Later, we can fork off a separate process to do that task, but for
now, we can just rely on the fact that unless the server was just
rebooted, the relevant messages will already be in that cache, and
even if it was, getting the needed messages once each is not a
particularly expensive query.
(imported from commit 6d08eba1b41237fd4e1204e181ce8f227573930d)
It can easily cost an entire line, and the information is available by
clicking on the sender's name. Plus on a phone, you can't hover
anyways.
Annoyingly, this tends to put the popover partly off the window, but
there must be a way to fix that separately.
(imported from commit 19334cb067981b323e300245654c83c8e545fb2e)
It's possible for selected_message_id to not exist in zhome. For
instance, when you open the page to a narrowed URL, there is a race
between loading the narrowed messages and loading all your
messages. If the narrowed message request completes first, it will
update selected_message_id to your initial_pointer if the latter
happens to be among the narrowed messages. Even if it is not, you can
select a message by clicking on it before the zhome request finishes.
Then the old code would never add that selected_message_id to the
zhome table, so it wouldn't show up in the Home view or if you
re-narrowed.
I'm pretty sure there are still cases where the selected_message_id
might be outside the range of messages in zhome, in which case adding
to zhome might put old messages at the bottom. I observed this twice
on staging but was unable to reproduce it consistently.
(imported from commit 162feff3090f8806cc67140db0cfabb6e965aece)
The schema change made this no longer a guarantee, and creating duplicate
PUs is not harmful.
(imported from commit ca05b4dad7d3cc12a35548817f98c0a072c14c8f)
This involved splitting off into a separate function the second step of
the two-step registration process.
(imported from commit bd1ec9b5abb15d29b0585873daaa54a8d4e26556)
Added field invited_at on zephyr.PreregistrationUser, with a one-time
default of Jan 1, 1970 for existing objects.
Added M2M table for streams on zephyr.PreregistrationUser
Deleted unique constraint for ['email'] on zephyr.PreregistrationUser
(imported from commit 85247acb488201f8fc51dfaae354423c27eddcb0)
This allows us to handle the return_messages_immediately part of
get_updates requests without having to talk to the database.
(imported from commit ed0b7742d359efb21a0a4960f4fc25f4337e9ad4)
This commit has the effect of eliminating all of the non-UserActivity
database queries from the Tornado process -- at least in the uncached
case.
This is safe to do, if a bit fragile, since our Tornado code only
accesses these objects (as opposed to their IDs) in a few places that
are all fine with old data, and I don't expect us to add any new ones
soon:
* UserActivity logging, which I plan to move out of Tornado entirely
* Checking whether we're authenticated in our decorators (which could
be simplified -- the actual security check is just whether the
Django session object has a particular field)
* Checking the user realm for whether we should sync to the client
notices about their Zephyr mirror being up to date, which is quite
static and I think we can move out of this code path.
But implementation constraints around mapping the user_ids to
user_profile_ids mean that it makes sense to get the actual objects
for now.
This code is not what I want to do long-term. I expect we'll be able
to clean up the dual User/UserProfile nonsense once we integrate the
upcoming Django 1.5 release, with its support for pluggable User
models, and after that I change, I expect it'll be fairly easy to make
the Tornado code only work with the user ID, not the actual objects.
(imported from commit 82e25b62fd0e3af7c86040600c63a4deec7bec06)
Otherwise one gets:
AttributeError: 'module' object has no attribute 'time'
when trying to use the time module from inside zephyr.lib.
(imported from commit 645368672a3eff68320278dd480edeed56721fcc)
We suspect that these seem to be causing a regression where scrolling
in narrowed views gets really sluggish, but we haven't totally been
able to figure out why since it's challenging to reproduce locally.
(It currently manifests itself on staging but not prod.)
So for now we'll back them out. Here's the full set of things:
Revert "Cause update_floating_recipient_bar to get called less frequently."
This reverts commit a6c1518c4001a2dde44d7b512236795da3ccd351.
Revert "Remove double-scroll in un-narrowing code."
This reverts commit 3dde6c27ffa1e8afa1a084b1b2baee3bc0512962.
Revert "Reset our scroll position if we change our hash to "#"."
This reverts commit 925b44d770c96dafaabebc9e0114f9a3b8f53c4d.
