Summary: This adds some basic per user / per room theming for Conpherence, which should hopefully let users identify rooms from just the sidebar color.
Test Plan: Lots of threads with different colors.
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin
Differential Revision: https://secure.phabricator.com/D17747
Summary:
Ref T12562. I think the pre-send-on-enter behavior was: disable textarea, send message, clear area on response?
That got changed but not completely, maybe. There's currently an issue here:
- Add a `sleep(3)` to `UpdateController`.
- Type "AAA".
- Press enter.
- Real fast, type "BBB".
- When the "AAA" arrives, your "BBB" is lost. Sad!
Test Plan:
- Did the thing described above; no longer lost "BBB".
- Switched threads, sent messages, couldn't find anything else this breaks. It dates from a long time ago so I think it's just pre-SOE stuff.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12562
Differential Revision: https://secure.phabricator.com/D17742
Summary:
Fixes T12599. If you're faster than the network request, we don't actually resolve your query when the data arrives.
Instead, when the data shows up, run the query if the user has typed something.
Test Plan: Pasted a filename in real fast, got results. (Previously: no results until you press something else.)
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T12599
Differential Revision: https://secure.phabricator.com/D17739
Summary:
Ref T11476. This is a bit hacky, but makes `Application` extend `LiskDAO` so we can apply transactions to it with an `Editor` class.
Also fixes schema stuff so builds should produce a clean bill of health again.
This might only get you slightly further, yell if you run into more trouble.
Test Plan:
- Ran `bin/storage upgrade -f` and got no warnings.
- Browsed around, nothing exploded?
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T11476
Differential Revision: https://secure.phabricator.com/D17738
Summary: Part of the groundwork for T11476.
Test Plan: ran `./bin/storage upgrade` and observed expected DB tables
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin
Maniphest Tasks: T11476
Differential Revision: https://secure.phabricator.com/D17736
Summary:
Pathway to D17685. This column is (mostly) a denormalization of `dateModified` on the thread.
Just use a JOIN instead.
This isn't //exactly// the same: we'll bump threads to the top now for non-message changes (e.g., a topic or title change). That seems fine, but we could put a `lastMessageDate` on Thread later if we want to refine it.
Also got rid of a lot of other unused stuff. There's a big garbage TODO here, I'll fix that in the next change.
Test Plan:
- Grepped for `dateTouched`.
- Grepped for `participantCursor`.
- Grepped for `ConpherenceParticipantQuery::LIMIT`.
- Looked for callsites to `setOrder()`, found none.
- Added a message to an older thread, saw it bump up to the top.
Reviewers: chad
Reviewed By: chad
Differential Revision: https://secure.phabricator.com/D17731
Summary:
Pathway to D17685. This column is a very complicated cache of: is participant.messageCount equal to thread.messageCount?
We can just ask this question with a JOIN instead and simplify things dramatically.
Test Plan:
- Ran migration.
- Browsed around.
- Sent a message, saw unread count go up.
- Read the message, saw unread count go down.
Reviewers: chad
Reviewed By: chad
Differential Revision: https://secure.phabricator.com/D17730
Summary: Pathway to D17685. Nothing reads this field and it has no use or value.
Test Plan:
- Ran migration.
- Grepped for `behindTransactionPHID`.
Reviewers: chad
Reviewed By: chad
Differential Revision: https://secure.phabricator.com/D17729
Summary: A few more mp3s to choose from for Conpherence.
Test Plan: Test each sound in a new room.
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin
Differential Revision: https://secure.phabricator.com/D17734
Summary: Minor pixel shifts with new header ui in place.
Test Plan: Desktop, Mobile, Tablet, with and without search and participants open
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin
Differential Revision: https://secure.phabricator.com/D17725
Summary:
Ref T12566. Ref T12563. This fixes three bugs with Aphlict replay stuff:
First, Conphernece would try to repaint the UI even if no thread was open. Only repaint when a thread is open.
Second, although we deduplicate JX.Leader messages, we didn't deduplicate actual notification messages. If you browsed the leader window, then it re-elected itelf as a leader and replayed history, it could rebroadcast notifications and other windows could show doubles. Deduplicate notifications to prevent this.
Third, we always replayed the last 60 seconds of history. When you browsed the leader window, whichever window became the new leader (possibly the one you just browsed) could replay messages from before it had opened, leading to duplicate messages. Particularly, after receiving a message and then browsing you could see that message again. Instead, only replay history as far back as when the window first opened.
