Re: Adding unwrap to gnuplot

[Jon G. <jo...@th...>]

> > s/x/$1/
> 
> Hmm. 
> "x" works for me.  Unlesss "set parametric" is active, in which case "t" works.
> But $1 is also correct.

Ah, yes, it works for me with CVS as well.
It wasn't working when I tried my patch on 4.6.

> > Personally, I think what makes most sense is to not deal with
> > plotting sampled functions as a special case.
> 
> Agreed.  I was originally worried that the unwrapping might apply only
> to the data inside a selected zoom box, in which case it would suffer
> the same shift of origin as you see for sampled functions.  But your
> code reevaluates the entire set of input data rather than just the 
> currently INRANGE data, so that problem is avoided.

Yes, I thought it best to do it that way precisely to avoid the data
jumping around too much. I suppose it might make the operation more
costly for large datasets, but considering unwrapping is a fairly
straightforward operation, I'm not too concerned.

> > One separate note about the patch is that the line:
> >     lasty = M_PI;
> > should possibly be changed to
> >     lasty = 0;
> 
> Origin at 0 seems more obvious in the absence of data.
> I have no idea which convention is more common for real world data sets.

I've changed it to 0 in the patch now.
Both conventions ([0,2π) and (-π,π]) are common in the real world.
Wikipedia also points this out when they discuss unwrapping with regard
to instantaneous phase[1]; they say "When phi(t), is constrained to
its principal value, either the interval (-π, π]  or  [0, 2π),  it is
called the wrapped phase."

I think sticking to 0 is marginally better because it will help move the
y-intercept closer to the x-axis, but that's about it. Choosing 0 also
means we don't have to justify choosing π.

> Could you write up brief section for the documentation/help file?

Done.
Full patch against CVS attached.