Summary:
the JS is fragile with respect to the tokenizer coming in from the people widget. make sure to always try to load this up. Note this generally needs to get cleaned up where the server should only send down the *exact* bits the client needs. This is all TODO as part of getting this on mobile perfectly. Also note this fragility exists still in that you can break conpherence by clicking quickly before the initial tokenizer load loads.
For old bad data, at some point we weren't updating participation as well as we do today in the editor class. the result is with the migration and code change some conversation participants have bad "last seen message" counts. the simplest case is the test user talking to themselves -- threads before the editor code fixes / changes will have the entire thread as unread for these folks. The other buggy case I saw was where the "last reply" to a thread wasn't being count. These issues showed up for threads February and older which is old.
Test Plan: edited conpherence meta data and no JS bugs. pontificated and no JS bugs. added a person and no JS bugs.
Reviewers: chad, epriestley
Reviewed By: epriestley
CC: aran, Korvin
Differential Revision: https://secure.phabricator.com/D5622
Summary: also re-enables the updating of the widgets and "cleans up" the javascript a tad. Ref T2867
Test Plan: all sorts of conpherence fun like adding people to threads, adding files, pontificating, etc
Reviewers: epriestley, chad
Reviewed By: epriestley
CC: aran, Korvin
Maniphest Tasks: T2867
Differential Revision: https://secure.phabricator.com/D5595
Summary:
Refs T1048; Fixes T2902 - (Probably) fixes @vrana's issue. I managed
to repro it on Ubuntu FF (though on Windows with 1.0GHz/512MB it's
really worse...).
This revises the approach to the graceful degradation for
`too-far-to-the-screen-edge-edge-case`.
I noticed that `x` could become very negative, up to about ~`-170px`.
This is due to the //"already-on-the-left-side"// nature of object tags.
`50 - 200 - 20` = `negative`. Adding `100px` (node.dimension.x / 4) to
that won't really help, as the hovercard would still be offscreen.
Instead, display them left-aligned with the object tags on the left edge
per default, and offer centerization in center cases. This is also better
for Pholio, Phriction, which have a way lower min-x than Maniphest,
Differential.
I also disabled placing the hovercard below the tag in case there's not
enough space on the north side. The hovercard would not display 99.99% of
the times after being hidden (and it retains the flickering behaviour).
Another reason is also our current hide-behaviour, which only assumes
north-side alignment. Adding south-side didn't really work (I'm bad with
JS), so I didn't bother with it. Disabling this is //acceptable//, since
it only really affects Pholio, Phriction. And nobody places object tags
in the first line, anyway. Except for my test pages, of course :/
Btw, this also removes the weird jaggy horizontal shifts for object of
various lengths (e.g. `{D4356}`, `{rP32ofhw0842obw}` etc.). I think
that's only good.
Test Plan:
Hovered in Pholio, Phriction in Chrome, FF. Did not touch
left screen border.
Hovered in Maniphest, Differential. Tags farther to the left were
aligned left, tags more in the center were pretty centered.
Reviewers: epriestley, vrana
Reviewed By: epriestley
CC: aran, Korvin
Maniphest Tasks: T1048, T2902
Differential Revision: https://secure.phabricator.com/D5612
Summary:
Refs T1048, T2902 - This //should// fix flickering hovercards. Probably does not fix displaced hovercards, though could (because of the flickering doing something bad to rendering).
I'm bad with JS D:
Test Plan: Tried out plenty of reload & hover combos. No flickering anymore. Never again.
Reviewers: epriestley, vrana
Reviewed By: epriestley
CC: aran, Korvin
Maniphest Tasks: T1048
Differential Revision: https://secure.phabricator.com/D5607
Summary:
Refs T1048 - Take this setup:
D123D321
Hover with the mouse over one object, and move to another one before it got loaded. You'll have two hovercards. Of these you can't get rid of one anymore. Not through resizing, not through showing another hovercard. (You may have to try a few times). I think it only happens when #2 loads faster than #1. #2 is displayed, then #1 gets loaded and draws.
Or something like that. You could have a race condition where one draws after another and overwrites the current node/anchor, without hiding it. This renders the previous card unhideable, even on resize / mouse far away.
Test Plan:
Did a repro by hovering a lot.
Applied this change. Could not repro anymore.
