Summary:
Depends on D19121. Ref T13083. Group transactions and show groups in the debugging view.
Fix some of the most obvious issues with fact generation:
- No more 0-point facts.
- Engine can now generate at least one of every type of fact.
Test Plan: Generated facts, viewed them in the debugging view, fact generation largely appeared to align with reality. No more "no facts in storage" facts.
Subscribers: yelirekim
Maniphest Tasks: T13083
Differential Revision: https://secure.phabricator.com/D19122
Summary:
Depends on D19120. Ref T13083. When you write a fact engine, it's currently somewhat difficult to figure out exactly what it's doing. It would also be difficult to diagnose bugs or report them to the upstream.
To ease this, add a page which shows all the facts an object generates. This allows you to iterate on an engine quickly without needing to reanalyze facts, take a screenshot, easily compare the timeline to the fact view, etc.
Test Plan: Viewed the object fact page for several objects.
Subscribers: yelirekim
Maniphest Tasks: T13083
Differential Revision: https://secure.phabricator.com/D19121
Summary:
Depends on D19119. Ref T13083. This is probably still very buggy, but I'm planning to build support tools to make debugging facts easier shortly.
This generates a large number of datapoints, at least, and can render some charts which aren't all completely broken in an obvious way.
Test Plan: Ran `bin/fact analyze --all`, got some charts with lines that went up and down in the web UI.
Subscribers: yelirekim
Maniphest Tasks: T13083
Differential Revision: https://secure.phabricator.com/D19120
Summary:
Ref T13083. Facts has a fair amount of weird hardcoding and duplication of responsibilities. Reduce this somewhat: no more hard-coded fact aggregates, no more database-driven list of available facts, etc. Generally, derive all objective truth from FactEngines. This is more similar to how most other modern applications work.
For clarity, hopefully: rename "FactSpec" to "Fact". Rename "RawFact" to "Datapoint".
Split the fairly optimistic "RawFact" table into an "IntDatapoint" table with less stuff in it, then dimension tables for the object PHIDs and key names. This is primarily aimed at reducing the row size of each datapoint. At the time I originally wrote this code we hadn't experimented much with storing similar data in multiple tables, but this is now more common and has worked well elsewhere (CustomFields, Edges, Ferret) so I don't anticipate this causing issues. If we need more complex or multidimension/multivalue tables later we can accommodate them. The queries a single table supports (like "all facts of all kinds in some time window") don't make any sense as far as I can tell and could likely be UNION ALL'd anyway.
Remove all the aggregation stuff for now, it's not really clear to me what this should look like.
Test Plan: Ran `bin/fact analyze` and viewed web UI. Nothing exploded too violently.
Subscribers: yelirekim
Maniphest Tasks: T13083
Differential Revision: https://secure.phabricator.com/D19119
Summary: See D19117. Instead of automatically figuring this out inside `phutil_tag()`, explicitly add rel="noreferrer" at the application level to all external links.
Test Plan:
- Grepped for `_blank`, `isValidRemoteURIForLink`, checked all callsites for user-controlled data.
- Created a link menu item, verified noreferrer in markup.
- Created a link custom field, verified no referrer in markup.
- Verified noreferrer for `{nav href=...}`.
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Differential Revision: https://secure.phabricator.com/D19118
Summary: Ref T13077. The context object wasn't being passed into the engine properly here, affecting relative link rendering in Phriction.
Test Plan: Viewed rendered Phriction documents with relative links, got clean renders.
Maniphest Tasks: T13077
Differential Revision: https://secure.phabricator.com/D19115
Summary: Ref T13077. This content extraction rule wasn't right and caused rendering on Phriction pages to extract context improperly.
Test Plan: Viewed pages in Phriction with relative links to other documents.
Maniphest Tasks: T13077
Differential Revision: https://secure.phabricator.com/D19114
Summary: Ref T13077. Freeze "phriction.info" in favor of the more modern "phriction.document.search".
Test Plan: Reviewed older method in web UI, saw frozen markers.
