GNOME Bugzilla – Bug 705034
Area plots are broken
Last modified: 2013-08-05 11:54:05 UTC
Created attachment 250311 [details] How things look like in curent git. I broke things when I added interpolation support. Actually things are broken when some series are shorter than others in stacked area plots. Missing values before the last one are replaced with 0, but we don't add 0 at the end to get the same length for all series. I'll explain later why padding with zeros is not so good.
Created attachment 250313 [details] How it looks in 1.10.17
Created attachment 250314 [details] How it looked in 1.8.4 when interpolation had not yet been implemented.
Created attachment 250316 [details] How it looks in LO when saved as .ods (strict conformance) Note that we don't use the same series order (I think we use the same as xl). Also we have a style issue. When saved to .xls, the order and style are good. Andreas, should I file a bug report against the ODF plugin? Anyway, looks like LO interprets the missing values as 0. I can't test with excel, but if somebody can, I'll provide the .xls file.
Created attachment 250317 [details] How it would look with padding 0 and splines If we add the missing zeros, we get something like this. I suppose we might add just one zero (or none, as we used to). Currently, things are broken because we fill the series with it's path and the path of the previous series. We should use only part of this last path along with parts of previous paths, and this needs some involved API additions. The easy way to fix is to add as many 0 as needed, but things might look weird as seen is this sample.
Jean, if you find anything that may be a problem in the ODS export, please file a bug report, but remember that if an 'automatic' setting is used, for example for the fill colour, then LO may automatically choose different colours, different sizes etc. That behaviour is expected, so you will normally want to test with specified colours.
Regarding the images. I really think that the 1.8.4 behaviour is usually the desired behaviour.
I note that currently things look even weirder for stacked areas if the first series has less values.
The issue with the 1.8.4 behavior is that if missing data are not at series end, they are interpreted as zero. Either we interpret missing data as zeros (may be using previous series interpolation if we have several consecutive missing data), or we skip all missing data, or we provide an option to choose between these, but we should not just replace some and not others.
I think I might be confused: I think that interpolation happens between (inter) data points. The examples all consider a situation beyond (extra) the last data point. Moreover, I don't think there is any common 'interpolation' method that insert's 0s. If we have 1, missing, 4 and linear interpolation I would expect 'missing' to become 2.5 but not 0.
For line plots, we might interpolate missing data. For now they are just ignored. But for area plots, they can't be interpolated, they must be zero, especially for stacked plots.
In stacked area plots, lines have only half of their width visible, except the top most series. LOCalc does the same weird thing. I'd prefer to show the whole line.
This problem has been fixed in our software repository. The fix will go into the next software release. Thank you for your bug report. We are back to 0.6 behavior. Showing the whole line width is just not possible.