` (the space between `core` and `team` missing).
The cause of this bug was that the immediate container was `div` with
`display: flex` and so spaces at the ends of its immediate child
elements: text nodes and `strong` elements - were not respected. This
is fixed now by wrapping all the `description_html` text in a `span`
element (while the optional user_pill_context stays in another `span`
as before).
This is consistent with Jest and other standard test runners, and is
necessary to support asynchronous tests.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
We forgot to change the selector here when we switched from a fa globe
icon to a .zulip-icon, which resulted in the new globe icon only having
the default color.
This commit makes the appropriate change.
Because rate_limit_request_by_ip is the only caller of it, it is safe
for us to inline RateLimitedIpAddr and remove this helper. This ensures
that we have consistent internals for rate limiting functions, which all
have a should_rate_limit check.
Signed-off-by: Zixuan James Li <p359101898@gmail.com>
This allows us to use them with HttpResponse objects returned by
calling a view function directly.
Signed-off-by: Zixuan James Li <p359101898@gmail.com>
This change incorporate should_rate_limit into rate_limit_user and
rate_limit_request_by_ip. Note a slight behavior change to other callers
to rate_limit_request_by_ip is made as we now check if the client is
eligible to be exempted from rate limiting now, which was previously
only done as a part of zerver.lib.rate_limiter.rate_limit.
Now we mock zerver.lib.rate_limiter.RateLimitedUser instead of
zerver.decorator.rate_limit_user in
zerver.tests.test_decorators.RateLimitTestCase, because rate_limit_user
will always be called but rate limit only happens the should_rate_limit
check passes;
we can continue to mock zerver.lib.rate_limiter.rate_limit_ip, because the
decorated view functions call rate_limit_request_by_ip that calls
rate_limit_ip when the should_rate_limit check passes.
We need to mock zerver.decorator.rate_limit_user for SkipRateLimitingTest
now because rate_limit has been removed. We don't need to mock
RateLimitedUser in this case because we are only verifying that
the skip_rate_limiting flag works.
To ensure coverage in add_logging_data, a new test case is added to use
a web_public_view (which decorates the view function with
add_logging_data) with a new flag to check_rate_limit_public_or_user_views.
Signed-off-by: Zixuan James Li <p359101898@gmail.com>
This allows us to avoid importing from zilencer conditionally in
zerver.lib.rate_limiter, as we make rate limiting self-contained now.
Signed-off-by: Zixuan James Li <p359101898@gmail.com>
- RateLimitTestCase.get_ratelimited_view is replaced by a view
function directly decorated by public_json_view.
- the META dict is initialized with "PATH_INFO": "test" because now the
tests cover the process_client codepath;
- HostRequestMock is initialized with host="zulip.testserver" to pass
the validate_account_and_subdomain check;
- check_rate_limit_public_or_user_views replaces both
test_rate_limiting_happens_in_normal_case and
test_rate_limiting_happens_by_ip_if_unauthed.
Overall, we deduplicate the test cases in this change, and make sure
that they also cover the view function decorators for authentication.
Signed-off-by: Zixuan James Li <p359101898@gmail.com>
The test setup for some of the test cases are largely similar, so it
would be cleaner to be able to reuse them.
Note that we use "check" in the name of this helper because later we
will extend it to take a flag to set whether rate limiting is expected.
Signed-off-by: Zixuan James Li <p359101898@gmail.com>
This refactors the test case alongside, since normal views accessed by
remote server do not get rate limited by remote server anymore.
Signed-off-by: Zixuan James Li <p359101898@gmail.com>
The reason for the flake was we were not waiting enough
time for the deactivation row to render.
To fix this, We are relying on the input from the user
deactivation screen to focus.
The reason for the flake was we were not waiting for the
settings modal to open completely.
To fix this we now wait until the profile tab is focused to
detect if the settings modal is open completely (i.e ready
for the interaction).
This is not a feature intended to be used outside zulip.com, since it
just sets your server to have the zulip.com landing pages. I think
it's only been turned on by people who were confused by this text.
Updates `stream_events.js` to use the subscription update event that
is now being sent for the `is_muted` stream property, instead of the
event sent for the `in_home_view` property.
Because the server is still sending events for `in_home_view`, keeps
it as a property processed by `update_property` function, but does
nothing with the event. Adds temporary test for coverage.
In Zulip 2.1.0, the `is_muted` stream subscription property was
added and replaced the `in_home_view` property. But the server has
still only been sending subscription update events with the
`in_home_view` property.
Updates `do_change_subscription_property` to send a subscription
update event for both `is_muted` and `in_home_view`, so that
clients can fully migrate away from using `in_home_view` allowing
us to eventually remove it completely.
When the user chose to send the composebox message on pressing ctrl +
enter instead of just enter, it only worked in writing mode but not in
the preview mode.
This change makes ctrl + enter send the message even in preview mode,
when that setting is chosen.
Fixes: #21670.
I'm not sure whether the bug this fixes was a regression resulting
from d6d3683da0, or an old bug, but
focusing the compose box is not sufficient to end preview mode; we
should be calling the function that's explicitly for that.
Because the typeahead.js list items are currently just text, a user's
full name and avatar should be displayed in `input_pill`. To use
`input_pill`, a separate Handlebars partial view was created to
provide a mandatory container (`<div class="pill-container">`) for
`input_pill` and a flex container (`<div class="search_list_item">`)
for vertically aligning the text.
The description of each suggestion (i.e `description_html`) is
rendered as raw HTML, so every special character (e.g. whitespace)
should be HTML-escaped. This enables highlighting the substring in
each search suggestion that matches the query.
Fixes: #20267
Anders Kaseorg
Tim Abbott
Julia Bichler
evykassirer
Aman Agrawal
Mateusz Mandera
Sahil Batra
N-Shar-ma
YashRE42
Zixuan James Li
Riken Shah
Lauryn Menard
Kartik Srivastava
Oliver Pham