RFR: 8193031: Collections.addAll is likely to perform worse than Collection.addAll
Сергей Цыпанов
github.com+10835776+stsypanov at openjdk.java.net
Thu Dec 17 10:19:02 UTC 2020
Hello, I feel like this was previously discussed in https://mail.openjdk.java.net/pipermail/core-libs-dev/ but since I cannot find original mail I post this here.
Currently `Collections.addAll()` is implemented and documented as:
/**
* ...
* The behavior of this convenience method is identical to that of
* {@code c.addAll(Arrays.asList(elements))}, but this method is likely
* to run significantly faster under most implementations.
*/
@SafeVarargs
public static <T> boolean addAll(Collection<? super T> c, T... elements) {
boolean result = false;
for (T element : elements)
result |= c.add(element);
return result;
}
But it practice the notation `this method is likely to run significantly faster under most implementations` is completely wrong. When I take this [benchmark](https://github.com/stsypanov/benchmarks/blob/master/benchmark-runners/src/main/java/com/luxoft/logeek/benchmark/collection/CollectionsAddAllVsAddAllBenchmark.java) and run it on JDK 14 I get the following results:
(collection) (size) Score Error Units
addAll ArrayList 10 37.9 ± 1.9 ns/op
addAll ArrayList 100 83.8 ± 3.4 ns/op
addAll ArrayList 1000 678.2 ± 23.0 ns/op
collectionsAddAll ArrayList 10 50.9 ± 1.1 ns/op
collectionsAddAll ArrayList 100 751.4 ± 47.4 ns/op
collectionsAddAll ArrayList 1000 8839.8 ± 710.7 ns/op
addAll HashSet 10 128.4 ± 5.9 ns/op
addAll HashSet 100 1864.2 ± 102.4 ns/op
addAll HashSet 1000 16615.5 ± 1202.6 ns/op
collectionsAddAll HashSet 10 172.8 ± 6.0 ns/op
collectionsAddAll HashSet 100 2355.8 ± 195.4 ns/op
collectionsAddAll HashSet 1000 20364.7 ± 1164.0 ns/op
addAll ArrayDeque 10 54.0 ± 0.4 ns/op
addAll ArrayDeque 100 319.7 ± 2.5 ns/op
addAll ArrayDeque 1000 3176.9 ± 22.2 ns/op
collectionsAddAll ArrayDeque 10 66.5 ± 1.4 ns/op
collectionsAddAll ArrayDeque 100 808.1 ± 55.9 ns/op
collectionsAddAll ArrayDeque 1000 5639.6 ± 240.9 ns/op
addAll CopyOnWriteArrayList 10 18.0 ± 0.7 ns/op
addAll CopyOnWriteArrayList 100 39.4 ± 1.7 ns/op
addAll CopyOnWriteArrayList 1000 371.1 ± 17.0 ns/op
collectionsAddAll CopyOnWriteArrayList 10 251.9 ± 18.4 ns/op
collectionsAddAll CopyOnWriteArrayList 100 3405.9 ± 304.8 ns/op
collectionsAddAll CopyOnWriteArrayList 1000 247496.8 ± 23502.3 ns/op
addAll ConcurrentLinkedDeque 10 81.4 ± 2.8 ns/op
addAll ConcurrentLinkedDeque 100 609.1 ± 26.4 ns/op
addAll ConcurrentLinkedDeque 1000 4494.5 ± 219.3 ns/op
collectionsAddAll ConcurrentLinkedDeque 10 189.8 ± 2.5 ns/op
collectionsAddAll ConcurrentLinkedDeque 100 1660.0 ± 62.0 ns/op
collectionsAddAll ConcurrentLinkedDeque 1000 17649.2 ± 300.9 ns/op
addAll:·gc.alloc.rate.norm ArrayList 10 160.0 ± 0.0 B/op
addAll:·gc.alloc.rate.norm ArrayList 100 880.0 ± 0.0 B/op
