Summary:
I was building a new `IGListBindingSectionController` subclass and accidentally used two view models with the same ID. Was seeing strange results and realized we're not removing dups or asserting here.
Adding a call to `objectsWithDuplicateIdentifiersRemoved` when the view models are first requested.
Reviewed By: rnystrom
Differential Revision: D8303601
fbshipit-source-id: 42c62adc401feaec2c7dce2a83cfc6533599752b
Summary:
If an `IGListAdapterUpdater` was unable to retrieve a collection view in `[IGListAdapterUpdater performBatchUpdatesWithCollectionViewBlock:]`, it would return early without cleaning state. This would sometimes cause future updates to crash, as this `IGListAdapterUpdater`'s state would now be out-of-sync.
This change also ensures that the `IGListAdapterUpdater`'s completion blocks run when `[IGListAdapterUpdater performBatchUpdatesWithCollectionViewBlock:]` and `[IGListAdapterUpdater performReloadDataWithCollectionViewBlock:]` return early due to having a nil `UICollectionView`.
Reviewed By: rnystrom
Differential Revision: D8056539
fbshipit-source-id: 1af7b675ec6805c2d8031f32d8a4c8e8929564e6
Summary:
I'd like to be able to access the current collection view scrolling traits inside section controllers. These are expressed as three properties from `UIScrollView`:
`isTracking`, `isDragging`, and `isDecelerating`.
My approach is to add a new struct `IGListCollectionScrollingTraits` with these three values, and expose this to section controllers through `IGListCollectionContext`.
Reviewed By: rnystrom
Differential Revision: D7986814
fbshipit-source-id: 19e9bd3b89545b10238dd060a5af8c5a0f39eb82
Summary:
Adding a fix to the `IGListAdapterUpdater` that requests the `UICollectionView` to perform updates on until just-before we update. This way if the `UICollectionView` is changed between update-queue and execution (b/c updates are async), we guarantee the update is performed on the correct view.
See the accompanying unit test that fails w/out the fix enabled.
Differential Revision: D7889908
fbshipit-source-id: 7178677f34951a1e42986b0289fc4abc708d6946
Summary:
Inspired by recent performance investigations in Instagram, adding an experimental feature to defer requesting objects from the `IGListAdapterDataSource` until //just before// diffing is executed. If //n// updates are coalesced into one, this results in just a single request for objects from the data source.
This is a **breaking change** to the public API, but since writing custom `IGListUpdatingDelegate`s is discouraged, I'm less worried about saving this for a major version launch.
Reviewed By: manicakes
Differential Revision: D7744518
fbshipit-source-id: f0001263cddda383e3dedd1d350a83d2cfb8362a
Summary:
Updating the pod and project version to 3.3.0 in preparation for a new release.
Closes https://github.com/Instagram/IGListKit/pull/1150
Differential Revision: D7691115
Pulled By: rnystrom
fbshipit-source-id: 145bc0930eb4a04809f0f4a0e7fc2a326fbd366c
Summary:
Issue fixed: #1111
As discussed in the issue's comments, this seems to be the most expedient fix if we accept that `cellForItemAtIndex:` should be allowed within update blocks. I am not familiar enough with `IGListAdapter` to know whether this is the best solution but am willing to make any changes requested!
- [x] All tests pass. Demo project builds and runs.
- [x] I added tests, an experiment, or detailed why my change isn't tested.
- [x] I added an entry to the `CHANGELOG.md` for any breaking changes, enhancements, or bug fixes.
- [x] I have reviewed the [contributing guide](https://github.com/Instagram/IGListKit/blob/master/.github/CONTRIBUTING.md)
Closes https://github.com/Instagram/IGListKit/pull/1144
Differential Revision: D7585125
Pulled By: rnystrom
fbshipit-source-id: 29a0e68da33bb4b94c2eb5e68844e537a7b05868
Summary:
Testing out new coalescence ideas after talking with @[100001606943892:tweilu]. Theory is that we can reduce stalls and improve scroll performance by batching more updates into one.
Need to test this out first.
Differential Revision: D7565003
fbshipit-source-id: b2e9fa39d52cc0b7aa59c2bfad709804ba0a0b63
Summary:
Issue fixed: #1117
I adding a new constructor for making a `IGListCollectionViewLayout` instance that can always show sticky header although section data is empty.