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, Korvin
Maniphest Tasks: T1048
Differential Revision: https://secure.phabricator.com/D5594
Summary: See discussion in D5563. This loads hovercards on-demand, and shows them once they load. Ref T1048.
Test Plan: Made a preview comment like `T1`, waved my mouse over it, got a hovercard shortly thereafter.
Reviewers: AnhNhan
Reviewed By: AnhNhan
CC: aran
Maniphest Tasks: T1048
Differential Revision: https://secure.phabricator.com/D5588
Summary:
Refs T1048; Depends on D5557, D5558
They hover right above your eyes. Try it out at home :D (in that case, don't forget to checkout D5557 and D5558)
Worth a lot of improvement. But it's great for now after a little bit of styling.
Scrape load (search current document for all hovercards and pre-load them) and lazy load (e.g. previewing comments which is not covered by scrape load) implemented.
Added some seemingly useful graceful situations. Too much to the left, too much to the top.
Test Plan:
Tested on Ubuntu, Chrome and FF. No Windows yet, since I have it live at no place. Works pretty fine, at least it will appear.
Could test left graceful fallback. Don't know what happens if you place it too far to the top. I expect either good results or placement about the center of the screen in case of glitches.
For now, they'll disappear right away once the mouse leaves the object tag. Intended is when leaving both tag and hovercard.
Reviewers: epriestley, chad, btrahan
CC: aran, Korvin
Maniphest Tasks: T1048
Differential Revision: https://secure.phabricator.com/D5563
Conflicts:
src/__celerity_resource_map__.php
Test Plan: go to the new task page, go to the 'Assign To' field, assign the task to someone and try to do a Shift-Tab to return to the title field
Reviewers: igoleo, btrahan
Reviewed By: btrahan
CC: aran, Korvin
Maniphest Tasks: T2760
Differential Revision: https://secure.phabricator.com/D5547
Summary:
Apparently I am crazy and didn't test D5537 propertly at all. In particular:
- Currently, the update sends back new "people" and "files" widgets. The "people" widget has a tokenizer, which fatals when the behavior initializes without the widget in the DOM. For now, disable widget updates on replies. I'll fix this in a future diff.
- Currently, we don't update the "last_transaction_id" in the form itself, so the first reply sends back 1 message, the next 2 messages, etc. Update the input.
- The transaction paging doesn't and has never worked, I am crazy. Make it actually work.
Test Plan:
computers are too hard
(also, this is why I hate Javascript)
Reviewers: btrahan
Reviewed By: btrahan
CC: aran
Differential Revision: https://secure.phabricator.com/D5538
Summary: Fixes T2861. This used to work but I think I broke it when I made the notification popup thing work. Merge it into the drag-and-drop since 90% of the code is the same anyway.
Test Plan: Pasted files into Chrome and got uploads. This doesn't work in Safari and Firefox; as far as I know it isn't supported in those browers.
Reviewers: blc, btrahan, chad
Reviewed By: btrahan
CC: aran
Maniphest Tasks: T2861
Differential Revision: https://secure.phabricator.com/D5530
Summary:
adds ye olde people widget. Features include
- handle-based display, so we get status for free. (Note less pretty than M14 would have it!)
- can add a person
- can remove a person
- can see the people already in the conpherence
Test Plan: added and removed people and noted they joined / re-added as appropriate. Tried to add someone already in the conpherence and got a "transaction has no effect" message
Reviewers: epriestley, chad
Reviewed By: epriestley
CC: aran, Korvin
Maniphest Tasks: T2399
Differential Revision: https://secure.phabricator.com/D5466
Summary: Fixes T2811. Chrome requires this, but only for files dragged from the download shelf, I guess? Not entirely sure what's going on here. I couldn't find much documentation on this and figured it out mostly by process of elimination.
Test Plan: Drag and drop still works everywhere, plus from the Chrome download shelf.
Reviewers: blc, btrahan, skrul
Reviewed By: skrul
CC: aran
Maniphest Tasks: T2811
Differential Revision: https://secure.phabricator.com/D5528
Summary: We may execute the Conpherence behavior before the initial device change, in which case we'll get a desktop -> desktop device event. This currently causes us to double-load the thread list. Instead, don't do anything if the device is the same as our current understanding of device state.
Test Plan: No double thread list in profiler.