Maniphest Tasks: T13077
Differential Revision: https://secure.phabricator.com/D19113
Summary: Ref T13077. Adds a "paths" constraint to the API query.
Test Plan: Used paths constraint to fetch documents.
Maniphest Tasks: T13077
Differential Revision: https://secure.phabricator.com/D19112
Summary: Depends on D19107. Ref T13077. The underlying datasource may need some adjustment but this appears to work properly locally.
Test Plan: Typed `[[ por` locally, was suggested "Porcupine Facts". Typed `[[ / ]]`, saw it render as a reference to the wiki root instead of the install root.
Maniphest Tasks: T13077
Differential Revision: https://secure.phabricator.com/D19108
Summary:
Depends on D19106. Fixes T5941. Ref T13077. Allows you to find Phriction documents as suggestions from global quick search.
Also supports `w` to jump to Phriction and `w query` to query Phriction.
The actual query logic for the datasource may need some tweaking after it collides with reality, but seems to produce fairly reasonable results in local testing against synthetic data.
Test Plan: Searched for "Porcupine Facts", "Travel Companions", and other useful local pages. Searched for `w`. Searched for `w zebra facts`.
Maniphest Tasks: T13077, T5941
Differential Revision: https://secure.phabricator.com/D19107
Summary:
Depends on D19105. Ref T13077. Fixes T12344.
The `[[ ... ]]` syntax accepts and handles characters which would require URL encoding if they appeared in URIs. For example, `[[ 100% Natural Cheese Dust ]]` is a legitimate, supported piece of remarkup syntax, and does not need to be written as `... 100%25 Natural ...`.
Likewise, `[[ BUY $DOGE ]]` is legitimate and does not need to be written as `[[ BUY %24DOGE ]]`. This piece of syntax creates a link to `/w/buy_$doge/`. This may or may not appear in your browser's URL bar as `/w/buy_%24doge/`, but internally "$" is a valid slug character and you'll see `buy_$doge` over Conduit, etc.
However, since users may reasonably copy paths from their browser URL bar, they may have unnecessary URL encoding. The syntax `[[ buy_$doge ]]` is legitimate, but a user copy/pasting may write `[[ buy_%24doge ]]` instead.
Currently, this extra URL encoding causes links to break, since `[[ buy_%24doge ]]` is a treated as link to `/w/buy_24doge/`, just like `[[ Fresh 100%AB Blood ]]` is a link to `/w/fresh_100_ab_blood/`.
To fix this:
- When the target for a link can be URL decoded, try to do lookups on both the un-decoded and decoded variations.
- If the un-decoded variation works, great: use it. This preserves behavior for all existing, working links.
- If the un-decoded variation fails but the decoded variation works, okay: we'll assume you copy-pasted a URL-encoded version and strip URL encoding.
- If both fail, same behavior as before.
Also, use a different spelling for "existent".
See T13084 for some "attacks" based on this behavior. I think the usability affordance this behavior provides greatly outweighs the very mild threat those attacks represent.
Test Plan:
- Created links to existing, nonexisting, and existing-but-not-visible documents, all of which worked normally.
- Created links to `[[ $doge ]]` and `[[ %24doge ]]`, saw them both go to the right place.
- Performed the "attacks" in T13084.
Maniphest Tasks: T13077, T12344
Differential Revision: https://secure.phabricator.com/D19106
Summary:
Fixes T10969. Ref T13077. When you create a Phriction document with a relative link (`[[ ./path/to/page ]]`) the initial email currently points to the wrong place.
This is because the context object (the page) isn't passed to the markup engine. Without this context, the relative link is rendered as though it appeared somewhere else (like a task or revision) where relative links don't make sense.
Test Plan: Created a new Phriction document with a relative link to `[[ ./porcupine_facts/starmap ]]`, saw a usable link in the resulting email.
Maniphest Tasks: T13077, T10969
Differential Revision: https://secure.phabricator.com/D19105
Summary:
See PHI370. Support the "Affected packages" and "Affected package owners" Herald fields in pre-commit hooks.
I believe there's no technical reason these fields aren't supported and this was just overlooked.