Revert "Properly update floating subject bar when you are at top of page."
This reverts commit 6633cc8a81aedcbb31b30d7c1f27816f8808c700.
(imported from commit a273730581cef30c33bedf701659ee084434f8ad)
Note that on local dev servers, this will print out every half second, as
Tornado polls for file changes for autoreloading. In production it will only
print out when network events occur.
(imported from commit adfe88879e4e446b7dfa6ee69e0a9ad013e9c4d4)
This should remove a database query invoked from Tornado by our
decorators in the common case.
We use hashlib.sha1() on the key because memcached doesn't support
keys with whitespace in them.
(imported from commit 351ef4c76bc68e2d53f0e7ddca91dcf95e7bb2a6)
Previously it wouldn't work due to using the wrong port numbers.
This commit also has the side effect of fixing the fact that our
frontend tests would send real emails and log events to the real
message log.
(imported from commit f2cf400e6061c089627acba2759d588981ecf5bb)
This should save a database query when we later need to access fields
such as the user's realm name in format_updates_response.
(imported from commit ceef726db9e917cfb0b47061130d7299ee64890d)
This code is from a previous protocol where the users array was posted
as a comma-separate list, rather than encoded via json.
(imported from commit 3a2edfcbf250a7eef305b2e98018c1361cc2fffe)
This is for consistency with the rest of our code dealing with message
delivery, which always uses the user_profile_id.
(imported from commit 5bf10bb9b994b0a98d3a22bd0bd86e542ab8a2ee)
Putting update_floating_recipient_bar in the old location caused it to
be called on every single keypress, which is unnecessarily
expensive. Instead, just call it once when we think we might actually
need it: after initiating a narrow.
(imported from commit a6c1518c4001a2dde44d7b512236795da3ccd351)
Watching new users, I've seen them not notice the pointer and
mix. Give them a little more hinting about what message is selected.
(imported from commit c98e22dcef881ed7400071ec438a6e91d6cd3d9e)
select_message_by_id with then_scroll: true already recenter_views
on the selected message; no need to also call scroll_to_selected.
(imported from commit 3dde6c27ffa1e8afa1a084b1b2baee3bc0512962)
Changing the hash to "#" causes Chrome to jump to the top of the page
on Mac OS X. This commit doesn't actually fix any bug, but it
is necessary for my *next* commit, where otherwise you'd have to
ensure that the scroll code came *after* the hashchange code.
(imported from commit 925b44d770c96dafaabebc9e0114f9a3b8f53c4d)
There's this very edge-case issue which is: if you go to the top of
the page and narrow to something other than the top message, the
floating subject bar does not update.
Why? Well, the way that the narrowing code works is that it sets up
narrowing and then calls
select_message_by_id(target_id, {then_scroll: true});
so that our selected message is in the view.
This in turn calls select_message, which calls recenter_view as
appropriate. This usually causes a scroll action, which in turn causes
the floating recipient bar to be updated.
But when we're at the top of the page, recenter_view doesn't need
to scroll at all! So the bar remains un-updated. Here we explicitly
update it to guard against that case.
This fixes Trac #651.
(imported from commit 6633cc8a81aedcbb31b30d7c1f27816f8808c700)
I don't view this as a complete solution to
Trac #466 - Make the fact that you are narrowed more apparent,
but it's a start.
(I think a real solution would have to give you something that
helps you make the determination of "Is this view narrowed?"
when you come back to your computer, and this animation does not
help you do that.)
(imported from commit eb3646f3f3a4e25a43266e9146308633fd997eb2)
This commit just moves around some lines so that the code that
hides the main view and shows the filtered view, or vice versa,
are together so that it's easier to reason about the sequence
of things that's happening.
(imported from commit 7e99f45293c0e1a4cdfa1a08f41f8c770c370d6c)
This used to be a button that let us un-narrow, I guess.
A git grep for it after this commit turns up no actual
references to it.
(imported from commit 05acb4bb40da1b032f548c511fbae5b2b20874a8)
And change the color to a more thematically appropriate blue.
The shadow pointer is sort of confusing; we should really provide some
different sort of indication that your pointer is potentially moving
on narrow-and-unnarrow. (I think my fade-in-fade-out later in this
commit series is a not-bad first crack at this.)
