Previously we only used bulk queries when adding many users to a
single stream, resulting in very slow performance when subscribing
users to large numbers of streams (as happens when setting up a new
MIT realm user).
(imported from commit 849fa7b2a1a146c0a9adc1c727c20c9fbfb7b425)
This comment was only ever accurate for prototype versions of
bulk_add_subscriptions prior to it being committed to master.
(imported from commit 89b9dc49423c45553cb6c810d97eea4583ff0f69)
The message_stream_count() function uses Django's count() method,
which is more efficient than doing len() on array of full objects.
(imported from commit 9c20a89a2cd02d9d39341132330d03a7f6c8be25)
This change removes an "if True:" that was
introduced to make the prior commit a bit more readable.
It also combines two loops, since the second loop is no
longer conditional.
(imported from commit df58f1e5de72d5669f6468fbff54fb62cd22cedb)
The tests in GetUpdatesTest had some callback logic that has
been dead code for at least three months. We now fully exercise
the callback codepath and make sure that the callbacks do happen.
(imported from commit f5d8fbab28ecc34dc81d3d0c29058b66c10f378f)
These contain timerender spans with unique IDs. If this string is cached
and re-used in a narrowed view, there are elements with duplicate IDs,
and only one of them is updated at midnight.
(imported from commit 29469fb5f0d8a9b7fe7988849d2936c49d4a038d)
As of f69d01b and e4a9f80, timerender only deals with the date part.
Since the time does not change when the friendly dates are updated,
just append the time outside of the span.timerender.
(imported from commit 18ec93550782c6fa8447ebc608f95da79d28dc8d)
QTBUG-3467 prevents non-normal-face @font-face fonts from being used when
defined as such in CSS. To work around this, the desktop applications now
ship the Humbug font themselves, and this commit causes the server to no
longer send the problematic CSS rules to those clients.
We have some duplication insofar as we now have two minified CSS files, but
this is better than conditionally applying the CSS at page runtime.
(imported from commit 9a887f9fb8002d44171d366d1249ebbf21cc9c77)
Trac #1403.
This shows the 5 most recent subjects, as well as any others with unread messages. This
requires tracking all subjects and filtering at display time, rather than filtering when
building the subject list.
(imported from commit 8bda7d50e6785a6e70abea4b3af4d03a16d076d3)
(The file-input widgets that come with browsers are ugly and
nonstandard across browsers, so it is a common technique to
have your own button that controls the file upload, and it
delegates to a hidden copy of the browser file-input widget.
We also allow you to clear the file.)
(imported from commit b55ef655e75746330dc3cc396cb908670e5019cc)
The add-bots form used to have a landscape alignment, as it was
integrated into the same HTML table that showed your existing bots.
This became unwieldy once we allowed users to upload avatars.
(imported from commit 246a35be77ce1679d595271e6911dc339a6813ab)
Caveats:
- Since Chrome has trouble using W3C Notification when it's not
initiated by a user gesture, we try to use webkitNotification first.
- FF doesn't allow iconUrl to be of a different origin, so it won't display
our gravatars
(imported from commit c4f99ce6927a0d203d9f220d50b06737779bd7f8)
Previously, every function which had the @json_* decorators on it
would independently call get_client when views.py was imported; we
resolve this using a small bit of in-memory caching, which is safe
since Client objects are immutable.
This also more generally cleans up the code path to be more common
between the API and JSON views.
(imported from commit 674ea327fc4cb5fa982a75e388f0b604bae66567)
Previously some database queries that happened prior to the request
could be counted as part of the request's in-database processing time,
resulting in confusing output where get_events spent more time doing
database queries than the total time spent processing the request.
(imported from commit 509247462d276d6020a2970f62edb33a1d633087)
As does the hotkey "i". It's somewhat less appropriate because it's all actions,
but it's where our "info" menu used to be, and we can workshop a better one. "a"
feels weird to me, but maybe it's just me.
You can also hover on the .message_content to see a popover with extra message details
This is for Trac #1334.
(imported from commit f8fbf70c8502370a78159e24f3cf9589fb9d384f)
This mostly applies to the previous two commits. In principle,
this could be teased out and merged into them, but the hassle
here doesn't totally seem to be worth it.
