That gets a single int, not an array as I said I wanted; that
    also doesn't actually do any file I/O as I also said I wanted.
    To be yet more specific, I need to do:

        Object readShortArray( RandomAccessFile raf, long size ) {
           byte[] bytes = new byte[ size ];
           raf.read( bytes );
           return (short[])bytes; // illegal, but what I want
        }

        void write( RandomAccessFile raf, Object obj ) {
           byte[] bytes = (byte[])obj;
           raf.write( bytes );
        }

        void callWrite() {
           short[] shorts = // ...
           write( raf, shorts );
        }

    Again, I want to read/write an entire array of int/short/long in
    a *single* I/O call.

    It would be nice if ByteBuffer had read( short[] ) but it
    doesn't.

    If this were C, it would be a no-brainer.

    - Paul

    I did prior to posting.

    I don't believe ByteBuffers will work.  A MappedByteBuffer, or
    more specifically mmap(2), is designed to be able to read large
    files, possible randomly, on an as-needed based, i.e., you fetch
    a byte: if it's not in memory, a page-fault occurs, and it's
    swapped in.  The file is transparently treated as though it *is*
    memory.

    For writing a disk cache, I don't think the mmap(2) model works
    too well.  When I decide I want to write an array to disk, I
    want it written immediately.  Once written, its in-memory
    counterpart is freed immediately.

    Using mmap(2), and therefore MappedByteBuffer, I'm at the mercy
    of the Java garbage collector to decide what and when to release
    from memory.  If memory gets tight (which is why I want a disk
    cache in the first place), there's no guarantee that the GC will
    free what I want it to free.  It might very well decide to free
    other, unrelated stuff in memory, e.g., other data structures
    using WeakReferences, and not the arrays I want it to free.

    - Paul

    The requested URL /applets.html was not found on this server.

    - Paul