Resolves Trac #472 - Dual pointers in narrowed view can be confusing
(imported from commit 2450517d99de85ade1c0e98c5510b59e70282451)
tornado.web already does this, based on the setting of the 'debug' kwarg.
Dropping this in production saves us waking up twice a second to stat()
a bunch of files.
We already explicitly restart the server on deploys.
(imported from commit 283bb0da609acb2699a04111a74c13224fe5124c)
If you narrow to a view that only has one or two message, sometimes
the grey box gets cut off and doesn't go to the bottom of the
page. This fixes that.
(imported from commit 55724d03aa30922d91bd33fab4447d889be78889)
CasperJS can't handle them; window.webkitNotifications.requestPermission()
throws a type error. We can revisit this when we want to write tests for the
notification code.
(imported from commit 90f4d6ac3ddb387e74051b9af2c230698fa94479)
We apparently cannot rely on Iago to consistently be subscribed to
"Denmark", so make this determination some other way.
(imported from commit 2a75b345c2d82097ab44538942af89536aac09ed)
Previously, if last was None, we wouldn't check dont_block,
server_generation, or any of the other reasons that get_updates might
return immediately, and just unconditionally entered longpolling mode.
In the process, this reorders return_messages_immediately to have
fewer cases and thus be easier to read.
(imported from commit 67803b8bfc7d9c9c1a4d6916eb2fb62664fb35a9)
This check was a workaround for the fact that the browser client
submitted a "last" value of -1.
(imported from commit a668f6a4e7a0c027f1214166a9bbf40d29b5daeb)
We shouldn't deploy this change until strictly after we deploy
"Fix website improperly submitting a last value of -1."
or we will break website clients.
(imported from commit 7f682ab0f7060b677f53f0a0073faef216f45d00)
This view lives at /accounts/accept_terms, and (after getting an acceptance
from the user) sends an email to all@ documenting the acceptance.
(imported from commit 8f64286ab02887fd6544fa274b2967f6499b6dbc)
So, I got annoyed that our test suite was taking forever to run:
real 2m13.443s
user 1m32.630s
sys 0m3.748s
Some quick profiling determined that the test suite is spending all of
its time loading the fixtures files (zephyr/fixtures/messages.json)
that it loads for each test case (3s to load that for each test case).
To improve this situation, I cut out from the test database used by
the test suite most of the users, subscriptions, etc. that aren't
being used directly by the test cases. The impact is a quite
significant speedup:
real 0m15.176s
user 0m9.161s
sys 0m0.508s
We're still spending over a quarter of a second per test, which isn't
great -- but this is at least no longer unbearable.
This commit doesn't make any changes to the populate_db output if you
don't pass the new --test-suite option.
(imported from commit 2334ba5399b33edab3d29ff269fde4ea77ccd48e)
The initial rationale for hiding the floating recipient bar
was that it duplicated information that was in the "narrowbar".
Now that this no longer exists, let's *always* show the
floating recipient bar.
(Yes, there is some duplication of this information in the
search area, but I think the situation is fundamentally
different now and would basically like to see it everywhere.)
(imported from commit 6fd4506c2f48caade9496139e580e6550252ce8c)
Apparently after you call waitForText, if you don't specify
a callback function, you need to put the next stuff after
a casper.then() -- not doing so caused some tests to fail
if, e.g. the subscription list did not load super-promptly
(because we checked for the text even before the waitFor
expired; in other words, we were not blocking on it.)
(imported from commit c71d543db0aba0c27b5136b92bb6e28e63278ac5)
Inspection of the postgres slow queries log showed that the "narrow to
personals with a particular user" database queries were taking a long
time to run (0.5s+). Further investigation determined that the OR
gate construction used here was causing the entire zephyr_message
table to be scanned; primarily I think because we were using the
implicit constraint that the logged in user had received messages.
This change makes that query explicit (improving performance), while
cleaning up the code to avoid an unnecessary query and read a little
more clearly.
After this change, the relevant database query takes 10s of milliseconds.
(imported from commit 020f5af5846c958386615e37ea9318383bf99ca0)
Alternatively the server could return a successful result with an empty list of
messages. But I prefer the solution in this commit, because it would allow us
in the future to warn the user about the problem. It does allow users to
determine if a given stream exists, but we haven't tried to hide that
information so far.