(imported from commit ee2469ca3762c50c6db49b93eec02b32589eafe3)
* Switch to a 2-column layout by removing the old unused message tools
column and the pointer columns.
* Change column spans to 2 instead of 4.
* Change recipient bar to have a colorblock on the left.
* Add a special "last_message" class to messages that appear last
in a message block.
(imported from commit 55036587445c699d5c55d52b0236daf402a80cff)
Previously, we'd highlighted the top of the block. The theory here is
that it will make things less visually cluttered/noisy, which should
improve readability.
(imported from commit f94ec6b8e55aef9c2413718aea23b1124f5308e7)
This allows us to do some nice styling on the final message in a group
(in this commit series, this leads to some gently rounded corners.)
(imported from commit 86e7e8586dd4ee3cd8d4ce3be43fd2abab3f1d16)
Showing a user's bots in alphabetical order leads to a mildly
confusing experience when we append a new bot to the end of the
table, but then you hit refresh, and the new bot goes to a different
position. Since any given user is unlikely to have zillions of
bots, I don't think we need alphabetical order to help them find
old bots.
(imported from commit 4f19dbd7a016e7d867e88248190849dcd52c6d71)
(and remove the colon after the sender name to make them more
consistent with our desktop notifications.)
(imported from commit f2f9e1ecf7766c717fe7afb0559c4771081c255b)
This stop words file is just the default Postgres english stop file
with all the rest of the letters of the alphabet added. Adding the
extra letters ensures that, e.g., "bed" doesn't get transformed into
"bed | b".
(imported from commit 0be3ef9a43eb524ed4f081d5081a786cf602c487)
Compare two user objects by id to prevent false negatives
when the objects are fetched thru different paths.
(imported from commit a41f30d27e2b8021600d89f32d6526f48677fd95)
This also has the property of fixing this for all forms-that-contain-modals
since the selector is more generic.
(imported from commit 18fd26dd94fe4f6f3fe9a01884b867a9d37993d0)
Since in the future we might want requests to add subscriptions to
include things like colors, in_home_view, etc., we're changing the
data format for the add_subscriptions API call to pass each stream as
a dictionary, giving a convenient place to put any added options.
The manual step required here is updating the API version in AFS
available for use with the zephyr_mirror.py system.
(imported from commit 364960cca582a0658f0d334668822045c001b92c)
Previously, it always failed because we had hooked up the API endpoint
to a function that doesn't exist.
(imported from commit b5269f6d8e385facae4362742fe69a422f6315b7)
This way we can return properties of the streams other than just their
names in future versions of the API without breaking old clients.
The manual step required is to deploy the updated version of
sync-public-streams on zmirror.humbughq.com when we deploy this code
to prod.
(imported from commit 42b86d8daa5729f52c9961dd912c5776a25ab0b4)
The previous API was very redundant, which meant that any refactoring
would result in lots of unnecessary changes.
(imported from commit e04f6cbd87c8f65d4eebbe6972d26998faa28a56)
I believe this should require no special work on deploy, since some
grepping of logs suggests we are not currently using this API query.
(imported from commit 240086f900c6680cbc90bf6a2f334a9e1f172df6)
Trying to check whether a Django model object is inside a set of other
Django models is not correct in general, e.g.:
UserProfile.objects.only("id").get(id=17) in set([UserProfile.objects.get(id=17)])
returns False.
This bug appears twice in the function, once when computing which
users were mentioned and again when pushing the flags through to
Tornado.
(imported from commit b09ed550258f9df2611e1b0a60f87c48a51830f8)
Previously, we were only checking whether the message was sent by you
for PMs, not for streams that you marked as notifiable.
(imported from commit 668e102e885aa347cceff376c1873c1c62aff997)
(Before it had been disabled only on prod/staging, but we are
removing it everywhere, motivated by making tests run faster.
In particular, the call to embedly_client.is_supported() was
expensive, as it went over the Internet.)
(imported from commit ea12bf6e7ae84ce7e8023a0d314ecc4c07cbc0a8)
This file is a dict mapping filenames like "min/api.js" to
"min/api.HEX_HASH_HERE.js". This will be useful later for figuring out
which hashed version of a generated file belongs to a specific deploy,
for reuse in future deploys with the same source files.
