Changes to RandomAccessFile in jdk9

[Alan Bateman]

I've re-read the discussion on this issue from 2016, both in 
JDK-8168628, and the mails here.

Looking at it now, I don't think RanomAccessFile.setLength should have 
been changed. Instead, the changes should have been limited to 
FileChannel.map to force disk space for the desired mapped region to be 
allocated first (irrespective of whether the mapping causes the file to 
be extended or not). This will also fix other cases with FileChannel 
that were missed. I'll create a bug for this.

Reading Ismael's mail then it will have no impact on the "memory mapped 
index file" case. If a sparse index file is memory mapped with 
FileChannel.map then the disk space for the mapped region will be 
allocated. If the "log segment files" are sparse then they will continue 
to be so unless they are memory-mapped.

-Alan