* Shrink ASDisplayNode from 1088 to 976 bytes, reduction of 11.48%
These objects accumulate in the heap, so reducing their size will allow more to accumulate before memory warnings.
Group the `BOOL`s into a struct. Shrink the various stored `enum`s to fit the size of their contents. Move the ivars around so that the smaller `enum` are near eachother and the bitfield struct.
* Forgot to bit-field-ify placeholderEnable; new small size is 968, old size is 1072 (measured consistentlyl on iPhoneSE simulator). 10.74% reduction.
* Remove ASEventLog. We'll focus on ASLog for the future, possibly making it triggerable on a per-node basis
* See ya!
* Hello semicolon
* Remove refs to gone tree methods
* Fix
* Remove irrelevant change, kick CI
* Clean up
We came across an infinite layout loop in `_u_measureNodeWithBoundsIfNecessary`. After requesting a layout from above, the sizes between pending and caluclated layout still do not match. We continue to prefer to use the pending layout and ask for another layout loop from above. We can’t seem to break out of this loop. The solution (thanks to Huy for the guidance) was to nil out the pending layout we get from requesting the layout from above.
I was only able to reproduce this when working with a node in a `UINavigationBarItem’s` `titleView`. I think that UIKit must be doing something sneaky with setting the frame on the view. While I was not able to create a unit test to catch this issue (I tried for a long time, and can post what I’ve come up with to see if anyone has any suggestions), I was able to create a pretty simple example project that shows the behavior:
https://github.com/rcancro/TextureLayoutLoopExample
* Rename ASDN C++ namespace to "AS." Referring to the framework as ASDisplayNode is pretty out-dated and verbose. See CoreAnimation which usees CA for their internal namespace.
More using
* More cases
* Optimize _assertSubnodeState
This method is actually pretty painful in today's world. In one iPad mini trace, the first page of nodes spent 6.6ms in this call, just in time profiler.
* Clean it up, check count
* Check the right value
* Lock up to yogaRoot during layout to avoid dead lock.
1) lock to root for tree
2) lock self to change parent (& consequently root)
3) Implement ASLocking (tryLock) on ASNodeController
4) add lockPair to try-lock node & controller together
5) lock controllers if they exist in lockToRoot...
Disable some asserts due to lock to root. :(
LL# No commands remaining.
* Add macro so non-Yoga still builds :)
* wut
* Improve separation of code for layout method types
* Address PR comments
- Delegate to layout spec engine if the node is a layout spec node but yoga engine was asked for calculate the layout
- Change ASLayoutType to ASLayoutEngineType
- Improve layout engine fall through code
We will revert #1023. The current solution introduces problems if we are unlocking before calling _completePendingLayoutTransition. _completePendingLayoutTransition needs to be happening in one transaction if called from _u_measureNodeWithBoundsIfNecessary.
* Thread safety for Yoga layout
* Support baseline alignments for ASYogaLayout
* Refactor ASLayoutElementYogaBaselineFunc to not require yogaParent (its parent style is set into a private var on ASLayoutElementStyle before layout instead)
* Only set the accessibility element if the view is loaded
* Add nodeWillCalculateLayout to ASNodeController
* Update Changelog
* Address first comments
* Simpler Huy fix for more efficient delayed flattening of the layout tree
* Nit
* Remove pbx changes
* Update CHANGELOG.md
* Add note about change in timing of _flattenedLayout capture
* [License] Simplify the Texture license to be pure Apache 2 (removing ASDK-Licenses)
With permission of the Facebook Open Source team, we are simplifying the Texture
license so that clients can rely on the Apache 2 terms that most of Texture is
already covered by. This means that code originally forked from AsyncDisplayKit
will be re-licensed from "BSD 3-clause + PATENTS v2" to Apache 2 without a
PATENTS file.
After getting confirmation that the updates to these core files look good, we'll
propagate this new license header to all files (in this same PR) and get sign-off
from all parties before landing.
* [License] Update all Texture source files to be pure Apache 2.
* Changelog entry for Apache 2 license update.
* Revert "[License] Update all Texture source files to be pure Apache 2."
This reverts commit ffa0fbbba9.
* [License] Update all Texture source files to be pure Apache 2, maintaining copyrights.
* [License] Update CONTRIBUTING, README, Podspec & Dangerfile.
- Rename `ASDisplayNodeAssertLockUnownedByCurrentThread` to `ASAssertUnlocked`, and `ASDisplayNodeAssertLockOwnedByCurrentThread` to `ASAssertLocked` -> shorter and hopefully easier to distinguish between the two.
- Add assertions to `_locked_` and `_u_` (i.e "unlocked") methods.
- Turn `CHECK_LOCKING_SAFETY` flag on by default. After #1022 and #1023, we're in a good shape to actually enforce locked/unlocked requirements of internal methods. Our test suite passed, and we'll test more at Pinterest after the sync this week.
