[MS]: When getting the data for [Optimized compilation of tcl], I met a few problems when compiling tcl with Intel's icc6.0 compiler for linux [http://www.intel.com/software/products/global/eval.htm]. This is how I (mostly) solved them.

   * '''Difficulty #1:''' icc writes to stderr even on succesful compilations without warnings, gcc does not. This was tripping part of the configure script.
''Solution: write a driving script that intercepts the offending writing.''

   * '''Difficulty #2:''' the configure script was producing an unusable set of macros (my guess is that the substitute headers provided with icc should also include a substitute for <features.h>).
''Solution: use the "-D_GNU_SOURCE" flag''

   * '''Difficulty #3: NOT THERE!''' This was a side-effect of my clumsy shell programming below (quoting issue, corrected now).
----

Difficulties #1 and #2 were solved by calling the compiler through the following script (which can certainly be improved to not use temp files by some better bash programmer than myself):

  #! /bin/sh

  if [ $# != 0 ]
  then
  
   ##
   ## REPLACEMENT for the icc script in order to compile tcl
   ##
   ## Replacement was needed because (AFAIU, which is not much):
   ##   (1) icc writes the filename to stderr even when it compiles
   ##       succesfully and with no warnings
   ##   (2) the configure script was producing an unusable set of flags,
   ##       and defining the macro _GNU_SOURCE fixed that. The problem
   ##       seems to be in the 64bit support headers; is this a 
   ##       problem with icc or with tcl's configure script?
   ##
  
   eval /opt/intel/compiler60/ia32/bin/icc  -D_GNU_SOURCE '$@' > icc.out 2> icc.err;
   errCode=$?
   cat icc.out
   if (test $errCode -ne 0 || test ` cat icc.err | wc -l` -gt 1);
   then
    cat icc.err 1>&2
   fi
   exit $errCode
  
   ## END REPLACEMENT
  
  else
   exec -a "/opt/intel/compiler60/ia32/bin/icc" /opt/intel/compiler60/ia32/bin/iccbin;
  fi

<<categories>> Porting