Realm emojis uploaded before the migration to store the emoji author
information was done don't have any author information. Such emojis
if listed on the settings page caused a traceback.
Fixes: #5133.
This covers all blueslip errors and warnings
in people.js. These do not need to be tested
to rigorously and just need to be covered to
get people.js to 100% coverage.
This fixes a regression where we removed a call to
unread_ops.process_visible() inside of stream_list.js. Now
we call it from within narrow.activate() in the the
maybe_select_closest() callback.
This test fails on self.assertTrue(delay < 0.001 * num_ids, error_msg)
randomly. This commit adds debug code to see what the real values of
paramters are.
When the emoji settings page was reopened after uploading a realm
emoji without doing a page refresh, the uploaded emoji disappeared
from the emoji list. This was so because the emoji settings page uses
`page_params.realm_emoji` to render the emojis which was not updated
when a emoji was added.
Fixes: #5130.
The settings dropdown was marked as having the "menu" ARIA role,
but contained no items with the "menuitem" role. I assigned this
role to the focusable link elements, and gave the intervening "li"
elements (and other "li" elements used as separators and such)
the "presentation" role to allow those to be gracefully ignored
when appropriate. Some of the links previously had the "button"
role, which I think was a holdover from a previous UI layout.
Fixes#5000.
We now specifically wait for the length to decrease by one. This seems
like a more deterministic condition to wait on.
Previously we were waiting till the id of the deleted message remained
visible; intuitively, this should have worked but it seems that there
is some race condition that was causing the test to fail sporadically.
Apparently, PyPI is very strict about package file names. Once you
upload files for 0.3.0, and only wish to make minor changes and
re-release it as the same version, it doesn't let you and complains
about identical file names.
This change should lead to clearer tracebacks when our
assumption about the stream list's list items get
violated, and we also short circuit some code in the
caller that tries to scroll to the active stream.
In stream_list.js we have some code to handle narrow activations,
and we were calling unread_ops.process_visible() only for
stream activations, not for PM-related activations, etc., so
our approach was inconsistent.
It also turns out that the call is redundant, since we call
unread_ops.process_visible() when the message pane scrolls as
part of updating the content.
Ideally, we want a more rigorous approach where we make this
call precisely when the new messages become visible to the user,
but the purpose of this fix is to de-clutter the stream_list
logic.