Test Plan:
- Clicked "Repaint" with a thread open, saw a repaint. Clicked "Repaint" with Conpherence open but no thread, no repaint and no 404 request to `/update/null/`.
- In browser A, opened three windows. In browser B, sent a notification. In browser A, browsed the leader window away twice in a row. Observed that the window which never became a leader doesn't duplicate notifications.
- In browser A, opened three windows. In browser B, sent a notification. In browser A, browsed the leader window away over and over again. Observed that replay requests issued with appropriate history windows.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12566, T12563
Differential Revision: https://secure.phabricator.com/D17722
Summary:
Deletion is a possibly time-intensive process, especially with large
files that are backed by high-latency, chunked storage (such as
S3). Even ~200mb objects take minutes to delete, which makes for an
unhappy experience. Fixes T10828.
Test Plan:
Delete a large file, and stare in awe of the swiftness with
which I am redirected to the main file application.
Reviewers: #blessed_reviewers, epriestley
Reviewed By: #blessed_reviewers, epriestley
Subscribers: thoughtpolice, Korvin
Maniphest Tasks: T10828
Differential Revision: https://secure.phabricator.com/D15743
Summary: Ref T12573. This sends a "ping" to the server, and a "pong" back to the client, every 15 seconds. This tricks ELBs into thinking we're doing something useful and productive.
Test Plan: Ran `bin/aphlict debug`, loaded Phabricator, saw ping/pong in logs.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12573
Differential Revision: https://secure.phabricator.com/D17717
Summary: Follows the outline in D15656 for implementing ngram search for names of File objects. Also created FileFullTextEngine, because without implementing `PhabricatorFulltextInterface`, `./bin/search` complains that `File` is not an indexable type.
Test Plan:
- ran `./bin/storage upgrade` to apply the schema change
- confirmed the presence of a new `file_filename_ngrams` table
- added a couple file objects
- ran `bin/search index --type file --force`
- confirmed the presence of rows in `file_filename_ngrams`
- did a few keyword searches and saw expected results
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin
Maniphest Tasks: T8788
Differential Revision: https://secure.phabricator.com/D17702
Summary: Ref T12566. When we reconnect, refresh the current thread even if we replayed notifications.
Test Plan:
- Clicked the "Repaint" button, saw the thread refresh.
- Clicked the "Reconnect" button, saw the thread reresh.
- Launched `aphlict debug`, killed it, restarted it, saw the thread refresh after reconnect.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12566
Differential Revision: https://secure.phabricator.com/D17713
Summary:
Fixes T12564. We already had some code which seems to deal with this properly, it just wasn't getting used.
Assign each application-level notification a unique ID, then ignore messages with duplicate IDs.
Test Plan:
- In browser A, loaded `/T123`.
- In browser B, loaded `/T123`.
- Made a comment as B.
- Saw notification as A.
- Mashed "Replay" a bunch.
- Before patch: piles of duplicate notifications.
- After patch: no duplicates.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12564
Differential Revision: https://secure.phabricator.com/D17710
Summary:
Fixes T12563. If we've ever seen an "open", mark all future connections as reconnects. When we reconnect, replay recent history.
(Until duplicate messages (T12564) are handled better this may cause some notification duplication.)
Also emit a reconnect event (for T12566) but don't use it yet.
Test Plan: {F4912044}
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12563
Differential Revision: https://secure.phabricator.com/D17708
Summary:
Fixes T12567. We currently retry after 2s, 4s, 8s, 16s, ...
If we connected cleanly once, retry the first time right away. There are a bunch of reasonable cases where this will work fine and we don't need to wait. Then we fall back: 0s, 2s, 4s, 8s, ...
Test Plan: {F4911905}
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12567
Differential Revision: https://secure.phabricator.com/D17706
Summary: Ref T12568. Ref T12567. Allows you to force a reconnect, and shows the reconnect delay on connection close/failure.
Test Plan: {F4911879}
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12568, T12567
Differential Revision: https://secure.phabricator.com/D17705
Summary:
Ref T12573. `JX.Leader` synchronizes the Aphlict connection across multiple windows.
Currently, we only test to see if the leader window has been closed every 16 seconds. Instead, test every 1.5 seconds.
Also, make windows keep trying to become the leader forever. This was removed previously (in D15806) but I think that change decreased robustness here.
Test Plan:
- Opened two windows to the "Realtime" tab in DarkConsole.
- Saw one become the leader and one become a follower.