(imported from commit a91e12c90b12d3c870c0b637c3f1d6d3cef88491)
It's cleaner if the filtering code recognizes only one value.
We can add this back in by converting in the parser.
(imported from commit 453b7b01e094955c6d66be63b5d997cc56b50a35)
Show the buttons iff
- the search input is focused,
- the search input has non-empty contents, or
- we are narrowed.
(imported from commit f5c98471a2db4ab522160960dd1271471a9db555)
We don't require that the parsed form be lower case; that's handled by
narrow.activate. However we unparse as lower case, in order to give the user a
hint that matching is not case sensitive.
(imported from commit 2882b440deb59a049b095db7a13cfc18e047caec)
Also removed .show()s for the alert on does-not-exist and not-subscribed, where
a blank error would display. This should fix the underlying issue with #166:
that hiding the composebox before send_message() was called would hide server
errors.
(imported from commit a8a50cdf82ddf1d15f1e405432ff3bbfdb7a491a)
This is needed to avoid exceptions trying to do internal_send_message
in any test against a simple populate_db database.
(imported from commit 36927f57cbbb7e30ae249b5f1a0549fb352827f5)
If you have a lot of subscriptions that you're trying to modify,
jumping back up to the top of the page is very disruptive. We still
show the success message, which has the effect of scrolling the page
and is thus surprising, but that's better than the user completely
losing their place.
We do need a story for informing users about failures to subscribe or
unsubscribe, though. We currently jump back to the top so they can
see the error, but that's not optimal.
(imported from commit 48d938ddc47f286a72e2147f4459b91ca5684e36)
This reverts commit a590bf6b8ee733893d3410ecb5eebe54141c48ea. This commit broke
the test suite because it was not tested after rebasing with Keegan's changes
to the tests.
(imported from commit 7248a55328609973c5303be6c85eeb5fbfc1475e)
GetOldMessagesTest had test methods that weren't included in the test suite
generated by Runner because they did not have "test" in their names. A few
bugs in these methods that were overlooked because of this were also fixed.
(imported from commit a590bf6b8ee733893d3410ecb5eebe54141c48ea)
For debugging in case this is ever different from platform.node(). I
think this would happen when using a CNAME, for example.
(imported from commit 47f6c3490712a3ac1c6a16f9146c2ef3ca8fc5e8)
This essentially reverts d900957e468815bcb99de67d570dfd7ce4413220.
This code was consuming up to 50-100ms per client recipient of a
message, so for any messages that would go to 50+ browser windows /
mobile devices, it would take several seconds to run, during which
time Tornado would be completely blocked.
In the future, we can re-fix #174 using a cache of recently delivered
messages, so that this code block doesn't go to the database and thus
can run instantaneously.
(imported from commit bdfa1664210429411737f70cde54ab5a56525341)
For whatever reason, specifying a percentage for bottom_whitespace
seems to cause issues in some browsers, including Firefox 17
and the Firefox Nightlies.
This is a bit confusing to me, since bottom_whitespace is basically
immediately resized by resizehandler initially anyway. But hey.
(imported from commit 93da101edeb6f16b01a92aed775e9117c0295086)
CSS height percentage was not working because parent div has an undefined
height, so instead it is set to 40% of the window height on resize (and initial
load) via JavaScript.
Fixes trac ticket #24.
(imported from commit 2c6a8489585c4bf70c44469ce8628264ec3fbc36)
Variables like stream, subject, and message were getting cleared from the DOM
when the compose box was closed. The "Create and send" button was trying to
access these variables to create a new stream, but they were gone. Now they
are cleared when a new compose is started.
Fixes trac ticket #568.
(imported from commit 39ccaaeacb3f92f4b1d771be1b34ff660e0d5883)
We now encase the request info in a preformatted block, which ensures we
won't accidentally trigger any formatting while being reasonably
space-efficient.
(imported from commit 7c69a6ff2b2abd9474aae08b5ba608bcb40cec56)
This should really be handled on a per-method basis, but in general we
don't want "password" or "key" to be sent to us for security reasons.
Addresses trac #569.
(imported from commit 1c246fce00f3740977c595641341ee36eb5ed831)
We were submitting a 'last' value of -1 to the server at startup,
which is invalid but normally ended up being OK because the user
usually had no messages whatsoever or had last be updated via
get_old_messages before the get_updates call went through.