It's working well and [this is demo project](https://github.com/marcuswu0814/IGListKit_ShowStickyHeaderWhenDataEmpty).
Bug I'm not sure is any good way to let this much more readability. Is there any good advice?
```objc
const CGRect headerBounds = (self.scrollDirection == UICollectionViewScrollDirectionVertical) ?
CGRectMake(insets.left,
(itemCount == 0) ? CGRectGetMaxY(rollingSectionBounds) : CGRectGetMinY(rollingSectionBounds) - headerSize.height,
paddedLengthInFixedDirection,
headerSize.height) :
CGRectMake((itemCount == 0) ? CGRectGetMaxX(rollingSectionBounds) : CGRectGetMinX(rollingSectionBounds) - headerSize.width,
insets.top,
headerSize.width,
paddedLengthInFixedDirection);
```
- [x] All tests pass. Demo project builds and runs.
- [x] I added tests, an experiment, or detailed why my change isn't tested.
- [x] I added an entry to the `CHANGELOG.md` for any breaking changes, enhancements, or bug fixes.
- [x] I have reviewed the [contributing guide](https://github.com/Instagram/IGListKit/blob/master/.github/CONTRIBUTING.md)
Closes https://github.com/Instagram/IGListKit/pull/1129
Differential Revision: D7551628
Pulled By: rnystrom
fbshipit-source-id: a60b65a92efcea5175c86aaed1de02686ea6d20a
Summary:
Issue fixed: #1141
This is a set of minimum fixes for Xcode 9.3 rather than #1142. Much smaller diffs 😃
The changes here were done by Xcode's FixIt. See also https://developer.apple.com/library/content/documentation/Cocoa/Conceptual/Strings/Articles/formatSpecifiers.html#//apple_ref/doc/uid/TP40004265-SW5.
> Platform Dependencies
>
> OS X uses several data types—NSInteger, NSUInteger,CGFloat, and CFIndex—to provide a consistent means of representing values in 32- and 64-bit environments. In a 32-bit environment, NSInteger and NSUInteger are defined as int and unsigned int, respectively. In 64-bit environments, NSInteger and NSUInteger are defined as long and unsigned long, respectively. To avoid the need to use different printf-style type specifiers depending on the platform, you can use the specifiers shown in Table 3. Note that in some cases you may have to cast the value.
- [x] All tests pass. Demo project builds and runs.
- [x] I added tests, an experiment, or detailed why my change isn't tested.
- [x] I added an entry to the `CHANGELOG.md` for any breaking changes, enhancements, or bug fixes.
- [x] I have reviewed the [contributing guide](https://github.com/Instagram/IGListKit/blob/master/.github/CONTRIBUTING.md)
Closes https://github.com/Instagram/IGListKit/pull/1143
Differential Revision: D7551617
Pulled By: rnystrom
fbshipit-source-id: b27ebe2b1a1c93ebfd1218dc38b7621fefee2e6b
Summary:
…instead of just allocating a vanilla UICollectionViewLayoutAttributes. I'm happy to add tests here, but don't really have much of an opinion either way on whether this should have an explicit test case or not.
Issue fixed: #1134
- [x] All tests pass. Demo project builds and runs.
- [x] I added tests, an experiment, or detailed why my change isn't tested.
- [x] I added an entry to the `CHANGELOG.md` for any breaking changes, enhancements, or bug fixes.
- [x] I have reviewed the [contributing guide](https://github.com/Instagram/IGListKit/blob/master/.github/CONTRIBUTING.md)
Closes https://github.com/Instagram/IGListKit/pull/1135
Differential Revision: D7551624
Pulled By: rnystrom
fbshipit-source-id: 930218c40b0082912454bb98bd02e7daa8257a3e
Summary: Prefix all private methods with an underscore. A private method is defined as any method not exposed via an interface or protocol. This will improve the readability of the code and hopefully reduce bugs.
Reviewed By: rnystrom
Differential Revision: D7421346
fbshipit-source-id: 536472d57ea7fd8990fe50f3e950907ca57b8e6d
Summary: Added the ability to register/dequeue a cell with a customized reuse identifier. This will allow for registering the same cell class for multiple reuse identifiers, as will be needed for `IGListItemView` cells.
