We no longer limit our list of candidate topics
to 300.
We continue to limit the topic results to 10, since
we don't want to overwhelm users or crowd out
non-topic-related suggestions.
We try to handle this is an efficient manner.
In message header search bar, users didn't use to get any typeahead
suggestions if a normal filter follows search filter.
E.g.: query => foo bar stream:D
In the above case, users didn't use to get any typeahead suggestions.
This was because we had set that the callers of 'parse' function can
assume that the 'search' operator is present in the last in the query.
Because of which `get_search_result` function (in search_suggestion.js)
didn't use to show any typeahead suggestions as it used to assume that
the latest typed query is for search filters.
Fixes part of #19435.
Long ago, we changed Zulip to inspect server-provided topic history in
our compose and search typeahead modules, but did not trigger a query
to the server to fetch data.
This could result in confusing experiences where someone thought that
a topic doesn't exist that just isn't present in recent history.
Once we merge this, we may want to adjust the messaging in typeahead
to advertise that the first option will create a new topic, since this
change may make it feel more like creating topics is a heavyweight
operation.
Fixes#9857.
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.
Use fully resolvable request paths because we need to be able to refer
to third party modules, and to increase uniformity and explicitness.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
For most functions that we were using __Rewire__ for,
it's better to just use the override helper, which
use __Rewire__ under the hood, but also resets
the reference at the end of run_tests.
Another nice thing about override() is that it reports
when you never actually needed the mock, and this
commit fixes the instances found here.
I didn't replace every call to __Rewire__. The
remaining ones fall under these categories:
* I looked for ") =>" in my code sweep,
so I missed stuff like "noop" helpers.
* Sometimes we directly update something
in a module that's not a function. This
is generally evil, and we should use setters.
* Some tests have setup() helpers or similar
that complicated this code sweep, so I
simply punted.
* Somes modules rely on intra-test leaks. We
should fix those, but I just punted for the
main code sweep.
We now just use a module._load hook to inject
stubs into our code.
For conversion purposes I temporarily maintain
the API of rewiremock, apart from the enable/disable
pieces, but I will make a better wrapper in an
upcoming commit.
We can detect when rewiremock is called after
zrequire now, and I fix all the violations in
this commit, mostly by using override.
We can also detect when a mock is needlessly
created, and I fix all the violations in this
commit.
The one minor nuisance that this commit introduces
is that you can only stub out modules in the Zulip
source tree, which is now static/js. This should
not really be a problem--there are usually better
techniques to deal with third party depenencies.
In the prior commit I show a typical workaround,
which is to create a one-line wrapper in your
test code. It's often the case that you can simply
use override(), as well.
In passing I kill off `reset_modules`, and I
eliminated the second argument to zrequire,
which dates back to pre-es6 days.
We still need to write to these globals with set_global because the
code being tested reads from them, but the tests themselves should
never need to read from them.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
Instead of prohibiting ‘return undefined’ (#8669), we require that a
function must return an explicit value always or never. This prevents
you from forgetting to return a value in some cases. It will also be
important for TypeScript, which distinguishes between undefined and
void.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
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>
Prettier would do this anyway, but it’s separated out for a more
reviewable diff. Generated by ESLint.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
Prettier would do this anyway, but it’s separated out for a more
reviewable diff. Generated by ESLint.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
This reverts part of b0d632577f.
The problem was that multiple queries were combined as a single
search pill. And since we create the pills then narrow / search,
we added a comma seperator between them for the typeahead lookups
as required by the logic in `input_pill.js`.
This however introduced a new bug where the search suggestions
were incorrect as the typeahead lookup table wasn't updated, so
every time an item from the type ahead was selected it updated
the input string with an invalid operator.
Thus to resolve the first problem, we follow a simpler approach
by extracting all operators from the search string using our
`Filter.parse` logic and next add the pills, one by one.
This is the exact same bug as observed in
02ab48a61e.
The bug is in the way we invoke `Filter.parse`.
`Filter.parse` returns a list of operators which
can contain only one 'search' term at max.
All strings with the 'search' operator present
in the query are combined to form this 'search'
term.
However on concatenating two filters we may get
two terms containing the 'search' operator. This
will lead to the search suggestions getting
generated based on only the last 'search' operator
term instead of all the terms having the 'search'
operator.
This is evident from the test change as suggestions
should be based on "s stream:of" but instead they
were based on just the latest query.
If typeahead is used, this adds comma separated search queries
so that multiple search pills don't get combined as one and the
search behaviour remains same as search_pills_enabled = False case.
If typeahead is not used, this prevent the typing of a single comma
after the pill gets created.
This makes it so that search_suggestion.js
does not depend on activity.js.
That dependency hasn't really been "elegant"
for quite some time, but it will become particularly
unnecessary when we go to remove the "Group PMs"
section from the right sidebar.
This commit introduces a temporary wart
where we have these two functions with the
same name in a sort of unnecessarily
complicated code stack:
activity.process_loaded_messages
huddle_data.process_loaded_messages
But we will eliminate the former function
very soon, and our message-related codepaths
will just call the `huddle_data` version
directly.
TESTING NOTES:
Now that `huddle_data` is a tiny leaf
module, it's super easy to just use the
real implementation of what was formerly
called `activity.get_huddles()` (and is
now in `huddle_data`).
When I first wrote this commit, introducing
the real implementation of `get_huddles` exposed
some bugs that I fixed in the immediately
prior commits to this.
When the tests were originally written,
I believe `activity.js` had some annoying
`jQuery` dependencies that made it hard
to unit test against. We've slimmed it over
time to be mostly just a "controller" module.
But even in its current state it would have
been a bit of a bloated dependency.
The other friction for using the actual
version of `get_huddles` was setting up
the message data, but that's pretty minor.
If you have a group PM where some users have
three-digit user_ids and some with four-digit
user_ids (or similar), a huddle could effectively
be ignored when determining the order of
search search suggestions.
Basically, we need a way to canonically sort
user_ids in "huddle" strings, and it's somewhat
arbitrary whether you sort lexically or sort
numerically, but you do need to be consistent
about it.
And JS is not exactly helpful here:
> [99, 101].sort()
[ 101, 99 ]
This is a pretty obscure bug with pretty low
user-facing consequences, and it was never
reported to us as far as I know, but the fix
here is pretty straightforward.
We have had similar bugs of slightly more consequence
in the past. The reason this bug has shown
up multiple times in our codebase is that every
component that deals with huddles has slightly
different forces that determine how it wants
to serialize the huddle. It's just one of those
annoying things. Plus, bugs with group PMs
do tend to escape detection, since most people
spend most of their time either on streams
or in 1:1 PMs.
Aman Agrawal
Steve Howell
Anders Kaseorg
akshatdalton
Riken Shah
Tim Abbott
Priyank Patel
Ryan Rehman
clarammdantas