[daip] Fwd: xterm?

[Eric Greisen]

Howard Powell writes:
 > Thanks for the reply Eric.
 > 
 > Friday after I emailed you I installed the binary version of 31DEC06,  
 > which of course had the same problems with the disappearing xterms.   
 > Following your suggestion below, I added -hold (there was no  
 > commented section, but I figured it out).  I see now that TEKSRV.EXE  
 > and MSGSRV.EXE are both looking for the libsvml.so shared library  
 > from the intel compilers.

  Both UNIXSERVERS and XASERVERS have the lines
#                                                  debug
#   AIPS_TEK_NAME_FLAG="$AIPS_TEK_NAME_FLAG -hold"

#                                                  debug
#   AIPS_MSG_NAME_FLAG="$AIPS_MSG_NAME_FLAG -hold"

The command issued to be executed in the xterm should expand the
$LD_LIBRARY_PATH before sending the xterm command and then the xterm
should set the LD_LIBRARY_PATH to the expanded string before
attempting to execute the load module.  If it is not doing that then
there is a bug in the xterm itself or in the OS that starts the xterm.

 > 
 > Our own version of the intel compilers are installed in /astro/intel/ 
 > compiler91/{lib, bin} rather than /opt/intel.  To get AIPS to work, I  
 > modified LOGIN.SH and LOGIN.CSH to include an LD_LIBRARY_PATH=/astro/ 
 > intel/compiler91/lib.  This path information isn't getting exported  
 > to the xterms when they fork.

You do not need and probably should not include your Intel libraries
(they may - will - not match versions).  The user should include in
their .csrrc, .bashrc, .tcshrc (etc) files references to the LOGIN.CSH
or LOGIN.SH which are executed even on non-interactive logins.  This
was the only way to make this work before we made the compound
statement we now ship.

 > 
 > Actually, the full $LD_LIBRARY_PATH after sourcing LOGIN.SH gets set to:
 > /astro/intel/compiler91/lib:/astro/intel/compiler91/lib:/astro/AIPS/ 
 > 31DEC06/LINUX/LIBR/INTELCMP:/astro/AIPS/31DEC06/LINUX/LIBR/INTELCMP
 > 
 > I see that libsvml.so is in the AIPS tree.
 > 
 > Any ideas why $LD_LIBRARY_PATH isn't getting exported properly to the  
 > subprocess?
 > 

Eric Greisen