Reviewers: btrahan
Reviewed By: btrahan
CC: aran
Differential Revision: https://secure.phabricator.com/D5525
Summary:
- Move `/N/` to `/thread/N/`.
- Move `/view/N/` to `/N`/.
- This makes the mobile "Conpherence -> click thread -> see thread" workflow work correctly. This also makes permalinks work correctly on mobile.
Ref T2421.
Test Plan: Clicked flows on desktop, mobile.
Reviewers: btrahan
Reviewed By: btrahan
CC: aran
Maniphest Tasks: T2421
Differential Revision: https://secure.phabricator.com/D5508
Summary:
When we access `/conpherence/view/2/` on the desktop, load the thread on the initial response and ajax in the thread list. Basically, the idea is:
- When we load the thread list, an individual thread, or the widget panel, we send back the piece the user asked for in the response.
- On mobile, we stop there: we don't ajax in anything else, and just hide the other parts of the layout.
- On Desktop, we fill out the other layout components via Ajax.
Ref T2421.
Test Plan: Hit `/conpherence/view/2/`, got a full page.
Reviewers: btrahan
Reviewed By: btrahan
CC: aran
Maniphest Tasks: T2421
Differential Revision: https://secure.phabricator.com/D5507
Summary: Currently, `/conpherence/` shows the widget panel on devices. Make it show the thread list instead.
Test Plan: {F38099}
Reviewers: btrahan
Reviewed By: btrahan
CC: aran, chad
Maniphest Tasks: T2421
Differential Revision: https://secure.phabricator.com/D5505
Summary:
Currently, `selected_conpherence_id` is actually a PHID. Instead, use ids everywhere.
Also, use replaceState instead of pushState. This means "back" takes you out of conpherence (not back to the last thread you looked at) but I think that's more consistent with user expectation.
Test Plan: Loaded thread list, loaded specific thread.
Reviewers: btrahan
Reviewed By: btrahan
CC: aran
Differential Revision: https://secure.phabricator.com/D5503
Summary: Ref T2421. For mobile, we don't want to select the first thread, since we'll just show a list of threads. Move this behavior to the client and do it when the page is shown on a desktop (or the view is changed to a desktop).
Test Plan: Viewed `/conpherence/`, saw first thread selected. Switched threads.
Reviewers: btrahan
Reviewed By: btrahan
CC: aran
Maniphest Tasks: T2421
Differential Revision: https://secure.phabricator.com/D5502
Summary: Simplifies the Pontificate button and makes Control + Enter work again.
Test Plan: Submitted the form by clicking, hitting return, hitting control+enter.
Reviewers: btrahan
Reviewed By: btrahan
CC: aran
Differential Revision: https://secure.phabricator.com/D5472
Summary:
Currently, this behavior binds a ton of IDs. This makes the behavior fragile: if it is invoked on a page without all the right elements, some `JX.$()` tends to explode.
Instead, don't assume anything is present on the page. This allows the behavior to be invoked on other pages (like the "New Conpherence" page) or pages without some elements (like some future thread-only mobile view) without creating JS errors.
Test Plan:
Added comments, added comments with files. Viewed "New conphenrece" page.
NOTE: Control+Enter is currently broken, but this diff didn't change that behavior.
Reviewers: btrahan
Reviewed By: btrahan
CC: aran
Differential Revision: https://secure.phabricator.com/D5468
Summary: Fixes T2833. Lock forms and prevent double submit on control + enter.
Test Plan: Mashed control + enter a whole bunch in Maniphest, got one comment instead of several.
Reviewers: btrahan, edward
Reviewed By: btrahan
CC: aran
Maniphest Tasks: T2833
Differential Revision: https://secure.phabricator.com/D5473
Summary: Pagination occurs once
Test Plan: Tested on conpherences I made with myself. Lot of bugs still remain, but shows older messages in gaps of 2.
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, Korvin, chad
Maniphest Tasks: T2428
Differential Revision: https://secure.phabricator.com/D5183
Conflicts:
src/applications/conpherence/controller/ConpherenceViewController.php
webroot/rsrc/js/application/conpherence/behavior-menu.js
Summary:
Hook @btrahan's Stripe form to the rest of Phortune.
- Users can add payment methods.
- They are saved to Stripe and associated with PhortunePaymentMethods on our side.
- Payment methods appear on account overview.
