Remove the ArrayDeque.subList().subList() test for comodification detection.
For types that extend AbstractList and to not override subList function,
the behavior varies depending on JDK version. With JKD8, the test fails for ArrayDeque.
While with JDK9+, it passes.
Test ArrayDeque with size 15. With the existing sizes the removeRange
implementation couldn't reach the maximum number of iterations - 3.
Benchmark results for ArrayDeque.subList().clear() that uses
removeRange() can be found here:
https://github.com/qurbonzoda/KotlinArrayDequeBenchmarks/tree/master/results
Provide an inefficient default implementation and make it "@PublishedApi internal" so it can be safely merged as is.
Intrinsics will be added later in follow-up PRs, and then the implementation will be adjusted accordingly
Merge-request: KT-MR-10843
Merged-by: Vsevolod Tolstopyatov <qwwdfsad@gmail.com>
Modification registration seems to changed in Java 9. Now if the asked
capacity in `ensureCapacity` function is less than current capacity,
no modification is registered.
If the found index is odd, retry search from index + 1.
Merge-request: KT-MR-10364
Merged-by: Abduqodiri Qurbonzoda <abduqodiri.qurbonzoda@jetbrains.com>
- keep offset in range (-1..+1) of time source units
- when adding big or infinite offset, saturate startedAt instead
- displace initial reading to zero, similar to MonotonicTimeSource
- fix the zero reading in TestTimeSource by calling markNow in constructor
- use more precise reading adjustment in TestTimeSource for big durations
- add a note about comparable time marks for AbstractLongTimeSource and TestTimeSource
Sequence::toSet and Sequence::toList both create a
collection, fill it with elements, and then try to optimize a result
by returning empty, singleton, or the allocated collection depending on
the elements count.
Instead of allocating the collection and then trying to return
an optimized instance it is worth checking the sequence's size
beforehand and return empty/singleton collection when possible.
Proposed change optimize performance of aforementioned functions and
also reduce allocation rate when a sequence consists of 0 or 1 elements.
^KT-55091 fixed
Improve the performance of reversed list view iterators by delegating to
underlying list iterators instead of using implementations provided by
the AbstractList.
Latter use AbstractList::get(index: Int) to implement next() and
previous() methods and depending on the underlying list's implementation
it may lead to worse performance.
The change also improves reversed list views test coverage.
Abduqodiri Qurbonzoda