On Fri, 18 Dec 2020 09:10:26 GMT, Alan Bateman <al...@openjdk.org> wrote:
> If there is a holder class for the MessageDigest then its initialisation can > fail, this would allow the orThrow method to go away As of allowing `Md5Digest` instead of explicitly throwing the exception from `orThrow` I think that the latter is more appropriate as in case of allowing the class initialization to fail all non-first calls will lead to `NoClassDefFoundError` which will be counterintuitive from users' perspective although the behaviour is simply unspecified for this in contract of `nameUUIDFromBytes` (Javadoc of `MessageDigest` requires the existence of `SHA-1` and `SHA-256` but nothing is said about `MD5` (should a bug-report be raised for the purpose of specifying this detail in the Javadoc?). > Are you planning to add a microbenchmark to demonstrate the issue? As for testing I am ready to write a benchmark but I am unsure what is the canonical way to write the one comparing the value before and after changeset. Or should it simply be a comparison of the very approach used here, not exactly the changed method (i.e. writing the benchmarks using the try/catch variant and holder variant and comparing these)? Could you please give any example of such if there are some. Thanks in advance! ------------- PR: https://git.openjdk.java.net/jdk/pull/1821