Reviewed By: rnystrom
Differential Revision: D7436575
fbshipit-source-id: d557836b5454c50611c5a5f9367f2f2e496c3878
Summary:
Issue fixed: #1105
- [x] All tests pass. Demo project builds and runs.
- [ ] ~~I added tests, an experiment, or detailed why my change isn't tested.~~
- [x] I added an entry to the `CHANGELOG.md` for any breaking changes, enhancements, or bug fixes.
Unable to add tests w/out hacking `UITouch` events on a collection view. Open to suggestions.
Closes https://github.com/Instagram/IGListKit/pull/1108
Differential Revision: D7362200
Pulled By: rnystrom
fbshipit-source-id: fed640c2c017f0ade0cefff0b0d2118564dda3b0
Summary:
if `stickyHeader` is true, the footer could be sticky to the top
Issue fixed: #1093
- [x] All tests pass. Demo project builds and runs.
- [x] I added tests, an experiment, or detailed why my change isn't tested.
- [x] I added an entry to the `CHANGELOG.md` for any breaking changes, enhancements, or bug fixes.
- [x] I have reviewed the [contributing guide](https://github.com/Instagram/IGListKit/blob/master/.github/CONTRIBUTING.md)
Closes https://github.com/Instagram/IGListKit/pull/1094
Differential Revision: D7246636
Pulled By: rnystrom
fbshipit-source-id: 9b7da52ab46229606b530a378c01d799767e27f2
Summary:
Copy objects when retrieving from datasource to account for the edge case where the returned data is a mutable array.
Issue fixed: #999
- [x] All tests pass. Demo project builds and runs.
- [x] I added tests, an experiment, or detailed why my change isn't tested.
- [x] I added an entry to the `CHANGELOG.md` for any breaking changes, enhancements, or bug fixes.
- [x] I have reviewed the [contributing guide](https://github.com/Instagram/IGListKit/blob/master/.github/CONTRIBUTING.md)
Closes https://github.com/Instagram/IGListKit/pull/1109
Differential Revision: D7246645
Pulled By: rnystrom
fbshipit-source-id: d9b0c2ba07983bf46327d4ee1ba0eba2e194ba19
Summary: Found a ~15% perf improvement on just insert/delete by removing copy. Saw traces from retaining the `NSMapTable` in profile.
Reviewed By: manicakes
Differential Revision: D6982523
fbshipit-source-id: 28c1539e06ecf3fe580bcccfecfc4915d4309714
Summary:
In an attempt to best other diffing libraries, I noticed that pure insert/delete diffing results would be almost 5x slower than changes between existing arrays. This is pretty much a benchmarking enhancement, but improves diffing performance by ~5x when the from-array or to-array is empty.
```
// OLD only inserts
avg: 0.007469, min: 0.006998, max: 0.016550, p50: 0.007254, p75: 0.007712, p90: 0.007899, p95: 0.008345, p99: 0.016550
// NEW
avg: 0.001392, min: 0.001256, max: 0.006772, p50: 0.001289, p75: 0.001348, p90: 0.001533, p95: 0.001614, p99: 0.006772
```
```
// OLD only deletes
avg: 0.005821, min: 0.005669, max: 0.006511, p50: 0.005766, p75: 0.005852, p90: 0.006030, p95: 0.006204, p99: 0.006511
// NEW
avg: 0.001184, min: 0.001096, max: 0.001673, p50: 0.001123, p75: 0.001212, p90: 0.001378, p95: 0.001467, p99: 0.001673
```
Note the average time improvements (seconds).
Benchmarking done on a 4s w/ this project:
https://pxl.cl/bLBB
Reviewed By: manicakes
Differential Revision: D6968683
fbshipit-source-id: 0d8e058f0aaa9ce756ca69326527d04504ac6429
Summary: Tiny, measurable perf gain by removing any retains on these params.
Reviewed By: manicakes
Differential Revision: D6968685
fbshipit-source-id: f0ef1ecac6367661d125ea85dc1c9989bf2e9659
Summary: Should also just be a c function instead of a block. Much simpler design.
