It should be very rare to discover new unread messages during a
message_fetch call. This can potentially happen due to races (fetching
just as a new message arrives), but it shouldn't be the common case.
Previously, we would trigger a full rerender of all UI displaying
unread messages every time a bulk message fetch operation returned
(including every time one narrowed), regardless of whether any actual
state had changed.
Fix this by actually checking if we discovered any new unread messages.
This also adds a comment noting a remaining performance bug we have in
this code path, namely fetching messages for a streams narrow will do
unnecessary work.
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>
f0c680e9c0 introduced a call to
message_helper.process_new_message without first calling
message_store.set_message_flags on the message.
This resulted in it being possible as a race, when loading the Zulip
app to a stream/topic/near narrow, for a message to have the
`historical` flag be undefined due to not being initialized.
That invalid state, in turn, resulted in the message_list_view code
path for rendering the message feed incorrectly displaying additional
recipient bars around the message.
We could fix this by just calling message_store.set_message_booleans
in this code path. However, this bug exposes the fact that it's very
fragile to expect every code path to call that function before
message_helper.process_new_message.
So we instead fix this by moving message_store.set_message_booleans
inside message_helper.process_new_message.
One call point of concern in this change is maybe_add_narrow_messages,
which could theoretically reintroduce the double set_message_flags
bugs detailed in 9729b1a4ad. However, I
believe that to not be possible, because that call should never
experience a cache miss.
The other existing code paths were already calling
set_message_booleans immediately before
message_helper.process_new_message. They are still changing here, in
that we now do a cache lookup before attempting to call
set_message_booleans. Because the message booleans do not affect the
cache lookup and the local message object is discarded in case of a
cache hit, this should have no functional impact.
Because I found the existing comment at that call site confusing and
almost proposed removing it as pointless, extend the block comment to
explicitly mention that the purpose is refreshing our object.
Fixes#21503.
This unfortunately requires somewhat ugly duplicated code, but I think
it's the best option for now.
I expect we will somewhat soon work on the transition to no longer
have two duplicate fetches here, and doing so will let us remove this
secondary code path.
Fixes#21304.
We split recent_topics module into recent_topics_(ui + data + util).
This allows us to reduce cyclical dependencies which were
created due to large list of imports in recent topics. Also, this
refactor on its own makes sense.
This data structure has never been one that we actually render into
the DOM; instead, its role is to support clicking into view that
contain muted streams and topics quickly.
This downgrade makes that situation much more explicit, and is also
useful refactoring to help simpify the upcoming changes in #16746.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
We move the message_store.add_message_metadata function
(and all its dependencies) into a new module called
message_helper and rename the function to process_new_message.
(It does a bit more than adding message metadata, such
as updating our message store.)
We also have a "protected" interface now in message_store,
so that message_helper can access the message store:
update_message_cache
get_cached_message
Because update_message_cache is identical to
the former create_mock_message, we just renamed it
in the tests.
Most callers should use these functions:
message_helper.process_new_message (setting)
message_store.get (getting)
It's slightly annoying that the setter interface
is in a different module than the getter interface,
but that's how you break a bunch of dependencies.
We also extract the tiny message_user_ids class:
user_ids()
add_user_ids()
All the code moves here are pretty trivial, and
the code that was moved maintains 100% line
coverage.
The module name `message_helper` is not ideal, but it's a single
function and it'll save time to just do the topology change now and
leave thinking through the right name to later.
This is a mostly verbatim extraction.
I re-phrased one line of code to work around a lint
false alarm. (Look for `preamble` in the diff.)
There are about 8 lines missing coverage here, so
the new module might be a good candidate to get
100% line coverage on.
Before this change, you would need to remove 74
edges from our dependency graph to make it
acyclic. Now it's 72.
Previously, the `muting_enabled` property of
MessageListData class was used to indicate whether
some messages in the message list need to be
filtered due to topic muting, depending on the
narrow. For example, we exclude messages belonging
to muted topics from stream narrows, but not from
search narrows.
The name `muting_enabled` is a bit confusing, and hence is
changed to `excludes_muted_topics`.
It is also important that the name be specific, since
a similar new property will be added for user mutes
in future commits.
This refactoring should have no functional effect for any call points,
but makes the function behave more naturally. The comments explain
the situation, but specifically:
* There's the page_params.narrow hack that affects both narrows and
home_msg_list.
* There's the shared data for home_msg_list and all_msg_list that
requires we modify the query from home_msg_list.data.public_operators().
And otherwise the logic should just use the operators associated with
the message_list.data object (allowing us to remove the force_fetch
hack added in the last commit).
Hopefully in some future refactoring, we'll be able to migrate those
hacks to live in the Filter object construction and eliminate this
block of conditionals entirely.
In commit ebea17b9a6,
we added an extra fetch to get accurate data for the top
items in recent topics table.
But the `narrow` parameter wasn't passed to the endpoint,
this resulted in fetching the user's overall message
history including the muted streams/topics which aren't
required by the recent topics table.
`operators` can be replaced as we set the same value for
the `narrow_state` module and the narrowed message list's
filter, when activating the narrow.
The comment explains the problem statement in some detail, but
basically this algorithm ensures that the top items in "Recent Topics"
on page load are always the very most recent topics the user has
received messages in (well, ignoring muted topics in this iteration).
ES and TypeScript modules are strict by default and don’t need this
directive. ESLint will remind us to add it to new CommonJS files and
remove it from ES and TypeScript modules.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
Tim Abbott
Alex Vandiver
Anders Kaseorg
Aman Agrawal
Steve Howell
Abhijeet Prasad Bodas
ryanreh99
Priyank Patel