Compile gnuplot (3.2) with the standard linux flags.
(Tommy Frandsen's patches)
If you dont want to end up testing the error
handling in the 387 then you should set VERYLARGE
to something smaller than DBL_MAX. (I have 1.79e+308).

Also you may want to add the compilation flag
-Dfloat=double in CFLAGS under LINUX_FLAGS
(it seems gnuplot source assumes that all unices
have doubles and floats of the same size)
or some code will refuse to compile with optimization
as then GCC evaluates constant expressions at compile time.

Here's a piece of the Makefile

LINUX_FLAGS = \
           CFLAGS="$(COPTS) $(OPTIONS) -Dfloat=double -DHUGE=1.79e+308\
                -DMEMCPY -DGETCWD -DGAMMA=lgamma" \
           LIBS="-lm" X11FLAGS="-fwritable-strings" X11INCLUDES="" \
           X11LIBS="$(X11LIBS)" PLOTXFLAG="-DX11"
GNUPLOT_X11="$(GNUPLOT_X11)" \
           X11INSTALL=$(X11INSTALL) \
           TERMFLAGS="-Iterm -DLINUX -DX11"

krishna