Edges in left-directed forks

[kopeckyf]

Dear Sašo,

I came across a bug in the edges library where the edge is not drawn correctly from the first parent to its children if grow=west is chosen. Everything is fine for grow=east:

\documentclass{article}

\usepackage[edges]{forest}

\begin{document}
	\begin{forest} for tree = {forked edges,grow=east}
		[Hello world
			[Bello World] 
			[Cello world]
		]
	\end{forest}\hfill
	\begin{forest} for tree = {forked edges,grow=west}
		[Hello world
		[Bello World] 
		[Cello world]
		]
	\end{forest}
\end{document}

[kopeckyf]

Inverting the sign for \forestoption{fork sep} in the edge path setting for the children appears to fix the issue:

\begin{forest} for tree = {forked edges,grow=west}
[Hello world
  [Bello World,edge path'={(!u.parent anchor) -- ++(-\forestoption{fork sep},0) |- (.child anchor)},] 
  [Cello world,edge path'={(!u.parent anchor) -- ++(-\forestoption{fork sep},0) |- (.child anchor)},]
]
\end{forest}

Usually, the forked edge' style would set:

edge path'={(!u.parent anchor) -- ++(\forestoption{fork sep},0) |- (.child anchor)},

(Alternatively, modify fork sep for the children or the entire tree:)

[Bello World, fork sep=-1ex]

[sasozivanovic]

Hi Felix!

Thanks for the bug report, and sorry for the late reply.

You're absolutely right, this is a bug. Actually, looking at the forked edge' code now, I don't know what I was thinking when I wrote it. :-)

Once one tries other growth directions, the problem becomes even worse. Here's a new definition of forked edge', which partially alleviates the problem.

\forestset{
  forked edge'/.style={
    edge=overlay,
    edge path'={%
      (!u.parent anchor) coordinate(forest@temp@pa)
      % the following coordinate forces the overlay
      ++(\forestoption{grow}:100pt) coordinate(forest@temp@grow)
      (.child anchor) coordinate(forest@temp@ca)
      ++(\forestoption{grow}:-\forestoption{fork sep}) coordinate(forest@temp@ca')
      (forest@temp@pa) -- ($ (forest@temp@pa)!(forest@temp@ca')!(forest@temp@grow) $)
      -- (forest@temp@ca') -- (forest@temp@ca)
    },
  },
}

This works for fortuitous combinations of grow and anchor. In particular, it works for a cardinal grow and the opposite anchor, for example like this:

\newcommand\test[1]{%
  \begin{forest}
    for tree={forked edges,delay={content+=\strut},#1}
    [\forestoption{grow}
      [Bello World] 
      [Cello World]
      [short]
      [long long long long long long long]
    ]
  \end{forest}%
}

\test{grow=west,anchor=east}

\test{grow=north,anchor=south}

Even this is ok:

\test{grow=45,anchor=west}

But it's easy to make all hell break loose (I marked the node anchors to make the issue clear):

\test{grow=-45, for children={tikz={\fill[red] (.anchor) circle[radius=2pt];}}}

The real source of the problem is that each edge is drawn separately, in the context of the child towards the parent, whereas in cases such as forked edges it would be way better to have the parent draw the edges towards all the children, because then the single forking point could be computed. I'll think about how to implement this in some future release, this is not the only example where this design makes trouble.

[kopeckyf]

Thank you for explaining the issue, and for looking into it!