If I recall correctly, fill creates patches; I am not sure about that.
If a patch object has any point which is 0 or negative on an axis, then when that axis is changed to log scale, the internal patch render algorithms give up trying to color it. This is the case even if the point is not within the current axis limits.
The work-around is not to use a non-negative coordinate -- even if it is eps() or even 1E-100, any non-negative value should do.
You might perhaps also be able to get around the problem by setting the figure rendering property to 'painters' or 'zbuffer'; I only routinely use OpenGL so I am not certain if the problem applies to the other renders.