(imported from commit d03c408b37a4e2d494abd4b1577a70b5ff5c2e34)
Whereas `SITE_ROOT` referred to the directory where settings.py is
located, *all* actual uses of `SITE_ROOT` were joining it with `..` to
get the root of the git checkout, a much more useful value.
`DEPLOY_ROOT` now represents the root of the git checkout.
(imported from commit 351437f9a5801e5c7c08a3a97619e863144e5cc8)
Previously we had an issue that every other update_active_status
request for a particular realm would result in doing the expensive
query to compute the list of active users in that realm. It turned
out this was because on every update_active_status request, we'd queue
an event that would have the effect of clearing the cache, even if
nobody's tatus changed. This fixes that issue, by only clearing the
cache for a realm if someone's status actually changed (or the 60s
timeout expires).
(imported from commit d5b829fe255a31c8cecb58458738f1e72a2cf6de)
This improves the performance of the get_old_messages queries when
loading the home page by about a factor of 2.5x.
(imported from commit 581840a6ed7b391c9d9fde67f368fce816e567e2)
This is in preparation for having a case in which we query the
database directly to get the message flags, without going through a
UserMessage object.
(imported from commit d5218974680b0c4b028a84f3aae1c8242ceb08ce)
This decreases the average size of the message dicts in memcached by
about half, without any significant change in the overall performance
of the query. Since these message dicts are a significant fraction of
what we put in memcached, this seems like a worthwhile optimization.
(imported from commit 3896328074aa4344b8ac7c7ba7685f0a167ec7ad)
memcached stores objects sent to it using pickling, which is very
slow. We work around this by sending memcached strings (i.e. JSON
dumps); pickling doesn't slow things down too much if all it is
getting is a string.
(imported from commit 0f0e534182eccb76c5731198e05a9324a1cef316)
This saves something like 15ms on our 1000 message get_old_messages
queries, and will save even more when we start sending JSON dumps into
our memcached system.
We need to install python-ujson on servers and dev instances before
pushing this to prod.
(imported from commit 373690b7c056d00d2299a7588a33f025104bfbca)
Previously, if you were narrowed when we reloaded you, you'd be
kicked out to the home view.
(imported from commit c00f1b92c9a4f559998463e141a402b69873fd56)
Some browsers don't support desktop notifications. Some people haven't allowed permissions
for humbug to have desktop notifications. This is a poor man's attempt at desktop notifications.
We are adding the bootstrap-notify package (http://nijikokun.github.io/bootstrap-notify/)
Resolves Trac #1336.
(imported from commit 6a54f7d1875e765dabd32d94ada8ebe4474a3d71)
We haven't been using this for months and not removing it before was
just an oversight.
(imported from commit d95c1911765a04a0c8713cc6c0dd346c123c97a3)
I noticed in our server logs that some users were sending
update_onboarding_steps requests to the server every time they sent a
message -- because they hadn't finished the onboarding process.
(imported from commit 8e1751a3344149f386022d96e7893e687d6631e8)
This reverts commit fec64815c879f9704847795cc163e2b517b4190d.
We've decided this experiment didn't work out.
(imported from commit 04f38e15c8ef19fbaa022a06bbcca3b0c67ec5ad)
This reverts commit 74fb298c711e61ae98c210d0ed11c875ce6ba591.
We've decided this experiment didn't work out.
(imported from commit 8acd3a030b692f9397155a20a9f89e63ed6a6cf7)
We had a few bugs where we were using a raw Django database query to
get a UserProfile object. This might seem OK, but going through
memcached is more efficient, and also guarantees that we get back the
.select_related() version of the object, so that if we later access
related fields like user_profile.realm.domain, we don't end up doing a
second database query as well.
Fixing these should in practice save a substantial number of database
queries on handling update_status_list requests, which happen very
often and access user_profile.realm.domain.
(imported from commit 0a2027da1b5bbc7a4f6c6927aca498530d7a4977)
Instead of calling test(result), we emulate the behavior of
calling SimpleTestCase.__call__, minus the results protocol.
(imported from commit e12e3d5caa8726c050a2a7cfa583ca4d571d6cbb)
Displaying the badge gets us a bunch of free mixpanel data points/month.
We only display the badge on the login page (which `/` redirects to)
and not on other portico pages.