(imported from commit df55ac1cdac443721c06ebed94a1c4b3ec7af2d1)
Importing zephyr.views here has the unfortunate side effect of
creating Client ids 1 and 2 automatically (via decorators.py
instantiating the two client objects it makes), before we go ahead and
delete all objects in the database as part of the populate_db startup.
(imported from commit da03cb7606334d5926e42f422ab94d1c884937b9)
This was not totally effective, and with the previous commit it is no longer
needed.
This reverts commit e86c0b653669cf86b0d8956c2c85eb7610fc342f.
(imported from commit 0de5bfec87147b1336f6f79c33d4e32493e1e508)
Previously, the StreamColor restore code didn't properly account for
the fact that most user subscriptions were in pending_subs and thus
not yet in the database.
(imported from commit 2e28c5a68aa045494b9336d7114c23f5c3706c28)
By processing UserMessage objects in batches as we go, this avoids
consuming a large amount of memory that is linear in the size of the
messages log.
(imported from commit 0c42d97f0863da9c079836c60bebcbaeec59f849)
This was causing issues with our ability to unsubscribe from
streams with " in their names.
The solution here is a bit hacky, since it depends on the JavaScript
being fairly aware of the layout of the DOM, which is not great.
But it works.
This fixes Trac #328.
(imported from commit a1b6c8e1f3a9daacdc48920a195717aa89b3a9a9)
This fixes Trac #522, which previously prevented you from
subscribing to a stream named
'"]'); alert('hi');
This does not fix#328, which is that you can't unsubscribe
from 'Waseem', among other things.
(imported from commit 869063cafa9e7e988aea993d072ca1ad880bcee1)
Unfortunately, this doesn't actually give us much performance gain
either; it's not really the calls to 'find' that are taking any time.
But I do find this a little cleaner as well.
Simply initializing 100 colorpickers with our options takes about 700ms.
Initializing ~100 colorpickers with the total default set of options
shaves that down to about 300-400ms (though obviously doesn't quite
achieve what we want).
(imported from commit 7084b35fb6e77600edfcdcfcc2761a11e6f38c03)
Rather than calling the template generating code once per
subscription, let's just do it in a batch when possible.
With about 100 subscriptions, the "fetch" call takes about 800ms to
render (while testing locally) both before and after this change,
which is somewhat disappointing.
But this *is* cleaner!
(imported from commit 9ba8819524da86c00a2508349be0ea0ddd48606b)
This is useful when testing the sigup workflow, as this script enables you
to run through a MIT signup without manually creating a new inactive user
in the database.
(imported from commit c22649cc7c561c2fbe8682d1b17d7e5aba9ac04e)
Hopefully this will make things slightly more discoverable;
the previous solution (putting a prompt in the initial text)
was not that discoverable.
(imported from commit f6a7fce1bfd27bda412522768e981b2ffc39f474)
This fixes https://trac.humbughq.com/ticket/546.
It's a little unpleasant that this special-casing lives in hotkey.js
-- instead you could imagine doing something where there was a whole
special set of hotkeys when a search is active, like we do with the
composebox, but this gets the job and is probably simplest.
I would have written this as a case that just falls through in the
else condition, but jslint doesn't like that.
(imported from commit 65a1b8aa1efc356b6690dc177058a4fb9e12745a)
We previously were only using it at the first loop through all
messages, which meant code accessing the message type copied from one
place to another would break (potentially subtly), because things
would work if and only if the very last message happened to have the
same type as what is expected in the relevant piece of code.
(imported from commit ad9ce5efdb200e0c0d5c3ffa6db33113fdad8c5a)
This addresses Greg Price's feedback in #527. We now distinguish
between normal pushes, force pushes, and branch deletions.
(imported from commit 0fab6055f63ffc7e6df283b8bb8ed9971000e6d5)
This cuts a 30s operation down to about 2s on my machine.
And also move the code to run before we print the "done" message and
have logging for how long it is taking.
(imported from commit 2f20f8ca3fee714735a50fe6c6cfd630df452768)
This is syntax like
Here's [a link][]
[a link]: http://google.com
This is not very useful for short chat-style messages. It will confuse users,
especially because we don't document it. And disabling it saves the effort of
applying the same link fixups as elsewhere.