- (Optionally, wait for 10 seconds here to test the "keep trying to become the leader" behavior.)
- Closed the leader.
- Saw the follower become the leader after ~1.5 seconds.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12573
Differential Revision: https://secure.phabricator.com/D17703
Summary: Some space is bleeding in here from two-column-css. Re-scope CSS.
Test Plan: Review creating a task on mobile with document preview present.
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin
Differential Revision: https://secure.phabricator.com/D17712
Summary: Cleans up the topic UI a little more, I think this feels nice for some reason.
Test Plan: visit a room with and without a topic, desktop, tablet, mobile
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin
Differential Revision: https://secure.phabricator.com/D17711
Summary: Fixes T12577. Some tweaks last week widened the default buttons, but these didn't get retouched.
Test Plan: Review calendar on desktop and mobile.
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin
Maniphest Tasks: T12577
Differential Revision: https://secure.phabricator.com/D17709
Summary: Fixes T12556 Uses more common components in ConpherenceThreadList by moving to PHUIListItemView. Reduces clutter by moving privacy into the header. Gets ride of "See More" double interchanges.
Test Plan:
I need to test this more, doesn't seem to auto-select top room any more, also might build a lipsum generator.
- Create lots of rooms with various policies
- Test clicking on policy object
- Click on different rooms
- Post in rooms
- Load up second account, see room numbers
- Clear room message count by clicking on room
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin
Maniphest Tasks: T12556
Differential Revision: https://secure.phabricator.com/D17698
Summary: Ref T12538. I missed this in D17695, which renamed the variable. The logic was also a little off since `jj` is an index, not a count.
Test Plan: Typed `con` in global search, which hits "Con-pherence", "Con-duit" and "Con-fig", plus a bunch of other stuff. Got results after patch.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12538
Differential Revision: https://secure.phabricator.com/D17700
Summary:
Fixes T12538. Instead of hiding "closed" results unless only closed results match, show closed results but sort them to the bottom.
This fixes the actual issue in T12538, and I think this is probably the correct/best behavior for the global search.
It also makes all other typeaheads use this behavior. They currently have a "bug" where enabled user `abcd` makes it impossible to select disabled user `abc`. This manifests in some real cases, where enabled function `members(abc)` makes it impossible to disabled user `abc` in some function tokenizers.
If ths feels worse, we could go back to filtering in the simpler cases and introduce a rule like "show closed results if only closed results would be shown OR if query is an exact match for the disabled result", but that gets dicier because "exact match" is a fuzzy concept.
(There are a lot of other minor bad behaviors that this doesn't try to fix.)
Test Plan:
Enabled project "instabug" no longer prevents bot user "instabug" from being shown:
{F4903843}
Disabled user "mmaven" is sorted below enabled user "mmclewis", in defiance of the otherwise alphabetical order. There's no visual cue that this user is disabled because of T6906.
{F4903845}
Same as above, but this source renders "disabled" in a more obvious way:
{F4903848}
Function selecting members of active project `members(instabug)` no longer prevents selection of bot user `instabug`:
{F4903849}
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12538
Differential Revision: https://secure.phabricator.com/D17695
Summary: We no longer display this any more in the UI, so go ahead and remove the callsites and db column.
Test Plan: New Room, with and without participants.
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin
Differential Revision: https://secure.phabricator.com/D17683
Summary: Will see how this goes in practice. Uses violet where color is used for non responsive peeps.
Test Plan: Create a user without email verification, test hover card, profile, mentions and lists.
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin
Differential Revision: https://secure.phabricator.com/D17678
Summary: This moves the count on the Conpherence Menu Item into a phui-list-item-count, and removes the CSS call to the entire Conphrence stack when durable column is open.
Test Plan: Test with and without the chat column, and a menu with a count
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin
Differential Revision: https://secure.phabricator.com/D17677
Summary: Does a few things. Turns off feed stories (again), removes "action" transactions from notificiations, and only updates message count on actual messages. This feels a bit cleaner and less spammy... I guess... I think @epriestley will really like it and do me a favor or something.
Test Plan: Pull up two windows. test a message, see message count on second screen. Edit a topic or title, get no notification. At all. Ever.
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin, epriestley
Differential Revision: https://secure.phabricator.com/D17674
Summary:
Depends on D17670. Fixes T12137. Fixes T12003. Ref T2632.
This shows users a readout of which terms were actually searched for.
This also drops those terms from the query we submit to the backend, dodging the weird behaviors / search engine bugs in T12137.
