Probable df_no_use_specs bug

[Daniel J S. <dan...@ie...>]

In attempting to make both plot3d.c use only df_readline(), even for 
matrix data, I think I've uncovered a bug, or perhaps more than one.

One of the plots following a use of PM3D with four user specs was 
failing.  Here is some code to illustrate the problem.

Launch gnuplot and type

splot 'binary2' binary u 1:2:3:($2+($1+$2)/10) w pm3d
splot sin(x) w l

The first plot shows a blank 3d plot because the palette isn't set. 
 However, the second plot should still work fine because it doesn't 
require the palette.  But it fails because inside of plot3d.c is a test

#ifdef PM3D
        if (df_no_use_specs==4 &&
        !(this_plot->lp_properties.use_palette ||
          pm3d.where[0] || this_plot->pm3d_where[0] || 
this_plot->plot_style & PM3DSURFACE))
        int_error(c_token, "`using` with 4 parameters but surface 
without `pm3d` or `palette`");
#endif

Notice that the first test is whether dn_no_use_specs==4, which it does. 
 The use specs were set by the 'binary2' command to be 4 inside 
df_open().  However, for the second plot, df_no_use_specs is still equal 
to 4 because df_open() was not called because a file is not being 
plotted in the second example.  [This is an argument for not using 
global variables in such a way... but I'll not press that issue.]

OK.  I think there are multiple bugs there because try the following:

set pm3d at s explicit solid
splot 'binary2' binary u 1:2:3:($2+($1+$2)/10) w pm3d
splot sin(x) w l

In this case the second plot works.  The above test fails because 
this_plot->lp_properties.use_palette is TRUE.  But notice, the second 
plot is a line plot.  There shouldn't be a color box for such a plot.

These problems aren't connected to "binary".  Any file read with 4 user 
specs would cause the same problems.

If someone else wants to tackle that one, fine.  I'm just going to fix 
it in the "image/binary" stuff so that it works for a single 
"df_readline()".  I should be done with this patch today some time.  It 
looks like I'm headed in the direction of weeding out the 
"df_3dmatrix()" routine and instead just using the df_readline() routine 
[which internally uses df_readascii() or df_readbinary()].  That then 
would be consistent usage throughout, and as far as I'd like to take the 
binary stuff.

I'll let people know when the patch is ready.

Dan