Writing a universal file sizeof() method.

Question

Hi,
I am writing an application that needs to compute the exact size of a file. I need for this function to work in Windows and UNIX.
One of my issues is that the length() method provided by the File class does not return the exact size of all files.
The essence of my problem is that I cannot properly compute the size of a directory or symbolic links in UNIX.
File.class() returns 0 for the size of directories regardless of OS. This appears to be correct on Windows (right-click -> properties on an empty directory in Windows). However this seems to be incorrect according to Linux.
When compared to the output of ls, the File.length() method will not return the proper length of directories (whether on a FAT or reiserfs file system) and symbolic links in UNIX. When the windows partition is mounted in Linux, the size of the directories on the windows partition are reported as being 4096 by file system utilities ls and du.
This contrasts the value of zero reported by the Microsoft 98 Windows Explorer. Moreover, the size of directories on reiserfs file systems seem to be unpredictable. (Whereas the size of a symbolic link is the length of the filename it is pointing to.)
Detailed examples of my problem can be found at the following thread: http://forum.java.sun.com/thread.jspa?threadID=574755
How can I compute a function that will be able to compute the size of any file under any operating system?
I would like to avoid his path:
    1. Determine if the class is being run in Windows or UNIX
    2. If running in UNIX,
           a. If a directory of symbolic link: execute : du -sb [file] | awk '{print $1}'
              or ls -al [file ] | awk ' { print  $5 }'
           b. Otherwise: File.length()
    3. If running Windows, simply class File.length() (?)By taking this route I need to perform an if() statement every time the size of any file is being computed, and also execute an expensive command for some special cases (which will common).
Thanks in advance for reading the entire article and thinking about potential solutions.

Answer 1

I am writing an application that will take an arbitrary number of directories, and split the data into the appropriately size chunks to be burned onto CDs.
The directories are passed as strings, and I am currently using Java classes to compute the size of the files. The computed size of the files must be accurate, as I am passing the segmented files into mkisofs (to create an image), and then burn that image onto a CD using cdrecord.
If the sizes are not computed properly, too many files may be passed to mkisofs (and the CD will be bad); or that the code is not entirely efficient (or correct).
As a "correct" value to use against validation, I use du on UNIX and the "properties" command on Windows.
My problem arises on UNIX where the File.length() method returns 0. My computed value is always less than the value given by du, since the values of directories are not properly computed because of the length() method.
This also makes the application difficult to debug. Specifially, verifying that all files passed in are actually all there. IE, are small directories missing or are there small files missing?