Reviewed By: manicakes
Differential Revision: D6968682
fbshipit-source-id: f3bac6c5e0f93deec46449eb7abcb42f5e4e6071
Summary: This is much simpler as a static function instead of a block. Also taking an object vs plucking from the array will lend itself to better performance and reusability later.
Reviewed By: manicakes
Differential Revision: D6968684
fbshipit-source-id: 6166473feb20937ce6a1337b4b99a5a240778f56
Summary:
I had a desire for interactive reordering in a personal project, so here's a first attempt at adding support in IGListKit.
I figured I might as well get a WIP PR up for comments before I continue further as there are a few aspects to interactive reordering that don't interplay perfectly with IGListKit.
As discussed in #291, I went after two prime use cases:
1. Moving items amongst a section
2. Rearranging whole sections
I also "disabled" moving items between sections by having those moves revert, to mimic interactive reordering cancellation as closely as possible.
You can see both in the Mixed Data example. Grid items can be moved within a section, while users can be moved to reorder whole sections. But trying to move a grid item out of a grid or a user item into a grid will auto-revert. The revert animation isn't as tight as it should be. It may be more desirable to disable the animation - though you lose the visual cue.
There is a also a new example, `ReorderableViewController`, that demonstrates 2 in its pure form (likely the most desired use case), where all sections are reorderable single rows.
Happy to take feedback -- this is my first experience working on IGListKit, so I would expect there to be gaps. (Ex. I haven't used `IGListStackedSectionController`, and its tests failed as I hadn't implemented reordering delegates for it. Those are simply stubbed out for now.)
Issue fixed: #291
- [x] All tests pass. Demo project builds and runs.
- [x] I added tests, an experiment, or detailed why my change isn't tested.
- [x] I added an entry to the `CHANGELOG.md` for any breaking changes, enhancements, or bug fixes.
- [x] I have reviewed the [contributing guide](https://github.com/Instagram/IGListKit/blob/master/.github/CONTRIBUTING.md)
- [x] Proper support in `IGListStackedSectionController`
Closes https://github.com/Instagram/IGListKit/pull/976
Differential Revision: D6674493
Pulled By: rnystrom
fbshipit-source-id: cd53c5fdc6fb59636edc4747c4bbd0f81a4610e5
Summary:
The original DEBUG was put inside a for loop, which caused a quadratic looping while creating bigger and bigger NSSet.
This is very inefficient and I verified that by profiling with 10000 objects in the array and it caused significant amount of CPU. The main thread is pretty much unresponsive.
Closes https://github.com/Instagram/IGListKit/pull/1084
Reviewed By: rnystrom
Differential Revision: D6903043
Pulled By: lorixx
fbshipit-source-id: 311e8a402eb8d5574fce0eabd626a674b6a5e8c5
Summary:
project:
- fix file target membership issues in framework targets and test targets
- fix private/internal header imports, which shouldn't be `<IGListKit/` apparently
- fix static analyzer errors
travis:
- always install latest swiftlint
- ~~don't cache bundler, attempts to fix #1060~~
- remove markdown link check
swiftlint:
- make script non-failing if *any* version of swiftlint is installed
- warning if incorrect version is installed
- fail if not installed
- remove `scripts/generate_ci_yaml.rb`, we can just set the config file path directly
Closes https://github.com/Instagram/IGListKit/pull/1068
Differential Revision: D6885575
Pulled By: rnystrom
fbshipit-source-id: 51b7baa73feefcea71d870c1220d0382df484199
Summary:
This reverts commit 7ccf5a286ab4ee63bf72273142857496ce62b6f4
bypass-lint
An infra SEV is better than not reverting this diff.
If you copy this password, see you in SEV Review!
cause_a_sev_many_files
Differential Revision:
D6871289
Original commit changeset: 7ccf5a286ab4
fbshipit-source-id: e75cc7bd4a378e1092f4a01be3104fddb336759a
Summary:
Issue fixed: #1071
- [x] All tests pass. Demo project builds and runs.
- [x] I added tests, an experiment, or detailed why my change isn't tested.
- [x] I added an entry to the `CHANGELOG.md` for any breaking changes, enhancements, or bug fixes.
- [x] I have reviewed the [contributing guide](https://github.com/Instagram/IGListKit/blob/master/.github/CONTRIBUTING.md)
Closes https://github.com/Instagram/IGListKit/pull/1076
Differential Revision: D6871289
Pulled By: rnystrom
fbshipit-source-id: 7ccf5a286ab4ee63bf72273142857496ce62b6f4
Summary:
Adding more metadata about the state of the data source at the time of crash.
