fixes issue #781

[ktsaou]

As described in #781, the library makes wrong decisions when the y-range is zero.
In such cases (y0 === y1), this PR changes:

y0 -= 0.5
y1 += 0.5

so that the chart is properly rendered.

[coveralls]

Coverage decreased (-0.1%) to 90.096% when pulling b5b0d36 on ktsaou:master into 6611837 on danvk:master.

[danvk]

Thanks for the PR! This will be good to merge if you make the requested change for log scales and add a unit test.

[danvk]

For a log scale, this could drop y below zero, which will produce NaNs. In the logscale branch of the conditional on line 2546, you can halve y0 and double y1 instead.

[ktsaou]

There are 3 cases where y0 === y1:

1. they are both zero
2. they are both positive
3. they are both negative

I think this will properly handle all cases, for both normal and logscale rendering:

      if (y0 === y1) {
      	if(y0 === 0) {
      	  y1 = 1;
      	} else {
      	  var delta = Math.abs(y0 / 2);
          y0 -= delta;
          y1 += delta;
      	}
      }

So, only when it is zero, set y1 = 1 (from the charts I render, I see this must be a default for other cases too - for example when stacked charts have only zeros, the library already visualizes 0 to 1).

Otherwise provide a range of -1/2 to +1/2 of what already the number is. So, do not change the sign. If it is negative keep it negative, if it is positive keep it positive.

Do you agree?

Of course I agree there should be a unit test for this part of the code - its absence is what caused the bug in the first place. However, I need some guidance - I don't have a clue how your unit testing works, or how to configure one.