%% forest-doc.tex %% `forest-doc.tex` is a main documentation source of package `forest`. %% %% Copyright (c) 2015 Saso Zivanovic %% (Sa\v{s}o \v{Z}ivanovi\'{c}) %% saso.zivanovic@guest.arnes.si %% %% This work may be distributed and/or modified under the %% conditions of the LaTeX Project Public License, either version 1.3 %% of this license or (at your option) any later version. %% The latest versionx of this license is in %% %% http://www.latex-project.org/lppl.txt %% %% and version 1.3 or later is part of all distributions of LaTeX %% version 2005/12/01 or later. %% %% This work has the LPPL maintenance status `author-maintained'. %% %% This file is a part of package `forest'. For the list of files %% constituting the package see main source file of the package, %% `forest.dtx', or the derived `forest.sty'. %% \documentclass{ltxdoc} \usepackage[a4paper, top=2cm,bottom=2.5cm, reversemarginpar,marginparwidth=2cm, right=1.5cm, nohead ]{geometry} %\immediate\write18{makeindex -s forest-doc.ist forest-doc} \def\indexlettertotoc#1\relax{% \phantomsection\addcontentsline{toc}{subsection}{#1}% #1% } \makeatletter \c@IndexColumns=2 \makeatother \IndexPrologue{Color legend: \textcolor{red}{definition}, \textcolor{darkgreen}{example}, \textcolor{blue}{other}. If an entry belongs to a library, the library name is given in parenthesis. All page numbers are hyperlinks, and definitions in text are hyperlinked to this index.} \makeindex \usepackage[external]{forest} \useforestlibrary{edges} \useforestlibrary{linguistics} % \tikzexternalize % forest-tikzexternalize either contains \tikzexternalize or not ... make adjusts this \IfFileExists{forest-doc.tex-externalize.tex}{% \input forest-doc.tex-externalize\relax }{} \tikzset{ external/prefix={forest.for.dir/}, external/system call={ pdflatex \tikzexternalcheckshellescape -halt-on-error -interaction=nonstopmode -jobname "\image" "\texsource"}, external/up to date check=simple, } \usepackage{forest-doc} \usepackage{hyperref} \hypersetup{unicode=true,colorlinks=true,linkcolor=blue,citecolor=blue,hyperindex=false} \usepackage{url} \usepackage[numbers]{natbib} \usepackage[multiple]{footmisc} \usepackage{tipa} \usepackage[inline,shortlabels]{enumitem} \newlist{compactitem}{itemize}{2} \setlist[compactitem]{nosep,label=\textbullet} \newlist{arguments}{enumerate}{1} \setlist[arguments]{nosep,label=\textbf{\#\arabic*}} \renewcommand\DescribeMacro[1]{\texttt{\detokenize{#1}}\marginpar{\hfill\small\strut\texttt{\detokenize{#1}}}} \usepackage{printlen} \usepackage{tabularx} \DeleteShortVerb\| \newcommand\OR{\ensuremath{\,|\,}}% \usepackage{multicol} %\usepackage[silent]{trace-pgfkeys} % silent, stack, trace, verbose %\pgfkeystracelevel{trace} %%% begin listing region: index_macro_style \forestset{ detokenize/.style={result=\protect\detokenize{##1}}, tt/.style={result=\protect\texttt{##1}}, macro/.style={detokenize, +result={\char\escapechar}, tt, print in color, hyper}, } %%% end listing region: index_macro_style \forestset{ pgfkey/.style={tt,print in color,hyper}, path/.style={+result={/},pgfkey}, handler/.style={+result={.},pgfkey}, option value/.style={ format={tt,result+={ value}}, for first={format=pgfkey}, }, environment/.style={ detokenize, tt, print in color, if stage={index}{}{ if index entry type={definition}{ temptoksa/.register=result, hyper, result/.wrap 2 pgfmath args={% \texttt{\string\begin\char`\{##1\char`\}}% \forestoption{argument}% \texttt{\string\end\char`\{##2\char`\}}% }{result}{temptoksa}, }{ hyper } } }, empty/.style={result={{}}}, meta/.style={result=\protect\meta{##1},print in color,hyper}, index library/.style={ if index entry type={definition}{ for first leaf'={ index format+={ result+'={\space {\protect\scriptsize (\protect\texttt{\protect\hyperrefnocolor[sec:library-#1]{#1}})% }% } } }, }{} }, margin/.style={print format={result=\marginpar{\strut##1}}}, also margin/.style={print format={result=##1\marginpar{\strut##1}}}, } \let\origmeta\meta \renewcommand\meta[1]{\origmeta{\rm\ignorespaces#1}} \renewcommand\rstyle{\color{red}} \newcommand\hyperrefnocolor[2][]{{\hypersetup{linkcolor=black}\hyperref[#1]{#2}}} \newcommand\librarysection[1]{% \subsection{\protect\texttt{#1}}% \label{sec:library-#1}% \forestset{every index end/.style={set={index library={#1}}}}% } \newcommand\reflibrary[1]{\hyperref[sec:library-#1]{\texttt{#1}}} \index[not print,not index,set={ index key=unknown, form={\textbf{unknown!!}}, for first={format={ result/.expanded=\noexpand\textbf{\forestregister{result}??}, print in color}}} ]>{@unknown} \indexset [print form={},for first={format=pgfkey}] >{node key} \indexset [for first={format={+result={/tikz/},pgfkey}}] >{tikz key} \indexset [for first={format=pgfkey}] >{option,readonly option,bracket key,nodewalk key,register,propagator,style,nodewalk style,stage,dynamic tree,step,loop,conditional,anchor,package option} %%% begin listing region: index_macro_category \indexset [for first={format=macro}] >{macro} %%% end listing region: index_macro_category \indexset [for first={format=environment}] >{environment} \indexset [format=empty]>{empty} \indexset [index form={type (of options and registers)},for first={format=meta}] >{type} \indexset [for first={format=handler}] >{handler} \indexset [format={aggregate function},for first={format=handler}] >{aggregate} \indexset [form={\texttt{pgfmath} function},for first={format={pgfkey}}] >{pgfmath} \indexset [for first={format=path}] >{path} \indexset[ form={}, for first={ for first={ key+/.pgfmath=argument("!parent"), index format={ result+/.wrap pgfmath arg={\protect\meta{##1 option}}{argument("!parent")} }, print format={ if argument={}{}{ result+/.option=argument }, }, format={ pgfkey }, } } ]>{prefix} \indexset[ form={}, for first={ for first={ +key/.pgfmath=argument("!parent"), index format={ +result/.wrap pgfmath arg={\protect\meta{##1 option}}{argument("!parent")} }, print format={ if argument={}{}{ +result/.option=argument }, }, format={ pgfkey }, } } ]>{suffix} \indexset[ form={}, for first={ for first={ key/.wrap 2 pgfmath args={##1 ##2}{key}{argument("!parent,parent")}, index format={ result+={\space}, result+/.wrap pgfmath arg={\protect\meta{##1}}{argument("!parent,parent")} }, print format={ if argument={}{}{ result+={\space}, result+/.option=argument } }, format={ pgfkey }, } } ]>{word prefix} \indexset[ form={}, for first={ for first={ key/.wrap 2 pgfmath args={##2 ##1}{key}{argument("!parent,parent")}, index format={ +result={\space}, +result/.wrap pgfmath arg={\protect\meta{##1}}{argument("!parent,parent")} }, print format={ if argument={}{}{ +result={\space}, +result/.option=argument } }, format={ pgfkey }, } } ]>{word suffix} \indexset[%option values index key format={result/.expanded=\forestoption{argument} value}, format={result/.expanded=\noexpand\texttt{\forestoption{argument}} value}, for first={format=pgfkey} ]>{value} \indexset[ index key format={result/.expanded=\forestoption{argument} option}, format={result/.expanded=\noexpand\texttt{\forestoption{argument}} option}, for first={format=pgfkey} ]>{option of} \indexset [form={},for first={format=meta}] >{meta} \indexset[ for first={+key={{cs }},format={pgfkey}} ]>{forest cs} \indexset[ for first={+key={{ss }},format={pgfkey}} ]>{short step} \indexset [for first={format=pgfkey}] >{named nodewalk} {\makeatletter % an dirty patch: \lst@nolig can sneak in the name... \gdef\myexampleindex#1{{\def\lst@nolig{}\lstaspectindex{#1}{}}} } \lstset{indexstyle={[1]\myexampleindex}} %%% end lst-related stuff \EnableCrossrefs \setlength\hfuzz{15pt} % dont make so many \hbadness=7000 % over and under full box warnings \usetikzlibrary{intersections} \tikzset{>=latex} \forestset{ background tree/.style={ for tree={text opacity=0.2,draw opacity=0.2,edge={draw opacity=0.2}}} } \title{\FoRest;: a \PGF;/\TikZ;-based package for drawing linguistic trees\\\normalsize\forestversion} \author{Sa\v so \v Zivanovi\'c\footnote{e-mail: \href{mailto:saso.zivanovic@guest.arnes.si}{saso.zivanovic@guest.arnes.si}; web: \href{http://spj.ff.uni-lj.si/zivanovic/}{http://spj.ff.uni-lj.si/zivanovic/}}} %\usepackage[trace]{trace-pgfkeys} % silent, stack, trace, verbose %\pgfkeystracelevel{trace} \begin{document} \typeout{)}% hacking auctex's error finding routing % short verbatim: | (changes spaces into _) \DeleteShortVerb\| {\catcode`\_=12 \def\marshal{% \lstMakeShortInline[basicstyle=\ttfamily,literate={_}{ }1 {__}{_}1]}% \expandafter}\marshal | \IfFileExists{forest-doc-test.tex}{% \input forest-doc-test\relax }{} \newbox\treebox \newbox\codebox {\settodayfromforestdate\maketitle} \begin{abstract} \FoRest; is a \PGF;/\TikZ;-based package for drawing linguistic (and other kinds of) trees. Its main features are \begin{enumerate*}[(i)] \item a packing algorithm which can produce very compact trees; \item a user-friendly interface consisting of the familiar bracket encoding of trees plus the key--value interface to option-setting; \item many tree-formatting options, with control over option values of individual nodes and mechanisms for their manipulation; \item a powerful mechanism for travesing the tree; \item the possibility to decorate the tree using the full power of \PGF;/\TikZ;; \item an externalization mechanism sensitive to code-changes. \end{enumerate*} \end{abstract} \vspace{1cm} \begin{forestexample}[basicstyle=\ttfamily\scriptsize,layout=tree on top,no label,v sep=1cm,index={content,pgfmath,propagator>if,repeat,append,before drawing tree,y,alias,for step,step>children,edge,before typesetting nodes,tree,s sep,dimen+,l,ancestors,typeset node,nodewalk,readonly option>level,sort by,step>min}] \pgfmathsetseed{14285} \begin{forest} random tree/.style n args={3}{% #1 = max levels, #2 = max children, #3 = max content content/.pgfmath={random(0,#3)}, if={#1>0}{repeat={random(0,#2)}{append={[,random tree={#1-1}{#2}{#3}]}}}{}}, before typesetting nodes={for tree={draw,s sep=2pt,rotate={int(30*rand)},l+={5*rand}, if={isodd(level())}{fill=green}{fill=yellow}}}, important/.style={draw=red,line width=1.5pt,edge={red,line width=1.5pt}}, before drawing tree={sort by=y, for nodewalk={min=tree,ancestors}{important,typeset node}} [,random tree={9}{3}{100}] \end{forest} \end{forestexample}% \begin{center}\mbox{}\box\treebox\\\box\codebox\end{center} \newpage {% \parskip 0pt \begin{multicols}{2} \tableofcontents \end{multicols} } \newpage \section{Introduction} Over several years, I had been a grateful user of various packages for typesetting linguistic trees. My main experience was with |qtree| and |synttree|, but as far as I can tell, all of the tools on the market had the same problem: sometimes, the trees were just too wide. They looked something like the tree on the left, while I wanted something like the tree on the right. \begin{center} \begin{forest} baseline, for tree={parent anchor=south,child anchor=north,l=7ex,s sep=10pt}, for children={fit=rectangle} [CP [DP [D][NP[N][CP[C][TP[T][vP[v][VP[DP][V'[V][DP]]]]]]]] [TP [T][vP[v][VP[DP][V'[V][DP]]]]] ] \end{forest} \hfill \begin{forest} baseline, for tree={parent anchor=south,child anchor=north,l=7ex,s sep=10pt}, [CP [DP [D][NP[N][CP[C][TP[T][vP[v][VP[DP][V'[V][DP]]]]]]]] [TP [T][vP[v][VP[DP][V'[V][DP]]]]] ] \end{forest} \end{center} Luckily, it was possible to tweak some parameters by hand to get a narrower tree, but as I quite dislike constant manual adjustments, I eventually started to develop \foRest;. It started out as |xyforest|, but lost the |xy| prefix as I became increasingly fond of \PGF;/\TikZ;, which offered not only a drawing package but also a `programming paradigm.' It is due to the awesome power of the supplementary facilities of \PGF;/\TikZ; that \foRest; is now, I believe, the most flexible tree typesetting package for \LaTeX\ you can get. The latest stable version of \FoRest; is \href{http://www.ctan.org/pkg/forest}{available at CTAN}. Development version(s) can be found \href{https://github.com/sasozivanovic/forest}{at GitHub}. Comments, criticism, suggestions and code are all very welcome! If you find the package useful, you can show your appreciation by making a PayPal donation to \url{saso.zivanovic@guest.arnes.si}. \section{Tutorial} \label{sec:tutorial} This short tutorial progresses from basic through useful to obscure \dots \subsection{Basic usage} \label{tut:basic-usage} A tree is input by enclosing its specification in a \index{forest} environment. The tree is encoded by \emph{the bracket syntax}: every node is enclosed in square brackets; the children of a node are given within its brackets, after its content. {\lstdefinelanguage[my]{TeX}[LaTeX]{TeX}{keywords=forest, otherkeywords={[,]},keywordstyle=\pstyle,texcsstyle={}}% \lstset{language={[my]TeX}}% \begin{forestexample} \begin{forest} [VP [DP] [V' [V] [DP] ] ] \end{forest} \end{forestexample}} Binary trees are nice, but not the only thing this package can draw. Note that by default, the children are vertically centered with respect to their parent, i.e.\ the parent is vertically aligned with the midpoint between the first and the last child. \begin{forestexample} \begin{forest} [VP [DP[John]] [V' [V[sent]] [DP[Mary]] [DP[D[a]][NP[letter]]] ] ] \end{forest} \end{forestexample} Spaces around brackets are ignored --- format your code as you desire! \begin{forestexample} \begin{forest} [VP[DP][V'[V][DP]]] \end{forest} \quad \begin{forest}[VP [DP ] [ V'[V][ DP]] ]\end{forest} \end{forestexample} If you need a square bracket as part of a node's content, use braces. The same is true for the other characters which have a special meaning in the \foRest; package, like comma |,| and equality sign |=|. \begin{forestexample} \begin{forest} [V{P,} [{[DP]}] [V' [V] [{===DP===}]]] \end{forest} \end{forestexample} Macros in a node specification will be expanded when the node is drawn --- you can freely use formatting commands inside nodes! \begin{forestexample} \begin{forest} [VP [{~\textbf~{DP}}] [V' [V] [DP]]] \end{forest} \end{forestexample} All the examples given above produced top-down trees with centered children. The other sections of this manual explain how various properties of a tree can be changed, making it possible to typeset radically different-looking trees. However, you don't have to learn everything about this package to profit from its power. Using styles, you can draw predefined types of trees with ease. For example, a phonologist can use the \index{GP1} style from library \reflibrary{linguistics} to easily typeset (Government Phonology) phonological representations. The style is applied simply by writing its name before the first (opening) bracket of the tree. \begin{forestexample}[label=ex:gp1-frost,code prefix={\def\usepackage[##1]##2{}}] \usepackage[~linguistics~]{forest} \begin{forest} ~GP1~ [ [O[x[f]][x[r]]] [R[N[x[o]]][x[s]]] [O[x[t]]] [R[N[x]]] ]\end{forest} \end{forestexample} Of course, someone needs to develop the style --- you, me, your local \TeX nician \dots\@ Fortunately, designing styles is not very difficult once you get the hang of \foRest;, if you write one, please contribute! Some macros relating to various fields are collected in \emph{libraries} that are distributed alongside the main package. This is the case for the \index{GP1} style used above, which is defined in the |linguistics| library. The simplest way to load a library is as shown in the example, by loading the package with an optional argument. For more information on loading libraries, see \S\ref{ref:package-options}. (Style \index{GP1} used above is defined in the \keyname{linguistics} library of the \foRest; package.) \subsection{Options} \label{tut:options} A node can be given various options, which control various properties of the node and the tree. For example, at the end of section~\ref{tut:basic-usage}, we have seen that the \index{GP1} style vertically aligns the parent with the first child. This is achieved by setting option \index{calign} (for \emph{c}hild-\emph{align}ment) to \index{value=calign>first} (child). Let's try. Options are given inside the brackets, following the content, but separated from it by a comma. (If multiple options are given, they are also separated by commas.) A single option assignment takes the form \meta{option name}|=|\meta{option value}. (There are also options which do not require a value or have a default value: these are given simply as \meta{option name}.) \begin{forestexample}[label=ex:numerals-simple,index={calign,value=calign>first}] \begin{forest} [\LaTeX\ numerals, ~calign=first~ [arabic[1][2][3][4]] [roman[i][ii][iii][iv]] [alph[a][b][c][d]] ] \end{forest} \end{forestexample} The experiment has succeeded only partially. The root node's children are aligned as desired (so \index{calign}|=|\index{value=calign>first} applied to the root node), but the value of the \index{calign} option didn't get automatically assigned to the root's children! \emph{An option given at some node applies only to that node.} In \foRest;, the options are passed to the node's relatives via special keys, called \emph{propagators}. What we need above is the \index{for step=\index{tree}} propagator. Observe: \begin{forestexample}[label=ex:numerals-manual] \begin{forest} [\LaTeX\ numerals, ~for tree~={calign=first} [arabic[1][2][3][4]] [roman[i][ii][iii][iv]] [alph[a][b][c][d]] ] \end{forest} \end{forestexample} The value of propagator \index{for step=\index{tree}} is the option string that we want to process. This option string is propagated to all the nodes in the subtree\footnote{It might be more precise to call this option \texttt{for subtree} \dots\@ but this name at least saves some typing.} rooted in the current node (i.e.\ the node where \index{for step=\index{tree}} was given), including the node itself. (Propagator \index{for step=\index{descendants}} is just like \index{for step=\index{tree}}, only that it excludes the node itself. There are many other \index{for step=\meta{step}} propagators; for the complete list, see sections~\ref{ref:spatial-propagators} and \ref{ref:nodewalks}.) Some other useful options are \index{option>parent anchor}, \index{option>child anchor} and \index{tier}. The \index{option>parent anchor} and \index{option>child anchor} options tell where the parent's and child's endpoint of the edge between them should be, respectively: usually, the value is either empty (meaning a smartly determined border point \citep[see][\S16.11]{tikzpgf2.10}; this is the default) or a compass direction \citep[see][\S16.5.1]{tikzpgf2.10}. (Note: the \index{option>parent anchor} determines where the edge from the child will arrive to this node, not where the node's edge to its parent will start!) Option \index{tier} is what makes the skeletal points $\times$ in example (\ref{ex:gp1-frost}) align horizontally although they occur at different levels in the logical structure of the tree. Using option \index{tier} is very simple: just set |tier=tier_name| at all the nodes that you want to align horizontally. Any tier name will do, as long as the tier names of different tiers are different \dots\@ (Yes, you can have multiple tiers!) \begin{forestexample}[point={tier},index={option>parent anchor,option>child anchor,tier},label=ex:tier-manual] \begin{forest} [VP, for tree={~parent anchor~=south, ~child anchor~=north} [DP[John,tier=word]] [V' [V[sent,tier=word]] [DP[Mary,tier=word]] [DP[D[a,tier=word]][NP[letter,tier=word]]] ] ] \end{forest} \end{forestexample} Before discussing the variety of \foRest;'s options, it is worth mentioning that \foRest;'s node accepts all options \citep[see \S16]{tikzpgf2.10} that \TikZ;'s node does --- mostly, it just passes them on to \TikZ;. For example, you can easily encircle a node like this:\footnote{If option \texttt{draw} was not given, the shape of the node would still be circular, but the edge would not be drawn. For details, see \cite[\S16]{tikzpgf2.10}.} \begin{forestexample} \begin{forest} [VP,~circle~,~draw~ [DP][V'[V][DP]] ] \end{forest} \end{forestexample} Let's have another look at example (\ref{ex:gp1-frost}). You will note that the skeletal positions were input by typing |x|s, while the result looks like this: $\times$ (input as |\times| in math mode). Obviously, the content of the node can be changed. Even more, it can be manipulated: added to, doubled, boldened, emphasized, etc. We will demonstrate this by making example (\ref{ex:numerals-manual}) a bit fancier: we'll write the input in the arabic numbers and have \LaTeX\ convert it to the other formats. We'll start with the easiest case of roman numerals: to get them, we can use the (plain) \TeX\ command |\romannumeral|. To change the content of the node, we use option \index{content}. When specifying its new value, we can use |#1| to insert the current content.\footnote{This mechanism is called \emph{wrapping}. \index{content} is the only option where wrapping works implicitely (simply because I assume that wrapping will be almost exclusively used with this option). To wrap values of other options, use handler \index{wrap value}; see~\S\ref{ref:handlers}.} \begin{forestexample}[point={content,delay},index={for step,step>children,content,delay},label=ex:romannumeral] \begin{forest} [roman, delay={for children={content=\romannumeral#1}} [1][2][3][4] ] \end{forest} \end{forestexample} This example introduces another option: \index{delay}. Without it, the example wouldn't work: we would get arabic numerals. This is so because of the order in which the options are processed. First, the processing proceeds through the tree in a depth-first, parent-first fashion (first the parent is processed, and then its children, recursively; but see \index{processing order}). Next, the option string of a node is processed linearly, in the order they were given. (Option \index{content} is specified implicitely and is always the first.) If a propagator is encountered, the options given as its value are propagated \emph{immediately}. The net effect is that if the above example contained simply |roman,for_children={content=...}|, the \index{content} option given there would be processed \emph{before} the implicit content options given to the children (i.e.\ numbers |1|, |2|, |3| and |4|). Thus, there would be nothing for the |\romannumeral| to change --- it would actually crash; more generally, the content assigned in such a way would get overridden by the implicit content. Key \index{delay} is true to its name. It delays the processing of the keylist given as its argument until the whole tree was processed. In other words, it introduces cyclical option processing. Whatever is delayed in one cycle, gets processed in the next one. The number of cycles is not limited --- you can nest \index{delay}s as deep as you need. Unlike \index{for step=\meta{step}} keys we have met before, \index{delay} is not a spatial, but a temporal propagator. Several other temporal propagators options exist, see \S\ref{ref:stages}. We are now ready to learn about simple conditionals. Every node option has the corresponding \index{if option=\meta{option}} and \index{where option=\meta{option}} keys. \index{if option=\meta{option}}|=|\meta{value}\meta{true options}\meta{false options} checks whether the value of \meta{option} equals \meta{value}. If so, \meta{true options} are processed, otherwise \meta{false options}. The \index{where option=\meta{option}} keys are the same, but do this for the every node in the subtree; informally speaking, |where| = |for_tree| + |if|. To see this in action, consider the rewrite of the \index{tier} example (\ref{ex:tier-manual}) from above. We don't set the tiers manually, but rather put the terminal nodes (option \index{n children} is a read-only option containing the number of children) on tier \keyname{word}.\footnote{We could omit the braces around \texttt{0} because it is a single character. If we were hunting for nodes with 42 children, we'd have to write \texttt{where n children=\{42\}...}.} \begin{forestexample}[index={tier,where option,n children}] \begin{forest} ~where n children~=0{tier=word}{} [VP [DP[John]] [V' [V[sent]] [DP[Mary]] [DP[D[a]][NP[letter]]] ] ] \end{forest} \end{forestexample} Note that you usually don't want to embed a \index{where} conditional in a \index{for step=\index{tree}}, as this will lead to a multiple traversal of many nodes, resulting in a slower execution. If you're inside a \index{for step=\index{tree}}, you probably want to use \index{if}. Finally, let's talk about styles. Styles are simply collections of options and/or other keys. (They are not actually defined in the \foRest; package, but rather inherited from |pgfkeys|.) If you often want to have non-default parent/child anchors, say south/north as in example (\ref{ex:tier-manual}), you would save some typing by defining a style. Styles are defined using \PGF;'s handler |.style|, like shown below.\footnote{Style \index{sn edges} is actually already defined by library \reflibrary{linguistics}. The definition there is a bit more generic.} \begin{forestexample}[index={tier,option>parent anchor,option>child anchor}] \begin{forest} ~sn edges~/~.style~={for tree={ parent anchor=south, child anchor=north}}, ~sn edges~ [VP, [DP[John,tier=word]] [V' [V[sent,tier=word]] [DP[Mary,tier=word]] [DP[D[a,tier=word]][NP[letter,tier=word]]]]] \end{forest} \end{forestexample} If you want to use a style in more than one tree, you have to define it outside the \index{forest} environment. Use macro \index{forestset} to do this. \begin{lstlisting} ~\forestset~{ sn edges/.style={for tree={parent anchor=south, child anchor=north}}, background tree/.style={for tree={ text opacity=0.2,draw opacity=0.2,edge={draw opacity=0.2}}} } \end{lstlisting} You might have noticed that in the last two examples contain, some keys occurred even before the first opening bracket, contradicting was said at the beginning of this section. This is mainly just syntactic sugar (it can separate the design and the content): such \emph{preamble} keys behave as if they were given in the root node, the only difference (which often does not matter) being that they get processed before all other root node options, even the implicit \index{content}. If you find yourself writing the same preamble for every tree in your document, consider redefining style \index{default preamble}, which is implicitely included at the beginning of every preamble. \subsection{Decorating the tree} \label{tut:decorating} The tree can be decorated (think movement arrows) with arbitrary \TikZ; code. \begin{forestexample} \begin{forest} [XP [specifier] [X$'$ [X$^0$] [complement] ] ] ~\node at (current bounding box.south) [below=1ex,draw,cloud,aspect=6,cloud puffs=30] {\emph{Figure 1: The X' template}};~ \end{forest} \end{forestexample} However, decorating the tree would make little sense if one could not refer to the nodes. The simplest way to do so is to give them a \TikZ; name using the \index{option>name} option, and then use this name in \TikZ; code as any other (\TikZ;) node name. \begin{forestexample}[point=name,index={option>phantom,option>name}] \begin{forest} [CP [DP,name=spec CP] [\dots [,phantom] [VP [DP] [V' [V] [DP,name=object]]]]] \draw[->,dotted] ~(object)~ to[out=south west,in=south] ~(spec CP)~; \end{forest} \end{forestexample} It gets better than this, however! In the previous examples, we put the \TikZ; code after the tree specification, i.e.\ after the closing bracket of the root node. In fact, you can put \TikZ; code after \emph{any} closing bracket, and \foRest; will know what the current node is. (Putting the code after a node's bracket is actually just a special way to provide a value for option \index{tikz} of that node.) To refer to the current node, simply use an empty node name. This works both with and without anchors \citep[see][\S16.11]{tikzpgf2.10}: below, |(.south east)| and |()|. \begin{forestexample}[index={option>phantom,option>name}] \begin{forest} [CP [DP,name=spec CP] [\dots [,phantom] [VP [DP] [V' [V] [DP,draw] ~{~ \draw[->,dotted] ~()~ to[out=south west,in=south] (spec CP); \draw[<-,red] ~(.south east)~--++(0em,-4ex)--++(-2em,0pt) node[anchor=east,align=center]{This guy\\has moved!}; ~}~ ]]]] \end{forest} \end{forestexample} Important: \emph{the \TikZ; code should usually be enclosed in braces} to hide it from the bracket parser. You don't want all the bracketed code (e.g.\ |[->,dotted]|) to become tree nodes, right? (Well, they probably wouldn't anyway, because \TeX\ would spit out a thousand errors.) \bigskip Finally, the most powerful tool in the node reference toolbox: \emph{relative nodes}. It is possible to refer to other nodes which stand in some (most often geometrical) relation to the current node. To do this, follow the node's name with a \index>{!} and a \emph{node walk} specification. A node walk is a concise\footnote{Actually, \foRest; distinguishes two kinds of steps in node walks: long-form and short-form steps. This section introduces only short-form steps. See \S\ref{ref:nodewalks}.} way of expressing node relations. It is simply a string of steps, which are represented by single characters, where: \index{ss u} stands for the parent node (up); \index{ss p} for the previous sibling; \index{ss n} for the next sibling; \index{ss s} for \emph{the} sibling (useful only in binary trees); \index{ss 1}, \index{ss 2}, \index[not print]{ss 3,ss 4,ss 5,ss 6,ss 7,ss 8}\dots\ \index{ss 9} for first, second, \dots\ ninth child; \index{ss l}, for the last child, etc. For the complete specification, see section~\ref{ref:short-form-steps}. To see the nodewalk in action, consider the following examples. In the first example, the agree arrow connects the V node, specified simply as |()|, since the \TikZ; code follows |[V]|, and the DP node, which is described as ``a sister of V's parent'': |!us| = up + sibling. \begin{forestexample}[index>={!}] \begin{forest} [VP [DP] [V' [V] {\draw[<->] ~()~ .. controls +(left:1cm) and +(south west:0.4cm) .. node[very near start,below,sloped]{\tiny agree} ~(!us)~;} [DP] ] ] \end{forest} \end{forestexample} \begingroup \footnotesize \begin{forestexample}[index={option>phantom,tikz,fit to,tree},layout=export,basicstyle=\footnotesize\ttfamily] \begin{forest} [CP [DP$_1$] [\dots [,phantom] [VP,tikz={\node [draw,red,inner sep=0,~fit to~=tree]{};} [DP$_2$] [V' [V] [DP$_3$] ]]]] \end{forest} \end{forestexample}% \endgroup The second example uses \TikZ;'s fitting library to compute the smallest rectangle containing node VP, its first child (DP$_2$) and its last grandchild (DP$_3$). The example also illustrates that the \TikZ; code can be specified via the ``normal'' option syntax, i.e.