Re: drawing graphs with dia

[Lars Clausen <lars raeder dk>]

On Sat, 2005-03-12 at 20:11 +0100, Grégoire Dooms wrote:
> Lars Clausen wrote:
>
> > First option is exactly what I had in mind for after 0.95.  We have code
> > that does line-gap well for almost all objects, but only with lines.  The
> > various straight lines would work well with it, but getting arcs and
> > beziers to do the right thing would be tricky.
> >  
> For arc and bezier, I'm not sure what the right thing to do is.
> I can suggest two things:
> - Connect the endpoint to the central connection point,
>   trim the curve at the edge of the connected object
>   and draw the arrow like if the endpoint was there
> - Preserve the slope the user set when the endpoint was connected to the
>   central CP,
>   connect the curve at the edge such that it aims at the central CP 
> (with the
>   same slope).

Let's not worry about the arc yet.

> For the line, the gap is perfect for my intended use, why did you 
> comment out the GUI ?

It was a rather clunky half-solution to a number of problems, and only
implemented for line.  I wanted something more complete.  I know,
perfectionist syndrome.  

> In a general setting with an auto-gap, what do you think about this:
> instead of absolute-gap and fractional_gap we could use gap_offset and 
> gap_proportion (liner factor) where this latter could apply to either 
> the line/curve length or the autogap length.
>
> To sum up:
> offset / prop / target
> .10 / 0 / self : constant gap of .10
> 0 / .1 / self : gap is 10% of length
> .10 / .1 / self : gap is 10% of length + .10
> 0 / 0 /auto : arrow touches edge
> .10 / 0 / auto : arrow is 10 away from edge
> 0 / -.5 / auto : arrow is halfway inside the object
>
> positive values always increase the gap, negative decrease the gap.
>
> > The main thing that's been holding me off is that I wanted to let all
> > objects have the virtual central CP.  Some objects have a dynamic number
> > of CPs, which makes it trickier.  We have some notes on this problem on
> > the TWiki.
> >
> >  
> Why does it need to be virtual ?  ie. Why not a real central CP and 
> multiple
> lines connected to it like with the ellipse object ?

The idea is to not have it be a normal CP, but a 'default' CP.  When a
line end is dragged onto the object, but not onto one of the line ends,
it connects to the virtual CP in the center, and the line ends at the
edge of the object.  This would make it possible to make many diagrams
without having to worry about CPs or having to rearrange connections.

Maybe I've been overengineering it in my plans, and it really should
just be that CP 0 for elements is special, and all the element objects
just need updating to place it in the middle.

> > I would love it if you started looking at this.  I'd want to keep it
> > entirely #ifdef'ed until 0.95 is out (unless you code well and fast), but
> > I'm very interested in seeing some action on that front.  I do feel,
> > though, that it would be unfair not to mention the Graphviz package, which
> > is made for making graphs.  Doesn't have the editing abilities of Dia, but
> > is really good at autolayout.
> >  
> Yeah I use it a lot too. The layout engines allow to output the graph in 
> xdot
> format (dot format with positions and parmaters for the curves). But 
> there is
> one awkward thing: if you modify the xdot file there is no way/option  
> to use
> the graphviz renderer to display the modified graph as it always 
> re-autolayout
> it (AFAIK).

Have you noticed that recent versions support Dia format?  Haven't tried
it out yet.

> > > PS: I tried to compile CVS HEAD on 2005-03-11 and got this error:
> > > widgets.c:30:27: dia-lib-icons.h: No such file or directory
> > > (I did cvs up -d to update my cvs tree) should I check-out from scratch ?
> > >    
> >
> > Is that from today?  There's something about how automake doesn't quite
> > force the creation of it, but if you 'make dia-list-icons.h' in lib it's
> > created.
> >  
> Works fine, I also had to autogen.sh in order to rebuild the configure 
> script.

Have to investigate a little further.  You're not the only one to notice
this.

-Lars

-- 
Lars Clausen <lars raeder dk>