* Build ASElementMap and use it
* Keep building!
* Clean up
* Restrict ASDataController subclassing
* Clean up more
* More rearranging
* Dear lord it's time for bed
* Make things clearer
* Sadly remove subclassing restriction for ASDataController
* Remove dead initializer
* Refactor ASDataController
Check optional methods in ASDataControllerSource
* Reimplement reloadData
* Refactor new code
- No more new/inserted contexts flag
- Encapsulate code shared between reloadData and updateWithChangeSet
- Remove dataControllerWillDeleteAllData delegate method
- Hierarchy changes no longer needs to conform to NSCopying
- Reword TODOs
* Forgot to call completion block of reloadData :P
* Completion block of -[ASDataController reloadDataWithCompletion:] is nullable
* Data queried from ASCollectionNode and ASTableNode should be in UIKit index space
- This helps to avoid immature node allocation, especially when node virtualization is a thing
- However, this means that -reloadDataInitiallyIfNeeded in ASCollectionNode and ASTableNode must wait until all updates are finished.
* ASDataController shouldn't assume that allocated nodes were also laid out
* Revert "Data queried from ASCollectionNode and ASTableNode should be in UIKit index space"
This reverts commit 7bc977b3808a92f484b297781d0f5b30aa258e17.
* -nodeAtIndexPath: of ASDataController now forces node allocation, with the assumption that clients absolutely need it.
- Revisit this when node virtualization is implemented.
* ASDataController only grab changeSet.completionHandler when needed because it'll be niled out
* Fix ASTableViewTests related to reloadData
* Fix testThatDeletedItemsAreMarkedInvisible in ASCollectionViewTests
* Minor changes in ASCollectionView and ASTableView
* ASCollectionView and ASTableView shouldn't call [super reloadData] before their data controller does anything
* Address comments
* Fuse reloadData into -updateWithChangeSet: of ASDataController
* reloadData shouldn't be called as if it's inside a batch
- It can't be used in conjuntion with other updates.
- Calling it inside a batch update during the initial load can cause data inconsistency thrown by UICollectionView/UITableView
* Refactor ASDataControllerDelegate and ASRangeControllerDelegate
- Replace delegate methods in these protocols with -willUpdateWithChangeSet and -didUpdateWithChangeSet.
- ASRangeController, ASCollectionView and ASTableView are simplified because of this.
* Fix mismatch between sorting orders in ASDataController
* Forgot to call completion handler of reload change sets
* Make sure ASCollectionView is compatible with the behavior of UICollectionView's reloadData
- Since UICollectionView's reloadData doesn't requery data source but defers until the next layout pass, we need to wait until then to update range controller and do batch fetching.
- `-[ASCollectionView waitUntilAllUpdatesAreCommited]` needs to force a layout pass to make sure
everything is ready after it returns.
* testSectionIndexHandling of ASTableViewTests should only check visible nodes. Other nodes will be re-measured later.
* ASTableView is not ready until the first layout pass finished
* Address comments
* Bug fixes
* Allow UICollectionViewLayout to give us a layout inspector, always call the didChangeDelegate/didChangeDataSource on binding
* Make an assertion
* Update the tests
* Tests use actual layout inspector
* Be more consistent
Caching the envTraitCollection upon creation of `ASIndexedNodeContext` can lead to it becoming out of date. Ask the `id<ASEnvironment>` object for the trait collection when we need it.
* Reimplement IGListKit support in a cleaner way
* Rename and fix some stuff
* Fix supplementaries more
* Update docs
* Update test
* Cleanup minor things
* Tweak it
* Indentation
* Remove irrelevant changes
* Break out cell into its own file
* Fix indentation
* Address feedback
* Rejigger our flow layout supplementary support to make more sense.
