Re: clipping filled polygons

[Ethan A M. <merritt@u.washington.edu>]

On Friday 11 January 2008 05:47, Gerard Cats wrote:
> 
> I encountered a problem when I tried to fill a closed polygon which 
> extended beyond the xrange, yrange. This is conform the documentation, 
> which says:
> "Zoom of a filled curve drawn from a datafile may produce empty or 
> incorrect area because gnuplot is clipping points and lines, and not areas."
> I use Version 4.2 patchlevel 2.
> I guess that it is desirable to remove this "feature" from gnuplot.

I think you are correct that it would be worth re-visiting the generic
filled-area code in gnuplot.  I suspect that the existing code was written
in terms of clipping points and lines just because those clipping routines
were already implemented.  But yes, the code in graphics.c 
 plot_filledcurves()
 finish_filled_curve()
 fill_missing_corners()
and maybe also
 plot_betweencurves()

is quite tangled, and could probably be entirely replaced by a polygon-
clipping algorithm such as you suggest below.  As a side benefit it would
reduce the size of the code by ~500 lines.

That would be a nice project for someone.  


	Ethan

> A well-known  algorithm to clip a filled polygon is by Sutherland and 
> Hodgman. Some very ingenious implementations exist, e.g.
> http://www.codeguru.com/Cpp/misc/misc/graphics/article.php/c8965/, but I 
> wrote a simple one, in Perl, with which I was able to clip my polygons 
> myself, before sending them to gnuplot. The advantage of my code is that 
> it is simple and very fast, but its disadvantage is that it needs to 
> store both the input and the output polygons, so it needs more memory 
> than more sophisticated solutions. Yet, with current memory prices, this 
> should not really be a problem anymore. My implementation is 
> specifically to clip to rectangular areas.
> 
> I include my Perl program. Perhaps a gnuplot code guru is interested to 
> implement Sutherland/Hodgman clipping; if so, (s)he may find it useful 
> to have a look at my Perl program.
> I myself cannot write C (easily), but I think conversion of Perl to C 
> code will not be a great challenge. The real challenge lies in adjusting 
> the code to gnuplot data structures.
> 

-- 
Ethan A Merritt