[Tex/LaTex] Package forest: use of forked edge and forked edges

Question

In my answer to question improving-tree-diagram i wrongly use forked edges:

\documentclass[border=3mm]{standalone}
\usepackage[edges]{forest}

\begin{document}
    \begin{forest}
    for tree={
      font=\footnotesize,
      draw, semithick, rounded corners,
            align = center,
        inner sep = 3mm,
             edge = {draw, semithick, -stealth},
             grow = east,
    forked edges, % <-- wrong, it should not be inside scope of "for tree"
            l sep = 12mm% level distance
                }
    [Multimedia\\ RSs
      [Multimedia\\ Content]
      [Multimedia\\ Content]
    ]
  \end{forest}
\end{document}

which gives:

However, with correct forked edge instead of forked edges the result is unacceptable:

It is corrected to first one, if I added to for tree options:

parent anchor = east,
child anchor = west,

or if I use forked edges outside of tree, for example:

  ...
 \begin{forest}
     forked edges,
   for tree={
      font=\footnotesize,
      draw, semithick, rounded corners,
            align = center,
        inner sep = 3mm,
             edge = {draw, semithick, -stealth},
             grow = east,
            l sep = 12mm% level distance
                }
 ...

I wonder, why the forked edge and forked edges gives different result?

One more question: why the point of the branch of edges is not on the middle of level distance if it is defined, for example l sep=12mm regardless if parents/child anchors are defined?

I read forest documentation, but didn't find anything usable about mentioned (maybe I need read it again and again … and also look into code) .

Answer 1

Here are the definitions for forked edge and forked edges from forest-lib-edges.sty.

\forestset{
  declare dimen={fork sep}{0.5em},
  forked edge/.style={
    edge={rotate/.pgfmath=grow()},
    edge path'={(!u.parent anchor) -- ++(\forestoption{fork sep},0) |- (.child anchor)},
  },
  forked edges/.style={
    for tree={parent anchor=children},
    for descendants={child anchor=parent,forked edge}
  },
}

So we see that forked edges not only applies forked edge to all descendants, but adjusts the parent anchor for the tree and the child anchor for the descendants as well.

I don't think that forked edge is correct here. It is true that you don't want forked edges because then you'll have

for tree={
  for tree={
    ...

which can cause problems.

However, if you use \forked edge without setting the parent/child anchors, then Forest defaults to using the (<node>) itself and TikZ then selects a border anchor in its usual way. However, the initial part of the forked path is drawn relative to the starting point. The starting point is also (<node>) but this time TikZ is measuring relative to rather than tracing a path from. So rather than using the node's border anchor, it uses the node's default anchor, which is .center by default.

So although the path starts from the appropriate border anchor, its initial segment takes it towards a location measured relative to the .center anchor and not measured relative to the border anchor.

Here's a visualisation:

The path starts from (a.east), which is right, but the point ++(.5em,0) from (a) is measured from (a.center). This is shown in red. The first part of the path, therefore, is the part shown in blue which is not .5em, of course. The remainder of the path is then in green.

We can see the same thing in the default growth direction where the edge takes rotate=-90.

I don't understand your second question. Why should l sep affect the branching point? I think you are asking why you don't get something like the following.

This can be done by adjusting fork sep.

\begin{forest}
  forked edges,
  for tree={
    font=\footnotesize,
    draw,
    semithick,
    rounded corners,
    align = center,
    inner sep = 3mm,
    edge = {semithick, -stealth},
    grow = east,
    l sep = 12mm,
    fork sep=6mm,
  }
  [Multimedia\\ RSs
    [Multimedia\\ Content]
    [Multimedia\\ Content]
  ]
\end{forest}

But note also, that l sep is a minimum distance between the closest border point of the parent and the closest border point of each child. That distance may end up being larger than the specified dimension. Forest just stops it being any smaller. (Unless it is altered later, of course, either for the whole tree or more locally.)

So you can't guarantee the midpoint in this way. To do that, you would need to alter fork sep later, as the tree is packed because you'd need, I think, to change it for each level before the sub-trees of the nodes on that level are packed. (Or you'd have to repack afterwards.) I'm not certain of this, however - it might be sufficient to do it after the entire tree is packed, at least in many cases. (Of course, you'd want to do it before the x and y values are computed.)

\documentclass[border=10pt,multi,tikz]{standalone}
\usepackage[edges]{forest}
\begin{document}

\begin{tikzpicture}
  \node [draw] (a) at (0,0) {a};
  \node [draw] (b) at (1,.25) {b};
  \node [draw] (c) at (1,-.25) {c};
  \foreach \i in {b,c}
  {
    \draw [rotate=0] (a) -- ++(.5em,0) |- (\i);
    \draw [rotate=0, green] (a) ++(.5em,0) |- (\i);
    \draw [rotate=0, blue] (a) -- ++(.5em,0);
  }
  \draw [red] (a.center) -- ++(.5em,0);
\end{tikzpicture}
\begin{tikzpicture}
  \node [draw] (a) at (0,0) {a};
  \node [draw] (b) at (-.25,-1) {b};
  \node [draw] (c) at (.25,-1) {c};
  \foreach \i in {b,c}
  {
    \draw [rotate=-90] (a) -- ++(.5em,0) |- (\i);
    \draw [rotate=-90, green] (a) ++(.5em,0) |- (\i);
    \draw [rotate=-90, blue] (a) -- ++(.5em,0);
  }
  \draw [rotate=-90, red] (a.center) -- ++(.5em,0);
\end{tikzpicture}

\begin{forest}
  forked edges,
  for tree={
    font=\footnotesize,
    draw,
    semithick,
    rounded corners,
    align = center,
    inner sep = 3mm,
    edge = {semithick, -stealth},
    grow = east,
    l sep = 12mm,
    fork sep=6mm,
  }
  [Multimedia\\ RSs
    [Multimedia\\ Content]
    [Multimedia\\ Content]
  ]
\end{forest}

\end{document}