Note that this will be a breaking change to the public repo, but it's on a much-less used API, and since this is very high-pri for us at the moment, I'm willing to make the breakage.
Reviewed By: manicakes
Differential Revision: D6863683
fbshipit-source-id: e979aea445abc1ea556182cb69758703499e161a
Summary:
Issue fixed: #1072
Changes don't need new tests.
- [x] All tests pass. Demo project builds and runs.
- [x] I added tests, an experiment, or detailed why my change isn't tested.
- [x] I added an entry to the `CHANGELOG.md` for any breaking changes, enhancements, or bug fixes.
- [x] I have reviewed the [contributing guide](https://github.com/Instagram/IGListKit/blob/master/.github/CONTRIBUTING.md)
Closes https://github.com/Instagram/IGListKit/pull/1075
Differential Revision: D6858886
Pulled By: rnystrom
fbshipit-source-id: ec3816cc66aa2f10c21ccc7dea6078adc034e74e
Summary:
Optimize `SEL` search efficiency, reduced the time complexity from O (n) to O (1).
Not need tests.
- [x] All tests pass. Demo project builds and runs.
- [x] I added tests, an experiment, or detailed why my change isn't tested.
- [x] I added an entry to the `CHANGELOG.md` for any breaking changes, enhancements, or bug fixes.
- [x] I have reviewed the [contributing guide](https://github.com/Instagram/IGListKit/blob/master/.github/CONTRIBUTING.md)
Closes https://github.com/Instagram/IGListKit/pull/1055
Differential Revision: D6839942
Pulled By: rnystrom
fbshipit-source-id: 911755d6f8a4cd79b387423a51b6ea44cc7a2a07
Summary:
The current animation is the collection default animation - fade in/out: demo: https://pxl.cl/bj8R
Based on our design, we want the Shelf to be push down: demo https://pxl.cl/bj9R
1. Created **IGListCollectionViewDelegate** which inherits **UICollectionViewDelegateFlowLayout**
2. **IGListAdapter ** conforms **UICollectionViewDelegateFlowLayout**
3. add **transitionDelegate** to IGListSectionController
4. **IGFeedSectionController** sets transitionDelegate as itself and handles IGListCollectionViewDelegate methods
Reviewed By: rnystrom
Differential Revision: D6785726
fbshipit-source-id: bdf19f84fef05264ca0e082c6a326a31494a20da
Summary:
* Issue: we don't log the IGListKit updates
* Cause: IGListAdapterUpdater clears the ongoing updates before we call the delegate (https://fburl.com/76w5wvpl)
* Fix: keep a hold of the updates before calling executeCompletionBlocks(...)
Reviewed By: rnystrom
Differential Revision: D6750327
fbshipit-source-id: a30572873ab753b067977288cdb465a2a222e715
Summary:
* Clarify assert message by adding item width/height
* Before: Width of item 1 in section 2 must be less than container 375 accounting for section insets {0, 0, 0, 0}
* After: Width of item 1 in section 2 (415 pt) must be less than or equal to container (375 pt) accounting for section insets {0, 0, 0, 0}
Reviewed By: rnystrom
Differential Revision: D6711829
fbshipit-source-id: be99e83fd68cd345cb6a05acdde27c8252e59650
Summary: Looking at crash logs, the new high-firing crash has 2 item inserts in it. Test deduping the insert. Testing b/c I want to make sure there aren't any weird side effects.
Differential Revision: D6605474
fbshipit-source-id: 522120074aed2ed4995104443d48d8254ddb4fec
Summary:
Added `-didDeselectSectionController:withObject:` to `IGListSingleSectionControllerDelegate`.
Not sure if it makes sense to make it non-optional in 4.0.0 or not (#909) - I can add the [same note](0f04a07319/Source/IGListBindingSectionControllerSelectionDelegate.h (L40)) to the docs if it is
- [x] All tests pass. Demo project builds and runs.
- [x] I added tests, an experiment, or detailed why my change isn't tested.
- [x] I added an entry to the `CHANGELOG.md` for any breaking changes, enhancements, or bug fixes.
