The PLplot development team have just implemented a demanding 2D fill
rendering test for the X stack where we modify our standard example 27
(see http://plplot.sourceforge.net/examples.php?demo=27) to change
from drawing the boundary line of "spirographic" (e.g., hypotrochoid
and epitrochoid) curves to filling those curves.  The boundaries
of the 9 test curves are illustrated in the first attachment.

The fills of those test curves (implemented using XFillPolygon with a
Complex shape and 20000 points in the polygon describing the boundary)
have rendering errors.  

If the fill is done with the EvenOddRule filling rule (set using
XSetFillRule) the rendering errors are severe. The first through 3rd
test cases show blank results (not illustrated) and the 4th test case
(second attachment) and remaining more complex fill cases (not
illustrated) show severe rendering errors with many fills missing and
some areas filled that should not be.  

If the fill is done with the WindingRule filling rule, the fill
rendering errors are much less severe than in the case of the
EvenOddRule filling rule. The first (attached), second (not
illustrated) , and 3rd (not illustrated) test cases look fine, but the
fourth (attached) and remaining (not illustrated) test cases have
minor rendering errors (areas filled which should not be on the
right-hand side of the figure so the total result is asymmetric rather
than symmetric as it should be for the symmetric hypotrochoid and
epitrochoid boundary curves of the fill.

Because most applications and libraries (including the PLplot one)
have great success with X filling with a small (3 or 4) number of
points in the boundary for the normal default EvenOddRule filling
rule, I tried the experiment of reducing the number of points from
20000 to 200, but that made no qualitative difference to the severely
bad fill rendering for the EvenOddRule filling rule.

My platform is Debian Squeeze with Intel (g33) video chipset.  I first reported this bug at http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=615491. My original report was for the intel driver, but at Michel Dänzer's suggestion, I also tried the fbdev and vesa drivers which also showed the identical issue.

The following steps (which need just a few minutes) are required to replicate this bug:

* Get latest svn trunk version of PLplot source code following
the directions (including appending '/trunk' to the HTTPS URL)
at http://sourceforge.net/scm/?type=svn&group_id=2915.  This
download should only take a few minutes.

* Optionally modify drivers/xwin.c to change XSetFillRule call from
WindingRule to EvenOddRule (if you want to see the horrendous results
from that filling rule as opposed to the mildly bad results from
WindingRule).

* Optionally modify examples/c/x27c.c (say to change the number of
points in the boundary from 20000 to 200 or 2000000).

* Build this particular PLplot x27c test and the xwin driver in an
initially empty build tree. This should only take a minute or so.

mkdir build_dir
cd build_dir
cmake -DBUILD_TEST=ON /PATH/TO/TOP/OF/SOURCE/TREE
make xwin x27c

* Run example to show fill problems for self-intersecting boundaries.
examples/c/x27c -dev xwin

You advance from one page of that test example to the next by hitting
the return/enter key.  The first set of pages should display just the
self-intersecting boundaries, and the last set of pages should display
the filled results for those boundaries.  For the EvenOddRule fill
rule, the first three "fill" pages are blank and the remaining pages
badly filled.  For the WindingRule fill rule, the first three "fill"
pages are fine, but the remaining ones have small fill errors (showing
as fill asymmetries on the right of the figure).

Alan W. Irwin