Test Plan: Wrote a rule which makes use of the new fields, pushed commits through it. Checked transcripts and saw sensible-looking values.
Differential Revision: https://secure.phabricator.com/D19104
Summary: Depends on D19100. Ref T13077. Adds a "content" attachment to get the actual page text. This works on both "phriction.document.search" and "phriction.content.search".
Test Plan: Called both API methods with the attachment, saw proper text content returned.
Maniphest Tasks: T13077
Differential Revision: https://secure.phabricator.com/D19103
Summary:
Depends on D19099. Ref T13077. Updates Phriction documents to string constants to make API interactions cleaner and statuses more practical to extend.
This does not seem to require any transaction migrations because none of the Phriction transactions actually store status values: status is always a side effect of other edits.
Test Plan: Created, edited, deleted, moved documents. Saw appropriate UI cues. Browsed and filtered documents by status in the index.
Maniphest Tasks: T13077
Differential Revision: https://secure.phabricator.com/D19100
Summary:
Depends on D19098. Ref T13077.
Phriction status constants currently use the "bag of statuses" approach typical of older code, and store integers in the database.
This fixes the "bag of statuses" stuff; a future change will fix the integers.
Also adds a skeleton for `phriction.document.search`, but doesn't implement the Conduit interface yet.
Test Plan: Searched for documents with various status constraints. Grepped for removed status constants. Viewed document list.
Maniphest Tasks: T13077
Differential Revision: https://secure.phabricator.com/D19099
Summary: Depends on D19097. Ref T13077. Freeze the older method now that the newer one is available.
Test Plan: Viewed the older method's page and saw it frozen; called it to make sure I didn't break it by accident.
Maniphest Tasks: T13077
Differential Revision: https://secure.phabricator.com/D19098
Summary: Depends on D19096. Ref T13077. Adds a new "v3" API method for Phriction document content, to replace the existing "phriction.history" call.
Test Plan: Made various calls via web API console.
Maniphest Tasks: T13077
Differential Revision: https://secure.phabricator.com/D19097
Summary:
Depends on D19095. Ref T6203. Ref T13077. This column is nullable in an inconsistent way. Make it non-nullable.
Also clean up one more content query on the history view.
Test Plan: Ran migration, then created and edited documents without providing a descriptino or hitting `NULL` exceptions.
Maniphest Tasks: T13077, T6203
Differential Revision: https://secure.phabricator.com/D19096
Summary: Depends on D19094. Ref T13077. Use modern infrastructure to perform these loads. I left a couple of calls in the older API methods unconverted.
Test Plan: Viewed documents. Viewed older versions. Viewed diffs. Did revert edits to older versions.
Maniphest Tasks: T13077
Differential Revision: https://secure.phabricator.com/D19095
Summary:
Depends on D19093. Ref T13077. Although content objects normally don't have any edges today, they may in the future.
Also implement Policy stuff properly.
Test Plan: Used `bin/remove destroy` to destroy a document, verified it also loaded and destroyed the correspoding Content correctly by looking at `--trace` and the database rows.
Maniphest Tasks: T13077
Differential Revision: https://secure.phabricator.com/D19094
Summary:
Depends on D19092. Ref T13077. This modernizes markup rendering for PhrictionContent.
This is a little messy because table of contents generation isn't straightforward.
Test Plan: Viewed Phriction documents with and without 3+ headers, saw ToC vs no ToC. Edited/previewed documents. Grepped for affected symbols. Checked DarkConsole for sensible cache behavior.
Maniphest Tasks: T13077
Differential Revision: https://secure.phabricator.com/D19093
Summary: Ref T13077. Prepares for modern API access to document history using standard "v3" APIs.
Test Plan: Ran migration, verified PHIDs appeared in the database. Created/edited a document, got even more PHIDs in the database.
Maniphest Tasks: T13077
Differential Revision: https://secure.phabricator.com/D19092
Summary: Ref T13073. The new log output from `bin/drydock lease` currently uses HTML handle rendering, but should render to text.