\ as a value to option \index{tikz}.\footnote{\label{fn:fit-to-tree}Actually, there's a simpler way to do this: use \index{fit to}\keyname{=}\index{tree}!\forestexampleimport} \begin{forestexample}[point=tikz,index={option>phantom,tikz},index>={!}] \begin{forest} [CP [DP$_1$] [\dots [,phantom] [VP,tikz={\node [draw,red,fit=~()(!1)(!ll)~] {};} [DP$_2$] [V' [V] [DP$_3$] ]]]] \end{forest} \end{forestexample} \subsection{Node positioning} \label{tut:node-positioning} \FoRest; positions the nodes by a recursive bottom-up algorithm which, for every non-terminal node, computes the positions of the node's children relative to their parent. By default, all the children will be aligned horizontally some distance down from their parent: the ``normal'' tree grows down. More generally, however, the direction of growth can change from node to node; this is controlled by option \index{grow}=\meta{direction}.\footnote{The direction can be specified either in degrees (following the standard mathematical convention that $0$ degrees is to the right, and that degrees increase counter-clockwise) or by the compass directions: \texttt{east}, \texttt{north east}, \texttt{north}, etc.} The system thus computes and stores the positions of children using a coordinate system dependent on the parent, called an \emph{ls-coordinate system}: the origin is the parent's anchor; l-axis is in the direction of growth in the parent; s-axis is orthogonal to the l-axis (positive side in the counter-clockwise direction from $l$-axis); l stands for \emph{l}evel, s for \emph{s}ibling. The example shows the ls-coordinate system for a node with |grow=45|. \begin{forestexample}[point=grow,index=grow] \begin{forest} background tree [parent, grow=45 [child 1][child 2][child 3][child 4][child 5] ] \draw[,->](-135:1cm)--(45:3cm) node[below]{$l$}; \draw[,->](-45:1cm)--(135:3cm) node[right]{$s$}; \end{forest} \end{forestexample} \begin{forestexample}[basicstyle=\scriptsize\ttfamily,layout=export] \newcommand\measurexdistance[5][####1]{\measurexorydistance{#2}{#3}{#4}{#5}{\x}{-|}{(5pt,0)}{#1}} \newcommand\measureydistance[5][####1]{\measurexorydistance{#2}{#3}{#4}{#5}{\y}{|-}{(0,5pt)}{#1}} \tikzset{dimension/.style={<->,>=latex,thin,every rectangle node/.style={midway,font=\scriptsize}}, guideline/.style=dotted} \newdimen\absmd \def\measurexorydistance#1#2#3#4#5#6#7#8{% \path #1 #3 #6 coordinate(md1) #1; \draw[guideline] #1 -- (md1); \path (md1) #6 coordinate(md2) #2; \draw[guideline] #2 -- (md2); \path let \p1=($(md1)-(md2)$), \n1={abs(#51)} in \pgfextra{\xdef\md{#51}\global\absmd=\n1\relax}; \def\distancelabelwrapper##1{#8}% \ifdim\absmd>5mm \draw[dimension] (md1)--(md2) node[#4]{\distancelabelwrapper{\uselengthunit{mm}\rndprintlength\absmd}}; \else \ifdim\md>0pt \draw[dimension,<-] (md1)--+#7; \draw[dimension,<-] let \p1=($(0,0)-#7$) in (md2)--+(\p1); \else \draw[dimension,<-] let \p1=($(0,0)-#7$) in (md1)--+(\p1); \draw[dimension,<-] (md2)--+#7; \fi \draw[dimension,-] (md1)--(md2) node[#4]{\distancelabelwrapper{\uselengthunit{mm}\rndprintlength\absmd}}; \fi} \end{forestexample} \input{\jobname.tmp} The l-coordinate of children is (almost) completely under your control, i.e.\ you set what is often called the level distance by yourself. Simply set option \index{option>l} to change the distance of a node from its parent. More precisely, \index{option>l}, and the related option \index{option>s}, control the distance between the (node) anchors of a node and its parent. The anchor of a node can be changed using option \index{option>anchor}: by default, nodes are anchored at their base; see \cite[\S16.5.1]{tikzpgf2.10}.) In the example below, positions of the anchors are shown by dots: observe that anchors of nodes with the same \index{option>l} are aligned and that the distances between the anchors of the children and the parent are as specified in the code.\footnote{Here are the definitons of the macros for measuring distances. Args: the x or y distance between points \#2 and \#3 is measured; \#4 is where the distance line starts (given as an absolute coordinate or an offset to \#2); \#5 are node options; the optional arg \#1 is the format of label. (Lengths are printed using package \texttt{printlen}.) \box\codebox} \begin{forestexample}[layout=tree on top,index={for step,tree,tikz,option>l,option>anchor},index>={!}] \begin{forest} background tree, for tree={draw,tikz={\fill[](.anchor)circle[radius=1pt];}} [parent [child 1, ~l~=10mm, ~anchor~=north west] [child 2, ~l~=10mm, ~anchor~=south west] [child 3, ~l~=12mm, ~anchor~=south] [child 4, ~l~=12mm, ~anchor~=base east] ] \measureydistance[\texttt{l(child)}=#1]{(!2.anchor)}{(.anchor)}{(!1.anchor)+(-5mm,0)}{left} \measureydistance[\texttt{l(child)}=#1]{(!3.anchor)}{(.anchor)}{(!4.anchor)+(5mm,0)}{right} \measurexdistance[\texttt{s sep(parent)}=#1]{(!1.south east)}{(!2.south west)}{+(0,-5mm)}{below} \measurexdistance[\texttt{s sep(parent)}=#1]{(!2.south east)}{(!3.south west)}{+(0,-5mm)}{below} \measurexdistance[\texttt{s sep(parent)}=#1]{(!3.south east)}{(!4.south west)}{+(0,-8mm)}{below} \end{forest} \end{forestexample} Positioning the chilren in the s-dimension is the job and \emph{raison d'etre} of the package. As a first approximation: the children are positioned so that the distance between them is at least the value of option \index{s sep} (s-separation), which defaults to double \PGF;'s |inner_xsep| (and this is 0.3333em by default). As you can see from the example above, s-separation is the distance between the borders of the nodes, not their anchors! A fuller story is that \index{s sep} does not control the s-distance between two siblings, but rather the distance between the subtrees rooted in the siblings. When the green and the yellow child of the white node are s-positioned in the example below, the horizontal distance between the green and the yellow subtree is computed. It can be seen with the naked eye that the closest nodes of the subtrees are the TP and the DP with a red border. Thus, the children of the root CP (top green DP and top yellow TP) are positioned so that the horizontal distance between the red-bordered TP and DP equals \index{s sep}. \begin{forestexample}[index={for step,tree,s sep}] \begin{forest} important/.style={name=#1,draw={red,thick}} [CP, ~s sep~=3mm, for tree=draw [DP, for tree={fill=green} [D][NP[N][CP[C][TP,important=left [T][vP[v][VP[DP][V'[V][DP]]]]]]]] [TP,for tree={fill=yellow} [T][vP[v][VP[DP,important=right][V'[V][DP]]]]] ] \measurexdistance[\texttt{s sep(root)}=#1] {(left.north east)}{(right.north west)}{(.north)+(0,3mm)}{above} \end{forest} \end{forestexample} Note that \foRest; computes the same distances between nodes regardless of whether the nodes are filled or not, or whether their border is drawn or not. Filling the node or drawing its border does not change its size. You can change the size by adjusting \TikZ;'s |inner_sep| and |outer_sep| \citep[\S16.2.2]{tikzpgf2.10}, as shown below: \begin{forestexample}[index={for step,tree,s sep}] \begin{forest} important/.style={name=#1,draw={red,thick}} [CP, s sep=3mm, for tree=draw [DP, for tree={fill=green,~inner sep~=0} [D][NP,important=left[N][CP[C][TP[T][vP[v] [VP[DP][V'[V][DP]]]]]]]] [TP,for tree={fill=yellow,~outer sep~=2pt} [T,important=right][vP[v][VP[DP][V'[V][DP]]]]] ] \measurexdistance[\texttt{s sep(root)}=#1] {(left.north east)}{(right.north west)}{(.north)+(0,3mm)}{above} \end{forest} \end{forestexample} (This looks ugly!) Observe that having increased |outer sep| makes the edges stop touching borders of the nodes. By (\PGF;'s) default, the |outer sep| is exactly half of the border line width, so that the edges start and finish precisely at the border. Let's play a bit and change the \index{option>l} of the root of the yellow subtree. Below, we set the vertical distance of the yellow TP to its parent to 3\,cm: and the yellow submarine sinks diagonally \dots\@ Now, the closest nodes are the higher yellow DP and the green VP. \begin{forestexample}[index={option>l,s sep,for step,tree}] \begin{forest} important/.style={name=#1,draw={red,thick}} [CP, s sep=3mm, for tree=draw [DP, for tree={fill=green} [D][NP[N][CP[C][TP [T][vP[v][VP,important=left[DP][V'[V][DP]]]]]]]] [TP,for tree={fill=yellow}, l=3cm [T][vP[v][VP[DP,important=right][V'[V][DP]]]]] ] \measurexdistance[\texttt{s sep(root)}=#1] {(left.north east)}{(right.north west)}{(.north)+(0,3mm)}{above} \end{forest} \end{forestexample} Note that the yellow and green nodes are not vertically aligned anymore. The positioning algorithm has no problem with that. But you, as a user, might have, so here's a neat trick. (This only works in the ``normal'' circumstances, which are easier to see than describe.) \begin{forestexample}[label=ex:l*,index={dimen*=l,option>phantom,for step,tree}] \begin{forest} [CP, for tree=draw [DP, for tree={fill=green},~l*~=3 [D][NP]] [TP,for tree={fill=yellow} [T][VP[DP][V'[V][DP]]]] ] \end{forest} \end{forestexample} We have changed DP's \index{option>l}'s value via ``augmented assignment'' known from many programming languages: above, we have used |l*=3| to triple \index{option>l}'s value; we could have also said |l+=5mm| or |l-=5mm| to increase or decrease its value by 5\,mm, respectively. This mechanism works for every numeric and dimensional option in \foRest;. Let's now play with option \index{s sep}. \begin{forestexample}[index={s sep,option>l,dimen*=l,for step,tree}] \begin{forest} [CP, for tree=draw, ~s sep~=0 [DP, for tree={fill=green},l*=3 [D][NP]] [TP,for tree={fill=yellow} [T][VP[DP][V'[V][DP]]]] ] \end{forest} \end{forestexample} Surprised? You shouldn't be. The value of \index{s sep} at a given node controls the s-distance \emph{between the subtrees rooted in the children of that node}! It has no influence over the internal geometry of these subtrees. In the above example, we have set |s_sep=0| only for the root node, so the green and the yellow subtree are touching, although internally, their nodes are not. Let's play a bit more. In the following example, we set the \index{s sep} to: $0$ at the last branching level (level 3; the root is level 0), to 2\,mm at level 2, to 4\,mm at level 1 and to 6\,mm at level 0. \begin{forestexample}[label=ex:spread-s,point={level},index={readonly option>level,for step,tree,s sep},index>={!}] \begin{forest} for tree={~s sep~=(3-level)*2mm} [CP, for tree=draw [DP, for tree={fill=green},l*=3 [D][NP]] [TP,for tree={fill=yellow} [T][VP[DP][V'[V][DP]]]] ] \measurexdistance{(!11.south east)}{(!12.south west)}{+(0,-5mm)}{below} \path(md2)-|coordinate(md)(!221.south east); \measurexdistance{(!221.south east)}{(!222.south west)}{(md)}{below} \measurexdistance{(!21.north east)}{(!22.north west)}{+(0,2cm)}{above} \measurexdistance{(!1.north east)}{(!221.north west)}{+(0,-2.4cm)}{below} \end{forest} \end{forestexample} As we go up the tree, the nodes ``spread.'' At the lowest level, V and DP are touching. In the third level, the \index{s sep} of level 2 applies, so DP and V' are 2\,mm apart. At the second level we have two pairs of nodes, D and NP, and T and TP: they are 4\,mm apart. Finally, at level 1, the \index{s sep} of level 0 applies, so the green and yellow DP are 6\,mm apart. (Note that D and NP are at level 2, not 4! Level is a matter of structure, not geometry.) As you have probably noticed, this example also demostrated that we can compute the value of an option using an (arbitrarily complex) formula. This is thanks to \PGF;'s module |pgfmath|. \FoRest; provides an interface to |pgfmath| by defining |pgfmath| functions for every node option, and some other information, like the \index{readonly option>level} we have used above, the number of children \index{n children}, the sequential number of the child \index{readonly option>n}, etc. For details, see \S\ref{ref:pgfmath}. The final separation parameter is \index{l sep}. It determines the minimal separation of a node from its descendants. It the value of \index{option>l} is too small, then \emph{all} the children (and thus their subtrees) are pushed away from the parent (by increasing their \index{option>l}s), so that the distance between the node's and each child's subtree boundary is at least \index{l sep}. The initial \index{option>l} can be too small for two reasons: either some child is too high, or the parent is too deep. The first problem is easier to see: we force the situation using a bottom-aligned multiline node. (Multiline nodes can be easily created using |\\| as a line-separator. However, you must first specify the horizontal alignment using option \index{align} (see \S\ref{ref:node-appearance}). Bottom vertical alignment is achieved by setting \index{base}|=|\index{value=base>bottom}; the default, unlike in \TikZ;, is \index{base}|=|\index{value=base>top}). \begin{forestexample}[point={align,base},index={align,base}] \begin{forest} [parent [child] [child] [a very\\tall\\child, align=center, base=bottom] ] \end{forest} \end{forestexample} The defaults for \index{option>l} and \index{l sep} are set so that they ``cooperate.'' What this means and why it is necessary is a complex issue explained in \S\ref{tut:defaults}, which you will hopefully never have to read \dots\@ You might be out of luck, however. What if you needed to decrease the level distance? And nothing happened, like below on the left? Or, what if you used lots of parenthesis in your nodes? And got a strange vertical misalignment, like below on the right? Then rest assured that these (at least) are features not bugs and read \S\ref{tut:defaults}. \begin{forestexample}[layout=tree on top,v sep=-55ex,code left skip=-3cm,label=ex:misalignments,index={option>phantom,for step,step>children,fit,propagator>for,baseline,edge,descendants,content,pgfmath,ss 1}] \begin{forest} [,phantom,for children={l sep=1ex,fit=band, for 1={edge'=,l=0},baseline} [{l+=5mm},for descendants/.pgfmath=content [AdjP[AdvP][Adj'[Adj][PP]]]] [default [AdjP[AdvP][Adj'[Adj][PP]]]] [{l-=5mm},for descendants/.pgfmath=content [AdjP[AdvP][Adj'[Adj][PP]]]] ] \path (current bounding box.west)|-coordinate(l1)(!212.base); \path (current bounding box.west)|-coordinate(l2)(!2121.base); \path (current bounding box.east)|-coordinate(r1)(!212.base); \path (current bounding box.east)|-coordinate(r2)(!2121.base); \draw[dotted] (l1)--(r1) (l2)--(r2); \end{forest} \hspace{5cm} \begin{forest} [x forest, baseline [x[x[x[x[x[x[x[x[x[x[x[x[x]]]]]]]]]]]]] [(x)[(x)[(x)[(x)[(x)[(x)[(x)[(x)[(x)[(x)[(x)[(x)[(x)]]]]]]]]]]]]] ] \end{forest} \end{forestexample} \subsubsection{The defaults, or the hairy details of vertical alignment} \label{tut:defaults} In this section we discuss the default values of options controlling the l-alignment of the nodes. The defaults are set with top-down trees in mind, so l-alignment is actually vertical alignment. There are two desired effects of the defaults. First, the spacing between the nodes of a tree should adjust to the current font size. Second, the nodes of a given level should be vertically aligned (at the base), if possible. Let us start with the base alignment: \TikZ;'s default is to anchor the nodes at their center, while \foRest;, given the usual content of nodes in linguistic representations, rather anchors them at the base \cite[\S16.5.1]{tikzpgf2.10}. The difference is particularly clear for a ``phonological'' representation: \begin{forestexample}[index={for step,tree,option>anchor}] \begin{forest} for tree={anchor=center} [maybe[m][a][y][b][e]] \end{forest}\quad \begin{forest} [maybe[m][a][y][b][e]] \end{forest} \end{forestexample} The following example shows that the vertical distance between nodes depends on the current font size. \begin{forestexample} \hbox{\small A small tree \begin{forest} baseline [VP[DP][V'[V][DP]]] \end{forest} \normalsize and \large a large tree \begin{forest} baseline [VP[DP][V'[V][DP]]] \end{forest}} \end{forestexample} Furthermore, the distance between nodes also depends on the value of \PGF;'s |inner_sep| (which also depends on the font size by default: it equals 0.3333\,em). \[\index{l sep}=\mbox{height}(\mbox{strut})+\mbox{\texttt{inner ysep}}\] The default value of \index{s sep} depends on |inner_xsep|: more precisely, it equals double |inner_xsep|). \begin{forestexample}[index={baseline,for step,tree}] \begin{forest} baseline,for tree=draw [VP[DP][V'[V][DP]]] \end{forest} \pgfkeys{/pgf/inner sep=0.6666em} \begin{forest} baseline,for tree=draw [VP[DP][V'[V][DP]]] \end{forest} \end{forestexample} Now a hairy detail: the formula for the default \index{option>l}. \[\index{option>l}=\index{l sep}+2\cdot\mbox{\texttt{outer ysep}}+\mbox{total height}(\mbox{`dj'})\] To understand what this is all about we must first explain why it is necessary to set the default \index{option>l} at all? Wouldn't it be enough to simply set \index{l sep} (leaving \index{option>l} at 0)? The problem is that not all letters have the same height and depth. A tree where the vertical position of the nodes would be controlled solely by (a constant) \index{l sep} could result in a ragged tree (although the height of the child--parent edges would be constant). \begin{forestexample}[index={baseline,for step,step>children,no edge,option>name,descendants,option>l}] \begin{forest} [default,baseline,for children={no edge} [DP [AdjP[Adj]] [D'[D][NP,name=np]]]] \path (current bounding box.west)|-coordinate(l)(np.base); \path (current bounding box.east)|-coordinate(r)(np.base); \draw[dotted] (l)--(r); \end{forest} \begin{forest} [{l=0},baseline,for children={no edge} [DP,for descendants={l=0} [AdjP[Adj]] [D'[D][NP,name=np]]]] \path (current bounding box.west)|-coordinate(l)(np.base); \path (current bounding box.east)|-coordinate(r)(np.base); \draw[dotted] (l)--(r); \end{forest} \end{forestexample} The vertical misalignment of Adj in the right tree is a consequence of the fact that letter j is the only letter with non-zero depth in the tree. Since only \index{l sep} (which is constant throughout the tree) controls the vertical positioning, Adj, child of Ad\emph{j}P, is pushed lower than the other nodes on level 2. If the content of the nodes is variable enough (various heights and depths), the cumulative effect can be quite strong, see the right tree of example (\ref{ex:misalignments}). Setting only a default \index{l sep} thus does not work well enough in general. The same is true for the reverse possibility, setting a default \index{option>l} (and leaving \index{l sep} at 0). In the example below, the depth of the multiline node (anchored at the top line) is such that the child--parent edges are just too short if the level distance is kept constant. Sometimes, misalignment is much preferred \dots \begin{forestexample}[index={align,{value=align>center},for step,tree,l sep}] \mbox{}\begin{forest} [default [first child[a][b][c]] [second child\\[-1ex]\scriptsize(a copy), align=center[a][b][c]] ] \end{forest}\\ \begin{forest} for tree={l sep=0} [{\texttt{l sep}=0} [first child[a][b][c]] [second child\\[-1ex]\scriptsize(a copy), align=center[a][b][c]] ] \end{forest} \end{forestexample} Thus, the idea is to make \index{option>l} and \index{l sep} work as a team: \index{option>l} prevents misalignments, if possible, while \index{l sep} determines the minimal vertical distance between levels. Each of the two options deals with a certain kind of a ``deviant'' node, i.e.\ a node which is too high or too deep, or a node which is not high or deep enough, so we need to postulate what a \emph{standard} node is, and synchronize them so that their effect on standard nodes is the same. By default, \foRest; sets the standard node to be a node containing letters d and j. Linguistic representations consist mainly of letters, and in the \TeX's default Computer Modern font, d is the highest letter (not character!), and j the deepest, so this decision guarantees that trees containing only letters will look nice. If the tree contains many parentheses, like the right tree of example (\ref{ex:misalignments}), the default will of course fail and the standard node needs to be modified. But for many applications, including nodes with indices, the default works. The standard node can be changed using macro \index{forestStandardNode}; see \ref{ref:standard-node}. \subsection{Advanced option setting} \label{tut:advanced-option-setting} We have already seen that the value of options can be manipulated: in (\ref{ex:romannumeral}) we have converted numeric content from arabic into roman numerals using the \emph{wrapping} mechanism |content=\romannumeral#1|; in (\ref{ex:l*}), we have tripled the value of |l| by saying |l*=3|. In this section, we will learn about the mechanisms for setting and referring to option values offered by \foRest;. One other way to access an option value is using macro \index{forestoption}. The macro takes a single argument: an option name. (For details, see \S\ref{sec:option-reading}.) In the following example, the node's child sequence number is appended to the existing content. (This is therefore also an example of wrapping.) \begin{forestexample}[label=ex:forestoption,index={option>phantom,delay,for step,descendants,content,readonly option>n,forestoption}] \begin{forest} [,phantom,delay={for descendants={ content=#1$_{~\forestoption~{n}}$}} [c][o][u][n][t]] \end{forest} \end{forestexample} However, only options of the current node can be accessed using \index{forestoption}. To access option values of other nodes, \foRest;'s extensions to the \PGF;'s mathematical library |pgfmath|, documented in \citep[part VI]{tikzpgf2.10}, must be used. To see |pgfmath| in action, first take a look at the crazy tree on the title page, and observe how the nodes are rotated: the value given to (\TikZ;) option \texttt{rotate} is a full-fledged |pgfmath| expression yielding an integer in the range from $-30$ to $30$. Similiarly, \index{dimen+=l} adds a random float in the $[-5,5]$ range to the current value of \index{option>l}. Example (\ref{ex:spread-s}) demonstrated that information about the node, like the node's level, can be accessed within |pgfmath| expressions. All options are accessible in this way, i.e.\ every option has a corresponding |pgfmath| function. For example, we could rotate the node based on its content: \begin{forestexample}[index={delay,for step,tree,content}] \begin{forest} delay={for tree={~rotate=content~}} [30[-10[5][0]][-90[180]][90[-60][90]]] \end{forest} \end{forestexample} All numeric, dimensional and boolean options of \foRest; automatically pass the given value through |pgfmath|. If you need pass the value through |pgfmath| for a string option, use the \index{pgfmath} handler. The following example sets the node's content to its child sequence number (the root has child sequence number 0). \begin{forestexample}[index={delay,for step,tree,content,readonly option>n,pgfmath}] \begin{forest} delay={for tree={content/~.pgfmath~=int(n)}} [[[][][]][[][]]] \end{forest} \end{forestexample} As mentioned above, using |pgfmath| it is possible to access options of non-current nodes. This is achieved by providing the option function with a \index{relative node name} (see~\S\ref{ref:relative-node-names}) argument.\footnote{The form without parentheses \texttt{option\string_name} that we have been using until now to refer to an option of the current node is just a short-hand notation for \texttt{option\string_name()} --- note that in some contexts, like preceding \texttt{+} or \texttt{-}, the short form does not work! (The same seems to be true for all pgfmath functions with ``optional'' arguments.)} In the next example, we rotate the node based on the content of its parent. \begin{forestexample}[index={delay,for step,descendants,content,ss u},index>={!}] \begin{forest} delay={for descendants={rotate=content~("!u")~}} [30[-10[5][0]][-90[180]][90[-60][90]]] \end{forest} \end{forestexample} Note that the argument of the option function is surrounded by double quotation marks: this is to prevent evaluation of the relative node name as a |pgfmath| function --- which it is not. Handlers \index{wrap pgfmath arg} and \index{wrap $n$ pgfmath args} (for $n=2,\dots,8$) combine the wrapping mechanism with the |pgfmath| evaluation. The idea is to compute (most often, just access option values) arguments using |pgfmath| and then wrap them with the given macro. Below, this is used to include the number of parent's children in the index. \begin{forestexample}[index={option>phantom,delay,for step,descendants,content,readonly option>n,n children,wrap $n$ pgfmath args},index>={!}] \begin{forest} [,phantom,delay={for descendants={ ~content/.wrap 3 pgfmath args= {#1$_{#2/#3}$}{content}{n}{n_children("!u")}~}} [c][o][u][n][t]] \end{forest} \end{forestexample} Note the underscore |__| character in |n__children|: in |pgfmath| function names, spaces, apostrophes and other non-alphanumeric characters from option names are all replaced by underscores. As another example, let's make the numerals example (\ref{ex:numerals-simple}) a bit fancier. The numeral type is read off the parent's content and used to construct the appropriate control sequence (|\@arabic|, |\@roman| and |\@alph|). (Also, the numbers are not specified in content anymore: we simply read the sequence number \index{readonly option>n}. And, to save some horizontal space for the code, each child of the root is pushed further down.) \begin{forestexample}[index={delay,where option,readonly option>level,content,readonly option>n,for step,step>children,option>l,wrap $n$ pgfmath args},index>={!}] \begin{forest} delay={where level={2}{~content/.wrap 2 pgfmath args= {\csname @#1\endcsname{#2}}{content("!u")}{n}~}{}}, for children={l*=n}, [\LaTeX numerals, [arabic[][][][]] [roman[][][][]] [alph[][][][]] ] \end{forest} \end{forestexample} The final way to use |pgfmath| expressions in \foRest;: |if| clauses. In section~\ref{tut:options}, we have seen that every option has a corresponding \index{if option=\meta{option}} (and \index{where option=\meta{option}} key. However, these are just a matter of convenience. The full power resides in the general \index{conditional>if} conditional, which takes three arguments: |if=|\meta{condition}\meta{true options}\meta{false options}, where \meta{condition} can be any |pgfmath| expression (non-zero means true, zero means false). (Once again, propagator \index{propagator>where} is an abbreviation for \index{for step=\index{tree}}|={|\index{conditional>if}|=...}|.) In the following example, \index{conditional>if} option is used to orient the arrows from the smaller number to the greater, and to color the odd and even numbers differently. \forestset{random tree/.style n args={3}{% #1=max levels, #2=max children, #3=max content content/.pgfmath={random(0,#3)}, if={#1>0}{repeat={random(0,#2)}{append={[,random tree={#1-1}{#2}{#3}]}}}{}}} \begin{forestexample}[index={before typesetting nodes,for step,descendants,conditional>if,content,edge,edge label,tree},index>={!},point=if] \pgfmathsetseed{314159} \begin{forest} before typesetting nodes={ for descendants={ if={content()>content("!u")}{edge=->}{ if={content()={!}] \begin{forest} for tree={calign=fixed edge angles}, move/.style={ tikz={\draw[->] () to[out=south west,in=south] (#1);}, delay={~#1.content~={##1},content=$t$}}, [CP[][C'[C][\dots[,phantom][VP[DP][V'[V][DP,move=!r1]]]]]] \end{forest} \end{forestexample} In the following example, the content of the branching nodes is computed by \foRest;: a branching node is a sum of its children. Besides the use of the relative node setting, this example notably uses a recursive style: for each child of the node, style \keyname{calc} first applies itself to the child and then adds the result to the node; obviously, recursion is made to stop at terminal nodes. \begin{forestexample}[index={if option,n children,content,for step,step>children,delay,pgfmath},index>={!}] \begin{forest} calc/.style={if n children={0}{}{content=0,for children={ calc,~!u.content~/.pgfmath=int(content("!u")+content())}}}, delay=calc, [[[3][4][5]][[3][9]][8][[[1][2][3]]]] \end{forest} \end{forestexample} \subsection{Externalization} \label{tut:externalization} \FoRest; can be quite slow, due to the slowness of both \PGF;/\TikZ; and its own computations. However, using \emph{externalization}, the amount of time spent in \foRest; in everyday life can be reduced dramatically. The idea is to typeset the trees only once, saving them in separate PDFs, and then, on the subsequent compilations of the document, simply include these PDFs instead of doing the lenghty tree-typesetting all over again. \FoRest;'s externalization mechanism is built on top of \TikZ;'s |external| library. It enhances it by automatically detecting the code and context changes: the tree is recompiled if and only if either the code in the \index{forest} environment or the context (arbitrary parameters; by default, the parameters of the standard node) changes. To use \foRest;'s externalization facilities, say:\footnote{When you switch on the externalization for a document containing many \keyname{forest} environments, the first compilation can take quite a while, much more than the compilation without externalization. (For example, more than ten minutes for the document you are reading!) Subsequent compilations, however, will be very fast.}\index[not print]{external} \begin{lstlisting}[point=external] \usepackage[external]{forest} ~\tikzexternalize~ \end{lstlisting} If your \index{forest} environment contains some macro, you will probably want the externalized tree to be recompiled when the definition of the macro changes. To achieve this, use \index{forestset}|{|\index{external/depends on macro}|=|\cmdname{macro}|}|. The effect is local to the \TeX\ group. \TikZ;'s externalization library promises a |\label| inside the externalized graphics to work out-of-box, while |\ref| inside the externalized graphics should work only if the externalization is run manually or by |make| \citep[\S32.4.1]{tikzpgf2.10}. A bit surprisingly perhaps, the situation is roughly reversed in \foRest;. |\ref| inside the externalized graphics will work out-of-box. |\label| inside the externalized graphics will not work at all. Sorry. (The reason is that \foRest; prepares the node content in advance, before merging it in the whole tree, which is when \TikZ;'s externalization is used.) \subsection{Expansion control in the bracket parser} \label{tut:bracket} By default, macros in the bracket encoding of a tree are not expanded until nodes are being drawn --- this way, node specification can contain formatting instructions, as illustrated in section~\ref{tut:basic-usage}. However, sometimes it is useful to expand macros while parsing the bracket representation, for example to define tree templates such as the X-bar template, familiar to generative grammarians:\footnote{Honestly, dynamic node creation might be a better way to do this; see~\S\ref{ref:dynamic}.