We had a theory that get_user_id() errors were often due to race
conditions related to reloads, so we would only report missing
user ids if subsequent lookups failed 5 seconds later. It turns
out we still get the blueslip errors, and now we don't get
meaningful tracebacks. This change makes it so that errors
get reported immediately again.
In f75af94984 I added some
lines of code that made it so that live updates for avatar
urls would affect messages currently in the browser.
This change worked well when the live update actually happened,
but then the next time the user would reload, the avatar in
the message pane would regress back to showing the avatar urls
from the server (which could have caching issues of their own).
This fix removes a couple lines of code that had the intended
effect of making all of your messages from any given sender
show the same url (good) but which generally grabbed
the url from an old message (bad).
After this fix, we go back to having old messages possibly
showing the old avatar urls, but new messages will display the
new avatar.
(There are lots of moving parts in the avatar system, because
not only do browsers cache image urls, but our server caches
messages and recipient info, so there have been "fixes" to
avatars since this change that are valid fixes in their own
right but not directly relevant to this commit.)
We use to have client-side logic that would append timestamps
or random numbers to avatar URLs to force browsers to
refresh their cache.
We no longer need this now that the back end maintains
versions for avatar changes and puts the version in the URLs.
When we process messages for unread counts, we now call
people.pm_reply_user_string() to get a string of user ids,
rather than using emails that may have changed since the
message was originally created.
We now convert our pm-with search operand to a list of user ids
for matching against messages, rather than using emails. On the
message side we look at user ids from display_recipient.
We have added people.pm_with_url(message), which computes a
PM url from a private message using user ids rather than emails.
We call this in add_message_metadata(), since the slugs will
be valid even if emails change, so we don't need to compute
them on the fly during message rendering.
The function people.update_email() is not yet connected
to anything, but it sets the stage for upcoming changes.
When emails get updated, fundamentally we just update
the appropriate person object and add a new key to
people_dict. We sort of get a shim for free--old email
lookups will continue to work--but we add blueslip warnings
for stale lookups.
We now trigger an event in user_events.js, and we dynamically
build the list of names in pm_list.js by calling out to
people.get_recipients().
We have a few variations of functions that build lists of names
for huddles, which should be cleaned up eventually. They are
called at different times in the code path, so the different
functions, while doing mostly the same thing, start with different
data sources.
The function people.filter_by_search_terms() used
to return a JS object with emails as keys to represent
a set of users. Now we return a Zulip Dict() object
with user_ids as keys.
Most of the magic happens in message_live_update.update_avatar().
The prior code was buggy, as it was using person.id instead of
person.user_id, and it was not setting the image resolution.
This change is a partial bug fix for avatar live updates.
It makes it so that we prefer the person.avatar_url to
the message.avatar_url when rendering messages. Our live
update code was already populating person.avatar_url, but
we were ignoring it until now.
This commit does not affect messages that were already
rendered with the old url.
If we get a realm_user update for a user that is **not**
changing their full name, we no longer call
admin.update_user_full_name().
This was probably a fairly minor bug.
Earlier commits removed all uses of page_params.email outside
of people.js, and it turns out we have page_params.user_id, so
we don't even need page_params.email for seeding the data.
The local echo code now marks up mention buttons with user ids
instead of email. Our code in message_list_view.js deals with
either the old style or the new style of markup now to determine
which mention buttons need to be highlighted.
As part of this commit we extract mention_button_refers_to_me().
This commit doesn't change any behavior yet, but it starts us
down the road of deprecating page_params.email and allowing
people.js to control all access to the current user's email,
which will be important for email changes.
In people.emails_strings_to_user_ids_string, we just warn
for bad emails going forward.
Users can enter bad emails into the search location bar,
for example, and that causes us to compute a browser hash,
which in turn uses this function.
(It's possible that we should adjust the search code not
to compute hashes for narrowing when the narrow doesn't
make sense, but that could be a non-trivial fix.)
The slugs for PM-with narrows now have user ids in them, so they
are more resilient to email changes, and they have less escaping
characters and are generally prettier.
Examples:
narrow/pm-with/3-cordelia
narrow/pm-with/3,5-group
The part of the URL that is actionable is the comma-delimited
list of one or more userids.
When we decode the slugs, we only use the part before the dash; the
stuff after the dash is just for humans. If we don't see a number
before the dash, we fall back to the old decoding (which should only
matter during a transition period where folks may have old links).
For group PMS, we always say "group" after the dash. For single PMs,
we use the person's email userid, since it's usually fairly concise
and not noisy for a URL. We may tinker with this later.
Basically, the heart of this change is these two new methods:
people.emails_to_slug
people.slug_to_emails
And then we unify the encode codepath as follows:
narrow.pm_with_uri ->
hashchange.operators_to_hash ->
hashchange.encode_operand ->
people.emails_to_slug
The decode path didn't really require much modication in this commit,
other than to have hashchange.decode_operand call people.slug_to_emails
for the pm-with case.
When somebody changes their name, we will now update
the buddy list right away. The old code was trying
to do this through a code path that was designed for
true presence updates, but it was also passing in an
empty array, instead of undefined, which caused it to
fail to invoke the intended part of the codepath to
redraw the buddy list.
Now we just call the new activity.redraw() function,
which does the right thing for the buddy list.
The group PM list was live-updating before this change,
and it continues to live-update as part of the new
activity.redraw() function.
This commit extracts people.extract_people_from_message()
from message_store.add_message_metadata(), and now
add_message_metadata() extract people before it calls
process_message_for_recent_private_messages(), which
fixes a bug where we are trying to look up an email
for a deactivated user who was in the message but not
in the pre-loaded list of people.
Fixes#2701
This commit changes people.remove() to be people.deactivate(),
and it fixes a bug where deactivating users was causing tracebacks
in the PM list if somebody had PM'ed the deactivated user
recently.
We need this for node tests, so that you don't have to explicitly
remove every user between tests. (Also, people.remove() is about
to have different semantics.)
We now sort lists of users ids deterministically, and we also
sort list of emails deterministically and without regard to case.
This probably fixes the bug #2343, although I never got a great
repro on that.
fionabunny
hackerkid
Steve Howell
Tommy Ip
Tim Abbott
AZtheAsian