Test Plan: Ran `bin/drydock lease` and saw normal text in log output. Viewed the same logs from the web UI and saw HTML.
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13073
Differential Revision: https://secure.phabricator.com/D19101
Summary: Depends on D19089. Fixes T13079. This is likely not the final form of this, but creates a defensible extension point.
Test Plan: See T13079 for discussion.
Maniphest Tasks: T13079
Differential Revision: https://secure.phabricator.com/D19090
Summary:
Depends on D19088. Ref T13079.
> Any sufficiently complicated C or Fortran program contains an ad hoc, informally-specified, bug-ridden, slow implementation of half of Common Lisp.
> - Greenspun's Tenth Rule
Move us a step closer to this noble goal.
This doesn't implement any `viewer(project())` stuff but it looks like the API doesn't need to change to do that in the future.
Test Plan: Grimmaced in pain.
Maniphest Tasks: T13079
Differential Revision: https://secure.phabricator.com/D19089
Summary: Depends on D19087. Ref T13079. This still doesn't feel like the most clean, general system in the world, but is a step forward from hard-coded `switch()` stuff.
Test Plan:
- Jumped to `r`.
- Jumped to `a`.
- Jumped to `r poe` (multiple results).
- Jumped to `r poetry` (one result).
- Jumped to `r syzygy` (no results).
- Jumped to `p`.
- Jumped to `p robot` (multiple results); `p assessment` (one result).
- The behavior for `p <string>` has changed slightly but should be more powerful now (it's consistent with `r <string>`).
- Jumped to `s <symbol>` and `s <context>-><symbol>`.
- Jumped to `d`.
- Jumped to `f`.
- Jumped to `t`.
- Jumped to `T123`, `D123`, `@dog`, `PHID-DREV-abcd`, etc.
Maniphest Tasks: T13079
Differential Revision: https://secure.phabricator.com/D19088
Summary: Ref T13079. This recently-introduced Engine/EngineExtension are a good fit for adding more datasource functions in general, but we didn't think quite big enough in naming them.
Test Plan: Used quick search typeahead, hit applications/users/monograms/symbols/etc.
Maniphest Tasks: T13079
Differential Revision: https://secure.phabricator.com/D19087
Summary:
Depends on D19084. Fixes T13078. When `phabricator.silent` is enabled, immediately fail the "HTTP Request", "CircleCI" and "Buildkite" build steps.
This doesn't feel quite as clean as most of the other behavior of `phabricator.silent`, since these calls are not exactly notifications in the same way that email is, and failing to make these calls means that builds run differently (whereas failing to deliver email doesn't really do anything).
However, I suspect that this behavior is almost always reasonable/correct, and that we can probably get away with it until this grey area between "notifications" and "external service calls" is more clearly defined.
Test Plan:
- Created a build with HTTP, CircleCI, and Buildkite steps.
- Put install in `phabricator.silent` mode: all three steps failed with "declining, because silent" messages.
- Put install back in normal mode: all three steps made HTTP requests.
- Read updated documentation.
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13078
Differential Revision: https://secure.phabricator.com/D19085
Summary: Ref T13078. The `phabricator.silent` configuration flag should disable webhook calls, since this is consistent with the documented and desired behavior.
Test Plan: Enabled `phabricator.silent`, made test hook calls, saw them fail with a "silent" failure reason.
Maniphest Tasks: T13078
Differential Revision: https://secure.phabricator.com/D19084
Summary: Revisions with blocking reviewers had this stamp built incorrectly, which cascaded into trying to use `array()` as a PHID. Recover so these tasks succeed.
Test Plan: Will deploy production.
Differential Revision: https://secure.phabricator.com/D19082
Summary:
Depends on D19078. Ref T13073. Currently, there is a narrow window where we can acquire a resource after a reclaim has started against it.
To prevent this, briefly lock resources before acquiring them and make sure they're still good. If a resource isn't good, throw the lease back in the pool.
Test Plan:
This is tricky. You need:
- Hoax blueprint with limits and a rule where leases of a given "flavor" can only be satisfied by resources of the same flavor.
