I see jdk9 has a fast path for decoding zip file names when the charset is UTF-8 as we expect. I suspect we can do better - I expect most zip file metadata to be pure ASCII and for the JDK to support COMPACT_STRINGS, in which case we could even share the byte arrays if we use the String(byte[], coder) constructor. No need to convert the byte[] to char[] and then back to the identical byte[].

But I don't see any jigsaw-blessed way of calling that String constructor; current uses are all in java.lang. At least all of java.base should be able to make use of it somehow.