(imported from commit c23391465486db545302b79c084b4f9cd5cdcc6a)
This was a really cute bug where our layout messes up if you resize
the page while "Subscriptions" (or to a less visible extent,
"Settings") is active.
The problem here is that we compute the size of the top navbar
based on the size of main_div -- but when main_div is hidden,
it has a width of zero!
We need to instead look at the width of the pane that *is* active.
Resolves https://trac.humbughq.com/ticket/216
(imported from commit adbef00d190845f90c5cfdb46df4ec7b703635ef)
feedback-bot and zephyr_mirror will need to be updated and restarted
when this is deployed to prod.
(imported from commit fe2b524424c174bcb1b717a851a5d3815fda3f69)
Ironically, I think this might've bee introduced by
commit ca35321c02d5e79e4f9c439a662805c016a333ed,
'Fix "resizing window breaks in Firefox" issue'.
Basically, when the window is 776px wide according to
window.innerWidth, that's the width not including the
scrollbar. However, in Chrome, the media query seems to ignore the
width of the scrollbar, so from the CSS's perspective, the window is
actually ~766px wide, so it goes into condensed mode.
But the rest of our code doesn't, which causes the break.
A bit more on this browser-specific difference at:
http://www.456bereastreet.com/archive/201101/media_queries_viewport_width_scrollbars_and_webkit_browsers/
So the issue we have is, to match the CSS's behavior:
* In Firefox, we should be listening to window.innerWidth
* In Chrome, we should be listening to window.width
We fix this hopefully once and for all by using window.matchMedia --
aka the exact same query that the CSS itself uses. As discussed in my
last commit, this feature is unavailable in IE<10, so we provide a
potentially more fragile fallback, i.e. what we did before this
commit.
(imported from commit d8e6425b81c90c8e0fdda28e7273988c9bfd67ec)
Make it so the image is not squished, change some paragraphs into
headers, and wordsmith a bit.
(imported from commit 81295e1a8ddd4f1ecd4532c4dfb8a38467bb530e)
This is an interim strategy for user education that'll be a stopgap
until we build something in the app itself.
(imported from commit 9022d4ceffca98e127f7045f73c012857fe6fc54)
When we changed our stream name model to treat stream names as
case-insensitive, we didn't update populate_db to do the same.
This commit makes that update, which is to use the lower-cased stream
name for dictionary lookups and deduplication, but the original-case
stream name for actually creating streams.
(imported from commit fc32ec75a5ae286bce7ec86c6e6fb6893888cbd0)
bulk_create_streams was taking about 10 seconds to run with prod data;
this should be a basically immediate operation. The cause was a
missing select_related on one of the loops through all streams.
(imported from commit 8b82f0c41facc3999bb699dbc350708ac69797e9)
in the narrowed view, not messages older than the oldest message in
the home view
Tim provided most of the code for this patch
(imported from commit ec0bbfd344cac351f56a456fc560848603721135)
The transaction.commit() line inside the except IntegrityError clause
doesn't work unless we've entered transaction management.
(imported from commit 2ae520e05c9a19ec35af7c244631b01d4b9598d6)
This makes subscribing to zephyr classes for the zephyr class
mirroring bot a lot faster, since we don't need to subscribe to the
third of our streams on which no users will actually receive messages.
(imported from commit 029b7fb260b480db5599e3c9f9effc502f6d8b59)
The tabindex on this link doesn't actually do anything, because of the Safari
tab workaround. I added it anyway in case we remove that workaround later.
(imported from commit 11f31f2561907300b350c11732be88589d736f94)
This fixes a bug where you could click on the already-selected mode button, and
it would switch modes without changing which button is selected.
(imported from commit 263ee0b52ba844c52a3a60968bb1bbff73482412)
With the removal of process_compose_hotkey, the state machine now has only one
state. Everything else is based on things like "is a text box focused right
now", which is probably a better approach.
(imported from commit 0e39c03956d28e30d2bdbf3b285410ad0cacca3e)
Tim Abbott
Reid Barton
Luke Faraone
Waseem Daher
Jessica McKellar
Zev Benjamin
Jeff Arnold
Keegan McAllister
Tyler Hallada
Jacob Hurwitz