This might need some design tweaking.
Test Plan: {F4899825}
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12137, T12003, T2632
Differential Revision: https://secure.phabricator.com/D17672
Summary: Begin converting Conpherence to ModularTransactions, this converts title, topic, and picture to use modular transactions. Participants seems hairy so I'll do that in another diff
Test Plan: Create a room with a topic, change room name, topic. Add people, remove people. Set a room image. Unset topic.
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin
Differential Revision: https://secure.phabricator.com/D17668
Summary: Fixes T11730. Removes an old transaction that hasn't been used in a year.
Test Plan: Run sql, check various rooms.
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin
Maniphest Tasks: T11730
Differential Revision: https://secure.phabricator.com/D17666
Summary: Builds a Conpherence Profile Menu Item, complete with counts for the unreads. This allows pinning to home as well as swapping out thread list in Conpherence for pinning eventually.
Test Plan: Add a menu item, chat in room, log into other account, see room count. Room count disappears after viewing.
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin
Differential Revision: https://secure.phabricator.com/D17662
Summary: Drops the inset box shadow and bumps standard UI elements from 28px to 30px. more room for activities.
Test Plan:
Spaces, Editing tasks, typeaheads, mobile, desktop.
{F4897792}
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin
Differential Revision: https://secure.phabricator.com/D17660
Summary:
Fixes T12503.
- Users with creative usernames like `MMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMM` could overflow "Subscribers" in the curtain UI.
- Other content like packages could also overflow.
- Users with interesting and unique names like `WWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWW` who were also away or disabled could get a linebreak between their availability dot and their username.
Test Plan:
See T12503 for "before" screenshots. Also tested mobile, which looked fine, but didn't screenshot it.
{F4849900}
{F4849912}
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12503
Differential Revision: https://secure.phabricator.com/D17650
Summary: Fixes T12138. Test for the presence of being in fullscreen mode, and disable send on enter if present. Side note, I'd love a first class "hasClass" type Javelin function.
Test Plan:
- Go to Conpherence
- Type some smack, see it send on enter
- Go fullscreen like a boss
- Let the words flow
- Close fullscreen, then send on enter.
- (might be nice someday to add a "submit" button to fullscreen editor)
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin
Maniphest Tasks: T12138
Differential Revision: https://secure.phabricator.com/D17590
Summary: Fixes T12488. Some events appear to have survived earlier migrations without getting completely fixed. Fix them.
Test Plan:
- Ran migration locally with `bin/storage upgrade` (but: I could not reproduce this problem locally).
- Ran migration in production and saw ICS import stop fataling.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12488
Differential Revision: https://secure.phabricator.com/D17642
Summary: Ref T12509. This encourages code to move away from HMAC+SHA1 by making the method name more obviously undesirable.
Test Plan: `grep`, browsed around.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12509
Differential Revision: https://secure.phabricator.com/D17632
Summary:
Ref T12509. This adds support for HMAC+SHA256 (instead of HMAC+SHA1). Although HMAC+SHA1 is not currently broken in any sense, SHA1 has a well-known collision and it's good to look at moving away from HMAC+SHA1.
The new mechanism also automatically generates and stores HMAC keys.
Currently, HMAC keys largely use a per-install constant defined in `security.hmac-key`. In theory this can be changed, but in practice essentially no install changes it.
We generally (in fact, always, I think?) don't use HMAC digests in a way where it matters that this key is well-known, but it's slightly better if this key is unique per class of use cases. Principally, if use cases have unique HMAC keys they are generally less vulnerable to precomputation attacks where an attacker might generate a large number of HMAC hashes of well-known values and use them in a nefarious way. The actual threat here is probably close to nonexistent, but we can harden against it without much extra effort.
Beyond that, this isn't something users should really have to think about or bother configuring.
Test Plan:
- Added unit tests.
- Used `bin/files integrity` to verify, strip, and recompute hashes.
- Tampered with a generated HMAC key, verified it invalidated hashes.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12509
Differential Revision: https://secure.phabricator.com/D17630
Summary: Fixes T11630. Not sure what the max-width fixes, but I don't see anything off on various mobile, desktop.
Test Plan: Enable filetree in differential, drag navigation all over, see normal width calculations.
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin
Maniphest Tasks: T11630
Differential Revision: https://secure.phabricator.com/D17591
Summary:
Fixes T12479. If you end a line with a character like ":" in a context which can trigger autocomplete (e.g., `.:`), then try to make a newline, we swallow the keystroke.
