Created attachment 61321 [details]
screen shot

That is definitely unexpected.  Please tack on a sample file just in case there are some magic settings somewhere.

Created attachment 61324 [details]
gnumeric sample file

Hmm, if y==NAN, we use y=0 and if x==NAN, we use the index.
if (!gog_axis_map_finite (y_map, y))
	y = 0; /* excel is just sooooo consistent */
if (!gog_axis_map_finite (x_map, x))
	x = i;
So, the result is logical if not wanted. We clearly must use something more consistent.

"if y==NAN, we use y=0" is not unexpected and some users may in fact want this. I would prefer if the user had a choice between this behaviour and excluding that point from the graph.

"if x==NAN, we use the index" I can't imagine any user really wanting this unless no x value range is given at all. It would be less surprising to use 0 for missing values but again it would be nice if we had a choice.

So, personally, using index (rather than 0) is a bug. Having a choice is an enhancement request. 

Hide points, with 0 as an option seems the best choice. Anyway, when drawing lines these points are excluded.

Created attachment 61489 [details] [review]
proposed patch

I think you should use glade for user interface. We'll probably add other settings to it.

Also, could you adjust widget layout in order to fit existing dialogs.

Created attachment 61600 [details] [review]
updated patch

The same, but using libglade

Looks ok for me, except for the fact GladeXML is not freed in 
gog_xy_series_populate_editor.

Created attachment 62292 [details] [review]
new updated patch

Emmanuel, you are right, this patch should solve this problem.

Can we change the name to invalid-as-zero, please?  Other than that, go
ahead and commit.

Created attachment 62776 [details] [review]
the same with invalid-as-zero

commited. fixed in HEAD