I have never used any of the tar pakages for Java so the following is only an
educated guess.

I suspect that you'll have to supply the size of the tar file entry before you
start writing the data for that entry.

From the sound of the error message, the Apache code hasn't been told how long
the data is going to be and has defaulted to zero.  So when you call
TarOutputStream.write() it checks that you are not writing more data than it
has been told to expect, and -- finding that you are -- refuses to play.

In the tar file format the size of each file's data is embedded in the output
file /before/ the data itself.  So obviously the TarOutputStream needs to know
how much data you are going to supply in advance[*].  The ICE implementation of
TarGzipOuputStream has code to buffer-up the write()s in memory until the entry
is close()ed, at which time it updates the header, and then writes the real
data.  (I don't know why TarOutputStream doesn't do the same thing for
uncompressed tar files -- maybe I'm missing an option somewhere.)   Which is
presumably why your existing code worked with that package.

If the Apache equivalent of TarGzipOuputStream lacks that feature for some
reason, or if it doesn't work the same way, then maybe that's why you are
seeing this failure.

   -- chris

[*] Actually, it could just write the data as you supply it and then go back in
the output file and patch the size information, but the ICE implementation
doesn't try to do that -- and anyway it couldn't because you can't jump back
and forth in a gzip-compressed stream.