On Mon, 7 Mar 2022 15:11:50 GMT, Сергей Цыпанов <d...@openjdk.java.net> wrote:
> `List.of()` along with `Set.of()` create unmodifiable `List/Set` but with
> smaller footprint comparing to `Arrays.asList()` / `new HashSet()` when
> called with vararg of size 0, 1, 2.
>
> In general replacement of `Arrays.asList()` with `List.of()` is dubious as
> the latter is null-hostile, however in some cases we are sure that arguments
> are non-null. Into this PR I've included the following cases (in addition to
> those where the argument is proved to be non-null at compile-time):
> - `MethodHandles.longestParameterList()` never returns null
> - parameter types are never null
> - interfaces used for proxy construction and returned from
> `Class.getInterfaces()` are never null
> - exceptions types of method signature are never null
Calling `Arrays.asList` vs `List.of` on a switch on the array length seems like
an overkill for such a simple thing. Imo the safest changes are where we know
the input array length is going to be 1 or 2.
src/java.base/share/classes/java/nio/file/FileTreeIterator.java line 70:
> 68: throws IOException
> 69: {
> 70: this.walker = new FileTreeWalker(List.of(options), maxDepth);
Relates to https://bugs.openjdk.java.net/browse/JDK-8145048
src/java.base/share/classes/sun/reflect/annotation/AnnotationSupport.java line
79:
> 77: containerBeforeContainee(annotations,
> annoClass);
> 78:
> 79: result.addAll((indirectFirst ? 0 : 1), List.of(indirect));
This `indirect` is most likely to be of size > 2
-------------
PR: https://git.openjdk.java.net/jdk/pull/7729
