* Measure performance for ASCellNode layout
* Address Adlai and Levi's feedback:
* Move to ASDisplayNode level
* Lock around setting/getting measurement options and results
* Record all measurement passes and report times as an array
* Only add relevant entries to the performanceMetrics dictionary
* Rebase
* Store sum and count instead of array
* Rename ScopeTimerDataPoint to ScopeTimerSum
* Address Levi's feedback
* Address Adlai's feedback
The previous code used either a `@property` with `assign, nonatomic` semantics, or a `__unsafe_unretained` (non-atomic) ivar pointer to an instance for `asyncdisplaykit_node`.
This commit changes those access so that they have the equivalent of `weak, atomic` semantics.
- `_ASDisplayView.mm`
- Removes the `_node` ivar that was qualified with `__unsafe_unretained`.
- Removes `@synthesize asyncdisplaykit_node = _node;`.
- All direct uses of `_node` were replaced with:
- Creates a strong reference via `ASDisplayNode *node = _asyncdisplaykit_node;`.
- `s/_node/node/`.
- These changes were made even if there's a single use of `_asyncdisplaykit_node` as a consistency / defensive measure so that anyone editing this code in the future does not accidentally use a `weak` reference twice.
- `ASDisplayNode.mm`
- Getters and setters for `asyncdisplaykit_node` were changed from `nonatomic, assign` semantics to the equivalent of `atomic, weak` semantics.
- `weak` semantics were implemented by using a `ASWeakProxy` object.
- `objc_setAssociatedObject` was changed from `OBJC_ASSOCIATION_ASSIGN` to `OBJC_ASSOCIATION_RETAIN` (the `atomic` version of retain).
- `ASDisplayNode+FrameworkPrivate.h`
- Changed the `@property` declarations for `asyncdisplaykit_node` from `nonatomic, assign` to `atomic, weak`.
- The actual getters and setters are implemented in `ASDisplayNode.mm`, which were changed accordingly.
* Initial commit for adding a size constraint to ASLayoutable's
* Some more commits
* Fix sample projects in extra/
* Remove preferredFrameSize test of ASEditableTextNode
* Remove preferredFrameSize from examples_extra
* Add deprecation warning to -[ASDisplayNode preferredFrameSize]
* Add deprecation warning to -[ASDisplayNode measureWithSizeRange:]
* Commit
* Commit
* Remove ASRelativeSizeRange
* Make ASRelativeSize private
* Adjust examples
* Improve setting of -[ASLayoutable size] with points or fractions
* Add ASWrapperLayoutSpec
* Improve creation of ASRelativeDimension
* Add `preferredFrameSize` back and add deprecated logging
* Add `layoutSpecBlock` setter and getter and add locking for it
* Add better documentation and fix macros to create ASRelativeDimension
* Create new ASSizeRangeMake with just a CGSize as parameter
* Update Kitten and Social App Layout example
* Add layoutThatFits: and deprecate measure:
* Rename ASRelativeDimension to ASDimension
* Fix examples for ASDimension renaming
* Remove fancy height and width setter
* Fix ASDimension helper
* Rename -[ASLayout layoutableObject] to -[ASLayout layoutable]
* Update layout related methods and more clearer documentation around how to use it
* Deprecate old ASLayout class constructors
* Don't unnecessary recalculate layout if constrained or parent size did not change
* Use shared pointer for ASDisplayNodeLayout
* Fix rebase conflicts
* Add documentation and move implementation in mm file of ASDisplayNodeLayout
* Fix test errors
* Rename ASSize to ASLayoutableSize
* Address comments
* Rename setSizeFromCGSize to setSizeWithCGSize
* Improve inline functions in ASDimension
* Fix rebase conflicts
* Dealloc _pendingViewState if range managed
_ASPendingState objects can add up very quickly when adding
many nodes. This is especially an issue in large collection views
and table views. This needs to be weighed against the cost of
reallocing a _ASPendingState. So in range managed nodes we
delete the pending state, otherwise we just clear it.
* Adlai's comments
* Automatic measure pass in layout pass if not happened before
If no measure pass happened or the bounds changed between layout passes we manually trigger a measurement pass for the node using a size range equal to whatever bounds were provided to the node
* Add test method that ensures that on a second layout pass with same bounds, layoutSpecThatFits: / layoutSpecBlock is not called
* Attempt to reproduce supplementary crash
* Get closer with supplementary issue testing
* Alright! We have a repro!
* The investigation continues
* Fixed!
* Renamed range update callbacks
We finally settled on
didEnter/ExitDisplayState
didEnter/ExitPreloadState
didEnter/ExitVisibleState
This change is meant to unify the range update methods to relate to each
other and hopefully be a bit more self explanatory.
* Guarantee interface callbacks happen on main.
* move fetchData / clearFetchedData to default implementations
* Move deprecated methods to new deprecated category
* Don't bring in cocoapod change.
* Nits
* Capetalize
* Fixes a deadlock caused by walking up the heirarchy.
* Use scope locker / unlocker and add a comment.
* Add comment about calling __setNeedsLayout without the lock.