} \begin{forestexample}[index={bracketset,action character}] ~\bracketset{action character=@}~ \def\XP#1#2#3{#1P[#2][#1'[#1][#3]]} \begin{forest} [~@~\XP T{DP}{~@~\XP V{DP}{DP}}] \end{forest} \end{forestexample} In the above example, the |\XP| macro is preceded by the \emph{action character} |@|: as the result, the token following the action character was expanded before the parsing proceeded. The action character is not hard coded into \foRest;. Actually, there is no action character by default. (There's enough special characters in \foRest; already, anyway, and the situations where controlling the expansion is preferable to using the pgfkeys interface are not numerous.) It is defined at the top of the example by processing key \index{action character} in the |/bracket| path; the definition is local to the \TeX\ group. Let us continue with the description of the expansion control facilities of the bracket parser. The expandable token following the action character is expanded only once. Thus, if one defined macro |\VP| in terms of the general |\XP| and tried to use it in the same fashion as |\XP| above, he would fail. The correct way is to follow the action character by a braced expression: the braced expression is fully expanded before bracket-parsing is resumed. \begin{forestexample}[index={bracketset,action character}] \bracketset{action character=@} \def\XP#1#2#3{#1P[#2][#1'[#1][#3]]} \def\VP#1#2{\XP V{#1}{#2}} \begin{forest} [@\XP T{DP}{~@{~\VP{DP}{DP}~}~}] \end{forest} \end{forestexample} In some applications, the need for macro expansion might be much more common than the need to embed formatting instructions. Therefore, the bracket parser provides commands |@+| and |@-|: |@+| switches to full expansion mode --- all tokens are fully expanded before parsing them; |@-| switches back to the default mode, where nothing is automatically expanded. \begin{forestexample}[index={bracketset,action character}] \bracketset{action character=@} \def\XP#1#2#3{#1P[#2][#1'[#1][#3]]} \def\VP#1#2{\XP V{#1}{#2}} \begin{forest} ~@+~ [\XP T{DP}{\VP{DP}{DP}}] \end{forest} \end{forestexample} All the action commands discussed above were dealing only with \TeX's macro expansion. There is one final action command, |@@|, which yields control to the user code and expects it to call |\bracketResume| to resume parsing. This is useful to e.g.\ implement automatic node enumeration: \begin{forestexample}[index={bracketset,action character,option>phantom,delay,where option,readonly option>level,content}] \bracketset{action character=@} \newcount\xcount \def\x#1{~@@~\advance\xcount1 \edef\xtemp{[$\noexpand\times_{\the\xcount}$[#1]]}% \expandafter\bracketResume\xtemp } \begin{forest} phantom, delay={where level=1{content={\strut #1}}{}} ~@+~ [\x{f}\x{o}\x{r}\x{e}\x{s}\x{t}] \end{forest} \end{forestexample} This example is fairly complex, so let's discuss how it works. |@+| switches to the full expansion mode, so that macro |\x| can be easily run. The real magic hides in this macro. In order to be able to advance the node counter |\xcount|, the macro takes control from \foRest; by the |@@| command. Since we're already in control, we can use |\edef| to define the node content. Finally, the |\xtemp| macro containing the node specification is expanded with the resume command sticked in front of the expansion. \section{Reference} \label{sec:reference} This section documents all publicly exposed keys and macros defined by the core package. All other commands defined by the package (see the implementation typeset in |forest.doc|) are considered internal and might change without prior notice or compatibility support. \subsection{Package loading and options} \label{ref:package-options} Load the package by writing |\usepackage{forest}| in the document preamble. Field-specific definitions and defaults are stored in separate libraries. Use |\usepackage[|\meta{library name}|]{forest}| to load library \meta{library name} and its defaults alongside the main package. Loading several libraries in this way is allowed: however, if you need more control over loading the defaults, use the following macros. \begin{syntax} \indexitem{macro>useforestlibrary}[|*|][|[|\meta{options}|]|]|{|\meta{library}|,...}| Loads the given libraries. The starred version applies their defaults as well, while the starless does not. Multiple library names can be given, separated by commas. Libraries can receive \meta{options}. This macro can only be used in the preamble. \indexitem{macro>forestapplylibrarydefaults}|{|\meta{library name},...|}| Loads the default settings of \meta{library}. Multiple library names can be given, separated by commas. This macro can be used either in the preamble or in the document body. Its effect is local to the current \TeX\ scope. \end{syntax} For example, the |linguistics| library defines c-command related nodewalks, changes the default parent--child edges to south--north (the main package default is border--border) and sets the baseline to the root node. Thus, if you write |\usepackage[linguistics]{forest}| in your preamble, or use macro \cmdname{forestapplylibrarydefaults} like below, you get the following: \begin{forestexample}[tree bin=minipage,index={for step,c-commanded}] package defaults: \begin{forest} [VP % cannot use "for c-commanded" below! [DP, for sibling={for tree=red}] [V'[V][DP]] ] \end{forest}\\[1ex] |linguistics| library defaults: ~\forestapplylibrarydefaults{linguistics}~ \begin{forest} [VP [DP, ~for c-commanded~={red}] [V'[V][DP]] ] \end{forest} \end{forestexample} \begin{syntax} \indexitem(false){package option>external}|=|\alternative{true,false} Enable/disable externalization, see \S\ref{ref:externalization}. \indexitem(most){package option>compat}|=|\meta{keylist} Enter compatibility mode with previous versions of the package. If at all possible, each backwards incompatible change is given a key in the |compat| path, e.g.\ |compat=|\index{1.0-forstep} reverts to the old behaviour of spatial propagators \index{for step=\meta{step}}, where a propagator could not fail. While each compatibility feature can be enabled individually, they are grouped for ease of use. To load compatibility features since version |x|[|.y|[|.z|]], write |compat=x|[|.y|[|.z|]]|-all| or |compat=x|[|.y|[|.z|]]|-most|. The former enables all compatibility features since that release, the latter only those that are guaranteed to not disrupt any new functionality of the package. To load all compatibility features since the last major release (|x| in |x.y.z|), write |compat=|\indexdef{option of=compat>all}; to load most of them, write |compat=|\indexdef{option of=compat>most} or simply |compat|. Specifying this option also defines macro \indexdef{macro>forestcompat} (taking the same arguments as the package option) which can be used to enable compatibility features locally, within the document body. To enable compatibility mode but not enable any specific compatibility feature for the entire document, write |compat=|\indexdef{option of=compat>none} as a package option. For a list of compatibility features, see \S\ref{sec:changelog}. By default, the package warns when a compatibility feature is used. Disable this behaviour by |compat=|\indexdef{option of=compat>silent}. \indexitem(true){package option>tikzcshack}|=|\alternative{true,false} Enable/disable the hack into \TikZ;'s implicit coordinate syntax, see \S\ref{ref:relative-node-names}. \indexitem(true){package option>tikzinstallkeys}|=|\alternative{true,false} Install certain keys into the \keyname{/tikz} path. Currently: \index{fit to}. \indexitem{package option>debug}|=|\meta{debug category}[, \meta{debug category}]* Prints out some debugging info to the log file. When given no argument, prints out all the available information, otherwise only the information on the listed (comma-separated) debug categories. The available categories are listed below. \begin{syntax} \indexitem{option of=debug>nodewalks} \indexitem{option of=debug>dynamics} \end{syntax} \end{syntax} \subsection{Invocation} \label{sec:invocation} \begin{syntax} %%% begin listing region: forest_environment_doc \indexitem{environment>forest={[\texttt{(}\meta{config}\texttt{)}]\meta{tree}}} %%% end listing region: forest_environment_doc \indexitem{macro>Forest}[*][|(|\meta{config}|)|]\marg{tree} The environment and the starless version of the macro introduce a group; the starred macro does not, so the created nodes can be used afterwards, like in the example below. (Note that this will leave a lot of temporary macros lying around. This shouldn't be a problem, however, since all of them reside in the |\forest| ``namespace''.) {\tikzexternaldisable \begin{forestexample} We create a ~\Forest*~{ [DP,name=DP,baseline [D] [NP] ] } and merge it into a ~\Forest*~{ [VP,baseline [V] [,replace by=DP ] ] } \end{forestexample}} Using \meta{config}, it is possible to configure behaviour of the environment/macro prior to application of \index{style>stages}. The following keys are defined. \begin{syntax} \indexitem{option of=forest>stages}|=|\meta{style definition} overrides the contents of style \index{style>stages} for the current environment/macro. \end{syntax} \indexitem{macro>forestset}\marg{keylist} Execute \meta{keylist} (of node keys) with the default path set to \keyname{/forest}. \begin{advise} \item This macro is usually used to define \foRest; styles. \item Usually, no current node is set when this macro is called. Thus, executing most node keys in this place will fail. However, if you have some nodes lying around, you can use propagator \index{for step=\index{option>name}}|=|\meta{node name} to set the node with the given name as current. \end{advise} \end{syntax} \subsection{The bracket representation} \label{ref:bracket} A bracket representation of a tree is a token list with the following syntax: \begin{eqnarray*} \meta{tree}&=&\left[\meta{preamble}\right]\meta{node}\\ \meta{node}&=&\texttt{[}\left[\meta{content}\right]\left[\texttt{,}\meta{keylist}\right] \left[\meta{children}\right]\texttt{]}\meta{afterthought}\\ \meta{preamble}&=&\meta{keylist}\\ \meta{keylist}&=&\meta{key--value}\left[,\meta{keylist}\right]\\ \meta{key--value}&=&\meta{key}\OR\meta{key}\texttt{=}\meta{value}\\ \meta{children}&=&\meta{node}\left[\meta{children}\right] \end{eqnarray*} The \meta{preamble} keylist is stored into keylist register \index{preamble}. The \meta{keylist} of a \meta{node} is stored into keylist option \index{given options}. \meta{content} and \meta{afterthought} are normally stored by prepending and appending \index{content}|'=|\meta{content} and \index{afterthought}|=|\meta{afterthought} to \index{given options}, respectively; this is customizable via \index{content to} and redefining style \index{afterthought}. Normally, the tokens in the bracket representation are not expanded while the input is parsed. However, it is possible to control expansion. Expansion control sequences of \foRest;'s bracket parser are shown below. Note that by default, there is no \index{action character}. \begin{center} \begin{tabular}{ll} \rstyle\meta{action character}\texttt{-}&no-expansion mode (default): nothing is expanded\\ \rstyle\meta{action character}\texttt{+}&expansion mode: everything is fully expanded\\ \rstyle\meta{action character}\texttt{}\meta{token}&expand \meta{token}\\ \rstyle\meta{action character}\texttt{}\meta{\TeX-group}&fully expand \meta{\TeX-group}\\ \rstyle\meta{action character}\meta{action character}&yield control;\\&upon finishing its job, user's code should call \indexdef{macro>bracketResume} \end{tabular} \end{center} To customize the bracket parser, call \indexdef{macro>bracketset}\meta{keylist}, where the keys can be the following. \begin{syntax} \indexitem({{[%] }}){bracket key>opening bracket}|=|\meta{character} \indexitem({{{{%[ ]}}}}){bracket key>closing bracket}|=|\meta{character} \indexitem(none){bracket key>action character}|=|\meta{character} \end{syntax} By redefining the following two keys, the bracket parser can be used outside \foRest;. \begin{syntax} \indexitem{bracket key>new node}|=|\meta{preamble}\meta{node specification}\meta{csname}. Required semantics: create a new node given the preamble (in the case of a new root node) and the node specification and store the new node's id into \meta{csname}. \indexitem{bracket key>set afterthought}|=|\meta{afterthought}\meta{node id}. Required semantics: store the afterthought in the node with given id. \end{syntax} \subsection{The workflow} \label{sec:workflow} \subsubsection{Stages} \label{ref:stages} \FoRest; does its job in several stages. The default course of events is the following: \begin{enumerate} \item\label{step:parsing-bracket} The bracket representation of the tree (\S\ref{ref:bracket}) is parsed and stored in a data structure. \item\label{step:given-options} The keys given in the bracket representation are processed. In detail, \index{default preamble} is processed first, then the given \index{preamble} (both in the context of the (formal) root node) and finally the keylists given to individual nodes. The latter are processed recursively, in a depth-first, parent-first fashion. \item\label{step:typeset-nodes} Each node is typeset in its own |tikzpicture| environment, saved in a box and its measures are taken. \item\label{step:pack} The nodes of the tree are \emph{packed}, i.e.\ the relative positions of the nodes are computed so that the nodes don't overlap. That's difficult. The result: option \index{option>s} is set for all nodes. (Sometimes, the value of \index{option>l} is adjusted as well.) \item\label{step:compute-xy} Absolute positions, or rather, positions of the nodes relative to the root node are computed. That's easy. The result: options \index{x} and \index{y} are set. \item\label{step:draw-tree} The \TikZ; code that will draw the tree is produced and executed. (The nodes are drawn by using the boxes typeset in step~\ref{step:typeset-nodes}.) \end{enumerate} Stage~\ref{step:parsing-bracket} collects user input and is thus ``fixed''. However, the other stages, which do the actual work, are under user's control. First, hooks exist between the individual stages which make it possible (and easy) to change the properties of the tree between the processing stages. For a simple example, see example~(\ref{ex:adjustxy}): the manual adjustment of \index{y} can only be done after the absolute positions have been computed, so the processing of this option is deferred by \index{before drawing tree}. For a more realistic example, see the definition of style \index{GP1}: before packing, \texttt{outer xsep} is set to a high (user determined) value to keep the $\times$s uniformly spaced; before drawing the tree, the \texttt{outer xsep} is set to \texttt{0pt} to make the arrows look better. Second, the execution of the processing stages \ref{step:given-options}--\ref{step:draw-tree} is \emph{completely} under user's control. To facilitate adjusting the processing flow, the approach is twofold. The outer level: \foRest; initiates the processing by executing style \keyname{stages}, which by default executes the processing stages \ref{step:given-options}--\ref{step:draw-tree}, preceding the execution of each but the first stage by processing the keys embedded in temporal propagators \keyname{before ...} (see \S\ref{ref:temporal-propagators}). The inner level: each processing step is the sole resident of a stage-style, which makes it easy to adjust the workings of a single step. What follows is the default content of style \keyname{stages}, including the default content of the individual stage-styles. Both nicely readable and ready to copy-paste versions are given. \begin{syntax} \indexitem{style>stages} \begin{syntax} \item \index{for step=\index{root'}}|={| \item | |\index{process keylist register}|=|\index{default preamble}|,| \item | |\index{process keylist register}|=|\index{preamble} \item |}| \item \index{process keylist}|=|\index{given options} \item \index{process keylist}|=|\index{before typesetting nodes} \indexitem{style>typeset nodes stage}\hfill |{|\index{for step=\index{root'}}|=|\index{typeset nodes}|}| \item \index{process keylist}|=|\index{before packing} \indexitem{style>pack stage}\hfill |{|\index{for step=\index{root'}}|=|\index{pack}|}| \item \index{process keylist}|=|\index{before computing xy} \indexitem{style>compute xy stage}\hfill |{|\index{for step=\index{root'}}|=|\index{compute xy}|}| \item \index{process keylist}|=|\index{before drawing tree} \indexitem{style>draw tree stage}\hfill |{|\index{for step=\index{root'}}|=|\index{draw tree}|}| \end{syntax} \end{syntax} \lstinputregion{forest.dtx}{stages} Both style \keyname{stages} and the individual stage-styles may be freely modified by the user. Obviously, as a style must be redefined before it is processed, |stages| should be redefined (using macro \index{forestset}) outside the \index{forest} environment; alternatively, stages can be given as the (parenthesized) optional argument of the environment (see~\S\ref{sec:invocation}). A stage style can also be redefined in the preamble or in any of the keylists processed prior to entering that stage. Here's the list of keys used either in the default processing or useful in an alternative processing flow. \begin{syntax} \indexitem{stage>typeset nodes} \itemnosep \indexitem{stage>typeset nodes'} Typesets each node of the current node's subtree in its own |tikzpicture| environment. The result is saved in a box (which is used later, in the \index{draw tree stage}) and its measures are taken. In the |typeset_nodes'| variant, the node box's content is not overwritten if the box already exists. The order in which the nodes are typeset is controlled by nodewalk style \indexdef{nodewalk style>typeset nodes processing order} or, if this style is not defined, by \index{processing order}. \indexitem{node key>typeset node} Typesets the \emph{current} node, saving the result in the node box. This key can be useful also in the default \index{style>stages}. If, for example, the node's content is changed and the node retypeset just before drawing the tree, the node will be positioned as if it contained the ``old'' content, but have the new content: this is how the constant distance between $\times$s is implemented in the \index{GP1} style. \indexitem{stage>pack} The nodes of the tree are \emph{packed}, i.e.\ the relative positions of the nodes are computed so that the nodes don't overlap. The result: option \index{option>s} is set for all nodes; sometimes (in tier alignment and for some values of \index{calign}), the value of some nodes' \index{option>l} is adjusted as well. \indexitem{node key>pack'} ``Non-recursive'' packing: packs the children of the current node only. (Experimental, use with care, especially when combining with tier alignment.) \indexitem{stage>compute xy} Computes the positions of the nodes in the subtree relative to the current node. The results are stored into options \index{x} and \index{y}. The current node's \index{x} and \index{y} remain unchanged. \indexitem{stage>draw tree} \itemnosep \indexitem{stage>draw tree'} Produces and executed the \TikZ; code that draws the (sub)tree rooted in the current node. The procedure uses the node boxes typeset by \index{typeset nodes} or friends. The |'| variant includes the node boxes in the picture using \cmdname{copy}, not \cmdname{box}, thereby preserving them. For details and customization, see \S\ref{sec:draw-tree}. \indexitem{node key>draw tree box}|=|[\meta{\TeX\ box}] The picture drawn by the subsequent invocations of \index{draw tree} and \index{draw tree'} is put into \meta{\TeX\ box}. If the argument is omitted, the subsequent pictures are typeset normally (the default). \indexitem{node key>process keylist}|=|\meta{keylist option} For each node in the entire tree, the keylist saved in \meta{keylist option} of the node is processed (in the context of that node). Note that this key is not sensitive to the current node: it processes the keylists for the whole tree. Actually, it is possible to control which nodes are visited: \index{keylist option processing order=\meta{keylist option}} is walked if it is defined, otherwise \index{processing order}. In both cases, the processing nodewalk starts at the formal root of the tree (see \index{root'} and \index{set root}), which is reevaluated at the beginning of each internal cycle (see below). By default, \index{keylist option processing order=\meta{keylist option}} is indeed undefined, while the \index{processing order} defauls to \index{tree}, which means that \emph{all the nodes in the entire tree} are processed. Keylist-processing proceeds in cycles. In a given cycle, the value of option \meta{keylist option name} is processed for every node visited by the processing nodewalk. During a cycle, keys may be \emph{delayed} using key \index{delay}. Keys delayed in a cycle are processed in the next cycle. The number of cycles in unlimited. Dynamic creation of nodes happens between the cycles. The options given to the dynamically created nodes are implicitely delayed and thus processed at the end of the next cycle. The calls of this key should \emph{not} be nested, and it should not be embedded under \index{process keylist'} or \index{process keylist register}. When changing the processing nodewalk, note that delayed keys will be executed only for nodes visited by the processing nodewalk. Delayed spatially propagated keys will be remembered, though, and executed when the given keylist is processed for the target node. Using spatial propagators without delaying cannot result in a non-processed key. \indexitem{node key>process keylist'}|=|\meta{keylist option}\meta{nodewalk} This key is a variant of \index{process keylist}. The differences are as follows. The processing nodewalk is given explicitely (by \meta{nodewalk}) and starts at the current node (in each internal cycle). There is no dynamic creation of nodes between the delay cycles. Any dynamic node instructions will be remembered and executed after the next cycle of \index{process keylist}, or an explicit call to \index{do dynamics}. It is safe to embed this key within \index{process keylist} and (all) friends. \indexitem{node key>process keylist''}|=|\meta{keylist option}\meta{nodewalk} This key is a variant of \index{process keylist} which executes neither dynamic node operations nor delayed keys (there are thus no internal cycles). Any delayed keys will not be processed during the execution of this key. They will be remembered and executed at the end of the next cycle of \index{process keylist} or \index{process keylist'}. As for \index{process keylist'}, the processing nodewalk is given explicitely (by \meta{nodewalk}) and starts at the current node. It is safe to embed this key within \index{process keylist} and (all) friends. \indexitem{node key>process keylist register}|=|\meta{register} Process the keylist saved in \meta{register} in the context of the current node. Any delayed keys will not be processed during the execution of this key. They will be remembered and executed at the end of the next cycle of \index{process keylist} or \index{process keylist'}. It is safe to embed this key within \index{process keylist} or \index{process keylist'}. \indexitem{node key>process delayed}|=|\meta{nodewalk} Process delayed keys. Keylist \index{delay} cannot be processed using \index{process keylist} or \index{process keylist'}. Thus this key. Like \index{process keylist} or \index{process keylist'}, this key uses internal cycles. Thus, any embedded \index{delay}s will be processed. There is no dynamic creation of nodes between the delay cycles. Any dynamic node instructions will be remembered and executed after the next cycle of \index{process keylist} or \index{process keylist'}, or an explicit call to \index{do dynamics}. This key is safe to use within \index{process keylist}, \index{process keylist'} and \index{process keylist register}. \indexitem(tree){nodewalk style>processing order}|=|\meta{nodewalk} Redefine this style to change the default order in which \index{process keylist} processes a keylist option. For example, to process the nodes in a child-first fashion, write \begin{lstlisting} for nodewalk={processing order/.style={tree children first}}{} \end{lstlisting} Note that this is a \emph{nodewalk} style, so it must be defined during a nodewalk. \indexitem(processing order)[form={processing order}]{word suffix=keylist option>nodewalk style>processing order=\meta{keylist option}}|=|\meta{nodewalk} Redefine this style to change the \index{process keylist} processing order for a specific \meta{keylist option}. For example, to process \index{before drawing tree} options in the child-first fashion, leaving the processing of other |before ...| keylists untouched, write \begin{lstlisting} for nodewalk={before drawing tree processing order/.style={tree children first}}{} \end{lstlisting} \indexitem{node key>do dynamics} Experimental. Perform pending dynamic tree operations. Do not use this key within \index{process keylist} or \index{process keylist'}. \end{syntax} \subsubsection{Temporal propagators} \label{ref:temporal-propagators} Temporal propagators delay processing of given keys until some other point in the processing of the tree. There are three kinds of temporal propagators. Most of the |before_...| and |after_...| propagators defer the processing of the given keys to a hook just before some stage in the workflow (\S\ref{ref:stages}). \index{before packing node} and \index{after packing node} are special as they fire \emph{during} the packing stage. The \keyname{delay} propagator is ``internal'' to the current hook: the keys in the hook are processed cyclically, and \keyname{delay} delays the processing of the given keys until the next cycle. Formally, temporal propagators are keylist options (except \index{delay n}, which is a style), so augmented assignments are possible (\S\ref{sec:option-setting}). All temporal propagators can be nested without limit. \begin{advise} \item A note on typos. By default, all keys unknown to \foRest; are appended to keylist option \index{node options} The value of \index{node options} is fed to \TikZ; when typesetting a node, so any typos are caught by \TikZ;. However, as nodes are normally typeset in stage \index{typeset nodes stage}, any typos in keys temporally propagated past that stage will not be noticed, simply because noone will use the value of \index{node options} where they end up (the exception being nodes which are explicitely retypeset by the user using \index{typeset node}). To sum up, typos in any keys temporally propagated by \index{before packing}, \index{before packing node}, \index{after packing node}, \index{before computing xy} and \index{before drawing tree} will be silently ignored. This is probably not what you want, so double-check everything you write there. Using \index{unknown to}|=|\index{unknown key error}, it is possible to change the default behaviour. You will catch all typos if you append the command to \index{pack stage}, as shown below. This can be done either in the tree or by \index{forestset}. \begin{lstlisting} typeset nodes stage/.append style={unknown to=unknown key error} \end{lstlisting} Of course, this makes it impossible to write simply |before_drawing_tree={inner_sep=5pt,_typeset_node}|. Any |tikz|'s options must be given explicitely via \index{node options}: |before_drawing_tree={node_options={inner_sep=5pt},_typeset_node}|. \end{advise} \begin{syntax} \indexitem{propagator>delay}|=|\meta{keylist} Defers the processing of the \meta{keylist} until the next cycle. Internally, |delay| is a keylist option, so augmented operators of the \index{keylist} type can be used. To check whether any keys were delayed, use conditional \index{if have delayed}. \indexitem{propagator>delay n}|=|\meta{integer}\meta{keylist} Defers the processing of the \meta{keylist} for $n$ cycles. $n$ may be $0$, and it may be given as a |pgfmath| expression. \indexitem{propagator>given options} When \index{style>stages} processing starts, this list holds the keys given by the user in the bracket representation. \indexitem{propagator>before typesetting nodes}|=|\meta{keylist} Defers the processing of the \meta{keylist} to until just before the nodes are typeset. \indexitem{propagator>before packing}|=|\meta{keylist} \indexitem{propagator>before packing node}|=|\meta{keylist} Defers the processing of the \meta{keylist} given to the node to until just before/after the subtree of \emph{this specific node} is packed. Even before packing node, the (subtrees of the) children of the node have already been packed.\footnote{\FoRest; employs two variants of the packing algorithm: the faster one is used for (parts of) trees with uniform growth, i.e.\ subtrees where \index{grow} does not change; the slower, generic variant is used in where this is not the case. Now, the fast method works by dealing with l and s dimension separately, and it is able to do this for the entire (sub)tree, without needing to invoke the packing method for its constituents. The consequence is that there is no place where \index{before packing node} could be called meaningfully, as the node's constituents are not packed individually, ``just before packing the current node'' is the same as ``just before packing the tree'', and for many nodes packing is not called anyway in the fast method. As the rationale behind \index{before packing node} is to be able to adjust the options of the subtree based on the information gained by packing its constituents, specifying \index{before packing node} automatically switches to the generic method.