(imported from commit 75871173a10ae888fdb7c92fe8e20586ce60c6bf)
The refactoring to use the cache_get() method incorrectly didn't
remove the addition of KEY_PREFIX inside cache_with_key. The result
was that the KEY_PREFIX was being added twice, once by cache_with_key
and again inside cache_get.
This had the impact of causing pointer saves to not take effect,
because our attempts to update the memcached cache when we save the
UserProfile object were using the correct cache key, but the actual
code reading values out of the caceh wasn't.
(imported from commit dcea000833f00622bdc0249488de3b186a7417b2)
The test_emoji test makes sure that we build the correct image
tag for all of our emojis. It's the same underlying algorithm
for all 800+ images, so we now test it in O(1) times instead
of O(N) time. Randomly sampling the emojis makes the tests
slightly non-determinisitic in a theoretical sense, but not in
a practical sense.
(imported from commit c78b50072953f2c081fca0272b9ecf762a4802db)
(We don't need the fixtures code once we override Django's
run_suite() method not to clobber our test database, since
all our tests use the same data anyway, and since Django
runs all the tests in a transaction and rolls back their
changes.)
(imported from commit d9daf09be3967de2de30934173b882d3d6b9ea27)
In tools/test-backend, we build the test database before calling
"manage.py test, so there is no reason for Django to wipe it
clean.
(imported from commit 7b09e9cc1a9259900ec1cc4f9ebbcacab1efb265)
(We now have a full replacement for TestResult, but we don't
support the full API of TestResult. If folks were to add new
tests that caused TestCase to call our class with an unsupported
method, this diff will provide hopefully a more clear error
message.)
(imported from commit 92033b12a1a901bd649ecb03e08779aa76bfef7a)
We use a simplified version of run_suite in our subclass of
DjangoTestSuiteRunner so that we can have more control over failing
fast and indicating progress. This change also speeds up the tests
by taking roughly 200 lines unittest/Django code out of the picture.
(We basically reinvent a simpler TextTestRunner and TextTestResult.)
(imported from commit 5216b6b582b071bd7c9e923c9886c3bc61faf581)
(This commit will probably be squashed, but it makes the subsequent
diffs more clear with respect to how we remove functionality from
run_tests.)
(imported from commit 9d0edcd5282a0f90ebd38e0ac229ef874dc00504)
Otherwise code paths that use those keys, like get_old_messages, will
incorrectly use the prefix-included keys.
This bug in our KEY_PREFIX system results in our memcached caching for
get_old_messages always missing.
(imported from commit 506c13e06d6f266596ead0b381c324c256e576c3)
If you click on "Streams" from the gear menu, we want to focus the
textbox to create a new stream. But if the Streams page is brought up
programmatically (e.g. to jump to a specific stream's settings), this
commit makes it not focus the textbox.
(imported from commit 900bdafb701180eac1d284120a91ea2a84d7177b)
The sidebar link now uses a one-time event handler for a custom event,
subs_page_loaded.zephyr. If the streams page is already open, we can't
rely on the event so we expand the stream in question immediately.
(imported from commit 3c22e1791d238a3be4a73edcfb5456e392cee608)
Get data directly from the main user list, rather than maintaining a
separate list just for autocomplete.
Fixes trac #1362 -- Does not depend on historical messages, so
will do the correct autocomplete after a single reload.
(imported from commit 6b35a709dba3384530082e8cfacf0151f9e0eb26)
Based on https://github.com/twitter/bootstrap/issues/6579. Doesn't
look like upstream is going to fix that or its 6 duplicates any
time soon.
(imported from commit b3e887d929cc42124aeb1f13abd7503f38f92a1f)
Wrapping render_to_response never actually worked correctly. On the
login page, mixpanel_token would be missing, but we wouldn't get an
error because it is surrounded by double quotes, which meant that it
was still valid Javascript.
(imported from commit 820ee42fab8f679983e5a3a4309a2feaf690f20f)
I don't love the default reloading message, but I'm not sure how to
tell the user why we need to reload in several cases.
(imported from commit 5f2eecdb0d83f31bd60673b247e8aac8f08ce20f)
This is a pure refactoring. We set messages.sent_by_me inside
add_message_metadata(), which gets called early in a message's
lifetime.