Test Plan:
{F37548}
{F37549}
{F37550}
Reviewers: chad, btrahan
Reviewed By: btrahan
CC: aran
Maniphest Tasks: T2787
Differential Revision: https://secure.phabricator.com/D5438
Summary:
This is a major pain on Windows and the main reason why Phabricator doesn't work there and is hard to fix.
The sad part is that Windows support symlinks (via `MKLINK`) but Git on Windows doesn't use them.
Test Plan: Loaded Phabricator on Windows without JS errors.
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, Korvin
Differential Revision: https://secure.phabricator.com/D5458
Summary: D5426 removed mobile menu for messages but missed a few spots
Test Plan: successfully submitted pontifications without JS errors and the form freezing
Reviewers: chad, epriestley
Reviewed By: chad
CC: aran, Korvin
Differential Revision: https://secure.phabricator.com/D5453
Summary: Introduces a new settings panel for Conpherence specific settings.
Test Plan:
started a thread with a test user, thus two participants total. Replied to conpherence, toggling notification settings in between. Verified 1 or 2 emails were sent as appropos to the current toggle.
Toggled global setting and verified setting was updated in conpherences where nothing was specified. Verified setting conpherence setting overrides global setting.
Reviewers: epriestley, chad
Reviewed By: epriestley
CC: aran, Korvin
Maniphest Tasks: T2521
Differential Revision: https://secure.phabricator.com/D5391
Summary:
This isn't quite complete, but everything else is technical cleanup. Broadly:
- Removed checkboxes. Selected state is now indicated with CSS, and toggled with shift-click. When nothing is selected, the text reads "Shift-Click Tasks to Select" to let users discover this feature.
- Updated drag-to-reorder code to work with ObjectItemListView.
- Closed/resolved is now shown with a grey footer icon.
- Assigned is now shown with a user profile image handle icon, with a hover state.
This could probably use some more tweaks, but overall I think it looks pretty reasonable?
Test Plan: {F35897}
Reviewers: chad
Reviewed By: chad
CC: aran
Differential Revision: https://secure.phabricator.com/D5340
Summary:
does the title and also a few other small tweaks
- kills the init js behavior; now its part of menu where it belongs.
- adds the underline to the icon when its toggled in the widget menu
- fixed JS initialization errors on the "create conpherence" page. Note I still like keeping all that init stuff in one function because its typing the same data a bunch to be passed over to the JS layer. Other ways to accomplish this obvi...
Only fun wrinkle here is I think Chad intended me to display "when the file was attached". Instead, I display when the file was *uploaded*. I think this jives better with our version where you can't delete and all that. Files are pretty powerful, long-living objects in Phabricator land.
Test Plan: added files to a conpherence and noted widget loaded updated okay. added a file with no author (generated by the system) and verified it still rendered okay. switched between conpherences and verified proper data in each files widget. uploaded image and text files to check the icons were correct.
Reviewers: epriestley, chad
Reviewed By: epriestley
CC: aran, Korvin
Maniphest Tasks: T2530
Differential Revision: https://secure.phabricator.com/D5337
Summary: One can point to spesific image in a Mock set using {Mxx|yy} syntax
Test Plan: {F35373}
Reviewers: chad, epriestley
Reviewed By: epriestley
CC: aran, Korvin, AnhNhan
Maniphest Tasks: T2710
Differential Revision: https://secure.phabricator.com/D5322
Summary:
this diff does a few, not so exciting things
- changes "conpher" to "conpherence" where it snuck into CSS, spritemap, etc -- I believe we now consistently call it conpherence. Feel free to change it, just change it everywhere. :D
- puts the widget icons in the right order per M14
- makes the "mobile-only" widgets show the toggles only in the mobile view
- also made it so clicking them does nothing for now
- removes the tasks widget since we don't want it
...my time is getting chopped up funny (yay puppy) so this is just an attempt at something that can go into the codebase and not make it worse. Next up is making the widgets that show on desktop look right / not say "TODO"
Test Plan: played around in Conpherence
Reviewers: epriestley, chad
Reviewed By: epriestley
CC: aran, Korvin
Maniphest Tasks: T2530
Differential Revision: https://secure.phabricator.com/D5334
Summary:
Preload all the images when viewing a pholio. Fixes T2692.