} \indexitem{propagator>after packing node}|=|\meta{keylist} Defers the processing of the \meta{keylist} given to the node to until just after \emph{this specific node} is packed. \begin{forestexample}[index={after packing node,no edge,calign,value=calign>first,option>l,l sep,content format,content,if option,n children,pgfmath,option>s,max x,min x,for step,step>children,wrap pgfmath arg,typeset node,dimen+=s}] \forestset{box/.style={ draw, no edge, l=0, l sep=1.5ex, calign=first, anchor=base west, content format={\strut\forestoption{content}}, if n children=0{}{ ~after packing node~={ minimum width/.pgfmath= {s("!l")+max_x("!l")-s("!1")-min_x("!1")}, for children/.wrap pgfmath arg={s+={##1}}{0}, typeset node}}}} \begin{forest} for tree={box} [/ [home[saso[Download][TeX]][alja][joe]] [usr[bin][share]]] \end{forest} \end{forestexample} \begin{advise} \item Remember to typeset or pack the node using \index{pack'} if you have changed options influencing the typesetting or packing process. \end{advise} \indexitem{propagator>before computing xy}|=|\meta{keylist} Defers the processing of the \meta{keylist} to until just before the absolute positions of the nodes are computed. \indexitem{propagator>before drawing tree}|=|\meta{keylist} Defers the processing of the \meta{keylist} to until just before the tree is drawn. \end{syntax} \subsubsection{Drawing the tree} \label{sec:draw-tree} This section provides a detailed description of how \index{draw tree} and friends draw the tree. First, here's the default course of events. \index{draw tree} is called from style \index{draw tree stage} in the context of the formal root node. It does not draw the tree directly, but rather produces \TikZ; code that actually does the drawing. The tree-drawing instructions are enclosed in a |tikzpicture| environment and come in three parts: the (non-phantom) nodes are drawn first, followed by edges between the drawn nodes and finally the custom \TikZ; code (of all, including phantom nodes). Each of those is drawn for the entire (sub)tree of the current node, in recursive, depth-first parent-first first-child-first order. Most parts of the tree drawing procedure are customizable. Zooming in from the invocation of \index{draw tree} to the keys that produce the drawing code, the customization options are as follows. There are two ways the invocation of \index{draw tree} can differ from the default. First, \index{draw tree} can be called within the context of any node. As a first approximation, that node will become the root of the tree that is being drawn; for the whole truth, see \index{draw tree method}. Second, \index{draw tree} can be called not only at \index{draw tree stage}, but any time after the nodes to be drawn have been typeset (see \index{typeset nodes stage}) and their absolute coordinates (\index{option>x} and \index{option>y}) computed (see \index{compute xy stage}). \begin{syntax} \indexitem(\begin{tikzpicture}){node key>begin draw}|/.code=|\meta{toks: \TeX\ code} \vspace{-\parskip} \indexitem(\end{tikzpicture}){node key>end draw}|/.code=|\meta{toks: \TeX\ code} The code produced by \index{draw tree} is put in the environment specified by \keyname{begin draw} and \keyname{end draw}. Thus, it is this environment, normally a |tikzpicture|, that does the actual drawing. A common use of these keys might be to enclose the |tikzpicture| environment in a |center| environment, thereby automatically centering all trees; or, to provide the \TikZ; code to execute at the beginning and/or end of the picture. Note that \keyname{begin draw} and \keyname{end draw} are \emph{not} node options: they are |\pgfkeys|' code-storing keys \citep[\S55.4.3--4]{tikzpgf2.10}. \end{syntax} Repeating from (\S\ref{ref:stages}), there are two variants of \index{draw tree}, which differ in how they use the node boxes created by \index{typeset nodes}: \index{draw tree} includes them using \cmdname{box}, so they are gone; \index{draw tree'} uses \cmdname{copy}, so they are preserved. Next, setting \index{draw tree box} will cause the tree to be drawn in the given \TeX\ box. \begin{syntax} \indexitem{style>draw tree method} This is the heart of the tree-drawing procedure: it determines which parts of the tree are drawn and in what order. What this style does by default was already described above, but is actually best seen from the definition itself: \lstinputregion{forest.dtx}{draw_tree_method} This style may be modified by the user, but it is and should be invoked only within \index{draw tree}, by the package: \emph{do not execute this style directly!} The nodewalks occurring in the default definition of this style are, with the exception of \index{processing order}, not used anywhere else in the package. \indexitem{nodewalk style>draw tree nodes processing order} \itemnosep \indexitem{nodewalk style>draw tree edges processing order} \itemnosep \indexitem{nodewalk style>draw tree tikz processing order} For each of these nodewalk styles the following holds. If it is defined, it determines which nodes / edges / pieces of \index{tikz} code are drawn and in which order. If any of these styles is not defined, its function is taken over by \index{draw tree processing order}. By default, none of them are defined. \indexitem{nodewalk style>draw tree processing order} If this nodewalk is defined, it functions as a fallback for node-, edge- and tikz-code-specific nodewalks. If it is not defined (the default situation), it has its own fallback: \index{processing order} (which defaults to \index{tree}). \indexitem{node key>draw tree node} \itemnosep \indexitem{node key>draw tree node'} Draws the current node at location specified by \index{option>x} and \index{option>y}. The |'| variant draws the node even if it's \index{phantom}. These keys should only be used only within the definition of \index{draw tree method}. \indexitem{conditional>if node drawn}|=|\meta{nodewalk}\meta{true keylist}\meta{false keylist} Execute \meta{true keylist} if the node at the end of \meta{nodewalk} was already drawn in the current invocation of \index{draw tree}; otherwise, execute \meta{false keylist}. \indexitem{node key>draw tree edge} \itemnosep \indexitem{node key>draw tree edge'} Draws the edge from the current node to its parent, using the information in \index{edge path} and \index{edge}. The variant without |'| variant tries to be smart: it draws the edge only if both the current node and its parent have been drawn in the current invocation of \index{draw tree}. (This prevents drawing the edge from the root node and edges from or to phantom nodes.) The |'| variant is dumb. These keys should only be used only within the definition of \index{draw tree method}. \indexitem(draw tree tikz'){style>draw tree tikz} \itemnosep \indexitem{node key>draw tree tikz'} Executes the custom code stored in option \index{tikz} of the current node. By default, both keys execute the code without performing any checks. Specifically, \index{tikz} code of phantom nodes is executed. To change this behaviour easily, the user can redefine \index{draw tree tikz}, which is a style; probably, the definition will employ \index{draw tree tikz'}. For example, to execute \index{tikz} code only if the node is not \index{phantom}, write \begin{lstlisting} draw tree tikz/.style={if phantom={draw tree tikz'}{}} \end{lstlisting} These keys should only be used only within the definition of \index{draw tree method}. \end{syntax} \subsection{Node keys} \label{ref:node-keys} \FoRest; is mostly controlled using \PGF;'s key management utility |pgfkeys| \citep[\S55]{tikzpgf2.10}. Most of the keys can be given next to the content in the bracket representation of a tree (\S\ref{ref:bracket}): we call these \emph{node keys}. Some keys, notably \emph{nodewalk steps} (\S\ref{ref:nodewalks}), must be used as arguments of specific commands. Most node keys perform some operation on the \emph{current node}. When the keylist given after the content of a node is processed, the current node is set to that node. However, the current node can be temporarily changed, for example by spatial propagators (\S\ref{ref:spatial-propagators}) or, more genarally, nodewalks (\S\ref{ref:nodewalks}). The most common function that node keys perform is to set or modify an \emph{option} of the current node (\S\ref{sec:options-and-registers}), usually to determine the appearance or position of the node and its edge (\S\ref{sec:formatting-tree}), but there are also several kinds of more exotic keys like spatial (\S\ref{ref:spatial-propagators}) propagators, which temporarily change the current node, temporal (\S\ref{ref:temporal-propagators}) propagators, which delay the processing of the keylist until some other stage in the workflow, keys that dynamically create and move nodes (\S\ref{ref:dynamic}), keys that control the way \foRest; processes the tree (\S\ref{ref:stages}) etc. Finally, users can also define their own keys, either by defining |pgfkeys| styles\footnote{% Styles are a feature of the \keyname{pgfkeys} package. They are named keylists, whose usage ranges from mere abbreviations through templates to devices implementing recursion. To define a style, use \PGF;'s handler \keyname{.style} \citep[\S55.4.4]{tikzpgf2.10}: \meta{style name}\keyname{/.style=}\meta{keylist}.} \citep[\S55.4.4]{tikzpgf2.10} or using \foRest;'s option declaration mechanism (\S\ref{sec:option-declaring}). \begin{advise} \item The style definitions and option declarations given among the other keys in the bracket specification are local to the current tree (but note that \foRest;'s keylist processing, including temporal and spatial propagation, introduces no groups). To define globally accessible styles and options (well, they are always local to the current \TeX\ group), use macro \cmdname{forestset} outside the \index{forest} environment, e.g.\ in the preamble of the document. (Although \index{forestset}\meta{keylist} is currently equivalent to \cmdname{pgfkeys}\texttt{\{/forest,}\meta{keylist}\texttt{\}}, don't rely on this as it will change in some (near) future version of the package, as there is a plan to introduce namespaces \dots) \end{advise} The following subsections list the node keys which are not described elsewhere (see above): spatial propagators (\S\ref{ref:spatial-propagators}) and general-purpose node keys, i.e.\ those which don't deal with tree formatting (\S\ref{sec:node-keys-various}). \subsubsection{Spatial propagators} \label{ref:spatial-propagators} Spatial propagators pass the given \meta{keylist} to other node(s) in the tree. \FoRest; provides many spatial propagators. Almost all of them are built from long-form nodewalk steps using prefix \index{for step}. This is why the list below is so short: it only documents this prefix and the exceptions. For the list of nodewalk steps, see \S\ref{ref:nodewalks}. Spatial propagation does not change the current node: after visiting the nodes the keys are propagated to, a spatial propagator (silently) returns to the origin of the embedded nodewalk. \begin{syntax} \indexitem{word prefix=step>propagator>for=\meta{step}}|=|\meta{arg$_1$}|...|\meta{arg$_n$}\meta{keylist} Walks the (single- or multi-step) \meta{step} from the current node and executes the given \meta{keylist} at every visited node. The current node remains unchanged. \meta{step} must be a long-form nodewalk step. If it has any arguments, they should be given before the keylist, with one exception: \meta{keylist: every-step} arguments of embedded nodewalks should be omitted, as it makes no sense to provide the every-step keylist twice. For details on nodewalk steps, see \S\ref{ref:nodewalks}. Examples: \begin{itemize} \item |for_parent={l_sep+=3mm}| \item |for_n=2{circle,draw}| \item |for_nodewalk={uu2}{blue}| \item |for_tree={s_sep+=1em}| \end{itemize} \indexitem{propagator>for tree'}|=|\meta{keylist 1}\meta{keylist 2} A ``combination'' of \index{for step=\index{tree children-first}} and \index{for step=\index{tree}}. Passes the keylists to the current node and its the descendants. At each node, the \meta{keylist 1} is processed first; then, children are processed recursively; finally, \meta{keylist 2} is processed. For an example, see the definition of \index{draw brackets} from \reflibrary{linguistics}. \indexitem{propagator>for 1}, \dots\indexdef[not print]{propagator>for 2}\indexdef[not print]{propagator>for 3}\indexdef[not print]{propagator>for 4}\indexdef[not print]{propagator>for 5}\indexdef[not print]{propagator>for 6}\indexdef[not print]{propagator>for 7}\indexdef[not print]{propagator>for 8}, \indexdef{propagator>for 9}|=|\meta{keylist} \itemnosep \indexitem{propagator>for -1}, \dots\indexdef[not print]{propagator>for -2}\indexdef[not print]{propagator>for -3}\indexdef[not print]{propagator>for -4}\indexdef[not print]{propagator>for -5}\indexdef[not print]{propagator>for -6}\indexdef[not print]{propagator>for -7}\indexdef[not print]{propagator>for -8}, \indexdef{propagator>for -9}|=|\meta{keylist} Although \index{for step} normally cannot precede short forms of steps, an exception is made for \index{ss 1}\index[not print]{ss 2,ss 3,ss 4,ss 5,ss 6,ss 7,ss 8}, \dots, \index{ss 9}. (These keys will work even if the short steps are redefined.) |for |$n$ passes the \meta{keylist} to the $n$th child of the current node. |for -|$n$ starts counting at the last child. \indexitem{node key>Nodewalk}|=|\meta{keylist: config}\meta{nodewalk}\meta{keylist: every-step} Configures and executes the \meta{nodewalk}. This key is a nodekey-space copy of nodewalk step \index{step>Nodewalk}. \begin{advise} \item Use this key carefully as it can change the current node! \item The envisioned purpose of this key is to change the current node within the every-step keylist of (an outer) nodewalk, where only node keys are accepted. The config defaults (independent every-step, shared history) are set to facilitate that purpose. \end{advise} \indexitem{node key>node walk}|=|\meta{node walk} \textbf{Deprecated!!!} Please use \index{for step=\index{nodewalk}} in new code. From the old documentation: \begin{quote} This is the most general way to use a \meta{node walk}. Before starting the \meta{node walk}, key \indexdef{node key>node walk/before walk} is processed. Then, the \meta{step}s composing the \meta{node walk} are processed: making a step (normally) changes the current node. After every step, key \indexdef{node key>node walk/every step} is processed. After the walk, key \indexdef{node key>node walk/after walk} is processed. \keyname{node walk/before walk}, \keyname{node walk/every step} and \keyname{node walk/after walk} are processed with \keyname{/forest} as the default path: thus, \foRest;'s node keys can be used normally inside their definitions. \begin{advise} \item Node walks can be tail-recursive, i.e.\ you can call another node walk from \keyname{node walk/after walk} --- embedding another node walk in \keyname{node walk/before walk} or \keyname{node walk/every step} will probably fail, because the three node walk styles are not saved and restored (a node walk doesn't create a \TeX\ group). \item \keyname{every step} and \keyname{after walk} can be redefined even during the walk. Obviously, redefining \keyname{before walk} during the walk has no effect (in the current walk). \end{advise} \end{quote} \end{syntax} \subsubsection{Various} \label{sec:node-keys-various} \begin{syntax} \indexitem{style>afterthought}|=|\meta{toks} Provides the afterthought explicitely. This key is normally not used by the end-user, but rather called by the bracket parser. By default, this key is a style defined by |afterthought/.style={tikz+={#1}}|: afterthoughts are interpreted as (cumulative) \TikZ; code. If you'd like to use afterthoughts for some other purpose, redefine this style --- this will take effect even if you do it in the tree preamble. \indexitem{node key>autoforward}|=|\meta{option}\meta{keylist}, \indexdef{node key>autoforward register}|=|\meta{register}\meta{keylist} \itemnosep \indexitem{node key>autoforward'}|=|\meta{option}\meta{keylist}, \indexdef{node key>autoforward register'}|=|\meta{register}\meta{keylist} Whenever the value of an autoforwarded option or register is given or changed (via an augmented assignment), \meta{option}|=|\meta{new value} or \meta{register}|=|\meta{new value} is appended to \meta{keylist}. This can be used to ``intercept and remember'' \TikZ; options, like \index{option>anchor} and \index{rotate}. The |autoforward'| variant keeps only a single instance of \meta{option} in \meta{keylist}. If you ever need to use the non-forwarded version of the key, prefix it with word |autoforwarded|, e.g.\ |autoforwarded_rotate|. Autoforwarding is limited to the current \TeX\ group. \indexitem{node key>Autoforward}|=|\meta{option}\meta{style definition}, \indexdef{node key>Autoforward register}|=|\meta{register}\meta{style definition} This is a more generic variant of autoforwarding. After the value of an option or register autoforwarded with this key is changed, the style defined by \meta{style definition} is called with the new option/register value as its argument. \begin{forestexample}[index={Autoforward,content,node options}] \forestset{~Autoforward~={content}{node options={#1}}} \begin{forest} [red[blue][green]] \end{forest} \end{forestexample} \indexitem{node key>unautoforward}|=|\meta{option or register} Undoes the autoforwarding of the option or register made by any of the autoforwarding keys. \indexitem{node key>content to}|=|\meta{key} When parsing the bracket representation of the tree, store the given content using \meta{key}|=|\meta{content}. \indexitem{node key>copy command key}|=|\meta{pgfkey: source}\meta{pgfkey: destination} Copies the |pgf| key in a way that |.add code| and |.add style| handlers still work. \indexitem({{{{{}}}}}){register>default preamble}|=|\meta{keylist} \itemnosep \indexitem{register>preamble}|=|\meta{keylist} These registers hold the content of the default preamble and the preamble of the current tree. |preamble| is set by the bracket parser. Set |default preamble| outside the \index{forest} environment using \index{forestset}. As |default preamble| and |preamble| are not styles but keylist registers, the |#| characters do not need to be doubled: you can freely copy and paste your keylists between the node options of the root node, the preamble and the default preamble. The only difference will be the order of execution: first default preamble, then preamble, and finally the root node's options. \indexitem{node key>split}|=|\meta{toks}\meta{separator}\meta{keylist} \itemnosep \indexitem{node key>split option}|=|\meta{option}\meta{separator}\meta{keylist} \itemnosep \indexitem{node key>split register}|=|\meta{register}\meta{separator}\meta{keylist} Split \meta{toks} or the value of \meta{option} or \meta{register} at occurrences of \meta{separator} (which must be a single token), and process the keys in \meta{keylist} with the pieces of the split token list as arguments, in the order given. The keys in \meta{keylist} can be any valid keys, including augmented assignments, non-current option assignments, even \index{TeX}. The difference in the number of split values and given keys is handled gracefully. If there is not enough values, the superfluous keys are not processed; if there are too many values, the last key is called repeatedly. For an example, see \S\ref{sec:example-decision-tree}. \indexitem{node key>TeX}|=|\meta{toks: \TeX\ code} The given code is executed immediately. This can be used for e.g.\ enumerating nodes: \begin{forestexample}[point=TeX,index={TeX,delay,where option,tier,content,GP1},label=ex:enumerate] \newcount\xcount \begin{forest} GP1, delay={TeX={\xcount=0}, where tier={x}{TeX={\advance\xcount1}, content/.expanded={##1$_{\the\xcount}$}}{}} [ [O[x[f]]] [R[N[x[o]]]] [O[x[r]]] [R[N[x[e]]][x[s]]] [O[x[t]]] [R[N[x]]] ] \end{forest} \end{forestexample} \indexitem{node key>TeX'}|=|\meta{toks: \TeX\ code} This key is a combination of keys \index{TeX} and \index{TeX''}: the given code is both executed and externalized. \indexitem{node key>TeX''}|=|\meta{toks: \TeX\ code} The given code is externalized, i.e.\ it will be executed when the externalized images are loaded. The image-loading and \keyname{TeX'(')} produced code are intertwined. \indexitem{node key>typeout}|=|\meta{toks} A \foRest; version of \LaTeX\ macro |\typeout|. Useful for debugging, trust me on this one. \indexitem(node options){node key>unknown to}|=|\meta{key} Forward unknown keys to \meta{key}. \begin{advise} \item Do \emph{not} use handler |.unknown| to deal with unknown keys, as it is used internally by \foRest;, and is set up to make it possible to set options of non-current nodes (see \S\ref{sec:option-setting}). \end{advise} \indexitem{node key>unknown key error}|=|\meta{keyval} Produces an error. Write \index{unknown to}|=unknown key error| to produce an error when a key unknown to \foRest; is used. \end{syntax} \subsection{Options and registers} \label{sec:options-and-registers} \FoRest; introduces two types of data storage: \emph{node options} (or just \emph{options} for short) and \emph{registers}. Options store data related to particular nodes. Each node has its own set of option values, i.e.\ the value of an option at some node is independent of its value at other nodes: in particular, setting an option of a node does \emph{not} set this option for the node's descendants. Register values are not associated to nodes. Note that option and register keys share the same ``namespace'' (|pgfkeys| path and |pgfmath| function names) so it is not possible to have an option and a register of the same name! \subsubsection{Setting} \label{sec:option-setting} The simplest way to set the value of an option or a register is to use the key of the same name. \begin{syntax} \indexitem[index key={@@@opt},form={}]{prefix>assignment>current node=\meta{option}}|=|\meta{value} Sets the value of \meta{option} of the current node to \meta{value}. Note that option types \index{keylist} and \index{autowrapped toks} redefine this basic key. \indexitem[index key={@@@reg},form={}]{prefix>assignment>current node=\meta{register}}|=|\meta{value} Sets the value of \meta{register} to \meta{value}. Note that register types \index{keylist} and \index{autowrapped toks} redefine this basic key. \end{syntax} Options can also be set for the non-current node: \begin{syntax} % \indexitem[index key={@@@.},form={.},+print format={+result={\index(not print){relative node name}\index(not index)[+print format=definition]{relative node name}}},+index format={+result={\meta{relative node name}}}]{prefix>assignment>relative node=\meta{option}}|=|\meta{value} \indexitem[index key={@@@.},form={.},+print format={+result={\index{relative node name}}},+index format={+result={\meta{relative node name}}}]{prefix>assignment>relative node=\meta{option}}|=|\meta{value} Sets the value of \meta{option} of the node specified by \meta{relative node name} to \meta{value}. Notes: \begin{enumerate*}[(i)] \item\emph{\meta{value} is evaluated in the context of the current node.} \item In general, the resolution of \meta{relative node name} depends on the current node; see \S\ref{ref:relative-node-names}. \item \meta{option} can also be an ``augmented assignment operator'' (see below) or, indeed, any node key. \end{enumerate*} \end{syntax} Additional keys for setting and modifying the value of an option or a register exist, depending on its data type. Informally, you can think of these keys as \emph{augmented operators} known from various programming languages. \begin{syntax} \indexitem{type>toks} contains \TeX's \meta{balanced text} \citep[275]{texbook}. A toks \meta{option} additionally defines the following keys: \begin{syntax} \indexitem{suffix>augmented assignment=toks>+=\meta{option}}|=|\meta{toks} appends the given \meta{toks} to the current value of the option. \indexitem{prefix>augmented assignment=toks>+=\meta{option}}|=|\meta{toks} prepends the given \meta{toks} to the current value of the option. % \item \index{if in toks option=\meta{toks option}}| =|\meta{toks}\meta{true % keylist}\meta{false keylist} checks if \meta{toks} occurs in the option value; if it does, % \meta{true keylist} are executed, otherwise \meta{false keylist}. % \item \index{where in toks option=\meta{toks option}}| =|\meta{toks}\meta{true % keylist}\meta{false keylist} is a style equivalent to \index{for step=\index{tree}}|={|\keyname{if in }\meta{option}=\meta{toks}\meta{true keylist}\meta{false keylist}|}|: for every node in % the subtree rooted in the current node, \keyname{if in }\meta{option} is executed in % the context of that node. \end{syntax} \indexitem{type>autowrapped toks} is a subtype of \index{toks} and contains \TeX's \meta{balanced text} \citep[275]{texbook}. {\rstyle\meta{option}}|=|\meta{toks} of an autowrapped \meta{option} is redefined to \meta{option}|/|\index{wrap value}|=|\meta{toks} of a normal \meta{toks} option. Keyvals \indexdef{suffix>augmented assignment=autowrapped toks>+=\meta{option}}|=|\meta{toks} and \indexdef{prefix>augmented assignment=autowrapped toks>+=\meta{option}}|=|\meta{toks} are redefined to \index{toks+=\meta{option}}|/|\index{wrap value}|=|\meta{toks} and \index{+toks=\meta{option}}|/|\index{wrap value}|=|\meta{toks}, respectively. The normal toks behaviour can be accessed via keys \indexdef{suffix>augmented assignment=autowrapped toks>'=\meta{option}}, \indexdef{suffix>augmented assignment=autowrapped toks>+'=\meta{option}}, and \indexdef[set={print format={result/.expanded=+\forestoption{argument}',pgfkey},index form=+\meta{autowrapped toks option}',index format=pgfkey,index key=+'}]{augmented assignment>+autowrapped toks'=\meta{option}}. \indexitem{type>keylist} is a subtype of \index{toks} and contains a comma-separated list of \meta{key}[|=|\meta{value}] pairs. Augmented assignment operators \indexdef{suffix>augmented assignment=keylist>+=\meta{option}} and \indexdef{prefix>augmented assignment=keylist>+=\meta{option}} automatically insert a comma before/after the appended/prepended material. Augmented assignment operator \indexdef{suffix>augmented assignment=keylist>-=\meta{option}}|=|\meta{keylist} deletes the keys from keylist \meta{option}. \meta{keylist} specifies which keys to delete. If a key is given no value, all occurrences of that key will be deleted. If a key is given a value, only occurrences with that value will be deleted. To delete occurrences without value, use special value \indexdef{macro>forestnovalue}. (Note: if you include a key in \meta{keylist} more than once, only the last occurrence counts.) {\rstyle\meta{option}}|=|\meta{keylist} of a keylist option is redefined to \meta{option}\keyname{+}|=|\meta{keylist}. In other words, keylists behave additively by default. The rationale is that one usually wants to add keys to a keylist. The usual, non-additive behaviour can be accessed by \indexdef{suffix>augmented assignment=keylist>'=\meta{option}}|=|\meta{keylist}. Manipulating the keylist option using augmented assignments might have the side-effect of adding an empty key to the list. \indexitem{type>dimen} contains a dimension. The value given to a dimension option is automatically evaluated by |pgfmath|. In other words, {\rstyle\meta{option}}|=|\meta{value} is implicitly understood as \meta{option}|/|\index{pgfmath}|=|\meta{value}. For a \meta{dimen} option \meta{option}, the following additional keys (``augmented assignments'') are defined: \begin{syntax} \indexitem{suffix>augmented assignment=dimen>+=\meta{option}}|=|\meta{value} is equivalent to \meta{option}|=|\meta{option}|()+|\meta{value} \indexitem{suffix>augmented assignment=dimen>-=\meta{option}}|=|\meta{value} is equivalent to \meta{option}|=|\meta{option}|()-|\meta{value} \indexitem{suffix>augmented assignment=dimen>*=\meta{option}}|=|\meta{value} is equivalent to \meta{option}|=|\meta{option}|()*|\meta{value} \indexitem{suffix>augmented assignment=dimen>:=\meta{option}}|=|\meta{value} is equivalent to \meta{option}|=|\meta{option}|()/|\meta{value} \end{syntax} The evaluation of \meta{pgfmath} can be quite slow. There are two tricks to speed things up \emph{if} the \meta{pgfmath} expression is simple, i.e.\ just a \TeX\ \meta{dimen}: \begin{enumerate} \item |pgfmath| evaluation of simple values can be sped up by prepending |+| to the value \citep[\S62.1]{tikzpgf2.10}; \item use the key \indexdef{suffix>augmented assignment=dimen>'=\meta{option}}|=|\meta{value} to invoke a normal \TeX\ assignment. \end{enumerate} The two above-mentioned speed-up tricks work for the augmented assignments as well. The keys for the second, \TeX-only trick are: \indexdef{suffix>augmented assignment=dimen>'+}, \indexdef{suffix>augmented assignment=dimen>'-}, \indexdef{suffix>augmented assignment=dimen>'*}, \indexdef{suffix>augmented assignment=dimen>':} --- note that for the latter two, the value should be an integer. \indexitem{type>count} contains an integer. The additional keys and their behaviour are the same as for the \meta{dimen} options.