- [x] I have reviewed the [contributing guide](https://github.com/Instagram/IGListKit/blob/master/.github/CONTRIBUTING.md)
Closes https://github.com/Instagram/IGListKit/pull/954
Differential Revision: D6164711
Pulled By: rnystrom
fbshipit-source-id: aa8cfd2bb72a16cb525d875e2cad93888f13c641
Summary:
Assertions for item's size added to prevent negative values return from IGListAdapter
Issue fixed: #977
- [x] All tests pass. Demo project builds and runs.
- [x] I added tests, an experiment, or detailed why my change isn't tested.
- [ ] I added an entry to the `CHANGELOG.md` for any breaking changes, enhancements, or bug fixes.
- [x] I have reviewed the [contributing guide](https://github.com/Instagram/IGListKit/blob/master/.github/CONTRIBUTING.md)
Closes https://github.com/Instagram/IGListKit/pull/992
Reviewed By: ryanolsonk
Differential Revision: D6272307
Pulled By: rnystrom
fbshipit-source-id: 844affd914329e0ca04597ada2952f1a077897af
Summary:
adopt adjustedContentInset instead of contentInset on iOS 11
- [x] All tests pass. Demo project builds and runs.
- [x] I added tests, an experiment, or detailed why my change isn't tested.
- [x] I added an entry to the `CHANGELOG.md` for any breaking changes, enhancements, or bug fixes.
- [x] I have reviewed the [contributing guide](https://github.com/Instagram/IGListKit/blob/master/.github/CONTRIBUTING.md)
Closes https://github.com/Instagram/IGListKit/pull/1020
Reviewed By: manicakes
Differential Revision: D6513703
Pulled By: rnystrom
fbshipit-source-id: 5d718f2cb30361959ca5ba8238746427223cb5fd
Summary:
Update pod and examples so project builds when synced
Depends on D6545334
Reviewed By: rnystrom
Differential Revision: D6573821
fbshipit-source-id: b0d4246fa9c0d627ebcd34ac7abd5c8453964037
Summary: * IGListCollectionViewLayout can now keep track of the minimumInvalidatedSection given individual section updates.
Reviewed By: rnystrom
Differential Revision: D6545333
fbshipit-source-id: 64c9b5ff32339299b52059a4d990a42d9e6a1af5
Summary:
* Currently, we invalidate the entire layout whenever we make any updates, like inserting new rows at the bottom.
* This is one of the most common causes of frame drop on feed, so let's allow partial invalidation based on the minimum modified section.
* For example, if we delete section 10, move section 4, and insert section 12, we would re-calculate the layout starting at section 4.
* This gets us the majority of the performance gains and it's relatively simple. In the future, we can make further optimizations, like 1) index path level invalidation and 2) finding the smallest modified index path whose properties (ex: size) have actually changed.
Reviewed By: rnystrom
Differential Revision: D6510140
fbshipit-source-id: 6ff1766b400c5aa82abc29ae76ab96660c3bb106
Summary:
Followup to make sure that object type mismatches can't happen, even if identifiers collide (which is discouraged). Add assert when duplicates are detected.
Patched some unit tests while I'm in here.
Reviewed By: calimarkus
Differential Revision: D6439094
fbshipit-source-id: d669c01734e5ce9483e851051f548d9960b3087c
Summary:
It's hard to figure out what the problem is, if you don't know which viewModel causes the trouble at all.
Let's add the class name to the assertion message.
Differential Revision: D6419460
fbshipit-source-id: 58edebca7839871b6d48e463caad957d7920e129
Summary:
… source setup too
Issue fixed: #815
- [x] All tests pass. Demo project builds and runs.
- [x] I added tests, an experiment, or detailed why my change isn't tested.
- [x] I added an entry to the `CHANGELOG.md` for any breaking changes, enhancements, or bug fixes.
- [x] I have reviewed the [contributing guide](https://github.com/Instagram/IGListKit/blob/master/.github/CONTRIBUTING.md)
Closes https://github.com/Instagram/IGListKit/pull/993
Reviewed By: manicakes
Differential Revision: D6388270
Pulled By: rnystrom
fbshipit-source-id: e5e7e047bad5f21b81b562ebd586f7f5036325ff
