After JDK-8333843, the javadoc for MemorySegment#getString suggests the following pattern for reading strings of known length:

    byte[] bytes = new byte[length];
    MemorySegment.copy(segment, JAVA_BYTE, offset, bytes, 0, length);
    return new String(bytes, charset);

However this is less efficient than what the implementation of getString does, after JDK-8362893 it uses JavaLangAccess::uncheckedNewStringNoRepl to avoid the copy.

It is possible that in the future C2 will be able to optimize away the copy for the known length pattern (see JDK-8364418), but it can't do that for current JDK versions, and it sounds like there are some challenges to doing that optimization.

Given all that, I think it makes sense to reconsider the possibility of adding an overload of MemorySegment#getString to read strings with known lengths, which would offer the same performance as the null-terminated getString, and avoid the copy of the byte[].

Additionally, the setString implementation uses optimizations in StringSupport for the case where the underlying bytes are compatible, but only supports null-terminated strings. It would be nice to offer an overload of setString as well for strings of known length.