% \indexdef[not print]{suffix>augmented assignment=count>+}% \indexdef[not print]{suffix>augmented assignment=count>-}% \indexdef[not print]{suffix>augmented assignment=count>*}% \indexdef[not print]{suffix>augmented assignment=count>:}% \indexdef[not print]{suffix>augmented assignment=count>'}% \indexdef[not print]{suffix>augmented assignment=count>'+}% \indexdef[not print]{suffix>augmented assignment=count>'-}% \indexdef[not print]{suffix>augmented assignment=count>'*}% \indexdef[not print]{suffix>augmented assignment=count>':}% \indexitem{type>boolean} contains $0$ (false) or $1$ (true). In the general case, the value given to a \meta{boolean} option is automatically parsed by |pgfmath| (just as for \meta{count} and \meta{dimen}): if the computed value is non-zero, $1$ is stored; otherwise, $0$ is stored. Note that |pgfmath| recognizes constants |true| and |false|, so it is possible to write \meta{option}|=true| and \meta{option}|=false|. If key \meta{option} is given no argument, pgfmath evaluation does not apply and a true value is set. To quickly set a false value, use key \indexdef{word prefix=boolean option>>not=\meta{option}} (with no arguments). \end{syntax} \subsubsection{Reading} \label{sec:option-reading} Option and register values can be accessed using the four macros listed below, handlers \index{handler>option} and \index{handler>register} (\S\ref{ref:handlers}) and |pgfmath| functions (\ref{ref:pgfmath}). \begin{syntax} \indexitem{macro>forestoption}|{|\meta{option}|}| \itemnosep \indexitem{macro>foresteoption}|{|\meta{option}|}| \itemnosep \indexitem{macro>forestregister}|{|\meta{register}|}| \itemnosep \indexitem{macro>foresteregister}|{|\meta{register}|}| These macros expand to the value of the given option or register. Note that \index{macro>forestoption} and \index{macro>foresteoption} expand to the value of the given option of the \emph{current node}; to access option values of a non-current node, use |pgfmath| functions. In the context of |\edef|, \cmdname{forestoption} and \cmdname{forestregister} expand precisely to the token list of the option value, while \cmdname{foresteoption} and \cmdname{foresteregister} fully expand the value. \begin{advise} \item These macros can be useful in \TeX\ code introduced by \index{TeX} or \PGF;'s handler |.expanded| \citep[\S55.4.6]{tikzpgf2.10}. \end{advise} \end{syntax} \subsubsection{Declaring} \label{sec:option-declaring} Using the following keys, users can also declare their own options and registers. The new options and registers will behave exactly like the predefined ones. Note that the declaration of an option must provide a default value, while the declaration of a register must not do that (registers are initialized to the empty string, |0pt| or |0|, as appropriate for the type). The default value of an option will be assigned to any newly created nodes; the existing nodes are not affected. \begin{syntax} \indexitem{node key>declare toks}|=|\meta{option name}\meta{default value} Declares a \meta{toks} option. \indexitem{node key>declare autowrapped toks}|=|\meta{option name}\meta{default value} Declares an \meta{autowrapped toks} option. \indexitem{node key>declare keylist}|=|\meta{option name}\meta{default value} Declares a \meta{keylist} option. \indexitem{node key>declare dimen}|=|\meta{option name}\meta{pgfmath: default value} Declares a \meta{dimen} option. \indexitem{node key>declare count}|=|\meta{option name}\meta{pgfmath: default value} Declares a \meta{count} option. \indexitem{node key>declare boolean}|=|\meta{option name}\meta{pgfmath: default value} Declares a \meta{boolean} option. \indexitem{node key>declare toks register}|=|\meta{register name} Declares a \meta{toks} register. \indexitem{node key>declare autowrapped toks register}|=|\meta{register name} Declares an \meta{autowrapped toks} register. \indexitem{node key>declare keylist register}|=|\meta{register name} Declares a \meta{keylist} register. \indexitem{node key>declare dimen register}|=|\meta{register name} Declares a \meta{dimen} register. \indexitem{node key>declare count register}|=|\meta{register name} Declares a \meta{count} register. \indexitem{node key>declare boolean register}|=|\meta{register name} Declares a \meta{boolean} register. \end{syntax} Several scratch registers are predefined: \begin{syntax} \indexitem{register>temptoksa}, \indexdef{register>temptoksb}, \indexdef{register>temptoksc}, \indexdef{register>temptoksd} Predefined \meta{toks} registers. \indexitem{register>tempkeylista}, \indexdef{register>tempkeylistb}, \indexdef{register>tempkeylistc}, \indexdef{register>tempkeylistd} Predefined \meta{keylist} registers. \indexitem{register>tempdima}, \indexdef{register>tempdimb}, \indexdef{register>tempdimc}, \indexdef{register>tempdimd}, \indexdef{register>tempdimx}, \indexdef{register>tempdimy}, \indexdef{register>tempdiml}, \indexdef{register>tempdims}, \indexdef{register>tempdimxa}, \indexdef{register>tempdimya}, \indexdef{register>tempdimla}, \indexdef{register>tempdimsa}, \indexdef{register>tempdimxb}, \indexdef{register>tempdimyb}, \indexdef{register>tempdimlb}, \indexdef{register>tempdimsb} Predefined \meta{dimen} registers. \indexitem{register>tempcounta}, \indexdef{register>tempcountb}, \indexdef{register>tempcountc}, \indexdef{register>tempcountd} Predefined \meta{count} registers. \indexitem{register>tempboola}, \indexdef{register>tempboolb}, \indexdef{register>tempboolc}, \indexdef{register>tempboold} Predefined \meta{boolean} registers. \end{syntax} \subsection{Formatting the tree} \label{sec:formatting-tree} \subsubsection{Node appearance} \label{ref:node-appearance} The following options apply at stage \index{typeset nodes}. Changing them afterwards has no effect in the normal course of events. \begin{syntax} \indexitem({{{{{}}}}}){option>align}|=|\index{value=align>left}\OR\index{value=align>center}\OR\index{value=align>right}\OR\meta{toks: tabular header} Creates a left/center/right-aligned multiline node, or a tabular node. In the \index{content} option, the lines of the node should separated by |\\| and the columns (if any) by |&|, as usual. The vertical alignment of the multiline/tabular node can be specified by option \index{base}. \begin{forestexample}[index={l sep,align,base}] \begin{forest} l sep+=2ex [special value&actual value\\\hline \indexdef{value=align>left}&||\texttt{@\{\}l@\{\}}\\ \indexdef{value=align>center}&||\texttt{@\{\}c@\{\}}\\ \indexdef{value=align>right}&||\texttt{@\{\}r@\{\}}\\ ,~align~=ll,draw [top base\\right aligned, ~align~=right,~base~=top] [left aligned\\bottom base, ~align~=left,~base~=bottom] ] \end{forest} \end{forestexample} Internally, setting this option has two effects: \begin{enumerate} \item The option value (a |tabular| environment header specification) is set. The special values \keyname{left}, \keyname{center} and \keyname{right} invoke styles setting the actual header to the value shown in the above example. \begin{advise} \item If you know that the \keyname{align} was set with a special value, you can easily check the value using \index{if in toks option=\index{align}}. \end{advise} \item Option \index{content format} is set to the following value: \begin{lstlisting} \noexpand\begin{tabular}[\forestoption{base}]{\forestoption{align}}% \forestoption{content}% \noexpand\end{tabular}% \end{lstlisting} As you can see, it is this value that determines that options \keyname{base}, \keyname{align} and \keyname{content} specify the vertical alignment, header and content of the table. \end{enumerate} \indexitem(t){option>base}|=|\meta{toks: vertical alignment} This option controls the vertical alignment of multiline (and in general, \texttt{tabular}) nodes created with \index{align}. Its value becomes the optional argument to the \texttt{tabular} environment. Thus, sensible values are \indexdef{value=base>t} (the top line of the table will be the baseline) and \indexdef{value=base>b} (the bottom line of the table will be the baseline). Note that this will only have effect if the node is anchored on a baseline, like in the default case of \index{option>anchor}|=base|. For readability, you can use \indexdef{value=base>top} and \indexdef{value=base>bottom} instead of \keyname{t} and \keyname{b}. (\keyname{top} and \keyname{bottom} are still stored as \keyname{t} and \keyname{b}.) \indexitem({{{{{}}}}}){option>content}|=|\meta{autowrapped toks} The content of the node. Normally, the value of option \keyname{content} is given implicitely by virtue of the special (initial) position of content in the bracket representation (see~\S\ref{ref:bracket}). However, the option also be set explicitely, as any other option. \begin{forestexample}[index={for step,tree,if option,readonly option>n,readonly option>n'},point={content,delay},index={content,delay}] \begin{forest} delay={for tree={ if n=1{content=L} {if n'=1{content=R} {content=C}}}} [[[][][]][[][][]]] \end{forest} \end{forestexample} Note that the execution of the \keyname{content} option should usually be delayed: otherwise, the implicitely given content (in the example below, the empty string) will override the explicitely given content. \begin{forestexample}[index={for step,tree,if option,readonly option>n,readonly option>n',content},point={content}] \begin{forest} for tree={ if n=1{content=L} {if n'=1{content=R} {content=C}}} [[[][][]][[][][]]] \end{forest} \end{forestexample} \indexitem(\forestoption{content}){option>content format}|=|\meta{toks} When typesetting the node under the default conditions (see option \index{node format}), the value of this option is passed to the \TikZ; \texttt{node} operation as its \meta{text} argument \citep[\S16.2]{tikzpgf2.10}. The default value of the option simply puts the content in the node. This is a fairly low level option, but sometimes you might still want to change its value. If you do so, take care of what is expanded when. Most importantly, if you use a formatting command such as |\textbf| in the default setting of \index{node format}, be sure to precede it with |\noexpand|. For details, read the documentation of option \index{node format} and macros \index{forestoption} and \index{foresteoption}; for an example, see option \index{align}. \indexitem{node key>math content} Changes \index{content format} so that the content of the node will be typeset in a math environment. \indexitem{node key>plain content} Resets \index{content format} to the default value. \indexitem{option>node format}|=|\meta{toks} \hfill|\noexpand\node(\forestoption{name})|\\ \mbox{}\hfill|[\forestoption{node options}]{\foresteoption{content format}};| The node is typeset by executing the expansion of this option's value in a |tikzpicture| environment. Important: the value of this option is first expanded using |\edef| and only then executed. Note that in its default value, \index{content format} is fully expanded using \index{foresteoption}: this is necessary for complex content formats, such as |tabular| environments. This is a low level option. Ideally, there should be no need to change its value. If you do, note that the \TikZ; node you create should be named using the value of option \index{option>name}; otherwise, parent--child edges can't be drawn, see option \index{edge path}. \indexitem{node key>node format'}|=|\meta{toks} Sets \index{node format}, automatically wrapping the given \meta{toks} by |\noexpand\node(\forestoption{name})| and |;|. Only the node options and content must therefore be given. \indexitem(anchor=base){option>node options}|=|\meta{keylist} When the node is being typeset under the default conditions (see option \index{node format}), the content of this option is passed to \TikZ; as options to the \TikZ; |node| operation \citep[\S16]{tikzpgf2.10}. This option is rarely manipulated manually: almost all options unknown to \foRest; are automatically appended to \keyname{node options}. Exceptions are (i) \index{label} and \index{pin}, which require special attention in order to work; and (ii) \index{option>anchor}, which is saved in order to retain the information about the selected anchor. \begin{forestexample}[index={for step,descendants,option>anchor,option>child anchor,option>parent anchor,grow,l sep,tree,propagator>where,delay,content,node options,pgfmath}] \begin{forest} for descendants={anchor=east,child anchor=east}, grow=west,anchor=north,parent anchor=north, l sep=1cm, for tree={~fill=yellow~},where={n()<=3}{~draw=red~}{}, delay={for tree={content/.pgfmath=~node_options~}} [root,rotate=90, [,~fill=white~] [,~node options'~] [] [] [,~node options~={~ellipse~}] ] \end{forest} \end{forestexample} \indexitem(false){option>phantom}|=|\meta{boolean} A phantom node and its surrounding edges are taken into account when packing, but not drawn. (This option applies in stage \index{draw tree}.) \begin{forestexample}[point=phantom,index={option>phantom}] \begin{forest} [VP[DP][V',phantom[V][DP]]] \end{forest} \end{forestexample} \end{syntax} \subsubsection{Node position} \label{ref:ref-node-position} Most of the following options apply at stage \index{pack}. Changing them afterwards has no effect in the normal course of events. (Options \index{option>l}, \index{option>s}, \index{x}, \index{y} and \index{option>anchor} are exceptions; see their documentation for details). \begin{syntax} \indexitem(base){option>anchor}|=|\meta{toks: \foRest; anchor} While this option is saved by \foRest;, it is essentially an option of \TikZ;'s |\node| command \citep[see][\S16.5.1]{tikzpgf2.10}. \FoRest; \index{autoforward}s it to keylist option \index{node options}, which is passed on to \TikZ;'s |\node| command when the node is typeset. (Option |anchor| thus normally applies in stage \index{typeset nodes}.) In the \TikZ; code, you can refer to the node's anchor using \FoRest;'s anchor \index{anchor>anchor}; this anchor is sometimes also called the node anchor in this documentation, to distinguish it clearly from parent and child anchors. \meta{toks: \foRest; anchor} can be any \TikZ; anchor. Additionally, \FoRest; defines several tree hierarchy related anchors; for details, see \S\ref{sec:anchors}. The effect of setting the node anchor is twofold: \begin{itemize} \item during packing, the anchors of all siblings are \index{option>l}-aligned; \item some \index{calign} methods use node anchors (of the parent and/or certain children) to \index{option>s}-align the block of children to the parent. \end{itemize} \indexitem(center){option>calign}|=|\alternative{child,child edge,midpoint,edge midpoint,fixed angles,fixed edge angles}\\\alternative{first,last,center}. The packing algorithm positions the children so that they don't overlap, effectively computing the minimal distances between the node anchors of the children. This option (\keyname{calign} stands for child alignment) specifies how the children are positioned with respect to the parent (while respecting the above-mentioned minimal distances). The child alignment methods refer to the primary and the secondary child, and to the primary and the secondary angle. These are set using the keys described just after \keyname{calign}. \let\outerleftmargin\leftmargin \begin{syntax} \item\keyname{calign}|=|\indexdef{value=calign>child} s-aligns the node anchors of the parent and the primary child. \item\keyname{calign}|=|\indexdef{value=calign>child edge} s-aligns the parent anchor of the parent and the child anchor of the primary child. \item \keyname{calign}|=|\indexdef{value=calign>first} is an abbreviation for |calign=child,calign_child=1|. \item \keyname{calign}|=|\indexdef{value=calign>last} is an abbreviation for |calign=child,calign_child=-1|. \item\keyname{calign}|=|\indexdef{value=calign>midpoint} s-aligns the parent's node anchor and the midpoint between the primary and the secondary child's node anchor. \item\keyname{calign}|=|\indexdef{value=calign>edge midpoint} s-aligns the parent's parent anchor and the midpoint between the primary and the secondary child's child anchor. \item \keyname{calign}|=|\indexdef{value=calign>center} is an abbreviation for\\ |calign=midpoint,| |calign_primary_child=1,| |calign_secondary_child=-1|. \begin{forestexample} \begin{forest} [center,calign=center[1] [first,calign=first[A][B][C]][3][4][5][6] [last,calign=last[A][B][C]][8]] \end{forest} \end{forestexample} \item\keyname{calign}|=|\indexdef{value=calign>fixed angles}: The angle between the direction of growth at the current node (specified by option \index{grow}) and the line through the node anchors of the parent and the primary/secondary child will equal the primary/secondary angle. To achieve this, the block of children might be spread or further distanced from the parent. \item\keyname{calign}|=|\indexdef{value=calign>fixed edge angles}: The angle between the direction of growth at the current node (specified by option \index{grow}) and the line through the parent's parent anchor and the primary/secondary child's child anchor will equal the primary/secondary angle. To achieve this, the block of children might be spread or further distanced from the parent. \begin{forestexample}[point=calign,index={calign,fixed edge angles,calign primary angle,calign secondary angle,for step,tree,option>l},index>={!}] \begin{forest} calign=fixed edge angles, calign primary angle=-30,calign secondary angle=60, for tree={l=2cm} [CP[C][TP]] \draw[dotted] (!1) -| coordinate(p) () (!2) -| (); \path ()--(p) node[pos=0.4,left,inner sep=1pt]{-30}; \path ()--(p) node[pos=0.1,right,inner sep=1pt]{60}; \end{forest} \end{forestexample} \end{syntax} \indexitem{node key>calign child}|=|\meta{count} is an abbreviation for \index{calign primary child}|=|\meta{count}. \indexitem(1){option>calign primary child}|=|\meta{count} Sets the primary child. (See \index{calign}.) \meta{count} is the child's sequence number. Negative numbers start counting at the last child. \indexitem(-1){option>calign secondary child}|=|\meta{count} Sets the secondary child. (See \index{calign}.) \meta{count} is the child's sequence number. Negative numbers start counting at the last child. \indexitem{node key>calign angle}|=|\meta{count} is an abbreviation for: \index{calign primary angle}|=-|\meta{count}, \index{calign secondary angle}|=|\meta{count}. \indexitem(-35){option>calign primary angle}|=|\meta{count} Sets the primary angle. (See \index{calign}.) \indexitem(35){option>calign secondary angle}|=|\meta{count} Sets the secondary angle. (See \index{calign}.) \indexitem{node key>calign with current} s-aligns the node anchors of the current node and its parent. This key is an abbreviation for:\indexex[not print]{wrap pgfmath arg,calign,child,calign primary child} \begin{lstlisting} for parent/.wrap pgfmath arg={calign=child,calign primary child=##1}{n}. \end{lstlisting} \indexitem{node key>calign with current edge} s-aligns the child anchor of the current node and the parent anchor of its parent. This key is an abbreviation for:\indexex[not print]{wrap pgfmath arg,calign,child edge,calign primary child} \begin{lstlisting} for parent/.wrap pgfmath arg={calign=child edge,calign primary child=##1}{n} \end{lstlisting} \indexitem(tight){option>fit}|=|\alternative{tight,rectangle,band} \begin{forestexample}[layout=export,basicstyle=\footnotesize\ttfamily] \makeatletter\tikzset{use path/.code={\tikz@addmode{\pgfsyssoftpath@setcurrentpath#1} \appto\tikz@preactions{\let\tikz@actions@path#1}}}\makeatother \forestset{show boundary/.style={ before drawing tree={get min s tree boundary=\minboundary, get max s tree boundary=\maxboundary}, tikz+={\draw[red,use path=\minboundary]; \draw[red,use path=\maxboundary];}}} \end{forestexample} \input{\jobname.tmp} This option sets the type of the (s-)boundary that will be computed for the subtree rooted in the node, thereby determining how it will be packed into the subtree rooted in the node's parent. There are three choices:\footnote{Below is the definition of style \keyname{show boundary}. The \keyname{use path} trick is adjusted from \TeX\ Stackexchange question \href{http://tex.stackexchange.com/questions/26382/calling-a-previously-named-path-in-tikz}{Calling a previously named path in tikz}. \box\codebox} \begin{itemize} \item\keyname{fit}|=|\indexdef{value=fit>tight}: an exact boundary of the node's subtree is computed, resulting in a compactly packed tree. Below, the boundary of subtree L is drawn. \begin{forestexample}[point={fit,tight},index={fit,{value=fit>tight},delay,for step,tree,option>name,content,pgfmath}] \begin{forest} delay={for tree={name/.pgfmath=content}} [root [L,fit=tight, % default show boundary [L1][L2][L3]] [R] ] \end{forest} \end{forestexample} \makeatletter\tikzset{use path/.code={% \tikz@addmode{\pgfsyssoftpath@setcurrentpath#1}% \appto\tikz@preactions{\let\tikz@actions@path#1}% }}\makeatother \item\keyname{fit}|=|\indexdef{value=fit>rectangle}: puts the node's subtree in a rectangle and effectively packs this rectangle; the resulting tree will usually be wider. \begin{forestexample}[point={fit,rectangle},index={fit,{value=fit>rectangle},delay,for step,tree,option>name,content,pgfmath}] \begin{forest} delay={for tree={name/.pgfmath=content}} [root [L,fit=rectangle, show boundary [L1][L2][L3]] [R] ] \end{forest} \end{forestexample} \item\keyname{fit}|=|\indexdef{value=fit>band}: puts the node's subtree in a rectangle of ``infinite depth'': the space under the node and its descendants will be kept clear. \begin{forestexample}[point={fit,band},index={fit,{value=fit>band},delay,for step,tree,option>name,content,pgfmath}] \begin{forest} delay={for tree={name/.pgfmath=content}} [root [L[L1][L2][L3]] [C,fit=band] [R[R1][R2][R3]] ] \draw[thin,red] (C.south west)--(C.north west) (C.north east)--(C.south east); \draw[thin,red,dotted] (C.south west)--+(0,-1) (C.south east)--+(0,-1); \end{forest} \end{forestexample} \end{itemize} \indexitem(270){option>grow}|=|\meta{count}, \indexdef{node key>grow'}|=|\meta{count}, \indexdef{node key>grow''}|=|\meta{count} The direction of the tree's growth at the node. The growth direction is understood as in \TikZ;'s tree library \citep[\S18.5.2]{tikzpgf2.10} when using the default growth method: the (node anchor's of the) children of the node are placed on a line orthogonal to the current direction of growth. (The final result might be different, however, if \index{option>l} is changed after packing or if some child undergoes tier alignment.) This option is essentially numeric (|pgfmath| function \keyname{grow} will always return an integer), but there are some twists. The growth direction can be specified either numerically or as a compass direction (|east|, |north east|, \dots). Furthermore, like in \TikZ;, setting the growth direction using key \keyname{grow} additionally sets the value of option \index{reversed} to |false|, while setting it with \keyname{grow'} sets it to |true|; to change the growth direction without influencing \index{reversed}, use key \index{grow''}. Between stages \index{pack} and \index{compute xy}, the value of \keyname{grow} should not be changed. \begin{forestexample}[index={delay,where in toks option,content,for step,current,grow,grow',grow'',pgfmath}] \begin{forest} delay={where in content={~grow~}{ for current/.pgfmath=content, content=\texttt{#1} }{} } [{~grow~=south} [{~grow'~=west}[1][2][3] [{~grow''~=90}[1][2][3]]] [2][3][4] [{~grow~=east}[1][2][3] [{~grow''~=90}[1][2][3]]]] \end{forest} \end{forestexample} \indexitem(false){option>ignore}|=|\meta{boolean} If this option is set, the packing mechanism ignores the node, i.e.\ it pretends that the node has no boundary. Note: this only applies to the node, not to the tree. Maybe someone will even find this option useful for some reason \dots \indexitem(false){option>ignore edge}|=|\meta{boolean} If this option is set, the packing mechanism ignores the edge from the node to the parent, i.e.\ nodes and other edges can overlap it. (See \S\ref{sec:bugs} for some problematic situations.) \begin{forestexample}[index={ignore edge,option>l,dimen*=l}] \begin{forest} [A[B[B][B][B][B]][C [\texttt{not ignore edge},l*=2]]] \end{forest} \begin{forest} [A[B[B][B][B][B]][C [\texttt{ignore edge},l*=2,~ignore edge~]]] \end{forest} \end{forestexample} \indexitem{option>l}|=|\meta{dimen} The l-position of the node, in the parent's ls-coordinate system. (The origin of a node's ls-coordinate system is at its (node) anchor. The l-axis points in the direction of the tree growth at the node, which is given by option \index{grow}. The s-axis is orthogonal to the l-axis; the positive side is in the counter-clockwise direction from |l| axis.) The initial value of \keyname{l} is set from the standard node. By default, it equals: \[\index{l sep}+2\cdot\mbox{\texttt{outer ysep}}+\mbox{total height(standard node)}\] The value of \keyname{l} can be changed at any point, with different effects. \begin{itemize} \item The value of \keyname{l} at the beginning of stage \index{pack} determines the minimal l-distance between the anchors of the node and its parent. Thus, changing \keyname{l} before packing will influence this process. (During packing, \keyname{l} can be increased due to parent's \index{l sep}, tier alignment, or \index{calign} methods \index{fixed angles} and \index{fixed edge angles}. \item Changing \keyname{l} after packing but before stage \index{compute xy} will result in a manual adjustment of the computed position. (The augmented assignment operators can be useful here.) \item Changing \keyname{l} after the absolute positions have been computed has no effect in the normal course of events. \end{itemize} \indexitem{option>l sep}|=|\meta{dimen} The minimal l-distance between the node and its descendants. This option determines the l-distance between the \emph{boundaries} of the node and its descendants, not node anchors. The final effect is that there will be a \keyname{l sep} wide band, in the l-dimension, between the node and all its descendants. The initial value of \keyname{l sep} is set from the standard node and equals \[\mbox{height}(\mbox{strut})+\mbox{\texttt{inner ysep}}\] Note that despite the similar name, the semantics of \keyname{l sep} and \keyname{s sep} are quite different. \indexitem(false){option>reversed}|=|\meta{boolean} If |false|, the children are positioned around the node in the counter-clockwise direction; if |true|, in the clockwise direction. See also \index{grow}. \indexitem(0){option>rotate}|=|\meta{count} This option is saved and \index{autoforward}ed to \TikZ;'s |\node| command via \index{node options}. \indexitem{option>s}|=|\meta{dimen} The s-position of the node, in the parent's ls-coordinate system. (The origin of a node's ls-coordinate system is at its (node) anchor. The l-axis points in the direction of the tree growth at the node, which is given by option \index{grow}. The s-axis is orthogonal to the l-axis; the positive side is in the counter-clockwise direction from |l| axis.) The value of \keyname{s} is computed in stage \index{pack stage} and used in stage \index{compute xy stage}, so it only makes sense to (inspect and) change it in \index{before computing xy} and during packing (\index{before packing node} and \index{after packing node}). \emph{Any value given before packing is overridden, and changing the value after computing xy has no effect.} For example, consider the manual correction below. By default, B is closer to A than C because packing proceeds from the first to the last child --- the position of B would be the same if there was no C. Adjusting \keyname{s} at the right moment, it is easy to center B between A and C. \begin{forestexample}[tree bin=minipage,point=s,index={before computing xy,option>s}] \begin{forest} [no manual correction of B [A[1][2][3][4]] [B] [C[1][2][3][4]] ] \end{forest} \begin{forest} [manual correction of B [A[1][2][3][4]] [B,before computing xy={s/.average={s}{siblings}}] [C[1][2][3][4]] ] \end{forest} \end{forestexample} \indexitem{option>s sep}|=|\meta{dimen} The subtrees rooted in the node's children will be kept at least \keyname{s sep} apart in the s-dimension. Note that \keyname{s sep} is about the minimal distance between node \emph{boundaries}, not node anchors. The initial value of \keyname{s sep} is set from the standard node and equals $2\cdot\mbox{\texttt{inner xsep}}$. Note that despite the similar name, the semantics of \keyname{s sep} and \keyname{l sep} are quite different. \indexitem({{{{{}}}}}){option>tier}|=|\meta{toks} Setting this option to something non-empty ``puts a node on a tier.'' All the nodes on the same tier are aligned in the l-dimension. Tier alignment across changes in growth direction is impossible. In the case of incompatible options, \foRest; will yield an error. Tier alignment also does not work well with \index{calign}|=|\index{fixed angles} and \index{calign}|=|\index{fixed edge angles}, because these child alignment methods may change the l-position of the children. When this might happen, \foRest; will yield a warning. \indexitem{option>x}=\meta{dimen} \vspace{-\parskip} \indexitem{option>y}=\meta{dimen} \keyname{x} and \keyname{y} are the coordinates of the node in the ``normal'' (paper) coordinate system, relative to the root of the tree that is being drawn. So, essentially, they are absolute coordinates. The values of \keyname{x} and \keyname{y} are computed in stage \index{compute xy}. It only makes sense to inspect and change them (for manual adjustments) afterwards (normally, in the \index{before drawing tree} hook, see \S\ref{ref:stages}.) \keyname{x} and \keyname{y} of the (formal) root node are exceptions, as they are not changed in stage \index{compute xy}. \begin{forestexample}[label=ex:adjustxy,index={y,dimen-,grow',option>l,for step,tree,before drawing tree}] \begin{forest} for tree={grow'=45,l=1.5cm} [A[B][C][D,before drawing tree={~y-~=4mm}[1][2][3][4][5]][E][F]] \end{forest} \end{forestexample} \end{syntax} \subsubsection{Edges} \label{ref:ref-edge} These options determine the shape and position of the edge from a node to its parent. They apply at stage \index{draw tree}. \begin{syntax} \indexitem({{{{{}}}}}){option>child anchor}|=|\meta{toks: \foRest; anchor} See \index{option>parent anchor}. \indexitem(draw){option>edge}|=|\meta{keylist} When \index{edge path} has its default value, the value of this option is passed as options to the \TikZ; |\path| expression used to draw the edge between the node and its parent. Also see key \index{no edge}. \begin{forestexample}[point=edge,index={edge,no edge,for step,tree,grow',option>l,option>anchor,option>child anchor}] \begin{forest} for tree={grow'=0,l=2cm,anchor=west,child anchor=west}, [root [normal] [none,~no~ edge] [dotted,edge=dotted] [dashed,edge=dashed] [dashed,edge={dashed,red}] ] \end{forest} \end{forestexample} \indexitem({{{{{}}}}}){option>edge label}|=|\meta{toks: \TikZ; code} When \index{edge path} has its default value, the value of this option is used at the end of the edge path specification to typeset a node (or nodes) along the edge. The packing mechanism is not sensitive to edge labels. \begin{forestexample}[index={edge label}] \begin{forest} [VP [V,~edge label~={node[midway,left,font=\scriptsize]{head}}] [DP,~edge label~={node[midway,right,font=\scriptsize]{complement}}] ] \end{forest} \end{forestexample} \indexitem{option>edge path}|=|\meta{toks: \TikZ; code} \hfill |\noexpand\path[|\index{forestoption}|{|\index{edge}|}]|\\ \mbox{}\hfill |(|\index>{!}|u.|\index{anchor>parent anchor}|)--(.|\index{anchor>child anchor}|)|\index{forestoption}|{|\index{edge label}|};| This option contains the code that draws the edge from the node to its parent. By default, it creates a path consisting of a single line segment between the node's \index{option>child anchor} and its parent's \index{option>parent anchor}. Options given by \index{edge} are passed to the path; by default, the path is simply drawn. Contents of \index{edge label} are used to potentially place a node (or nodes) along the edge. When specifying the edge path, the values of options \index{edge} and \index{edge label} can be used. Furthermore, two anchors, \index{anchor>parent anchor} and \index{anchor>child anchor}, are defined, to facilitate access to options \index{option>parent anchor} and \index{option>child anchor} from the \TikZ; code. The node positioning algorithm is sensitive to edges, i.e.