Re: Matlab compatible legends for FLTK/Gnuplot backends?

[Ben Abbott]

On Sep 4, 2010, at 5:47 PM, David Bateman wrote:

> Ben Abbott wrote:
>> 
>> Implementing the legend the Matlab way with Gnuplot will get even messier, 
>> if we need to handle the outerposition property in the manner Matlab does. I 
>> recommend we continue to implement the legend for the Gnuplot backend using 
>> Gnuplot's key.
>> 
>> However, I see no need to keep the axes "key*" properties in place.
>> 
>> A Matlab compatible legend has additional properties ...
>> 
>>    'EdgeColor'
>>    'Interpreter'
>>    'Location'
>>    'Orientation'
>>    'String'
>>    'TextColor'
>> 
>> If we also include a list of handles for the objects listed in the legend, 
>> then it will be possible to parse the information in the front end (i.e. 
>> properties) and use Gnuplot's key to produce the desired legend.
>>  
> If we do this it means either one of two things has to be done that are quite 
> ugly
> 
> 1) In the gnuplot backend don't display the axes with a tag of "legend", but 
> rather try and recover the information here to recreate the legend with 
> gnuplot functions. This is ugly as essentially we are ignoring the text and 
> line values in the legend axis and trying to re-associate these to the 
> objects that they relate to.
> 
> 2) Put a gnuplot dependency in the frontend legend function that prevents the 
> legend axis being created in the first place. Its not good to make the 
> frontend dependent on the backend as then switching dynamically between 
> backends becomes problematic.
> 
> Another issue is that the I don't see how bug #30519 can be treated with the 
> gnuplot legend functions. There might be a way though as I haven't 
> investigated far.
> 
> D.

I agree. Some ugliness looks unavoidable.

I prefer 1) ... its the path of least resistance, and will please the Gnuplot 
fans.

Ben