The following are known bugs or problems:

Version 5.1.4
 
  - On "tungsten.ncsa.uiuc.edu" or other Linux platform (such as 
    boomer.ou.edu), jobs may abort with a segmentation violation 
    when the user is writing restart files.  If so, add the line:

        limit stacksize unlimited

    to the ".cshrc" file.  This applies to csh/tcsh users.  Users 
    of other shells (such as bash) should consult the shell man 
    page for the proper technique.

Version 5.1.0  (Do NOT need since Version 5.1.4)

  - On some 64bit machine, such as Compaq AlphaServer ES40 at TASC,
    The following changes are required for EXT2ARPS and MCI2ARPS 
    to work.
    
    1. uncomment the following line in include/mc_area.h
       /*#define long int              /* On some 64bit platform */ */
    2. Change the declaration of grbunit in src/ext2arps/rdnmcgrb.f90
       to "INTEGER*8 :: grbunit".
     

Version 5.0.0

-  "Arps_mpi" goes unstable quickly when compile with "ifc"
   version 6.0.  Use 6.0.1 or version 7.0.

-  When brcopt = 4 the symmetry test result is not exactly symmetric.

-  On SGI Origin machine, when compiled using "-Ofast" (default)
   and "prcout = 1" in arps.input, the program will abort with
   a bus error.

-  Possible WARNINGs/comments when compiled on Linux using ifc
   1. "!DIR$" directives
   2. constant truncated --- precision too great
   3. "GO TO" statement
   4. some statement functions
   5. some character string truncations
   6. "EQUIVALENCE" statement
   7. some declarations like, "REAL*8 ...", "INTEGER*2 ...", etc.

-  When compiles using ifc on Intel processors, option "-opt 3" does
   not work for ext2arps and adas. So makearps reset the option to 
   "-O2" whenever the option specified is higer than "2".

-  Inconsistent results exist between MPI run and no MPI run for Open 
   (radiation) boundary condition and periodic boundary condition. 
   The tests were done on CAPS Origin 2000. However, the difference 
   will disappear when using "-opt 3" or lower level optimization 
   option.


Version 4.5.1

- ARPS GRIB history data are not transferable beteen little and big
  endian machines.
- ARPS MP version: there is a boundary condition bug in the ARPS model
  which is causing the message passing version of the arps to produce
  incorrect results.