\ it will avoid a node overlapping an edge or two edges overlapping. However, the positioning algorithm always behaves as if the \keyname{edge path} had the default value --- \emph{changing the \keyname{edge path} does not influence the packing!} Sorry. (Parent--child edges can be ignored, however: see option \index{ignore edge}.) \indexitem{node key>edge path'}|=|\meta{toks: \TikZ; code} Sets \index{edge path}, automatically wrapping the given path by |\noexpand\path[\forestoption{edge}]| and |\forestoption{edge label};|. \indexitem({{{{{}}}}}){option>parent anchor}|=|\meta{toks: \foRest; anchor} (Information also applies to option \index{option>child anchor}.) \FoRest; defines anchors \index{anchor>parent anchor} and \index{anchor>child anchor} (which work only for \foRest; and not also \TikZ; nodes, of course) to facilitate reference to the desired endpoints of child--parent edges. Whenever one of these anchors is invoked, it looks up the value of the \keyname{parent anchor} or \keyname{child anchor} of the node named in the coordinate specification, and forwards the request to the (\TikZ;) anchor given as the value. The intended use of the two anchors is chiefly in \index{edge path} specification, but they can used in any \TikZ; code. \begin{forestexample}[index={anchor>parent anchor,anchor>child anchor,for step,tree},index>={!}] \begin{forest} for tree={~parent anchor~=south,~child anchor~=north} [VP[V][DP]] \path[fill=red] (.parent anchor) circle[radius=2pt] (!1.child anchor) circle[radius=2pt] (!2.child anchor) circle[radius=2pt]; \end{forest} \end{forestexample} The empty value (which is the default) is interpreted as in \TikZ;: as an edge to the appropriate border point. See also \S\ref{sec:anchors} for a list of additional anchors defined by \foRest;. \indexitem{node key>no edge} Clears the edge options (\index{edge}|'={}|) and sets \index{ignore edge}. \end{syntax} \subsubsection{Information about node} \label{ref:node-info} \label{ref:readonly-options} The values of these options provide various information about the tree and its nodes. \begin{syntax} \indexitem{node key>alias}|=|\meta{toks} \itemnosep \indexitem{node key>alias'}|=|\meta{toks} Sets the alias for the node's name. Unlike \index{option>name}, \keyname{alias} is \emph{not} an option: you cannot e.g.\ query it's value via a |pgfmath| expression. If the given alias clashes with an existing node name, |alias| will yield an error, while |alias'| will silently rename the node with this name to its default (|node@|\meta{id}) value. Use an empty argument to reset the node's name to its default value. Aliases can be used as the \meta{forest node name} part of a relative node name and as the argument to the \index{step>name} step of a node walk. The latter includes the usage as the argument of the \index{for step=\index{step>name}} propagator. Technically speaking, \foRest; alias is \emph{not} a \TikZ; alias! However, you can still use it as a ``node name'' in \TikZ; coordinates, since \foRest; hacks \TikZ;'s implicit node coordinate system to accept relative node names; see \S\ref{ref:forest-cs}. \indexitem{readonly option>id}|=|\meta{count} The internal id of the node. \indexitem{readonly option>level}|=|\meta{count} The hierarchical level of the node. The root is on level $0$. \indexitem{readonly option>max x}|=|\meta{dimen} \vspace{-\parskip} \indexitem{readonly option>max y}|=|\meta{dimen} \vspace{-\parskip} \indexitem{readonly option>min x}|=|\meta{dimen} \vspace{-\parskip} \indexitem{readonly option>min y}|=|\meta{dimen} Measures of the node, in the shape's coordinate system \citep[see][\S16.2,\S48,\S75]{tikzpgf2.10} shifted so that the node anchor is at the origin. In |pgfmath| expressions, these options are accessible as |max__x|, |max__y|, |min__x| and |min__y|. \indexitem{readonly option>n}|=|\meta{count} The child's sequence number in the list of its parent's children. The enumeration starts with 1. For a geometric root, \keyname{n} equals $0$. \indexitem{readonly option>n'}|=|\meta{count} Like \index{readonly option>n}, but starts counting at the last child. In |pgfmath| expressions, this option is accessible as |n__|. \indexitem(node@\meta{id}){option>name}|=|\meta{toks} \itemnosep \indexitem{node key>name'}|=|\meta{toks} Sets the name of the node. The expansion of \meta{toks} becomes the \meta{forest node name} of the node. The \TikZ; node created from the \foRest; node will get the name specified by this option. Node names must be unique. If a node with the given name already exists, |name| will yield an error, while |name'| will silently rename the node with this name to its default (|node@|\meta{id}) value. Use an empty argument to reset the node's name to its default value. \indexitem{readonly option>n children}|=|\meta{count} The number of children of the node. In |pgfmath| expressions, this option is accessible as |n__children|. \end{syntax} \subsubsection{Various} \begin{syntax} \indexitem{node key>baseline} The node's anchor becomes the baseline of the whole tree \citep[cf.][\S69.3.1]{tikzpgf2.10}. In plain language, when the tree is inserted in your (normal \TeX) text, it will be vertically aligned to the anchor of the current node. Behind the scenes, this style sets the alias of the current node to \keyname{forest@baseline@node}. \begin{forestexample}[index={baseline,use as bounding box'}] {\tikzexternaldisable Baseline at the \begin{forest} [parent,~baseline~,use as bounding box' [child]] \end{forest} and baseline at the \begin{forest} [parent [child,~baseline~,use as bounding box']] \end{forest}.} \end{forestexample} \indexitem>{tikz key>fit to}|=|\meta{nodewalk} Fits the \TikZ; node to the nodes in the given \meta{nodewalk}. This key should be used like \keyname{/tikz/fit} of the \TikZ;'s fitting library \citep[see][\S34]{tikzpgf2.10}: as an option to \emph{\TikZ;'s} |node| operation, the obvious restriction being that \keyname{fit to} must be used in the context of some \foRest; node. For an example, see footnote~\ref{fn:fit-to-tree}. This key works by calling \keyname{/tikz/fit} and providing it with the the coordinates of the subtree's boundary. The \meta{nodewalk} inherits its history from the outer nodewalk (if there is one). Its every-step keylist is empty. \indexitem{node key>get min s tree boundary}|=|\meta{cs} \vspace{-\parskip} \indexitem{node key>get max s tree boundary}|=|\meta{cs} Puts the boundary computed during the packing process into the given \meta{cs}. The boundary is in the form of \PGF; path. The |min| and |max| versions give the two sides of the node. For an example, see how the boundaries in the discussion of \index{fit} were drawn. \indexitem{option>label}|=|\meta{toks: \TikZ; node} The current node is labelled by a \TikZ; node. The label is specified as a \TikZ; option \texttt{label} \citep[\S16.10]{tikzpgf2.10}. Technically, the value of this option is passed to \TikZ;'s as a late option \citep[\S16.14]{tikzpgf2.10}. (This is so because \foRest; must first typeset the nodes separately to measure them (stage \index{typeset nodes}); the preconstructed nodes are inserted in the big picture later, at stage \index{draw tree}.) Another option with the same technicality is \index{pin}. \indexitem{option>pin}|=|\meta{toks: \TikZ; node} The current node gets a pin, see \citep[\S16.10]{tikzpgf2.10}. ¸ The technical details are the same as for \index{label}. \indexitem{node key>use as bounding box} The current node's box is used as a bounding box for the whole tree. \indexitem{node key>use as bounding box'} Like \index{use as bounding box}, but subtracts the (current) inner and outer sep from the node's box. For an example, see \index{baseline}. \indexitem({{{{{}}}}}){option>tikz}|=|\meta{toks: \TikZ; code} ``Decorations.'' The code given as the value of this option will be included in the |tikzpicture| environment used to draw the tree. The code given to various nodes is appended in a depth-first, parent-first fashion. The code is included after all nodes of the tree have been drawn, so it can refer to any node of the tree. Furthermore, relative node names can be used to refer to nodes of the tree, see \S\ref{ref:relative-node-names}. By default, bracket parser's afterthoughts feed the value of this option. See \index{afterthought}. \end{syntax} \subsection{Nodewalks} \label{ref:nodewalks} A \emph{nodewalk} is a sequence of \emph{steps} describing a path through the tree. Most steps are defined relative to the current node, for example \index{step>parent} steps to the parent of the current node, and \index{step>n}|=2| steps to the second child of the current node, where ``to make a step'' means to change the current node. Thus, nodewalk |parent,_parent,_n=2| describes the path which first steps to the parent of the \emph{origin} node, then to its grandparent and finally to the second child of the origin's grandparent. The origin of the nodewalk depends on how the nodewalk is invoked. When used after the \index>{!} in a relative node name (\S\ref{ref:relative-node-names}), the origin is the node with the name given before \index>{!}; when invoked by a spatial propagator such as \index{for step=\index{nodewalk}} (\S\ref{ref:spatial-propagators}), the origin is the current node. Formally, a \meta{nodewalk} is a list of |pgfkeys| key--value pairs. Steps in a nodewalk are thus separated by commas. However, \foRest; also recognizes \emph{short-form} steps, whose names consist of a single character and which do not need to be separated by a comma. For example, nodewalk |parent,_parent,_n=2| can be concisely written as |uu2|. Long and short forms can be mixed freely, like this: |next,_uu2,_previous|. Besides nodewalk keys, a \meta{nodewalk} can also contain node keys (or even \TikZ; keys).\footnote{The precise algorithm for keyname resulotion in nodewalks is as follows. \begin{itemize} \item First, \foRest; searches for the given \meta{keyname} in the \index{path>forest/nodewalk} path. If found (a long-form step or a nodewalk style), it is executed. \item Next, it is checked whether \meta{keyname} is a sequence of short-form steps; is so, they are executed. \item Otherwise, \meta{key} is executed in the \keyname{/forest} path. This includes both \foRest;'s and \TikZ;'s keys. The latter are usually forwarded to \TikZ; via \index{node options}. \end{itemize} There are some clashes between node key and nodewalk step names. For example, \keyname{l} is both a \meta{dimen} option and a short form of the step to the last child. According to the rules above, the nodewalk step will take precedence in case of a clash. Use nodewalk key \index{options} to execute a clashing node key.} These keys do their usual function, but within the context of the current node of the nodewalk: |parent, s=2em,| |parent, text=red| sets the parent's |s| to 2em and the grandparent's text color to red. It is worth noting that node keys include \index{TeX}, which makes it possible to execute any \TeX\ code while nodewalking. Some steps target a single node, like above-mentioned \index{step>parent} and \index{step>n}. Others, called multi-steps, describe mini-walks themselves: for example \index{step>children} visits each child of the node in turn, and \index{tree} visits each of the node's descendants (including the node itself). The path of many steps is determined by the geometric relations of the tree, or the value of some option. However, there are also keys for embedding nodewalks (\index{nodewalk key>nodewalk}, \index{branch}, etc.), saving and loading nodewalks, sorting them, or even re-walking the history of steps made (like in a web browser).\footnote{Note that nesting operation (\S\ref{ref:nodewalk-operations}) and history (\S\ref{ref:nodewalk-history}) steps, or embedding nodewalks under these steps doesn't work, for most combinations, as many of them internally manipulate nodewalk history.} Finally, if all this is not enough, you can define your own steps, see \S\ref{ref:defining-steps}. Each nodewalk has an associated \emph{every-step keylist}: a keylist of node keys\footnote{When executing the \index{register>every step} keylist, \foRest; switches into the \index{path>forest} path, which makes it impossible to directly include a nodewalk into the every-step keylist. The reason is performance. Every time a \index{path>forest/nodewalk} key is not found, the short-form nodewalk recognition algorithm is executed, and this algorithm is slow. As \index{register>every step} is used a lot (it is for example used every invocation of every spatial propagator) and the keys in \index{register>every step} are usually node options from \index{path>forest} path, \foRest; would spend way too much time checking if a given node option is actually a short-form nodewalk. If you need to execute nodewalk keys within the every-step keylist, use node key \index{node key>Nodewalk}.} which get executed after each step of the nodewalk. The every-step keylist of the current nodewalk is contained in register \index{register>every step} and can be changed at any point during the nodewalk. Its value at the start of the nodewalk depends on how the nodewalk was invoked. In most cases (e.g.\ \index{step>nodewalk} or prefix \index{for step}-based spatial propagators), it is given explicitely as an argument to the key that executes the nodewalk. However, see \index{step>Nodewalk} for information on how the every-step keylist of an embedded nodewalk can interact with the every-step keylist of its parent nodewalk. Each nodewalk step can be either \emph{real} or \emph{fake}. Fake steps only change the current node. Real steps also trigger execution of the every-step keylist and update of history. Fake steps are sometimes useful as a ``computational tool''. They can be introduced explicitely using \index{nodewalk key>fake}; some other keys (like some history nodewalk keys, \S\ref{ref:nodewalk-history}) introduce fake steps implicitely. In some cases, the nodewalk might step ``out of the tree''. (Imagine using \index{step>parent} at the root of the tree, or \index{readonly option>n}|=42| at a node with less that 42 children.) Our official term will be that the nodewalk stepped on an \emph{invalid node}; what happens formally is that the current node is changed to the node with \index{readonly option>id}|=0|. Normally, such an event raises an error. However, the full story is told by \index{nodewalk key>on invalid}. Nodewalks can be hard to follow, especially when designing styles. \FoRest; does its best to help. First, it logs the nodewalk stack in case of error. Second, if package option \index{debug} is given, it logs every step made. \subsubsection{Single-step keys} \label{ref:single-step-keys} Single-step nodewalk keys visit a single node. If the node a single-step is trying to visit does not exist, the current node is set to the invalid node (i.e.\ node with \index{readonly option>id}|=0|). This can be useful for looping, see \S\ref{ref:loops}. Use \index{last valid} or \index{last valid'} to return to the node that was current before the ``unsuccessful'' single-step. For each single-step key, propagator \index{for step=\meta{step}} is also defined. \index{for step=\meta{step}}|=|\meta{keylist} is equivalent to \index{propagator>for}|={|\meta{step}|}{|\meta{keylist}|}|. If the step takes an argument, then its \index{for step=\meta{step}} propagator takes two and the argument of the step precedes the \meta{keylist}. See also \S\ref{ref:spatial-propagators}. Linear order below means the order of nodes in the bracket representation, i.e.\ depth-first parent-first first-child-first. \begin{syntax} \indexitem{step>current} an ``empty'' step: the current node remains the same\footnote{While it might at first sight seem stupid to have an empty step, this is not the case. For example, using propagator \index{for step=\index{current}} derived from this step, one can process a \meta{keylist} constructed using \index[not index,print format+={result+={\index[not print]{wrap pgfmath arg}\index[not print]{wrap $n$ pgfmath args}}}]{handler>wrap ($n$) pgfmath arg(s)} or \index{wrap value}.} \indexitem{step>first} the first child \indexitem{step>first leaf}, \indexdef{step>first leaf'} the first leaf (terminal node) of the current node's descendants (|first leaf|) or subtree (|first leaf'|), in the linear order \indexitem{step>id}|=|\meta{id} the node with the given id; this step does not depend on the current node \indexitem{step>last} the last child \indexitem{step>last dynamic node} the last non-integrated (created/removed/replaced) node; see \S\ref{ref:dynamic} \indexitem{step>last leaf}, \indexdef{step>last leaf'} the last leaf (terminal node) of the current node's descendants (|last leaf|) or subtree (|last leaf'|), in the linear order \indexitem{step>n}|=|$n$ the $n$th child; counting starts at $1$\footnote{Note that \keyname{n} \emph{without} an argument is a short form of \keyname{next}.} \indexitem{step>n'}|=|$n$ the $n$th child, starting the count from the last child \indexitem{step>name}|=|\meta{name} the node with the given name or alias; this step does not depend on the current node \indexitem{step>next} the next sibling \indexitem{step>next leaf} the next node (in the linear order) which is a leaf (the current node need not be a leaf) \indexitem{step>next node} the next node of the entire tree, in the linear order \indexitem{step>next on tier}|=|\meta{tier} the next node (in the linear order) on the given tier; if no tier is given, assume the tier of the current node \indexitem{step>origin} the starting node of the nodewalk; note that the starting point does not automatically count as a step: if you want to step on it, use this key (or \index{current}, at the beginning of the nodewalk) \indexitem{step>parent} the parent \indexitem{step>previous} the previous sibling \indexitem{step>previous leaf} the previous node (in the linear order) which is a leaf (the current node need not be a leaf) \indexitem{step>previous node} the previous node of the entire tree, in the linear order \indexitem{step>previous on tier}|=|\meta{tier} the previous node (in the linear order) on the given tier; if no tier is given, assume the tier of the current node \indexitem{step>root} the root node, i.e.\ the ancestor of the current node without the parent; note that this key \emph{does} depend on the current node \indexitem{step>root'} the formal root node (see \index{set root} in \S\ref{ref:dynamic}); this key does not depend on the current node \indexitem{step>sibling} the sibling (don't use if the parent doesn't have exactly two children \dots) \indexitem{step>to tier}|=|\meta{tier} the first ancestor of the current node (or the node itself) on the given \meta{tier} \end{syntax} \subsubsection{Multi-step keys} \label{ref:multi-step-keys} Multi-step keys visit several nodes, in general. If a multi-step key visits no nodes, the current node remains unchanged. For each multi-step key, spatial propagator \index{for step=\meta{step}} is also defined, see \S\ref{ref:spatial-propagators}. Many of the keys below have a |reversed| variant. Those keys reverse the order of \emph{children}. Note that in general, this differs from operation key \index{reverse}, which reverses the order of the entire embedded nodewalk. Linear order below means the order of nodes in the bracket representation, i.e.\ depth-first parent-first first-child-first. \begin{syntax} \indexitem{step>children}, \indexdef{step>children reversed} Visit all the children of the current node. \indexitem{step>tree}, \indexdef{step>tree reversed} \itemnosep \indexitem{step>tree children-first}, \indexdef{step>tree children-first reversed} \itemnosep \indexitem{step>tree breadth-first}, \indexdef{step>tree breadth-first reversed} Visit the current node and all its descendants. The above keys differ in the order the nodes are visited. The basic key, \keyname{tree}, traverses the nodes in the depth-first, parent-first first-child-first order, i.e.\ the order in which they are given in the bracket representation: so it visits the parent before its children and it visits the children from the first to the last. |reversed| variants reverse the order of \emph{children}, visiting them from the last to the first (from the viewpoint of the bracket representation). |children-first| variants visit the children before the parent. |breadth-first| variants behave like |level| steps below: they first visit level |0| nodes, then level |1| nodes etc. \begin{forestexample}[index={}] \forestset{ enumerate/.style={ tempcounta=1, for #1={ content/.pgfmath=tempcounta, tempcounta+=1 } } } \newcommand\enumtree[1]{% \begin{forest} [#1,l sep=0,for n=1{ l=0,no edge,delay={enumerate=#1}} [[[[][]][[][]]][[[][]]]] ] \end{forest} } \renewcommand\arraystretch{2} \begin{tabular}{cc} \enumtree{tree}& \enumtree{tree reversed}\\ \enumtree{tree children-first}& \enumtree{tree children-first reversed}\\ \enumtree{tree breadth-first}& \enumtree{tree breadth-first reversed} \end{tabular} \end{forestexample} \indexitem{step>descendants}, \indexdef{step>descendants reversed} \itemnosep \indexitem{step>descendants children-first}, \indexdef{step>descendants children-first reversed} \itemnosep \indexitem{step>descendants breadth-first}, \indexdef{step>descendants breadth-first reversed} Visit all the descendants of the current node. Like the |tree| keys, but the current node is not visited. \indexitem[print form+=\spaces{9}]{step>relative level<}, \indexdef[print form+=\spaces{9}]{step>relative level}, \indexdef[set={form=relative level\protect\greaterthan}]{step>relative levelGT}\texttt{\spaces{9}}|=|\meta{count} \itemnosep \indexitem{step>relative level reversed<}, \indexdef{step>relative level reversed}, \indexdef[set={form=relative level reversed\protect\greaterthan}]{step>relative level reversedGT}|=|\meta{count} \itemnosep \indexitem[print form+=\spaces{9},+print form=\spaces{9}]{step>level<}, \indexdef[print form+=\spaces{9},+print form=\spaces{9}]{step>level}, \indexdef[set={form=level\protect\greaterthan},+print form=\spaces{9}]{step>levelGT}\texttt{\spaces{9}}|=|\meta{count} \itemnosep \indexitem[+print form=\spaces{9}]{step>level reversed<}, \indexdef[+print form=\spaces{9}]{step>level reversed}, \indexdef[set={form=level reversed\protect\greaterthan},+print form=\spaces{9}]{step>level reversedGT}|=|\meta{count} Visits the nodes in the subtree of the current node whose level (depth) is less than \emph{or equal to}, equal to, or greater than \emph{or equal to} the given level. The |relative| variants consider the level as relative to the current node: relative level of the current node is $0$; relative level of its children is $1$, of its grandchildren $2$, etc. The absolute variants consider the depth with respect to the (geometric) root, i.e.\ as returned by node option \index{readonly option>level}. The nodes are traversed in the breadth-first order. The |reversed| variants reverse the order of the children within each level, but the levels are still traversed from the highest to the deepest. \indexitem[+print form=\spaces{12},print form+=\spaces{9}]{step>preceding siblings}\quad\quad \indexdef[+print form=\spaces{12}]{step>following siblings}% \itemnosep \indexitem[print form+=\spaces{9}]{step>current and preceding siblings}\quad\quad \indexdef{step>current and following siblings}% \itemnosep \indexitem[+print form=\spaces{12}]{step>preceding siblings reversed}\quad\quad \indexdef[+print form=\spaces{12}]{step>following siblings reversed}% \itemnosep \indexitem{step>current and preceding siblings reversed}\quad\quad \indexdef{step>current and following siblings reversed}% \itemnosep \indexitem[+print form=\spaces{22},print form+=\spaces{9}]{step>siblings}% \itemnosep \indexitem[print form={current and \spaces{9} siblings\spaces{9}}]{step>current and siblings}% \itemnosep \indexitem[+print form=\spaces{22},print form+=\spaces{9}]{step>siblings reversed}% \itemnosep \indexitem[print form={current and \spaces{9} siblings reversed}]{step>current and siblings reversed}% Visit preceding, following or all siblings; visit the current node as well or not; visit in normal or reversed order. \indexitem{step>ancestors} \itemnosep \indexitem{step>current and ancestors} Visit the ancestors of the current node, starting from the parent/current node, ending at the root node. \indexitem[+print form=\spaces{12}]{step>preceding nodes}\quad\quad \indexdef[+print form=\spaces{12}]{step>following nodes}% \itemnosep \indexitem{step>current and preceding nodes}\quad\quad \indexdef{step>current and following nodes}% Visit all preceding or following nodes of the entire tree, in the linear order; visit the current node as well or not. \end{syntax} \subsubsection{Embedded nodewalks} \label{ref:embedded-walks} An embedded nodewalk is a nodewalk within a nodewalk. The embedded nodewalk has its own every-step keylist. While stepping through the inner nodewalk, the every-step keylist of the outer nodewalk is not executed. Modifying the every-step keylist of the inner walk (by \index{register>every step}) does not influence the outer nodewalk. After the embedded nodewalk finished, its final current node becomes the current node of the outer nodewalk. This is true even if the final node was reached by a fake step. \begin{advise} \item As node keys can be used within nodewalks, one can use spatial propagators (\S\ref{ref:spatial-propagators}) like \index{for step=\index{step>nodewalk}} to embed a nodewalk which (silently) returns to the current node of the outer nodewalk. \end{advise} An embedded nodewalk does not count as a step of the outer nodewalk (except if noted otherwise, see \index{group}). This means that the every-step keylist of the outer nodewalk is not executed at the final node of the embedded nodewalk. In other words, the step made by the embedded nodewalk as a unit is fake. Embedded nodewalks also have their own origin and history. Nodewalks can be nested to an arbitrary level. \begin{syntax} \indexitem{step>Nodewalk}|=|\meta{keylist: config}\meta{nodewalk}\meta{keylist: every-step} Walks an (embedded) \meta{nodewalk} starting at the current node. From the perspective of the outer nodewalk, the embedded nodewalk is an (implicitely) fake step (and cannot be made real by \index{real}): the last node visited by the embedded \meta{nodewalk} becomes the new current node of the outer nodewalk, even if it is invalid. The \meta{config} argument serves to specify the interaction between the outer and embedded nodewalk. It can contain the following keys: \let\outerleftmargin\leftmargin \begin{syntax} \indexitem(independent){option of=Nodewalk>every step}|={|\alternative{independent,inherited,shared}|}| \indexitem(shared){option of=Nodewalk>history}|={|\alternative{independent,inherited,shared}|}| The following table shows what happens to the every-step keylist and history depending on the value of \keyname{every step} and \keyname{history}, respectively. State B is \meta{every-step} for every step and empty for history. \begin{tabular}{l|ccc} &\keyname{independent}&\keyname{inherited}&\keyname{shared}\\\hline state of the outer nodewalk&A&A&A\\ initial state of the inner nodewalk&B&A&A\\ \dots\\ final state of the inner nodewalk&C&C&C\\ state of the outer nodewalk&A&A&C \end{tabular} \end{syntax} As shown in the table above, argument \meta{every-step} is used to initialize the embedded nodewalk's every-step keylist when it is independent of the outer nodewalk. In other cases, this argument is ignored (use |{}|). \begin{syntax} \indexitem(inherited){option of=Nodewalk>on invalid}|={|\alternative{error,fake,step,inherited}|}| Like \index{nodewalk key>on invalid}, but local to this nodewalk. The additional alternative \indexdef{value=on invalid>inherited} (which is the default) means to behave as specified by \index{nodewalk key>on invalid} (or the package default, \index{value=on invalid>error}, if no \index{nodewalk key>on invalid} was given). \end{syntax} \begin{forestexample} \begin{forest} for 2=calign with current, for children={for descendants={circle,draw}} [every step, [independent,for nodewalk={ 1,Nodewalk={~every step=independent~}{1,every step=fill,1}{},1}{draw=red}, [[[[]]]]] [inherited,for nodewalk={ 1,Nodewalk={~every step=inherited~}{1,every step=fill,1}{},1}{draw=red}, [[[[]]]]] [shared,for nodewalk={ 1,Nodewalk={~every step=shared~}{1,every step=fill,1}{},1}{draw=red}, [[[[]]]]]] \end{forest} \end{forestexample} \begin{forestexample} \begin{forest} mark/.style={tempcounta+=1,content+/.register=tempcounta,content+={,}}, [history:\\effect on the,align=center [inner nodewalk % uncommenting this would result in an error: % [independent, delay={for nodewalk={ % tempcounta=0,111, % Nodewalk={history=independent}{walk back=2}{mark,fill=yellow} % }{mark,draw=red}}, % [[[]]]] [inherited\\or\\shared, align=center,delay={for nodewalk={ tempcounta=0,111, Nodewalk={~history=inherited~}{walk back=2}{mark,fill=yellow} }{mark,draw=red}}, [[[]]]]] [outer nodewalk [inherited,delay={for nodewalk={ tempcounta=0,111, Nodewalk={~history=inherited~}{11}{mark,fill=yellow}, walk back=2 }{mark,draw=red}}, [[[[[]]]]]] [shared,delay={for nodewalk={ tempcounta=0,111, Nodewalk={~history=shared~}{11}{mark,fill=yellow}, walk back=2 }{mark,draw=red}}, [[[[[]]]]]]]] \end{forest} \end{forestexample} \indexitem{step>nodewalk}|=|\meta{nodewalk}\meta{keylist: every-step} This key is a shorthand for \begin{center} \index{step>Nodewalk}|={|\index{option of=Nodewalk>every step}|=independent,|\index{option of=Nodewalk>history}|=independent}|\meta{nodewalk}\meta{keylist: every-step} \end{center} \begin{advise} \item \index{for step=\index{nodewalk}} is also defined. \end{advise} \indexitem{step>nodewalk'}|=|\meta{nodewalk} This key is a shorthand for \begin{center} \index{step>Nodewalk}|={|\index{option of=Nodewalk>every step}|=inherited,|\index{option of=Nodewalk>history}|=independent}|\meta{nodewalk}\meta{keylist: every-step} \end{center} \begin{advise} \item \index{for step=\index{nodewalk'}} is also defined. \end{advise} \indexitem{step>group}|=|\meta{nodewalk} Treat \meta{nodewalk} as a single step of the (outer) nodewalk, i.e.