- Fix ASVideoNode to avoid calling `play` while holding the lock. That method inserts a subnode and must be called lock free.
- Simplify `_loaded(node)` to only nil-check `_layer` because regardless of whether the node is view or layer backed, the layer should always be set if loaded. Use it throughout.
- Other minor changes.
* Add move detection and support to ASLayoutTransition
...and NSArray+Diffing.
Add some tests.
* Update CHANGELOG.md
* Update CHANGELOG.md
* Update ASLayout+IGListKit.h
* Update ASLayout+IGListKit.mm
* Use std collections to avoid NSNumber boxing
* Update ASLayoutTransition.mm
* Code review updates.
* Use `unordered_multimap` on stack instead of unordered_map<id,queue> on heap
* Remove notFound BOOL (use NSNotFound sentinel value) and put some vars inside the if (insertions/moves) loop
* Don't copy defaultCompare block (redundant under ARC)
* Whitespace
* Remove unneeded mutableCopy-s in ArrayDiffingTests
* Code review updates.
* Type _subnodeMoves pair.first to ASDisplayNode * instead of id
* C++ enumeration
* unowned refs for adding previousLayout nodes to _subnodeMoves
* Remove unreleated ASDynamicCast that is probably right though
* Add commentary to NSArray+Diffing.h; make multimap elements unowned
* Use std::make_pair, optimize ASLayout+IGListKit
* Oops I thought I had added these headers but nope
* Simplify simplify
* Diff subnodes instead of sublayouts
* Another randomized test with actual ASLayouts
This is because committing the layout transition (aka `-_completeLayoutTransition:`) results in subnode insertions and removals which must be called lock-free.
* Make display node, layout spec, and style conform to NSLocking so that users/subclasses can access their locks
* Update the changelog
* Align slashes
* Put it back, when we're in ASDisplayNode
* Go a little further
* Put back the changes I didn't mean to commit
* Kick the CI
* Fix yoga build
* Put back non-locking change
* Address comments from Scott
* [ASDisplayNode layout] Fix an issue that causes a node's pending layout to not be applied
- Since the implementation of layout version (#428), if a node's pending and calculated layouts have the same current version as well as the same constrained size, the 2 layouts are considered equal and can be used interchangeably. A layout version check between the 2 layouts was added in #695. This PR adds a missing constrained size check.
- If the pending layout has the same version but a different constrained size compare to the calculated layout's, we can assume that the pending layout is newer and should be preferred over the calculated one. That is because layout operations always register their new layout as pending, which then (immediately or eventually) get applied to the node as calculated layout.
Before:
- Even if a pending layout was applied before, it'll be unnecessarily applied again in next layout passes and cause `-calculatedLayoutDidChange` being called multiple times.
After:
- If a pending layout was applied, the calculated layout will not be ignored but reused, if possible, in next layout passes.
Test plan: testSetNeedsLayoutAndNormalLayoutPass in #424.
* Be more aggressive at invalidating layouts during transitions, add a debug method, fix some build errors when verbose logging
* Add a changelog entry
* Invalidate the calculated layout transitioning using the same size range
- It's possible that -transitionLayoutWithSizeRange: can be called with the same size range used for previous layout passes. In that case, it's necessary to invalidate the current layout to avoid it being reused later on in the process.
* Fix CHANGELOG
* Fix infinite layout loop
- The problem will occur if a node does either of the followings:
1. Keeps using a stale pending layout over a calculated layout
2. Doesn't update the next layout's version after calling _setNeedsLayoutFromAbove.
* Update CHANGELOG
I believe this check is supposed to be here. Unit tests pass with it in place.
Without it, calling layoutThatFits: (as ASDataController does) and then calling
it again later (as ASCollectionNode does when answering the sizeForItem: call)
will recompute the layout, potentially on main.
This seems to have more of an impact / benefit for Yoga layouts, but I don't think
its benefit is exclusive to them.
This ensures that measure funcs are not set for container / empty spacer
nodes, because Yoga has more internal capabilities than layoutThatFits:
knows about.
Avoid need for the main layout pass to directly call setup for measure
funcs, by making it an implicit part of .yogaLayoutInProgress =.
Tear down the measure func after the layout pass to avoid retain cycle.
This has one important benefit: fixing the stretching behavior of spacer nodes.
In addition, it should help efficiency of Yoga and certainly minimize calls
to layoutThatFits:.
Next up for Yoga is a mostly-red diff, deleting the non-Contiguous code branches.
* [ASTraitCollection] Convert ASPrimitiveTraitCollection from lock to atomic.
This resolves a deadlock case: https://github.com/TextureGroup/Texture/issues/353
* [ASTraitCollection] Use assignment operator instead of .store() for C++ atomic.
Greg Bolsinga
Adlai Holler
Huy Nguyen
Michael Schneider
ricky
Tim Norman
Kevin
Michael Zuccarino
appleguy
Max Wang
Garrett Moon