We should probably use lower-resolution previews on mobile and only show full resolution when the user taps. We should also suppress the thumbnails from loading on mobile since they aren't visible. But neither is critical for the moment.
Test Plan: Added logging, verified everything loaded.
Reviewers: chad
Reviewed By: chad
CC: aran
Maniphest Tasks: T2692
Differential Revision: https://secure.phabricator.com/D5316
Summary: Pholio inline comments now have minimum size of 16x16 px
Test Plan: Made some inlines
Reviewers: epriestley
CC: aran, Korvin, AnhNhan
Maniphest Tasks: T2647
Differential Revision: https://secure.phabricator.com/D5250
Summary:
Ref 2700. No sexy animation yet, but allows you to swipe left and right to switch photos. Generally feels pretty good to me.
Also fixes a left/right but and a bug where taps could be interpreted as gestures and simplifies some touch code.
derpdog
Test Plan:
Swiped left and right in Pholio.
{F35239}
Reviewers: chad
Reviewed By: chad
CC: aran
Differential Revision: https://secure.phabricator.com/D5309
Summary:
Ref T2700. Allow JS to listen for swipes on devices.
There are a bunch of tricky cases here and I probably didn't get them all totally right, but this interaction broadly looks like this:
- We implement gesture recognition for the mouse in device modes (narrow browser), and for touch events from an actual device.
- The sigil `touchable` indicates that a node wants to react to touch events.
- When the user touches a `touchable` node, we start listening for moves. They might be tapping/clicking (in which case we don't care), but they might also be gesturing.
- Once the user moves their finger/pointer far enough away from the tap origin, we recognize it as a gesture. I hardcoded this at 20px; I wasn't able to find any "official" Apple value, but 20px seems like a common default.
- At this point, we look at where their finger has moved.
- If they moved it mostly up/down, we interpret the gesture as "scroll" and just stop listening. The device does its own thing.
- However, if they moved it mostly left/right, we interpret it as a "swipe". We start killing the moves so the device doesn't scroll.
- Once we've recognized that a gesture is underway, we send a "gesture.swipe.start" event and then "gesture.swipe.move" events for every move.
- When the user ends the gesture, we send "gesture.swipe.end".
- If the user cancels the gesture (currently, only by tapping with a second finger), we send "gesture.swipe.cancel".
- Gesture events have raw position data and some convenience fields.
Test Plan:
Wrote UI example and used it from the Desktop, iPhone simulator, and a real iphone.
- The code always seems to get "scroll" vs "swipe" correct (i.e., consistent with my intentions).
- The threshold feels pretty good to me.
- Tapping with a second finger cancels the action.
Reviewers: chad, btrahan
Reviewed By: chad
CC: aran
Maniphest Tasks: T2700
Differential Revision: https://secure.phabricator.com/D5308
Summary: Fixes T2711. Allows special keys (arrows, tab, return) to be bound as shortcuts. Binds left and right as shortcuts in Pholio.
Test Plan: Pressed left. Pressed right.
Reviewers: chad
Reviewed By: chad
CC: aran
Maniphest Tasks: T2711
Differential Revision: https://secure.phabricator.com/D5303
Summary: files widget updates as new files are added. made basically all edits "ajax" except for when you change the conpherence image which just does a reload. I will fix this in a future diff but it was starting to spiral out of control a bit.
Test Plan: commented on a task with files and noted the updated file widget. updated header image via drag and drop and dialogue -- noted a full reload. cropped image and re-titled conpherence - ajax updated as expected.
Reviewers: epriestley, chad
Reviewed By: epriestley
CC: aran, Korvin
Maniphest Tasks: T2530
Differential Revision: https://secure.phabricator.com/D5288
Summary: okay title. other apps can get this by implementing shouldAllowPublic and set(ting)RequestURI on TransactionsCommentView. note i put some css inline -- let me know if that belongs someplace else or needs better design.
Test Plan: viewed a mock logged out and saw new button. used new button and ended up on the mock logged in with a clean URI.
Reviewers: epriestley, chad
Reviewed By: chad
CC: aran, Korvin
Maniphest Tasks: T2653
Differential Revision: https://secure.phabricator.com/D5266
Summary:
I have blame enabled by default and displaying files with long history takes easily over 10 seconds.
Load the blame data by AJAX instead.
This is actually doing more work and the total response time is longer but it's worth it for me as I am interested just in the file contents quite often.