* Support old protocol
* Update
* Update deprecation message
* Undeprecate insetForSection method, because it actually _does_ still work
* Update the tests
* Remove irrelevant junk
* Remove cast, add pragma
* Start removing ASChangeSetDataController
* Continue removing ASChangeSetDataController
* Remove unnecessary change
* ASDataController is no longer an abstract class, remove its assertion
* Get back beginUpdates and endUpdatesAnimated:completion in ASCollectionNode
* Remove duplicate import
* Move `ASLayoutController` related code out of CoreGraphics+ASConvenience
* Further cleanup
* Move private layout files to Privat/Layout
* Move ASLayoutElementStylePrivate into Private/Layout
* Further cleanup
* Move tvOS related files to tvOS folder
* Further cleanup
* Add inverted property to ASTableNode
* Add inverted property to ASCollectionNode
* Add example
* De-inform data controller and use different CATransform3D
* Add logic to pending state
* Add accessor for inverted property
* ASDataController: Correctly handle fetching node at nil index path (return nil).
* Be more aggressive, in order to avoid needlessly waiting for updates to complete
* Test that the initial range bounds are what we expect
* Use the user's previous scroll direction in the range controller
* Remove flow layout assertion
* Improve visibility handling for range-managed nodes
* Tweak ASWeakSet & the test
* Put back a few things that are pending a different diff
* Add a test
* Ensure we update visible nodes, even if there are no new ones
* Implement mutex ownership and use it to check potential upward lock gathering
* Don't hold instance lock and call willEnterHierarchy/didExitHierarchy of ASDisplayNode
- This can cause deadlocks (e.g #2605) if subsequent methods, that are implemented by developers, walk up the node tree.
- This is a way to keep the optimizations introduced in 9e87813 while making sure the locking situation is a bit safer.
* More lock ownership assertions in ASDisplayNode
* Document main thread contract of -clearContents
* ASNetworkImageNode shoud not call setNeedsPreload while holding instance lock
- This helps to avoid potentially deadlocks caused if the node (esp in case it's a subclass of ASNetworkImageNode) walks up the tree in didEnterPreloadState, for example to build logging context.
* ASVideoNode should not call setNeedsPreload while holding instance lock
- This helps to avoid potentially deadlocks caused if the node (esp. if it's a subclass of ASVideoNode) walks up the tree in didEnterPreloadState, for example to build logging context.
* Don't hold instance lock for the entire insert subnode operation
- The recursive lock should not be held throughout `_insertSubnode:atSubnodeIndex:sublayerIndex:andRemoveSubnode:`. The method instead should manage the lock itself and acquire it as shortly as possible. The reason is that this method calls many methods outside the scope of `self`. `[subnode __setSupernode:self]` is especially troublesome because it causes the subnode to migrate to new hierarchy and interface states, which triggers `didEnter/Exit(.*)State` methods. These methods are meant to be overriden by subclasses. Thus they might walk up the node tree and potentially cause deadlocks, or they perform expensive tasks and cause the lock to be held for too long.
- Other methods that call this method should release the lock before doing so.
* Lock getter and setter of `synchronous` flag
* Address comment in ASVideoNode
* Add main thread assertions to methods that change asset and assetURL of ASVideoNode
* Explain CHECK_LOCKING_SAFETY flag
* More thread and locking assertions in ASVideNode
- It's not safe to call `-[subnode __setSupernode:self]` while holding instance lock of soon-to-be supernode (e.g `self`).
- It's also not safe to call `[subnode __setSupernode:nil]` while holding the instance lock of the old supernode (e.g `self`).
- did(Enter|Exit)(.*)State methods are always called on main. Add main thread assertions to indicate that.
* Minor change in explanation of CHECK_LOCKING_SAFETY flag
* [ASDisplayNode] Ensure all subclasses are using base class __instanceLock__ and not re-defining their own.
This also moves the @package definition of the instance variable to +FrameworkPrivate instead of Internal.h,
because Internal.h should ideally not be used outside of the ASDisplayNode file setup. This has greatly reduced
the number of imports of Internal.h.
* [ASDisplayNode] Add ASDisplayNode+FrameworkSubclasses.h to share __instanceLock__ definition.
* Improve handling of rasterize node interface states and testing
* Fix harder
* Finish fixes and move rasterization flag into beta header
* Re-enable rasterization in ASDKgram
* Re-enable working test
* Only do it in debug