addAll:·gc.alloc.rate.norm ArrayList 1000 8080.3 ± 0.0 B/op
collectionsAddAll:·gc.alloc.rate.norm ArrayList 10 80.0 ± 0.0 B/op
collectionsAddAll:·gc.alloc.rate.norm ArrayList 100 1400.2 ± 0.0 B/op
collectionsAddAll:·gc.alloc.rate.norm ArrayList 1000 15025.6 ± 0.1 B/op
addAll:·gc.alloc.rate.norm HashSet 10 464.0 ± 0.0 B/op
addAll:·gc.alloc.rate.norm HashSet 100 5328.5 ± 0.0 B/op
addAll:·gc.alloc.rate.norm HashSet 1000 48516.7 ± 0.1 B/op
collectionsAddAll:·gc.alloc.rate.norm HashSet 10 464.0 ± 0.0 B/op
collectionsAddAll:·gc.alloc.rate.norm HashSet 100 5328.5 ± 0.0 B/op
collectionsAddAll:·gc.alloc.rate.norm HashSet 1000 48516.6 ± 0.1 B/op
addAll:·gc.alloc.rate.norm ArrayDeque 10 112.0 ± 0.0 B/op
addAll:·gc.alloc.rate.norm ArrayDeque 100 560.0 ± 0.0 B/op
addAll:·gc.alloc.rate.norm ArrayDeque 1000 4160.5 ± 0.0 B/op
collectionsAddAll:·gc.alloc.rate.norm ArrayDeque 10 112.0 ± 0.0 B/op
collectionsAddAll:·gc.alloc.rate.norm ArrayDeque 100 1048.1 ± 0.0 B/op
collectionsAddAll:·gc.alloc.rate.norm ArrayDeque 1000 14929.4 ± 0.0 B/op
addAll:·gc.alloc.rate.norm CopyOnWriteArrayList 10 88.0 ± 0.0 B/op
addAll:·gc.alloc.rate.norm CopyOnWriteArrayList 100 448.0 ± 0.0 B/op
addAll:·gc.alloc.rate.norm CopyOnWriteArrayList 1000 4048.1 ± 0.0 B/op
collectionsAddAll:·gc.alloc.rate.norm CopyOnWriteArrayList 10 456.0 ± 0.0 B/op
collectionsAddAll:·gc.alloc.rate.norm CopyOnWriteArrayList 100 22057.2 ± 0.0 B/op
collectionsAddAll:·gc.alloc.rate.norm CopyOnWriteArrayList 1000 2020150.3 ± 7.3 B/op
addAll:·gc.alloc.rate.norm ConcurrentLinkedDeque 10 312.0 ± 0.0 B/op
addAll:·gc.alloc.rate.norm ConcurrentLinkedDeque 100 2472.1 ± 0.0 B/op
addAll:·gc.alloc.rate.norm ConcurrentLinkedDeque 1000 24073.7 ± 0.1 B/op
collectionsAddAll:·gc.alloc.rate.norm ConcurrentLinkedDeque 10 288.0 ± 0.0 B/op
collectionsAddAll:·gc.alloc.rate.norm ConcurrentLinkedDeque 100 2448.3 ± 0.0 B/op
collectionsAddAll:·gc.alloc.rate.norm ConcurrentLinkedDeque 1000 24051.4 ± 0.3 B/op
There's never a case when `Collections.addAll` is fater - on the contrary `c.addAll(Arrays.asList())` always wins. Pay attention especially to dramatic difference for array-based collection.
So I propose to reimplement the method by simply delegating to `Arrays.asList` because the spec declares identical behaviour and to remove perfomance notation from JavaDoc.
-------------
Commit messages:
- Collections:addAll: instead of one-by-one add elements in bulk
Changes: https://git.openjdk.java.net/jdk/pull/1764/files
Webrev: https://webrevs.openjdk.java.net/?repo=jdk&pr=1764&range=00
Issue: https://bugs.openjdk.java.net/browse/JDK-8193031
Stats: 7 lines in 1 file changed: 1 ins; 4 del; 2 mod
Patch: https://git.openjdk.java.net/jdk/pull/1764.diff
Fetch: git fetch https://git.openjdk.java.net/jdk pull/1764/head:pull/1764
PR: https://git.openjdk.java.net/jdk/pull/1764
More information about the core-libs-dev
mailing list