I know you were talking about building blame cache but until we have it...
I'm not sure if the AJAX loading indicator in bottom right corner is enough to inform the user that we are loading it on background.
Test Plan:
?view=highlighted
?view=plainblame
?view=blame
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, Korvin
Differential Revision: https://secure.phabricator.com/D5244
Summary: serving up for some feedback -- does anything fancy need to happen when new messages load (say highlight em and fade the highlight out) or just scrolling down okay? in JS I have a TODO about how come it doesn't work; that code works okay in my console if I print out various debugging values and enter them in manually.
Test Plan: pontificate with myself; note new message and message entry updates properly. pontificate as different user then as myself; note two messages load and message entry updates properly. pontificate as a user who isn't the last to reply; note new message and message entry updates properly
Reviewers: epriestley, chad
Reviewed By: epriestley
CC: aran, Korvin
Maniphest Tasks: T2522
Differential Revision: https://secure.phabricator.com/D5214
Summary:
Ref T2641. Currently, we scale images to fit horizontally, but let them have arbitrary vertical size. This is nice in theory but kind of sucks in practice because it makes everything below the stage jump around when you switch images. It would also make swiping through images on mobile super weird.
Instead, scale to fit in both dimensions. This feels a lot better and more application-like to me. (I also think most mocks are not especially tall?)
Test Plan:
{F34648}
(Note that the image is enormous.)
Reviewers: chad
Reviewed By: chad
CC: aran
Maniphest Tasks: T2641
Differential Revision: https://secure.phabricator.com/D5233
Summary:
Ref T2644. This isn't perfect, but makes Pholio more-or-less usable on mobile. In particular:
- Inline comments drop below the image.
- Clicking the image lightboxes a full size version so you can scroll around it and inspect details.
- Clicking it again dismisses it.
Things that aren't good:
- Can't add inline comments. This seems complicated and not critical.
- Can't easily figure out which inline comments go where since there's no hover. Maybe let you tap them? Not sure if we really need this.
- Thumbs are between image and image data. I'm planning to get rid of the thumbs and do swipe left/right instead.
- It would be nice to scroll the lightbox to center on the part of the image you tapped. I just thought of this, though.
Test Plan:
{F34640}
{F34641}
{F34642}
Reviewers: chad
Reviewed By: chad
CC: aran
Maniphest Tasks: T2644
Differential Revision: https://secure.phabricator.com/D5232
Summary: I broke the code which shows which thumb is currently selected when I made the URLs fancy, by changing the container from `div` to `a`.
Test Plan: Verified current thumb is now highlighted.
Reviewers: chad
Reviewed By: chad
CC: aran
Differential Revision: https://secure.phabricator.com/D5231
Summary:
Fixes T2669.
- Currently, making an inline comment does not focus the textarea. Instead, focus the textarea.
- Currently, the positioning is kind of buggy. Make it viewport-relative and put the dialog slightly below the inline reticle.
- Use `JX.Workflow` more and simplify some of the ajax stuff.
Test Plan: Created inline comments.
Reviewers: chad
Reviewed By: chad
CC: aran
Maniphest Tasks: T2669
Differential Revision: https://secure.phabricator.com/D5229
Summary:
Some recent CSS change tweaked these out a bit. They were also sort of weird looking after fixing the double-draw thing:
- Make them work properly;
- make the implementation a bit simpler;
- make them clearer visually.
Also fix a bug where "border" and "fill" were accidentally reversed.
Test Plan:
{F34625}
(The highlight on the reticle is a little hard to pick out in the screenshot, but very obvious in practice.)
Reviewers: chad
Reviewed By: chad
CC: aran
Differential Revision: https://secure.phabricator.com/D5228
Summary: Currently, if you click and drag in the checkered background, we'll create a 1px-wide or 1px-tall inline comment. Don't start inlines unless the user clicks on the image itself.
Test Plan: Clicked on and off the image.
Reviewers: chad
Reviewed By: chad
CC: aran
Differential Revision: https://secure.phabricator.com/D5227
Summary: Fixes T2646. Title and description are placeholders. Styles on this are a bit rough.
Test Plan: {F34623}
Reviewers: chad
Reviewed By: chad
CC: aran
Maniphest Tasks: T2646
Differential Revision: https://secure.phabricator.com/D5226