* Add default fade in / out layout transition
* Add example for layout transition
* Update for recent layout transition API changes
* To be able to do a layoutTransition the node needs to be loaded
* Rename layoutTransitionDuration to defaultLayoutTransitionDuration
* Expose default layout transition duration delay and options
* Use `UIViewAnimationOptionBeginFromCurrentState` for initial defaultLayoutTransitionOptions
The pending layout transition needs to stay alive at least until applySubnodeInsertions did finish execute as it can happen that with Implicit Hierarchy Management new nodes gonna be added that internally call setNeedsLayout what will invalidate and deallocate the transition in the middle of inserting nodes
* Revert "Revert "[ASLayoutSpec] Use childrenMap directly to prevent creating an NSArray within ASDK (#1937)""
This reverts commit 735b4ebd0872483044d98a5d05b43324e76fc8d4.
* Fix crash and add exception for mutating while using fast enumeration of ASLayoutSpec children
NSFastEnumeration is potentially quite dangerous in the wrong hands. In particular, it does not provide a safe mechanism for you to return temporary objects directly, and it does not provide any guarantee that you will be called when the enumeration has completed; therefore if we generate temporaries and store them in an instance variable, we will not necessarily be able to clean them up! This means fast enumeration methods should never be called within an autorelease pool or the autorelease pool be drained within the fast enumeration loop.
The reason is we store references to objects in the stackBuf struct by casting the child pointer to __autoreleasing id. If we pop the autorelease pool between calls to -countByEnumeratingWithState:objects:count:, it will die in a messy explosion of pointer dereferences and EXC_BAD_ACCESS.
* Add tests for ASDisplayNode and ASLayoutSpec fast enumeration
If a node was added to a supernode, the supernode could be in a layout pending state. All of the hierarchy state properties related to the transition need to be copied over as well as propagated down the subtree. This is especially important as with Implicit Hierarchy Management adding subnodes can happen while a transition is in fly
Something interesting going on here with ARC / Objective-C++ that we are investigating and will re-land.
This reverts commit c90ed08d1073701e2c7f8a2677d460c140f05264.
* [ASDisplayNode] Ensure that nil can never be returned from -measureWithSizeRange:
This can happen in rare cases when multiple relayouts occur while a transition is being measured.
* [ASDisplayNode] Use ternary operator style for nil check.
* Use childrenMap directly to prevent creating an NSArray in ASDK for ASLayoutSpec children
* Add locking for parent property in ASLayoutSpec
* Remove unnecessary import
* Add newline
* Add NSFastEnumeration to ASEnvironment and ASDisplayNode / ASLayoutSpec
* Change NSMutableArray initializer to arrayWithCapacity:
* Move ASLayoutSpec+Private.h into Private folder
Fixes building with Swift
* Remove lock for ASLayoutSpec parent
* Simplify applying layout transition in preparation for bigger layout transition API work
* Change from apply to complete in if layout transitions are involved and _applyLayout: to _setCalculatedLayout: for layout
* Change to applySubnodeInsertions and applySubnodeRemovals
* Change from completeTransition to commitTransition and flip logic around when to trampoline to the main thread for implicit hierarchy management
* More internal API improvements
* Fix merge conflicts
* Rename _layout to _calculatedLayout
* [ASEnvironmentTraitCollection] default user interface idiom to ASEnvironmentTraitCollection
UIUserInterfaceIdiomUnspecified is -1 so we were actually defaulting our trait collection to UIUserInterfaceIdiomPhone (which is 0).
* Fix a few places where we weren’t using the default method to create the base traits
[_ASHierarchyChangeSet] Oopsy daisy
[ASDataController] Tweak our update validation
[ASHierarchyChangeSet] Fix bugs
Finish up some stuff
[ASDataController] Put some stuff back
[ASChangeSetDataController] Always use changeset
[ASDataController] Put other stuff back
[_ASHierarchyChangeSet] Use fast enumeration
[_ASHierarchyChangeSet] Fix assertion format strings, return on fail so we don't crash in production
[ASDataController] Store data source item counts as vector rather than NSArray
[ASDataController] Build some tests for the update validation
[ASDataController] Fix issues with update validation
Get rid of new file
[ASDataController] Suppress changeset validation before initial reload
[ASDataController] Make invalid update log vs. exception publicly toggleable
* [ASDisplayNode] Short circuit measure calls that have a zero-area constrainedSize.
// If the constrainedSize is completely zero-area, then there is no possibility for layout calculations to be successful.
// This also avoids the issue of an inset being applied to 0, creating negative frame values.
* [ASDisplayNode] Fix to shouldMeasure change.
* One more fix.
* Add automatic measurement before layout
* Remove code not needed or addressed in a different PR
* Adjust comments and rename __layoutSublayouts to __layoutSubnodes
* Check before setting up a placeholder layer if the node should have a placeholder
* Always layout nodes on a background thread
* Remove semaphore in ASDataController for allocating nodes and layout
* Fix variable not used error
* Remove overhead to create subarray of contexts of nodes while layout nodes
* Remove extra allocation of allocatedNodes and indexPaths array
