If compress::decompress returns NULL, we need to clean up the allocated 
specialized compress_* objects -- but NOT delete the passed-in 
io_stream*.  The problem occurs Installer::installOne, but is corrected 
by modifying the behavior of compress::decompress and the various 
specialized decompression classes.

Detailed explanation:

 From Installer::installOne:

   if ((try_decompress = compress::decompress (pkgfile)) != NULL)
     {
       // if we get here, then try_decompress has taken ownership
       // of pkgfile

       if ((tarstream = archive::extract (try_decompress)) == NULL)
         {
           // error handling stuff
           delete try_decompress;
           // since try_decompress owns pkgfile, this cleans
           // up both try_decompress and pkgfile.
           return;
         }

        // this is success...tarstream now owns try_decompress,
        // which in turn owns pkgfile.  Eventually, tarstream
        // will be cleaned up along with both try_decompress and
        // pkgfile.
     }
   // however, when the IF statement above fails (returns NULL), then
   // without my patch the internal decompress object will NOT be
   // deleted before compress::decompress returns NULL. So,
   // compress::decompress() needs to delete that object, but since
   // we need pkgfile to still be valid in order to do the test below,
   // we have to tell the internal decompress object to relinquish
   // ownership of pkgfile, and NOT delete it during its own destructor.
   // That's why the code here in Installer::installOne is not changed,
   // but rather the fix is contained wholly in compress::decompress
   // and the specialized decompression classes.
   else if ((tarstream = archive::extract (pkgfile)) == NULL)
     {
       /* Not a compressed tarball, not a plain tarball, give up.  */
       delete pkgfile;
       ...


--
Chuck