Re: chickenpox: dia in LWN (again)

[Lars Clausen <lars raeder dk>]

On Fri, 2004-03-19 at 22:08, Cyrille Chepelov wrote:
> Le Fri, Mar 19, 2004, Ã  10:27:33PM +0100, Lars Clausen a écrit:
> 
> 
> > Chickenpox mode is ... interesting.  There's definitely scenarios where
> > it could be useful, but I think almost all needs would be covered by
> > three things:
> >   Our current connection points (for nice lining up)
> >   Virtual center connection point (for just pointing at the object)
> >   Dynamic connection points at the edges of objects (for spreading
> >     out zigzaglines pointing at an object)
> 
> This doesn't solve the problem Jonathan Corbet had on his kernel charts (the
> starts of the brown lines is NOT on the edge and still would need to be
> nicely arranged). You also need dynamic connection points on the body of
> objects (as an alternate to at the edges -- you need the ability to switch
> between current, virtual, edge and body. Which is basically what I described
> with chickenpox + what Philippe asked for, isn't it ? ;-))

More or less.  I just think that the virtual and edge CPs are much more
generally useful than the body CPs, and so should be looked at first. 
But then, they're also more complicated, aren't they?  I think the edge
connections are well enough served by ordinary connections for now, but
the virtual CPs have the big advantage of allowing extremely easy
diagram manipulation.  No more scrounging around for CPs, no more
adjusting the lines because you moved the object to where the line
crosses it.  And nice-looking, too.

> > I'm not sure I like the idea of explicitly erasing connectionpoints.  I
> > think dragging a line into an object might connect there (optionally
> > snapped to edge), but if you move the line away, the cp should just
> > disappear.
> 
> Makes a lot of sense. So, disconnect a dynamic CP, poof it goes. static CPs
> remain as they are.

Yes.  Dunno if it should really be a CP as such -- but it would ease the
saving, and also allow several lines to go to the same CP.

> > As for when this stuff should happen:  I really would like the virtual
> > center point before 1.0.  The chickenpox idea and variations I see as
> > somewhat more experimental, and would like to way till after 1.0.
> 
> OK. Will try to get some of the support code up as I get the time (one thing
> I'm pretty sure of -- ConnectionPoints need to be identified by ID, not by
> index like we do now, with all the contorsions in CPL; no matter how the UI
> looks like eventually, we'll need freedom in the way to access and refer to
> CPs). 

That is true.  But I suppose there's no problem in having the existing
CPs identified by, say, low integers:)

-Lars