(imported from commit 2f6270f92ecdf0e4a64401030f86d89a8fb7be2e)
This updates desktop notifications to use avatar_url from the
server (in case users have uploaded their own avatar). It
also removes the unneeded stamp parameter from the URL, and
it unifies URL handling with message_list.
(imported from commit 6bb43a25c01cc3d26f30fc167780a477d1c5d023)
Trigger a repaint only on the messages that change fade state,
rather than repainting all messages.
(imported from commit d3a225448c004a4b467e51ceb97fda06a7a6f04b)
This should substantially improve narrowing performance, while still
delivering enough messages that one should only need to contact the
server again if the user scrolls up a significant distance.
(imported from commit 2b0af0e5b9d3a449e7f195a87d57d517b47f0b3a)
This change backs off from the always-open-compose-box paradigm,
so that when a user "closes" the compose box, they can no longer
type text, attach files, etc., and instead, they have buttons
to reopen the compose box.
(This change does not affect the feature that when you're in reply
mode, the compose box stays open.)
(imported from commit 32305d5b35f457a83f126a2265defdd98e885a2b)
Disable the hotkeys that act on the current message, except Enter
immediately shows the cursor without moving it.
(imported from commit 74fb298c711e61ae98c210d0ed11c875ce6ba591)
Hopefully this time with fewer stack overflows. It sets the flag
that will break the cycle *before* triggering another focus event.
(imported from commit 5e3ed74f1eeab8a18e36525ae7c8f1f1c756eb58)
Some versions of IE (specifically on Windows Media Center PC, it
seems) lack a console object. For such browsers, we just black hole
any output we would have sent to the console.
(imported from commit 30151c60a68a47990bf8f9be4476b716352befab)
When a user sends a message, it should be considered to be "read"
by that same user, but all that logic is handled on the back end
now, so we can remove some of the front end code related to
saying that a message is unread.
(imported from commit e4263f86c666882db42d7ae3d399196803d700cd)
This cleans up most of our blocks of code that assume in any narrow
that the only operators present are the ones of interest and that they
always appear in the expected order.
(imported from commit 038707aefbe125b0c14f823fa93472fd40302e20)
This is essentially a bug fix. It was pretty clear that the
original author intended to stop polling once the gravatar was
updated, but they checked for the updated flag before the callback
completed, instead of inside the success callback, so it wouldn't
stop polling regardless of the update.
(imported from commit 7998c6890a26a008810b8a6d8e7998a53c6e175d)
Before, sidebar_li, used in rebuilding the streams list
during a sort, was set to some HTML rendered from a template.
Now it's set to a jquery object that is updated when the DOM is
updated.
This resolves Trac #1310
(imported from commit ba96d9da4deebf2f674f2c093e81b3f0032a3fe4)
This change makes it so we now keep track of full stream information
for both subscribed and unsubscribed streams in our frontend. Previously,
any unsubscribed streams had no associated data.
(imported from commit c445b19abe11c43c710c264fffcf3af5097deb6c)
This seems to only happen with Firefox for some reason. We've had
similar issues with other ajax endpoints, so this is probably not the
long-term solution, but I'd like to get this traceback fixed now.
(imported from commit aa552fb56882ae2c73e352c7baf9532a88c5cf0a)
Show user-uploaded avatars on the website for users who have
UserProfile.avatar_source == 'U'. (Continue to show gravatars
for other users.) This includes the home page, the visible-phone
div, and the settings page.
This fix does NOT address a few things:
* There is no GUI to actually upload user images yet on the website.
* The !gravatar syntax in bugdown will continue to show gravatar images
only.
* We are not changing identicon behavior.
(imported from commit 9f5ac0bbe21ba56528048233aab2430e4dd431aa)
I moved the list of internal users out of populate_db.py and
into settings.py, and I removed some dead code related to sqlite.
(imported from commit 1e080716dc296c05f51cdd229911082469de64bd)
Don't show an error banner for any uncaught JS exception, as often the app
will continue to work fine, and there's no way to dismiss it other than
reloading the page.
Also, don't show a transient "could not connect to humbug" error if
the check-for-messages-in-narrow request fails.