Instead, allow the keystroke through if the user hasn't typed anything else yet.
Test Plan:
- Autocompleted emoji and users normally.
- In an empty textarea, typed `.:<return>`, got a newline instead of a swallowed keystroke.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12479
Differential Revision: https://secure.phabricator.com/D17583
Summary:
Fixes T12460. Also ":)", ":(", ":/", and oldschool ":-)" variants.
Not included are variants with actual letters (`:D`, `:O`, `:P`) and obscure variants (`:^)`, `:*)`).
Test Plan: Typed `:3` (no emoji summoned). Typed `:dog3` (emoji summoned). Typed `@3` (user autocomplete summoned).
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12460
Differential Revision: https://secure.phabricator.com/D17577
Summary:
Ref T10967. This is explained in more detail in T10967#217125
When an author does "Request Review" on an accepted revision, void (in the sense of "cancel out", like a bank check) any "accepted" reviewers on the current diff.
Test Plan:
- Create a revision with author A and reviewer B.
- Accept as B.
- "Request Review" as A.
- (With sticky accepts enabled.)
- Before patch: revision swithced back to "accepted".
- After patch: the earlier review is "voided" by te "Request Review", and the revision switches to "Review Requested".
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T10967
Differential Revision: https://secure.phabricator.com/D17566
Summary: Fixes T11641. We're overbroad here (and this may need more scoping?) but this seems to resolve the immediate issue.
Test Plan: Upload a few diffs and ask disabled accounts to comment on them inline.
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin
Maniphest Tasks: T11641
Differential Revision: https://secure.phabricator.com/D17565
Summary: General CSS and usability touchup of the Remarkup bar states for fullscreen and preview. Larger fonts, more spacing, some hint of the underlying page. Disable buttons that can't be used in preview mode.
Test Plan:
Formal test coming with mobile, browsers. This is a kick the tires upload.
{F4283448}
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin
Differential Revision: https://secure.phabricator.com/D17563
Summary:
The goal is to make fulltext search back-ends more extensible, configurable and robust.
When this is finished it will be possible to have multiple search storage back-ends and
potentially multiple instances of each.
Individual instances can be configured with roles such as 'read', 'write' which control
which hosts will receive writes to the index and which hosts will respond to queries.
These two roles make it possible to have any combination of:
* read-only
* write-only
* read-write
* disabled
This 'roles' mechanism is extensible to add new roles should that be needed in the future.
In addition to supporting multiple elasticsearch and mysql search instances, this refactors
the connection health monitoring infrastructure from PhabricatorDatabaseHealthRecord and
utilizes the same system for monitoring the health of elasticsearch nodes. This will
allow Wikimedia's phabricator to be redundant across data centers (mysql already is,
elasticsearch should be as well).
The real-world use-case I have in mind here is writing to two indexes (two elasticsearch clusters
in different data centers) but reading from only one. Then toggling the 'read' property when
we want to migrate to the other data center (and when we migrate from elasticsearch 2.x to 5.x)
Hopefully this is useful in the upstream as well.
Remaining TODO:
* test cases
* documentation
Test Plan:
(WARNING) This will most likely require the elasticsearch index to be deleted and re-created due to schema changes.
Tested with elasticsearch versions 2.4 and 5.2 using the following config:
```lang=json
"cluster.search": [
{
"type": "elasticsearch",
"hosts": [
{
"host": "localhost",
"roles": { "read": true, "write": true }
}
],
"port": 9200,
"protocol": "http",
"path": "/phabricator",
"version": 5
},
{
"type": "mysql",
"roles": { "write": true }
}
]
Also deployed the same changes to Wikimedia's production Phabricator instance without any issues whatsoever.
```
Reviewers: epriestley, #blessed_reviewers
Reviewed By: epriestley, #blessed_reviewers
Subscribers: Korvin, epriestley
Tags: #elasticsearch, #clusters, #wikimedia
Differential Revision: https://secure.phabricator.com/D17384
Summary:
Ref T12271. Don't do anything with this yet, but store who accepted/rejected/whatever on behalf of reviewers.
In the future, we could use this to render stuff like "Blessed Committers (accepted by epriestley)" or whatever. I don't know that this is necessarily super useful, but it's easy to track, seems likely to be useful, and would be a gigantic pain to backfill later if we decide we want it.
Test Plan: Accepted/rejected a revision, saw reviewers update appropriately.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12271
Differential Revision: https://secure.phabricator.com/D17537