- Reduce the 3-minute "wait before resources can be released" to 3 seconds.
- Limit Hoaxes to 1.
- Allocate one "cherry" flavored Hoax and release the lease.
- Add a `sleep(15)` to `releaseResource()` in `DrydockResourceUpdateWorker`, after the `canReclaimResource()` check, with a `print`.
Now:
- Run `bin/phd debug task` in two windows.
- Run `bin/drydock lease --type host --attributes flavor=banana` in a third window.
- This will start to reclaim the existing "cherry" resource. Once one of the `phd` windows prints the "RECLAIMING" message run `bin/drydock lease --type host --attributes flavor=cherry` in a fourth window.
- Before patch: the "cherry" lease acquired immediately, then was released and destroyed moments later.
- After patch: the "cherry" lease yields.
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13073
Differential Revision: https://secure.phabricator.com/D19080
Summary:
Depends on D19077. Ref T13073. When we're using slot locks to enforce a limit (e.g., maximum of 5 simultaneous things) we currently load locks owned by the blueprint to identify which slots are likely to be free.
However, this isn't right: the blueprint doesn't own these locks. The resources do.
We still get the right behavior eventually, but we incorrectly identify that every slot lock is always free, so as the slots fill up we'll tend to guess wrong more and more often.
Instead, load the slot locks by name explicitly.
Test Plan: Implemented lock-based limiting on `HoaxBlueprint`, `var_dump()`'d the candidate locks, saw correct test state for locks. Acquired leases without releasing, got all of the slots filled without any slot lock collisions (previously, the last slot or two tended to collide a lot).
Subscribers: yelirekim, PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13073
Differential Revision: https://secure.phabricator.com/D19078
Summary:
Depends on D19076. Ref T13073. Blueprints are stored as an attribute and `setAttributes()` overwrites all attributes.
This is sorta junk but make it less obviously broken, at least.
Test Plan: Ran `bin/drydock lease --type working-copy --attributes x=y` without instantly getting a fatal about "no blueprint PHIDs".
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13073
Differential Revision: https://secure.phabricator.com/D19077
Summary:
Depends on D19075. Ref T13073. If a lease acquires a resource but finds that the resource builds directly into a dead state (which can happen for any number of reasonable reasons), reset the lease and throw it back in the pool.
This isn't the lease's fault and it hasn't caused any side effects or done anything we can't undo, so we can safely reset it and throw it back in the pool.
Test Plan:
- Created a blueprint which throws from `allocateResource()` so that resources never activate.
- Tried to lease it with `bin/drydock lease ...`.
- Before patch: lease was broken and destroyed after a failed activation.
- After patch: lease was returned to the pool after a failed activation.
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13073
Differential Revision: https://secure.phabricator.com/D19076
Summary: Ref T13073. Depends on D19074. Update icons and UI for resource status.
Test Plan: Viewed resources in detail view and list view, saw better status icons.
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13073
Differential Revision: https://secure.phabricator.com/D19075
Summary:
Depends on D19073. Ref T13073. Give leases a normal header tag and try to wrangle their status constants a bit.
Also, try to capture the "status class" pattern a bit. Since we target PHP 5.2.3 we can't use `static::` so the actual subclass is kind of a mess. Not exactly sure if I want to stick with this or not. We could consider targeting PHP 5.3.0 instead to get `static::` / late static binding.
Test Plan: Viewed leases and lease lists, saw better and more conventional status information.
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13073
Differential Revision: https://secure.phabricator.com/D19074
Summary:
Depends on D19072. Ref T13073. Currently, you can leave leases stranded by using `^C` to interrupt the script. Handle signals and release leases on destruction if they haven't activated yet.
Also, print out more useful information before and after activation.
Test Plan: Mashed ^C while runnning `bin/drydock lease ... --trace`, saw the lease release.
Subscribers: yelirekim, PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13073
Differential Revision: https://secure.phabricator.com/D19073
Summary: Depends on D19071. Ref T13073. While the daemons are supposedly doing things, show the user any logs they generate. There's often something relevant but unearthing it can be involved.