(imported from commit 2c634ba088b58c17fa5b2e3353b0589d40b8e357)
It makes the stream and subject sometimes not auto-fill when replying.
This reverts commit 86603aefbbcd5f766b0c397583483810948046de.
(imported from commit 934e991566fa7a082ab8e2ba661ec973bce46b85)
Treat shift-space like page-up. Let the browser handle
shift-page-up, shift-escape, and similar keystrokes.
(imported from commit 31d5c5eb1dd4af7228c5e7794fb4cffc4bd8e88b)
I extracted get_event_name(), which should help isolate
the problem of identifying keys from the specific mechanics of
dispatching actions for given keystroke events.
(imported from commit 058c0749016dc17cce554788e10ccb32438e9dfe)
ce4e860a introduced CSS `.alert{display:none;}` because alerts are
always included in `/signup/` and shown by JS. Use a new `.alert-hidden`
class for this purpose to avoid breaking other pages.
(imported from commit 199ba35dd3356bd4093aac2a54181331b3993ee8)
The new file can't be called logging.py because then it would be
annoying to import the system logging module within it.
(imported from commit 71d116e4be98d45b09dda049a43142a82647b727)
This, in effect, reverts ff0c27ccb177ddc69a31bf8997d31e7cfb5b78b5.
The rationale here is that actually we look pretty good with the
browser's own zoom/font-size-resize in Chrome and Firefox, and it's
better to let the browser handle these kinds of changes than us.
(imported from commit 5949b57bdaf20d4fdf2bbd7ed89d1285a8b8e453)
"(" and "↓" share the same e.which, but only "(" has a non-zero
charCode. This commit will start checking for non-zero charCodes
for directional keys.
(imported from commit bcb8c3c5ef2c13708fd04cca5f4d8b0f65beaa84)
1) When you send a message, restore the focus to the composebox, targeted at the same recipient
2) If the composebox is completely empty and you press up or down, have that close the composebox and take the appropriate action
3) If you started the compose via a reply option (r, enter, click), don't refocus the composebox if the cursor has changed.
(imported from commit 84545e49d06959eb62e7fd2b22e1387383df6d1d)
I tried to remove the line of code that removes the old
subjects as part of rebuilding the new ones, but that line
of code is still needed in places.
(imported from commit 97621553c267a79f33d34537a67101464bdac434)
Previously, we would only collapse the old subject list if
the new narrow had a stream operator.
(imported from commit 664f984d932d0968a9b901f2a09272e11138843d)
Before this fix, you could expand a stream, and then any
subjects that already had a zero count could not be
incremented when new messages came in, until you rebuilt
the subject list again.
(imported from commit 98c95e201f6ec745d7c857da6f42495c8bf88ee0)
(I also introduced a couple local variables that would have
made this and similar problems a bit more convenient to debug.)
(imported from commit 6793c16ffb17514fd9b5a069d384d2c74dac6111)
If you clicked on the unread counts span inside the right sidebar
links, e.target would not be the link itself but instead the count
span inside the link, so the extraction of the user's email address
was incorrect.
(imported from commit 559d93622078e4d909f60de794df3f039ea7e5f2)
The message_viewport_info() function encapsulates our logic
around the compose box and other elements blocking the viewport,
so viewport.js seems like a more logical home for it. It also
makes ui.js, one of our largest modules, a little bit smaller.
(imported from commit 7838668b28175e161b87a6d7a8124b73012f0ff3)
I've tried to do this in a way that's scalable and easily configured,
so that we can add new such filters for customers on-demand without
needing to add anything other than a bit of configuration.
Once we're confident in the arguments to this system, I think we'll
want to move the regular expression lists into the database so that we
don't need to do a prod push to modify the regular expression lists.
The initial set of regular expressions are:
(1) Linkifying e.g. "trac #224" in the Humbug realm, so we're exercising this code.
(2) The various ticket number things CUSTOMER7 uses for the CUSTOMER7 realm.
(imported from commit 992b0937b9012c15a7c2f585eb0aacb221c52e01)
Jeff Arnold
Tim Abbott
Steve Howell
Kevin Mehall
Luke Faraone
Jessica McKellar
Waseem Daher
acrefoot
Scott Feeney
Allen Rabinovich
Zev Benjamin
Leo Franchi