\ the outer every-step keylist is executed only at the end of the embedded nodewalk. The embedded \meta{nodewalk} inherits history from the outer nodewalk. Using this key is equivalent to writing \begin{center} \index{step>Nodewalk}|={|\index{option of=Nodewalk>every step}|=independent,|\index{option of=Nodewalk>history}|=inherited}|\meta{nodewalk}|{}, current| \end{center} \begin{advise} \item It is safe to modify the every-step keylist of \meta{nodewalk}. \item \index{for step=\index{group}} is also defined. \end{advise} \end{syntax} \subsubsection{Operations} \label{ref:nodewalk-operations} Operations internally walk the given nodewalk (except |..._in_nodewalk| variants, which operate on the portion of the current nodewalk that was already walked) to gather some information about visited nodes. Then, based on that information, they walk some steps in the context (most importantly, with \index{register>every step} keylist) of the current nodewalk. Most of the following keys have a |walk_and_...| variant, which first walks the given nodewalk and then performs the desired operation. All operation keys except |..._in_nodewalk| variants can be prefixed by \index{for step}. \begin{advise} \item The every-step keylist of the internal nodewalk is empty; or, more precisely, it contains only instructions to how collect the relevant information. You can safely append and prepend to \index{register>every step} keylist of the internal nodewalk, but you should not completely rewrite it. \end{advise} \begin{syntax} \indexitem{step>reverse}|=|\meta{nodewalk} \itemnosep \indexitem{step>walk and reverse}|=|\meta{nodewalk} Visits the nodes of the given \meta{nodewalk} in the reversed order. \indexitem{nodewalk key>save}|=|\meta{toks: name}\meta{nodewalk} \itemnosep \indexitem{step>walk and save}|=|\meta{toks: name}\meta{nodewalk} Saves the given \meta{nodewalk} under the given name. \indexitem{step>load}|=|\meta{toks: name} Walks the nodewalk saved under the given name. Note that it is node \index{readonly option>id}s that are saved: loading a named nodewalk with in a context of a different current node, or even with a tree whose geometry has changed (see \S\ref{ref:dynamic}) will still visit exactly the nodes that were visited when the nodewalk was saved. \indexitem{step>sort}|=|\meta{nodewalk} \itemnosep \indexitem{step>sort'}|=|\meta{nodewalk} \itemnosep \indexitem{step>walk and sort}|=|\meta{nodewalk} \itemnosep \indexitem{step>walk and sort'}|=|\meta{nodewalk} Walks the nodes of the nodewalk in the order specified by the last invokation of \index{sort by}. The \keyname{sort} variants sort in the ascending order, the \keyname{sort'} variants in the descending order. The \keyname{walk and sort} variants first visit the nodes in the order specified by the given \meta{nodewalk}. \indexitem{node key>sort by}|={|\meta{pgfmath}|,|\dots|,|\meta{pgfmath}|}| Sets the sorting order used by all keys comparing nodes: \index{step>sort}, \index{step>min} and \index{step>max} key families in the nodewalk namespace, and the \index{dynamic tree>sort} key family in the option namespace (dynamic tree). For each node being ordered, an ``$n$-dimensional coordinate'' is computed by evaluating the given list of |pgfmath| expressions in the context of that node.\footnote{Don't worry, lazy evaluation is used.} Nodes are then ordered by the usual sort order for multi-dimensional arrays: the first item is the most important, the second item is the second most important, etc. Simply put, if you want to sort first by the number of children and then by content, say |sort_by={|\index[+print format=pgfmath]{n children}|,_|\index{content}|}|. In the simplest case, the given \meta{pgfmath} expressions are simply node options. However, as any |pgfmath| expression is allowed in the sort key, you can easily sort by the product of the content of the current node and the content of its first child: |sort_by={|\index{content}|()*|\index{content}|("|\index>{!}|1")|. Note, however, that the value returned by these expression must be numeric, i.e.\ alphabetic comparison of strings is not implemented. \indexitem{step>min}|=|\meta{nodewalk}, \indexdef{step>max}|=|\meta{nodewalk} \itemnosep \indexitem{step>walk and min}|=|\meta{nodewalk}, \indexdef{step>walk and max}|=|\meta{nodewalk} \itemnosep \indexitem{step>mins}|=|\meta{nodewalk}, \indexdef{step>maxs}|=|\meta{nodewalk} \itemnosep \indexitem{step>walk and mins}|=|\meta{nodewalk}, \indexdef{step>walk and maxs}|=|\meta{nodewalk} Visit the node(s) in the given \meta{nodewalk} with a minimum/maximum value with respect to the sort order previously specified by \index{sort by}. Variants \keyname{mins}/\keyname{maxs} visit all the nodes that with the minimum/maximum value of the sorting key; variants \keyname{min}/\keyname{max} visit only the first such node (first in the order specified by the given nodewalk). \indexitem{step>min in nodewalk}, \indexdef{step>max in nodewalk} \itemnosep \indexitem{step>mins in nodewalk}, \indexdef{step>maxs in nodewalk} \itemnosep \indexitem{step>min in nodewalk'}, \indexdef{step>max in nodewalk'} These keys search for the minimum/maximum among the nodes that were already visited in the current nodewalk. Keys \keyname{mins in nodewalk} and \keyname{maxs in nodewalk} visits all nodes that reach the minimum/maximum, while keys \keyname{min in nodewalk} and \keyname{max in nodewalk} variants visit only the first such node. Keys \keyname{min in nodewalk'} and \keyname{max in nodewalk'} visit the first minimal/maximal node by moving back in the history, see \index{back}. \indexitem{step>filter}|=|\meta{nodewalk}\meta{pgfmath: condition} Visit the nodes of the given \meta{nodewalk} for which the given \meta{condition} is true. \indexitem{step>branch}|={|\meta{nodewalk$_1$}|,...,|\meta{nodewalk$_n$}|}| \itemnosep \indexitem{step>branch'}|={|\meta{nodewalk$_1$}|,...,|\meta{nodewalk$_n$}|}| Visit the nodes in a ``cartesian product'' of any number of nodewalks, where a cartesian product is defined as a nodewalk where at every step of \meta{nodewalk$_i$} ($1\leq iback}|=|\meta{count: n} \itemnosep \indexitem{step>jump back}|=|\meta{count: n} \itemnosep \indexitem{step>walk back}|=|\meta{count: n} Move $n$ steps back in the history. In the \keyname{back} variant, all steps are fake; in the \keyname{jump back} variant, the final step is real; and in the \keyname{walk back} variant, all steps are real. Note that as the origin is not a part of the history, these keys will \emph{not} step there (unless \index{current} was the first step of your nodewalk). (Use \index{origin} to move to the origin of the nodewalk.) \indexitem{step>forward}|=|\meta{count: n} \itemnosep \indexitem{step>jump forward}|=|\meta{count: n} \itemnosep \indexitem{step>walk forward}|=|\meta{count: n} Move $n$ steps forward in the history. In the \keyname{forward} variant, all steps are fake; in the \keyname{jump forward} variant, the final step is real; and in the \keyname{walk forward} variant, all steps are real. \indexitem{node key>save history}|=|\meta{toks: back name}\meta{toks: forward name} Saves the backwards and forwards history under the given names. (Load them using \index{load}.) The backwards history is saved in the reverse order of how it was walked, i.e.\ outward from the perspective of the current position in the nodewalk. \end{syntax} \subsubsection{Miscellaneous} \label{ref:nodewalk-miscellaneous} The following nodewalk keys are not steps. Rather, they influence the behaviour of nodewalk steps in various ways. The keys in this section having \meta{nodewalk} arguments do not start a new nodewalk in the sense of \S\ref{ref:embedded-walks}; the given nodewalk steps rather become a part of the current nodewalk. \begin{syntax} \indexitem{register>every step}|=|\meta{keylist} Contains the every-step keylist of the current nodewalk. \indexitem{nodewalk key>fake}|=|\meta{nodewalk} \itemnosep \indexitem{nodewalk key>real}|=|\meta{nodewalk} The \meta{nodewalk} embedded under |fake| consists of ``fake'' steps: while the current node is changed, every-step keylist is not executed and the history is not updated. Note that these keys do not introduce an embedded nodewalk. The given \meta{nodewalk} will not have its own history and every-step keylist. |real| undoes the effect of |fake|, but cannot make real the implicitely fake steps, such as the return to the origin in spatial propagators like \index{for step=\index{nodewalk}}. |fake| and |real| can be nested without limit. \indexitem{step>last valid} \itemnosep \indexitem{step>last valid'} If the current node is valid, these keys do nothing. If the current node of the nodewalk is invalid (i.e.\ its \index{readonly option>id} is $0$), they step to the last valid visited node. If there was no such node, they step to the origin of the nodewalk. The variant \emph{without} |'| makes a fake step. More precisely, it behaves as if both \index{nodewalk key>fake} and \index{nodewalk key>on invalid}|=fake| are in effect. \indexitem(error){nodewalk key>on invalid}|={|\alternative{error,fake,step}|}|\meta{nodewalk} This key determines what should happen if a nodewalk step landed on the invalid node, i.e.\ the node with \index{readonly option>id}|=0|. There is a moment within the step when the current node is changed but the step itself is not yet really done, is ``still fake'', i.e.\ the history is not yet updated and the every-step keylist is not yet executed. If the new current node is invalid, this key determines what should happen next. \let\outerleftmargin\leftmargin \begin{syntax} \item |on invalid=|\indexdef{value=on invalid>error} produces an error; \item |on invalid=|\indexdef{value=on invalid>fake} does nothing: history is not updated and the every-step keylist is not executed, thus making the step essentialy fake; \item |on invalid=|\indexdef{value=on invalid>step} makes a step, either real or fake, depending on the environment. \end{syntax} Loops with the implicit |id=0| condition (\S\ref{ref:loops}) automatically switch to |on_invalid=fake| mode. \indexitem{nodewalk key>options}=\meta{keylist: node keys} Execute the given node options in the context of the current node. There is not much need to use this key, as any keys that are not (long) steps or sequences of short steps are automatically used as \foRest; node options any way, but there are still usage cases, for example whenever the names of node options and (long) steps are the same, or in a style that wants to ensure there is no overlap. \indexitem{nodewalk key>strip fake steps}|=|\meta{nodewalk} If \meta{nodewalk} ends with fake steps, return to the last node current before those steps were made. For details, see \index{define long step}. \end{syntax} \subsubsection{Short-form steps} \label{ref:short-form-steps} All short forms of steps are one token long. When using them, there is no need to separate them by commas. Here's the list of predefined short steps and their corresponding long-form steps. \begin{syntax} \indexitem{short step>1}, \indexdef{short step>2}, \indexdef{short step>3}, \indexdef{short step>4}, \indexdef{short step>5}, \indexdef{short step>6}, \indexdef{short step>7}, \indexdef{short step>8}, \indexdef{short step>9} the first, \dots, ninth child --- \index{readonly option>n}|=1|,\dots,|,9| \indexitem{short step>l} the last child --- \index{step>last} \indexitem{short step>u} the parent (up) --- \index{step>parent} \indexitem{short step>p} the previous sibling --- \index{previous} \indexitem{short step>n} the next sibling --- \index{next} \indexitem{short step>s} the sibling --- \index{sibling} \indexitem{short step>P} the previous leaf --- \index{previous leaf} \indexitem{short step>N} the next leaf --- \index{next leaf} \indexitem{short step>F} the first leaf --- \index{first leaf} \indexitem{short step>L} the last leaf --- \index{last leaf} \indexitem[set={index key={GT} the next node on the current tier --- \index{next on tier} \indexitem{short step><} the previous node on the current tier --- \index{previous on tier} \indexitem{short step>c} the current node --- \index{current} \indexitem{short step>o} the origin --- \index{origin} \indexitem{short step>r} the root node --- \index{root} \indexitem{short step>R} the formal root node --- \index{root'} \indexitem{short step>b} back one fake step in history --- \index{back}|=1| \indexitem{short step>f} forward one fake step in history --- \index{forward}|=1| \indexitem{short step>v} last valid node in the current nodewalk, fake version --- \index{last valid} \indexitem{short step>*}\meta{count: n}\meta{keylist} repeat keylist $n$ times --- \index{repeat}|=|\meta{count: n}\meta{keylist} \item {\rstyle $\{$}\meta{keylist}{\rstyle $\}$} put keylist in a group --- \index{group}|=|\meta{keylist} \end{syntax} \subsubsection{Defining steps} \label{ref:defining-steps} You can define your own steps, both long and short, or even redefine predefined steps. Note, though, that it is not advisable to redefine long steps, as their definitions are interdependent; redefining short steps is always ok, however, as they are never used in the code of the package. \begin{syntax} \indexitem{node key>define long step}|=|\meta{name}\meta{options}\meta{nodewalk} Define a long-form step named \meta{name} as equivalent to \meta{nodewalk}. \meta{options} control the exact behaviour or the defined step. \let\outerleftmargin\leftmargin \begin{syntax} \item\indexdef{option of=define long step>n args}|=|\meta{number}\hfill |0| \item\indexdef{option of=define long step>make for}|=|\meta{boolean}\hfill |true| Should we make a \index{for step} prefix for the step? \item\indexdef{option of=define long step>strip fake steps}|=|\meta{boolean}\hfill |true| Imagine that \meta{nodewalk} ends with fake steps. Now, fake steps are usually just a computational tool, so we normally wouldn't want the current node after the walk to be one of them. As far as the outer world is concerned, we want the node to end at the last real step. However, simply appending \index{last valid} to our style will not work. Imagine that the nodewalk results in no steps. In this case, we'd want to remain at the origin of our empty nodewalk. However, imagine further that the (outer) step just before the empty nodewalk was fake. Then \index{last valid} will not step to the correct node: instead of staying at the origin, it will go to the node that the last real step prior to our nodewalk stepped to. In case there was no such real step, we'd even step to the invalid node (normally getting an error). Defining the step using |strip fake steps| ensures the correct behaviour described above. Set |strip fake steps=false| only when the fake steps at the end of the nodewalk at important to you. \begin{advise} \item See also nodewalk key \index{nodewalk key>strip fake steps}. \end{advise} \end{syntax} \begin{forestexample}[index={define long step,option of=define long step>n args,conditional>if,step>n,while,readonly option>n,next}] \forestset{ ~define long step~={children from to}{n args=2}{ if={#1>#2}{}{n=#1,while={n()<#2}{next}} } } \begin{forest} for children from to={2}{5}{draw} [root[1][2][3][4][5][6]] \end{forest} \end{forestexample} \indexitem{node key>define short step}|=|\meta{token: short step}\meta{n args}\meta{nodewalk} Define short step taking $n$ arguments as the given \meta{nodewalk}. Refer to the arguments in the usual way, via |#1|, \dots. To (re)define braces, |{}|, write |define short step={group}{1}{...}|. \indexitem{handler>nodewalk style}|=|\meta{nodewalk} A shorthand for |for nodewalk={|\meta{nodewalk}|}{}|. \end{syntax} \subsection{Conditionals and loops} \label{ref:conditionals} \label{ref:loops} All conditionals take arguments \meta{true keylist} and \meta{false keylist}, and all loops take an argument \meta{keylist}. The interpretation of the keys in these keylists depends on the environment the conditional/loop appears in. If it is a part of a nodewalk specification, the keys are taken to be nodewalk keys (\S\ref{ref:nodewalks}), otherwise node keys (\S\ref{ref:node-keys}). All the loops and conditionals can be nested safely. \begin{advise} \item Except in special circumstances, you probably don't want to embed keys from the \index{where option} family within a \index{for step=\index{tree}}, as this results in two nested loops. It is more usual to use an \index{conditional>if} family key there. For an example where using \index{where option} actually does the wrong thing, see question \href{http://tex.stackexchange.com/questions/167978/smaller-roofs-for-forest}{Smaller roofs for forest} on \TeX\ Stackexchange. \end{advise} \begin{syntax} \indexitem{conditional>if}|=|\meta{pgfmath condition}\meta{true keylist}\meta{false keylist} If \meta{pgfmath condition} evaluates to |true| (non-zero), \meta{true keylist} is processed (in the context of the current node); otherwise, \meta{false keylist} is processed. For a detailed description of |pgfmath| expressions, see \cite[part VI]{tikzpgf2.10}. (In short: write the usual mathematical expressions.) \indexitem{word prefix=option>conditional>if=\meta{option}}|=|\meta{value}\meta{true keylist}\meta{false keylist} A simple conditional is defined for every \meta{option}: if \meta{value} equals the value of the option at the current node, \meta{true keylist} is executed; otherwise, \meta{false keylist}. \indexitem{word prefix=boolean option>conditional>if=\meta{boolean option}}|=|\meta{true keylist}\meta{false keylist} Execute \meta{true keylist} if \meta{boolean option} is true; otherwise, execute \meta{false keylist}. \indexitem{word prefix=toks option>conditional>if in=\meta{toks option}}|=|\meta{toks}\meta{true keylist}\meta{false keylist} Checks if \meta{toks} occurs in the option value; if it does, \meta{true keylist} are executed, otherwise \meta{false keylist}. This conditional is defined only for \meta{toks} options, see \S\ref{sec:option-setting}. \indexitem{conditional>if nodewalk valid}|=|\meta{keylist: test nodewalk}\meta{true keylist}\meta{false keylist} If the test nodewalk finished on a valid node, \meta{true keylist} is processed (in the context of the current node); otherwise, \meta{false keylist} is processed. \indexitem{conditional>if nodewalk empty}|=|\meta{keylist: test nodewalk}\meta{true keylist}\meta{false keylist} If the test nodewalk contains no (real) steps, \meta{true keylist} is processed (in the context of the current node); otherwise, \meta{false keylist} is processed. \indexitem{conditional>if in saved nodewalk}|=|\meta{nodewalk}\meta{toks: nodewalk name}\meta{true keylist}\meta{false keylist} If the node at the end of \meta{nodewalk} occurs in the saved nodewalk, \meta{true keylist} is processed (in the context of the current node); otherwise, \meta{false keylist} is processed. \indexitem{propagator>if have delayed}|=|\meta{true keylist}\meta{false keylist} If any options were delayed in the current cycle (more precisely, up to the point of the execution of this key), process \meta{true keylist}, otherwise process \meta{false keylist}. (\index{delay n} will trigger ``true'' for the intermediate cycles.) This key assumes that the processing order of the innermost invocation of \index{process keylist} or \index{process keylist'} is given by \index{processing order}. If this is not the case, explicitely supply the processing order using \index{if have delayed'}. \indexitem{propagator>if have delayed'}|=|\meta{nodewalk}\meta{true keylist}\meta{false keylist} Like \index{if have delayed}, but assume the processing order given by \meta{nodewalk}. \indexitem{propagator>where}|=|\meta{value}\meta{true keylist}\meta{false keylist} \itemnosep \indexitem{word prefix=option>conditional>where=\meta{option}}|=|\meta{value}\meta{true keylist}\meta{false keylist} \itemnosep \indexitem{word prefix=toks option>conditional>where in=\meta{toks option}}|=|\meta{toks}\meta{true keylist}\meta{false keylist} \itemnosep \indexitem{step>where nodewalk valid}|=|\meta{toks: nodewalk name}\meta{true keylist}\meta{false keylist} \itemnosep \indexitem{step>where nodewalk empty}|=|\meta{toks: nodewalk name}\meta{true keylist}\meta{false keylist} \itemnosep \indexitem{step>where in saved nodewalk}|=|\meta{nodewalk}\meta{toks: nodewalk name}\meta{true keylist}\meta{false keylist} These keys execute their corresponding |if ...| conditional for every node in the subtree of the current node (including the node itself). In other words: \begin{center} |where ...|\meta{arg$_1$}\dots\meta{arg$_1$}|/.style={for tree={if ...=|\meta{arg$_1$}\dots\meta{arg$_1$}|}}| \end{center} \indexitem{loop>repeat}|=|\meta{number}\meta{keylist} The \meta{keylist} is processed \meta{number} times. The \meta{number} expression is evaluated using |pgfmath|. \indexitem{loop>while}|=|\meta{pgfmath condition}\meta{keylist} \itemnosep \indexitem{loop>do while}|=|\meta{pgfmath condition}\meta{keylist} \itemnosep \indexitem{loop>until}|=|\meta{pgfmath condition}\meta{keylist} \itemnosep \indexitem{loop>do until}|=|\meta{pgfmath condition}\meta{keylist} \keyname{while} loops cycle while the condition is true, \keyname{until} loops terminate when the condition becomes true. The \keyname{do} variants check the condition after processing the \meta{keylist}; thus, the keylist is executed at least once. The variants without the \keyname{do} prefix check the condition before processing the \meta{keylist}, which might therefore not be processed at all. When \meta{pgfmath condition} is an empty string, condition \index{valid} is implicitely used, and \meta{keylist} is implicitely embedded in \index{nodewalk key>on invalid}|=fake|. Thus, the |while| loops will cycle until they ``walk out of the tree'', and |until| loops will cycle until they ``walk into the tree.'' \begin{advise} \item If a loop ``walks out of the tree'', you can get it back in using \index{last valid} or \index{nodewalk key>strip fake steps}. \end{advise} \begin{forestexample}[index={root',content,step>n,delay,for step,nodewalk,do until}] \pgfmathsetseed{1234} \begin{forest} try/.style={root',content+={#1,\ },n=#1}, delay={ for nodewalk={~do until~={}{try/.pgfmath={random(1,20)}},draw}{}, }, [tried:\ [1][2][3][4][5]] \end{forest} \end{forestexample} \indexitem{loop>while nodewalk valid}|=|\meta{nodewalk}\meta{keylist} \itemnosep \indexitem{loop>do while nodewalk valid}|=|\meta{nodewalk}\meta{keylist} \itemnosep \indexitem{loop>until nodewalk valid}|=|\meta{nodewalk}\meta{keylist} \itemnosep \indexitem{loop>do until nodewalk valid}|=|\meta{nodewalk}\meta{keylist} \itemnosep \indexitem{loop>while nodewalk empty}|=|\meta{nodewalk}\meta{keylist} \itemnosep \indexitem{loop>do while nodewalk empty}|=|\meta{nodewalk}\meta{keylist} \itemnosep \indexitem{loop>until nodewalk empty}|=|\meta{nodewalk}\meta{keylist} \itemnosep \indexitem{loop>do until nodewalk empty}|=|\meta{nodewalk}\meta{keylist} \meta{nodewalk} is embedded within \index{nodewalk key>on invalid}|=fake|. \indexitem(0){loop>break}|=|$n$ Break out of the loop. The loop is only exited after all the keys in the current cycle have been processed. The optional argument $n$ ($n\geq0$) specifies which level of embedding to break from; the default is to break out of the innermost loop. \end{syntax} \subsection{Dynamic tree} \label{ref:dynamic} The following keys can be used to change the geometry of the tree by creating new nodes and integrating them into the tree, moving and copying nodes around the tree, and removing nodes from the tree. The \meta{node} that will be (re)integrated into the tree can be specified in the following ways: \begin{syntax} \item \meta{empty}: uses the last non-integrated (i.e.\ created/removed/replaced) node. \begin{advise} \item This node can also be referred to using nodewalk step \index{last dynamic node}. \item The list of all such nodes is automatically saved in named nodewalk \indexdef{named nodewalk>dynamic nodes}, to be \index{load}ed when needed. \end{advise} \item \meta{node}: a new node is created using the given bracket representation (the node may contain children, i.e.\ a tree may be specified), and used as the argument to the key. The bracket representation must be enclosed in brackets, which will usually be enclosed in braces to prevent them being parsed while parsing the ``host tree.'' \begin{advise} \item Unlike the bracket representation in a \index{environment>forest} environment, the bracket representation of a dynamically created node \emph{must} start with |[|. Specifically, it cannot begin with a preamble or the action character. \end{advise} \item \index{relative node name}: the node \meta{relative node name} resolves to will be used. \end{syntax} A dynamic tree operation is made in two steps: \begin{itemize} \item If the argument is given by a \meta{node} argument, the new node is created immediately, i.e.\ while the dynamic tree key is being processed. Any options of the new node are implicitely \index{delay}ed. \item The requested changes in the tree structure are actually made between the cycles of keylist processing. \end{itemize} \begin{advise} \item Such a two-stage approach is employed because changing the tree structure during the dynamic tree key processing would lead to an unmanageable order of keylist processing. \item A consequence of this approach is that nested dynamic tree keys take several cycles to complete. Therefore, be careful when using \index{delay} and dynamic tree keys simultaneously: in such a case, it is often safer to use \index{before typesetting nodes} instead of \index{delay}, see example (\ref{ex:append}). \item Examples: title page (in style |random tree|) and (\ref{ex:niceemptynodes}) (in style |xlist|). \end{advise} Here is the list of dynamic tree keys: \begin{syntax} \indexitem{dynamic tree>append}|=|\meta{empty}\OR|[|\meta{node}|]|\OR\meta{relative node name} The specified node becomes the new final child of the current node. If the specified node had a parent, it is first removed from its old position. \begin{forestexample}[label=ex:append,point=append,index={append,delay,for step,tree,readonly option>n,content,readonly option>n',repeat}] \begin{forest} before typesetting nodes={for tree={ if n=1{content=L} {if n'=1{content=R} {content=C}}}} [,repeat=2{append={[ ,repeat=3{append={[]}} ]}}] \end{forest} \end{forestexample} \indexitem{dynamic tree>create}|=[|\meta{node}|]| Create a new node. The new node becomes the last node. \indexitem{dynamic tree>create'}|=[|\meta{node}|]| Create a new node and process its given options immediately. The new node becomes the last node. \indexitem{dynamic tree>insert after}|=|\meta{empty}\OR|[|\meta{node}|]|\OR\meta{relative node name} The specified node becomes the new following sibling of the current node. If the specified node had a parent, it is first removed from its old position. \indexitem{dynamic tree>insert before}|=|\meta{empty}\OR|[|\meta{node}|]|\OR\meta{relative node name} The specified node becomes the new previous sibling of the current node. If the specified node had a parent, it is first removed from its old position. \indexitem{dynamic tree>prepend}|=|\meta{empty}\OR|[|\meta{node}|]|\OR\meta{relative node name} The specified node becomes the new first child of the current node. If the specified node had a parent, it is first removed from its old position. \indexitem{dynamic tree>remove} The current node is removed from the tree and becomes the last node. The node itself is not deleted: it is just not integrated in the tree anymore. Removing the root node has no effect. \indexitem{dynamic tree>replace by}|=|\meta{empty}\OR|[|\meta{node}|]|\OR\meta{relative node name} The current node is replaced by the specified node. The current node becomes the last node. If the specified node is a new node containing a dynamic tree key, it can refer to the replaced node by the \meta{empty} specification. This works even if multiple replacements are made. If \keyname{replace by} is used on the root node, the ``replacement'' becomes the root node (\index{set root} is used). \end{syntax} If given an existing node, most of the above keys \emph{move} this node (and its subtree, of course). Below are the versions of these operations which rather \emph{copy} the node: either the whole subtree (|'|) or just the node itself (|''|). \begin{syntax} \indexitem{dynamic tree>append'}, \indexdef{dynamic tree>insert after'}, \indexdef{dynamic tree>insert before'}, \indexdef{dynamic tree>prepend'}, \indexdef{dynamic tree>replace by'} Same as versions without |'| (also the same arguments), but it is the copy of the specified node and its subtree that is integrated in the new place. \indexitem{dynamic tree>append''}, \indexdef{dynamic tree>insert after''}, \indexdef{dynamic tree>insert before''}, \indexdef{dynamic tree>prepend''}, \indexdef{dynamic tree>replace by''} Same as versions without |''| (also the same arguments), but it is the copy of the specified node (without its subtree) that is integrated in the new place. \begin{advise} \item You might want to \index{delay} the processing of the copying operations, giving the original nodes the chance to process their keys first! \end{advise} \indexitem{dynamic tree>copy name template}|=|\meta{empty}\OR\meta{macro definition} \hfill\meta{empty} Defines a template for constructing the \index{option>name} of the copy from the name of the original. \meta{macro definition} should be either empty (then, the \index{option>name} is constructed from the \index{readonly option>id}, as usual), or an expandable macro taking one argument (the name of the original). \begin{forestexample} \begin{forest} copy name template={copy of #1} [CP,delay={prepend'=subject} [VP[DP,name=subject[D][NP]][V'[V][DP]]]] \draw[->,dotted] (subject)--(copy of subject); \end{forest} \end{forestexample} \indexitem{dynamic tree>set root}|=|\meta{empty}\OR|[|\meta{node}|]|\OR\meta{relative node name} The specified node becomes the new \emph{formal} root of the tree. Note: If the specified node has a parent, it is \emph{not} removed from it. The node becomes the root only in the sense that the default implementation of stage-processing will consider it a root, and thus typeset/pack/draw the (sub)tree rooted in this root. The processing of keys such as \index{for step=\index{step>parent}} and \index{for step=\index{root}} is not affected: \index{for step=\index{root}} finds the real, geometric root of the current node. To access the formal root, use nodewalk step \index{root'}, or the corresponding propagator \index{for step=\index{root'}}. \indexitem{dynamic tree>sort}, \indexdef{dynamic tree>sort'} Sort the children of the current node, using the currently active sort key specified in \index{sort by} (see \S\ref{ref:nodewalk-operations}). |sort| sorts in ascending and |sort'| in descending order. \end{syntax} \subsection{Handlers} \label{ref:handlers} \begin{syntax} \indexitem{handler>option}|=|\meta{option} The result is the value of \meta{option} at the current node. \indexitem{handler>register}|=|\meta{register} The result is the value of \meta{register}. \indexitem{handler>pgfmath}|=|\meta{pgfmath expression} The result is the evaluation of \meta{pgfmath expression} in the context of the current node. \begin{advise} \item If you only need to access an option or register value, using \index{handler>option} or \index{handler>register} is much faster than using \keyname{.pgfmath}. \end{advise} \indexitem{handler>wrap value}|=|\meta{macro definition} The result is the (single) expansion of the given \meta{macro definition}. The defined macro takes one parameter. The current value of the handled option will be passed as that parameter. \indexitem{handler>wrap $n$ pgfmath args}|=|\meta{macro definition}\meta{arg $1$}\dots\meta{arg $n$} The result is the (single) expansion of the given \meta{macro definition}. The defined macro takes $n$ parameters, where $n\in\{2,\dots,8\}$. Expressions \meta{arg $1$} to \meta{arg $n$} are evaluated using |pgfmath| and passed as arguments to the defined macro. \indexitem{handler>wrap pgfmath arg}|=|\meta{macro definition}\meta{arg} Like \index{wrap $n$ pgfmath args} would work for $n=1$. \indexitem{handler>process args}|=|\meta{argument types}\meta{arg $1$}\dots\meta{arg $n$} The result is a list of processed arguments. Arguments are processed individually according to their given type. The types are given sequentially in \meta{argument types}, one character (no spaces) for each argument. The meanings of the characters are given in the table below. The exception to the one character -- one argument rule are characters connected by |+|, which apply iteratively (in the given order) to a single argument. If the argument type list is shorter than the number of argument, the remaining arguments are assumed to require no processing. \begin{tabular}{ll} |__|&no processing\\ |o|&expand once\\ |x|&fully expand\\ |O|&argument is an option name\\ |R|&argument is a register name\\ |P|&argument is a pgfmath expression\\ |r|&the argument should be a comma-separated list: reverse it\\ |+|&chain processors)\\ \end{tabular} Example: \index{split option} is defined by preprocessing the first argument of \index{split} as an option. \lstinputregion{forest.dtx}{split_option} \begin{forestexample}[index={delay,append,strequal,split,process args}] \begin{forest} [ {1,2,3,4,5}, delay={ my append/.style={if={strequal("#1","")}{}{append={[#1]}}}, split/.