Test Plan: {F5427773}
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13073
Differential Revision: https://secure.phabricator.com/D19072
Summary: Depends on D19070. Ref T13073. Some messages contain an interesting story or a clever anecdote. Respect newlines during rendering to preserve authorial intent.
Test Plan:
Viewed a message with linebreaks and could still read it.
{F5427754}
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13073
Differential Revision: https://secure.phabricator.com/D19071
Summary:
Ref T13073. When a Blueprint says it will be able to allocate a resource but then throws an exception while attempting that allocation, we currently fail the lease permanently.
This is excessively harsh. This blueprint may have the best of intentions and have encountered a legitimately unforseeable failure (like a `vm.new` call to build a VM failed) and be able to succeed in the future.
Even if this blueprint is a dirty liar, other blueprints (or existing resources) may be able to satisfy the lease in the future.
Even if every blueprint is implemented incorrectly, leaving the lease alive lets it converge to success after the blueprints are fixed.
Instead of failing, log the issue and yield.
(In the future, it might make sense to distinguish more narrowly between "actually, all the resources are used up" and all other failure types, since the former is likely more routine and less concerning.)
Test Plan:
- Wrote a broken `Hoax` blueprint which always claims it can allocate but never actually allocates (just `throw` in `allocateResource()`).
- Used `bin/phd drydock lease` to acquire a Hoax lease.
- Before patch: lease abruptly failed permanently.
- After patch: lease yields after allocation fails.
{F5427747}
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13073
Differential Revision: https://secure.phabricator.com/D19070
Summary: Like "Commit Hook" rules, these also fire oddly and don't have an object PHID or a list of transactions.
Test Plan: Verified that "Call Webhooks" was no longer available from Diff rules, but still available from other rule types.
Differential Revision: https://secure.phabricator.com/D19069
Summary:
See PHI346. Ref T13054. If you have prototypes enabled on the server but use `master` / `stable` on the client and run `arc diff --plan-changes`, the transition is rejected because "Draft -> Changes Planned" isn't currently a legal transition.
Allow this transition if not coming from the web UI (to keep it out of the dropdown).
Test Plan:
- Ran `arc diff --plan-changes` on `master`, got a "Changes Planned" revision instead of a validation error.
- Ran `arc diff` without `--plan-changes`, got a draft, verified that "Plan Changes" still doesn't appear in the action dropdown.
Maniphest Tasks: T13054
Differential Revision: https://secure.phabricator.com/D19067
Summary:
Depends on D19065. Ref T13054. Instead of just updating `containerPHID` and hoping for the best, queue a proper BuildWorker to process a "your container has changed, update it" message.
We also need to remove a (superfluous) `withContainerPHIDs()` when loading active diffs for a revision.
Test Plan:
- Without daemons, created a revision and saw builds stick in "preparing" with no container PHID, but also stay in draft mode.
- With daemons, saw builds actually build and get the right container PHID.
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13054
Differential Revision: https://secure.phabricator.com/D19066
Summary:
Depends on D19064. Ref T13054. See that task for additional discussion.
When buildables are created by `arc` and have lint/unit messages, they can currently pass or fail before Herald triggers actual builds. This puts them in a pre-build state where they can't complete until Herald says it's okay.
On its own, this change intentionally strands `arc diff --only` diffs in the "PREPARING" stage forever.
Test Plan:
- Ran a build with `bin/harbormaster`, saw it build normally.
- Ran a build with web UI, saw it build normally.
- Ran a build with `arc diff`, saw it build normally.
- Ran a build with `arc diff --only`, saw it hang in "PREPARING" forever.
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13054
Differential Revision: https://secure.phabricator.com/D19065
Summary: Depends on D19063. Ref T13054. Prepare for the addition of a new `PREPARING` status by getting rid of the "scattered mess of switch statements" pattern of status management.
Test Plan: Searched/browsed buildables. Viewed buildables. Viewed revisions. Grepped for all affected symbols.
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13054
Differential Revision: https://secure.phabricator.com/D19064
