On Fri, 4 Dec 2020 16:02:33 GMT, Roger Riggs <rri...@openjdk.org> wrote:
>> This rewrites the doc of ArraysSupport.newLength, adds detail to the >> exception message, and adds a test. In addition to some renaming and a bit >> of refactoring of the actual code, I also made two changes of substance to >> the code: >> >> 1. I fixed a problem with overflow checking. In the original code, if >> oldLength and prefGrowth were both very large (say, Integer.MAX_VALUE), this >> method could return a negative value. It turns out that writing tests helps >> find bugs! >> >> 2. Under the old policy, if oldLength and minGrowth required a length above >> SOFT_MAX_ARRAY_LENGTH but not above Integer.MAX_VALUE, this method would >> return Integer.MAX_VALUE. That doesn't make any sense, because attempting to >> allocate an array of that length will almost certainly cause the Hotspot to >> throw OOME because its implementation limit was exceeded. Instead, if the >> required length is in this range, this method returns that required length. >> >> Separately, I'll work on retrofitting various call sites around the JDK to >> use this method. > > Nice clean description of the algorithm. For what it's worth, the following code: var list = Collections.nCopies(Integer.MAX_VALUE - 3, ""); var list2 = new ArrayList<>(list); list2.addAll(list); results in `java.lang.NegativeArraySizeException: -8` where the -8 is returned by `ArraysSupport.newLength`. This is a demonstration of the problem with overflow checking that is fixed by this change. ------------- PR: https://git.openjdk.java.net/jdk/pull/1617