~process args~={O+r}{content}{,}{my append}, } ] \end{forest} \end{forestexample} \end{syntax} \subsubsection{Aggregate functions} \label{ref:aggregate-functions} Aggregate functions are defined as handlers over options. They walk a nodewalk, use the values of visited options to calculate something, and store the result in the handled option. Aggregate functions cannot be nested. \begin{syntax} \indexitem{aggregate>count}|=|\meta{nodewalk} Store the number of nodes visited in the nodewalk into the handled option. \indexitem{aggregate>sum}|=|\meta{pgfmath}\meta{nodewalk}\itemnosep \indexitem{aggregate>average}|=|\meta{pgfmath}\meta{nodewalk}\itemnosep \indexitem{aggregate>product}|=|\meta{pgfmath}\meta{nodewalk}\itemnosep \indexitem{aggregate>min}|=|\meta{pgfmath}\meta{nodewalk}\itemnosep \indexitem{aggregate>max}|=|\meta{pgfmath}\meta{nodewalk} Calculate the value of \meta{pgfmath} expression at each visited node. Store the sum / average / product / minimum / maximum of these values into the handled option. \indexitem{aggregate>aggregate}|=|\meta{pgfmath: start value}\meta{pgfmath: every step}\meta{pgfmath: after walk}\meta{nodewalk} The generic aggregate function. First, the result is set to \meta{pgfmath: start value}. Then, the given nodewalk is walked. After each step of the \meta{nodewalk}, \meta{pgfmath: every step} expression is evaluated in the context of the new current node and stored as the result. After the walk, the current node is reset to the origin. \meta{pgfmath: after walk} expression is then evaluated in its context and stored as the result. Use (appropriately escaped) |#1| to refer to the current value of the result in \meta{pgfmath: every step} and \meta{pgfmath: after walk}. Use count register \index{aggregate n} to refer to the current step number (in \meta{pgfmath: every step}) or the total number of steps (in \meta{pgfmath: after walk}) Example 1: see \index{option>s}. Example 2: the definition of \index{aggregate>average} below. Note that as we are using |.aggregate| inside a style definition, |##1| must be used to refer to the current value of the result. \begin{lstlisting}[] /handlers/~.average~/.style 2 args={ /handlers/~.aggregate~= {0} {(##1)+(#1)} {##1/\forestregister{aggregate n}} {#2} } \end{lstlisting} Example 3: randomly generate the content of leaves; the content of a parent is the sum of it's children's content. \begin{forestexample}[index={delay,aggregate postparse,for step,tree children-first,if option,n children,content,pgfmath,sum,step>children}] \begin{forest} delay={ aggregate postparse=int, for tree children-first={ if n children=0 {content/.pgfmath={random(0,100)}} {content/~.sum~={content}{children}} } } [[[[][]][]][[][][[][][]]][[][[][]]]] \end{forest} \end{forestexample} \indexitem{register>aggregate n}|=|\meta{count} In the every-step expression of an aggregate function, refers to the (real) step number in the aggregate's \meta{nodewalk}. In the after-walk expression, refers to the total number of (real) steps made. \indexitem{node key>aggregate postparse}|=|\alternative{none,int,print,macro} Roughly speaking, how should the result be formatted? For details, see \citep[\S89]{tikzpgf3.0.0}. \begin{syntax} \item\indexdef{value=aggregate postparse>none} No formatting. \item\indexdef{value=aggregate postparse>int} The result is an integer. \item\indexdef{value=aggregate postparse>print} Use |pgf|'s number printing extension, see \citep[\S93]{tikzpgf3.0.0}. \item\indexdef{value=aggregate postparse>macro} Use a custom macro. Specify the macro using \indexdef{node key>aggregate postparse macro}|=|\meta{cs}. \end{syntax} \end{syntax} \subsection{Relative node names} \label{ref:relative-node-names} \begin{syntax} \indexitem{meta>relative node name}|=|[\meta{forest node name}][\indexdef[set={code}]>{!}\meta{nodewalk}] \meta{relative node name} refers to the \foRest; node at the end of the \meta{nodewalk} starting at node named \meta{forest node name}. If \meta{forest node name} is omitted, the walk starts at the current node. If \meta{nodewalk} is omitted, the ``walk'' ends at the start node. (Thus, an empty \meta{relative node name} refers to the current node.) The \meta{nodewalk} inherits its history from the outer nodewalk (if there is one). Its every-step keylist is empty. \end{syntax} Relative node names can be used in the following contexts: \begin{itemize} \item \FoRest;'s |pgfmath| option functions (\S\ref{ref:pgfmath}) take a relative node name as their argument, e.g.\ |content("!u")| and |content("!parent")| refer to the content of the parent node. \item An option of a non-current node can be set by \meta{relative node name}|.|\meta{option name}|=|\meta{value}, see \S\ref{sec:option-setting}. \item The |forest| coordinate system, both explicit and implicit; see \S\ref{ref:forest-cs}. \end{itemize} \subsection{The \texttt{forest} coordinate system} \label{ref:forest-cs} Unless package options \index{tikzcshack} is set to |false|, \TikZ;'s implicit node coordinate system \citep[\S13.2.3]{tikzpgf2.10} is hacked to accept relative node names.\footnote{Actually, the hack can be switched on and off on the fly, using \cmdname{i}\keyname{foresttikzcshack}.}. The explicit \texttt{forest} coordinate system is called simply |forest| and used like this: |(forest_cs:|\meta{forest cs spec}|)|; see \citep[\S13.2.5]{tikzpgf2.10}. \meta{forest cs spec} is a keylist; the following keys are accepted. \begin{syntax} \indexitem{forest cs>name}|=|\meta{node name} The node with the given name becomed the current node. The resulting point is its (node) anchor. \indexitem{forest cs>id}|=|\meta{node id} The node with the given name becomed the current node. The resulting point is its (node) anchor. \indexitem{forest cs>go}|=|\meta{nodewalk} Walk the given nodewalk, starting at the current node. The node at the end of the walk becomes the current node. The resulting point is its (node) anchor. The embedded \meta{nodewalk} inherits history from the outer nodewalk. \indexitem{forest cs>anchor}|=|\meta{anchor} The resulting point is the given anchor of the current node. \indexitem{forest cs>l}|=|\meta{dimen} \vspace{-\parskip} \indexitem{forest cs>s}|=|\meta{dimen} Specify the \index{option>l} and \index{option>s} coordinate of the resulting point. The coordinate system is the node's ls-coordinate system: its origin is at its (node) anchor; the l-axis points in the direction of the tree growth at the node, which is given by option \index{grow}; the s-axis is orthogonal to the l-axis; the positive side is in the counter-clockwise direction from |l| axis. The resulting point is computed only after both \index{option>l} and \index{option>s} were given. \item Any other key is interpreted as a \index{relative node name}[.\meta{anchor}]. \end{syntax} \subsection{Anchors} \label{sec:anchors} \FoRest; defines several anchors which can be used with any \TikZ; node belonging to a \FoRest; tree (manually added \TikZ; nodes are thus excluded). \begin{syntax} \indexitem{anchor>parent anchor}\itemnosep \indexitem{anchor>child anchor}\itemnosep \indexitem{anchor>anchor} These anchors point to coordinates determined by node options \index{option>parent anchor}, \index{option>child anchor} and \index{option>anchor}. \indexitem{anchor>parent}, \indexdef{anchor>parent'}\itemnosep \indexitem{anchor>parent first}, \indexdef{anchor>parent first'}\itemnosep \indexitem{anchor>first}, \indexdef{anchor>first'}\itemnosep \indexitem{anchor>children first}, \indexdef{anchor>children first'}\itemnosep \indexitem{anchor>children}, \indexdef{anchor>children}\itemnosep \indexitem{anchor>children last}, \indexdef{anchor>children last}\itemnosep \indexitem{anchor>last}, \indexdef{anchor>last}\itemnosep \indexitem{anchor>parent last}, \indexdef{anchor>parent last} Growth direction based anchors. \TikZ;'s ``compass anchors'' |east|, |north| etc.\ resolve to coordinates on the border of the node facing east, north etc.\ (for the shapes that define these anchors). The above \FoRest;'s anchors are similar in that they also resolve to coordinates on the border of the node. However, the ``cardinal directions'' are determined by the \index{grow}th direction of the tree in the node, so that \begin{itemize} \item anchor |parent| faces the parent node (or, in case of the root, where the parent would be); \item anchor |children| faces the children (or, in case of a node without children, where the children would be); \item anchor |first| faces the first child (or \dots\ you get it, right?); \item anchor |last| faces the last child (or \dots\ you know!). \end{itemize} Combinations like |parent_children| work like combinations of compass directions, e.g.\ |north_west|. These anchors work as expected even if the node is \index{rotate}d, or if the children are \index{reversed}. The |'| variants refer precisely to the point where the cardinal growth direction intersects the border. Variants without |'| snap to the closest compass anchor of the node. For simple examples, see definitions of \index{sn edges} and \index{roof}; for more involved examples, see the \reflibrary{edges} library. \end{syntax} \subsection{Additional \texttt{pgfmath} functions} \label{ref:pgfmath} For every option and register, \foRest; defines a |pgfmath| function with the same name, with the proviso that the name might be mangled in order to conform to |pgfmath|'s naming rules. Specifically, all non-alphanumeric characters in the option/register name and the initial number, if the name starts with one, are replaced by an underscore |__| in the |pgfmath| function name. Pgfmath functions corresponding to options take one argument, a \index{relative node name} (see~\S\ref{ref:relative-node-names}) expression, making it possible to refer to option values of non-current nodes. The \meta{relative node name} expression must be enclosed in double quotes in order to prevent pgfmath evaluation: for example, to refer to the content of the parent, write \index{content}|("|\index>{!}|u")|. To refer to the option of the current node, use empty parentheses: \index{content}|()|.\footnote{In most cases, the parentheses are optional, so \texttt{content} is ok. A known case where this doesn't work is preceding an operator: \texttt{l+1cm} will fail.} If the \index{relative node name} resolves to the invalid node, these functions will an return empty token list (for \meta{toks} options), 0pt (for \meta{dimen} options) or 0 (for \meta{count} options). Note that the nodewalk in the relative node name inherits its history from the outer nodewalk (if there is one), so strange but useful constructions like the following are possible. \begin{forestexample}[index={tikz,wrap pgfmath arg,nodewalk,back,b,ancestors,register>every step}] \begin{forest} for tree={no edge}, before typesetting nodes={ for nodewalk={ c, every step={ tikz/.wrap pgfmath arg= {\draw[<-] ()--(#1);} {name("~!b~")} }, 21{up1},ancestors }{} }, [1[2[3]][4[5]]] \end{forest} \end{forestexample} Boolean function \indexdef{pgfmath>valid} returns true if the node's \index{readonly option>id}$\neq0$, i.e.\ if the node is a real, valid node; see \S\ref{ref:spatial-propagators} and \S\ref{ref:nodewalks}. Boolean function \indexdef{pgfmath>invalid} is a negation of |valid|. \begin{syntax} \indexitem[set={+format={detokenize}}]{pgfmath>min_l}|=(|\meta{nodewalk: node}|,|\meta{nodewalk: context node}|)| \itemnosep \indexitem[set={+format={detokenize}}]{pgfmath>min_s}|=(|\meta{nodewalk: node}|,|\meta{nodewalk: context node}|)| \itemnosep \indexitem[set={+format={detokenize}}]{pgfmath>max_l}|=(|\meta{nodewalk: node}|,|\meta{nodewalk: context node}|)| \itemnosep \indexitem[set={+format={detokenize}}]{pgfmath>max_s}|=(|\meta{nodewalk: node}|,|\meta{nodewalk: context node}|)| These functions return the minimum/maximum value of \index{option>l}/\index{option>s} of node at the end of \meta{nodewalk: node} in the context (i.e.\ growth direction) of node at the end of \meta{nodewalk: context node}. \end{syntax} Three string functions are also added to |pgfmath|: \indexdef{pgfmath>strequal} tests the equality of its two arguments; \indexdef{pgfmath>instr} tests if the first string is a substring of the second one; \indexdef{pgfmath>strcat} joins an arbitrary number of strings. \bigskip Some random notes on |pgfmath|: \begin{enumerate*}[(i)] \item |&&|, \verb!||! and |!| are boolean ``and'', ``or'' and ``not'', respectively. \item The equality operator (for numbers and dimensions) is |==|, \emph{not} |=|. \end{enumerate*} And some examples: \begin{forestexample}[layout=tree on top,index={for step,tree,grow',calign,option>l,l sep,option>child anchor,option>anchor,fit,tier,readonly option>level,delay,before typesetting nodes,content,wrap $n$ pgfmath args,pgfmath},index>={!}] \begin{forest} for tree={grow'=0,calign=first,l=0,l sep=2em,child anchor=west,anchor=base west,fit=band,tier/.pgfmath=~level~()}, fullpath/.style={if n=0{}{content/.wrap 2 pgfmath args={##1/##2}{~content~("!u")}{~content~()}}}, delay={for tree=fullpath,content=/}, before typesetting nodes={for tree={content=\strut#1}} [ [home [joe [\TeX]] [saso [\TeX]] [a user with a long name [\TeX]]] [usr]] \end{forest} \end{forestexample} \begin{forestexample}[point=instr,index={delay,for step,tree,conditional>if,content,n children}] % mark non-phrasal terminal nodes \begin{forest} delay={for tree={if= {!instr("P",~content~) && ~n_children~==0} {fill=yellow} {} }} [CP[DP][C'[C][TP[DP][T'[T][VP[DP][V'[V][DP]]]]]]] \end{forest} \end{forestexample} \begin{forestexample}[point=instr,index={where option,n children,tier,content,no edge,tikz},index>={!}] % roof terminal phrases \begin{forest} where n children=0{tier=word, if={~instr~("P",~content~("!u"))}{no edge, tikz={\draw (!.north west)-- (!.north east)--(!u.south)--cycle; }}{} }{}, [VP[DP[Mary]][V'[V[loves]][DP[her cat]]]] \end{forest} \end{forestexample} \subsection{Standard node} \label{ref:standard-node} \begin{syntax} \indexitem{macro>forestStandardNode}\meta{node}\meta{environment fingerprint}\meta{calibration procedure}\meta{exported options} This macro defines the current \emph{standard node}. The standard node declares some options as \emph{exported}. When a new node is created, the values of the exported options are initialized from the standard node. At the beginning of every \index{forest} environment, it is checked whether the \emph{environment fingerprint} of the standard node has changed. If it did, the standard node is \emph{calibrated}, adjusting the values of exported options. The \emph{raison d'etre} for such a system is given in \S\ref{tut:defaults}. In \meta{node}, the standard node's content and possibly other options are specified, using the usual bracket representation. The \meta{node}, however, \emph{must not contain children}. The default: \texttt{[dj]}. The \meta{environment fingerprint} must be an expandable macro definition. It's expansion should change whenever the calibration is necessary. \meta{calibration procedure} is a keylist (processed in the |/forest| path) which calculates the values of exported options. \meta{exported options} is a comma-separated list of exported options. This is how the default standard node is created: \begin{lstlisting} \forestStandardNode[dj] {% \forestOve{\csname forest@id@of@standard node\endcsname}{content},% \the\ht\strutbox,\the\pgflinewidth,% \pgfkeysvalueof{/pgf/inner ysep},\pgfkeysvalueof{/pgf/outer ysep},% \pgfkeysvalueof{/pgf/inner xsep},\pgfkeysvalueof{/pgf/outer xsep}% } { l sep={\the\ht\strutbox+\pgfkeysvalueof{/pgf/inner ysep}}, l={l_sep()+abs(max_y()-min_y())+2*\pgfkeysvalueof{/pgf/outer ysep}}, s sep={2*\pgfkeysvalueof{/pgf/inner xsep}} } {l sep,l,s sep} \end{lstlisting} \end{syntax} \subsection{Externalization} \label{ref:externalization} Externalized tree pictures are compiled only once. The result of the compilation is saved into a separate |.pdf| file and reused on subsequent compilations of the document. If the code of the tree (or the context, see below) is changed, the tree is automatically recompiled. Externalization is enabled by: \begin{lstlisting} \usepackage[~external~]{forest} ~\tikzexternalize~ \end{lstlisting} Both lines are necessary. \TikZ;'s externalization library is automatically loaded if necessary. \begin{syntax} \indexitem{node key>external/optimize} Parallels \keyname{/tikz/external/optimize}: if |true| (the default), the processing of non-current trees is skipped during the embedded compilation. \indexitem{node key>external/context} If the expansion of the macro stored in this option changes, the tree is recompiled. \indexitem{node key>external/depends on macro}|=|\meta{cs} Adds the definition of macro \meta{cs} to \keyname{external/context}. Thus, if the definition of \meta{cs} is changed, the tree will be recompiled. \end{syntax} \foRest; respects or is compatible with several (not all) keys and commands of \TikZ;'s externalization library. In particular, the following keys and commands might be useful; see \cite[\S32]{tikzpgf2.10}. \begin{itemize} \item\keyname{/tikz/external/remake next} \item\keyname{/tikz/external/prefix} \item\keyname{/tikz/external/system call} \item\cmdname{tikzexternalize} \item\cmdname{tikzexternalenable} \item\cmdname{tikzexternaldisable} \end{itemize} \FoRest; does not disturbe the externalization of non-\foRest; pictures. (At least it shouldn't \dots) The main auxiliary file for externalization has suffix |.for|. The externalized pictures have suffices |-forest-|$n$ (their prefix can be set by \keyname{/tikz/external/prefix}, e.g.\ to a subdirectory). Information on all trees that were ever externalized in the document (even if they were changed or deleted) is kept. If you need a ``clean'' |.for| file, delete it and recompile. Deleting |-forest-|$n$|.pdf| will result in recompilation of a specific tree. Using \keyname{draw tree} and \keyname{draw tree'} multiple times \emph{is} compatible with externalization, as is drawing the tree in the box (see \index{draw tree box}). If you are trying to externalize a \index{forest} environment which utilizes \index{TeX} to produce a visible effect, you will probably need to use \index{TeX'} and/or \index{TeX''}. \setcounter{CodelineNo}{0} \DocInput{forest-libs.dtx} \section{Gallery} \label{sec:gallery} \subsection{Decision tree} \label{sec:example-decision-tree} The following example was inspired by a question on \TeX\ Stackexchange: \href{http://tex.stackexchange.com/questions/39103/how-to-change-the-level-distance-in-tikz-qtree-for-one-level-only}{How to change the level distance in tikz-qtree for one level only?}. The question is about |tikz-qtree|: how to adjust the level distance for the first level only, in order to avoid first-level labels crossing the parent--child edge. While this example solves the problem (by manually shifting the offending labels; see \texttt{elo} below), it does more: the preamble is setup so that inputing the tree is very easy. \begin{forestexample}[layout=tree on top,index={if option,readonly option>n,no edge,tikz,strequal,strcat,option>child anchor,option>parent anchor,option>anchor,anchor>anchor,calign,for step,tree,s sep,option>l,n children,declare toks,delay,content,before typesetting nodes,descendants,keylist+=tikz,wrap pgfmath arg,wrap $n$ pgfmath args,split option},index>={!}] \forestset{ declare toks={elo}{}, % Edge Label Options anchors/.style={anchor=#1,child anchor=#1,parent anchor=#1}, dot/.style={tikz+={\fill (.child anchor) circle[radius=#1];}}, dot/.default=2pt, decision edge label/.style n args=3{ edge label/.expanded={node[midway,auto=#1,anchor=#2,\forestoption{elo}]{\strut$\unexpanded{#3}$}} }, decision/.style={if n=1 {decision edge label={left}{east}{#1}} {decision edge label={right}{west}{#1}} }, decision tree/.style={ for tree={ s sep=0.5em,l=8ex, if n children=0{anchors=north}{ if n=1{anchors=south east}{anchors=south west}}, math content, }, anchors=south, outer sep=2pt, dot=3pt,for descendants=dot, delay={for descendants={~split option~={content}{;}{content,decision}}}, } } \begin{forest} decision tree [N,plain content [I;{p_1=0.5},plain content,elo={yshift=4pt} [{5,1};a] [II;b,plain content [{1,2};m] [{2,3};n] ] ] [II;{p_2=0.5},plain content,elo={yshift=4pt} [;c [{1,0};z] [{2,2};t] ] [;d [{3,1};z] [{0,0};t] ] ] {\draw[dashed](!1.anchor)--(!2.anchor) node[pos=0.5,above]{I};} ] \end{forest} \end{forestexample} \setcounter{CodelineNo}{0} \DocInput{forest-index.dtx} \section{Past, present and future} \label{sec:changes} \paragraph{Roadmap} What's planned for future releases? \begin{itemize} \item filling up the libraries \item faster externalization \item custom-edge aware packing algorithm and a more flexible (successor of) \index{calign} \item support for specialized |forest| environments, including: \begin{itemize} \item selectable input parser, \item namespaces (different function, different options), \item better support for different output types. \end{itemize} In short, everything you need to make \foRest; your favourite spreadsheet! ;-) \item code cleanup and extraction of sub-packages possibly useful to other package writers \end{itemize} \subsection{Changelog} \label{sec:changelog} First of all, the list of all \index{compat} keys for backward compatibility, and their groupings. \lstinputregion{forest-compat.sty}{compat_keys} \subsubsection{v2.0} \newcommand\compatitem[2]{\item[,,in \texttt{-#1}]\texttt{compat=}\indexdef{option of=compat>#2}} \begin{description} \item[v2.0.3 (2016/04/03)] \mbox{} Backward incompatible changes (with a \index{compat} key): \begin{syntax} \compatitem{all}{2.0.2-delayn} \itemnosep \compatitem{all}{2.0.2-wrapnpgfmathargs} This is really a bugfix: keys \index{delay n} and \index{wrap $n$ pgfmath args} (for $n\geq 2$) were introducing two layers of hash doubling. Now this confusing behaviour is gone, but as finding the correct number of hashes is always a tough job, \index{compat} keys are provided. \end{syntax} Improvements: \begin{itemize} \item Rework \index{draw tree edge} so that by default, an edge is drawn only it both its node and its node's parent are drawn. And yes, implement \index{if node drawn}. \item Implement circularity detection in dynamic node operations. \item Implement debug categories and debugging of dynamic node operations. \item Declare some further |tempdim...| registers. \item Make option \index{readonly option>id} accessible via \index{forestoption}. \end{itemize} Bugfixes: \begin{itemize} \item Execute \index{tikz} code for all (including phantom) nodes. (The feature of ignoring phantom nodes was introduced in v2.0.2, but turns out it was a bad idea: for example, having a phantom root with some tikz code is not uncommon.) \item Keys \index{label} and \index{pin} now \emph{append} to option \index{tikz}, as makes sense. \item Fix nodewalk steps \index{filter} and \index{branch} so that they can be embedded under nodewalk operations. (Uh, and recategorize them as operations themselves.) \item Execute \index{before packing node} even when the node has no children. \item \index{step>level}|<={0}{...}| now works as expected. \item Re-setting the node name to the same value doesn't yield an error anymore. \item Don't add the separator when adding the first element to a keylist option or register. \item Copy externalization files in TeX (don't rely on |\write18|). \item Consistently store dimen options and registers with |pt|s of catcode other. \item Properly initialize readonly count options (\index{readonly option>n}, \index{readonly option>n'}, \index{n children} and \index{readonly option>level}). \item Fix some typos. \end{itemize} \item[v2.0.2 (2016/03/04)] \mbox{} Backward incompatible changes: \begin{itemize} \item The semantics of the parenthesized optional argument to \index{forest} environment and \index{Forest} macro has changed. The argument was introduced in v2.0.0: if present, it redefined \index{style>stages} style for the current environment/macro. This argument is now generalized to allow further (pre-\index{style>stages}) customization in future versions of the package. To temporarily redefine \index{style>stages}, write |(|\index{option of=forest>stages}|={...})|. \end{itemize} New functionality: \begin{itemize} \item Key \index{last dynamic node} and named nodewalk \index{dynamic nodes}. \item An optional argument to \index{useforestlibrary} to pass package options to libraries. \item Handler \index{nodewalk style}. \item Keys \index{draw tree node'}, \index{draw tree edge'} and \index{draw tree tikz'}. \end{itemize} Bugfixes: \begin{itemize} \item Fixed \index{replace by} when applied to the root node. \item Registers are now initialized to an empty string, 0pt, or 0. \item Packing doesn't destroy the current pgfpath anymore. \item \index{forestStandardNode} now uses \index{name'}. \item \index{draw tree edge} now respects \index{phantom}. \end{itemize} \item[v2.0.1 (2016/02/20)] \mbox{} New functionality: \begin{itemize} \item \index{current and siblings}, \index{current and siblings reversed} \item Add |*| argument to \index{useforestlibrary}. \end{itemize} Bugfixes: \begin{itemize} \item Correctly mangle option/register names to pgfmath names (\S\ref{ref:pgfmath}). \item Refer to parent (not node) anchor in \index{calign}|=|\index{edge midpoint}. \item Accept key \index{history} in \index{step>Nodewalk} config. \end{itemize} \item[v2.0.0 (2016/01/30)] \mbox{}\footnote{The year of the release date in the package was wrong \dots \ 2015.} Backwards incompatible changes (\emph{without} a \index{compat} key --- sorry!): \begin{itemize} \item The unintended and undocumented way to specify defaults using |\forestset{.style={...}}| (see question \href{http://tex.stackexchange.com/questions/167972/making-a-certain-tree-style-the-default-for-forest}{Making a certain tree style the default for forest} at \TeX\ SE) does not work anymore. (Actually, it has never truly worked, and that's why it has not \index{compat} key.) Use \index{default preamble}. \item Renamed augmented assignment operator \meta{option}|-| for prepending to \meta{toks} and \meta{keylist} options \index{+toks=\meta{option}}\index[not print]{+keylist}. A new \index{keylist-=\meta{option}} is defined for keylist options and means ``delete key from keylist.'' \item Short nodewalk steps are not simply styles anymore: use \index{define short step} to define them. \end{itemize} Backwards incompatible changes with a \index{compat} key: \begin{syntax} \compatitem{all}{1.0-stages} Processing of \index{given options}, which is now exposed, and the new keylists \index{default preamble} and \index{preamble} is now included at the start of the default \index{style>stages} style. When changing \index{style>stages}, the instruction to process these keylists must now be given explicitely. \compatitem{all}{1.0-forstep} In v1.0, a spatial propagator \index{for step=\meta{step}} could never fail. This turned out to be difficult to debug. In this version, when a propagator steps ``out of the tree'', an error is raised by default. Check out \index{option of=Nodewalk>on invalid} to learn how to simulate the old behaviour without using this compatibility key. \compatitem{all}{1.0-rotate} This version of the package introduces option \index{rotate} and \index{autoforward}s it to \index{node options}. This is needed to handle the new \foRest; anchors (\S\ref{sec:anchors}). However, in some rare cases (like the tree on the title page of this manual) it can lead to a discrepancy between the versions, as the time when the value given to \index{rotate} is processed is different. |1.0-rotate| removes option \index{rotate}. \compatitem{all}{1.0-name} Documentation of v1.0 requested that node names be unique, but this was not enforced by the package, sometimes leading to errors. v2.0 enforces node name uniqueness. If this causes problems, use this compatibility key. In most cases using \index{name'} instead of \index{option>name} should fix the problem without using compatibility mode. \item These keys have been renamed: \begin{tabularx}{\linewidth}{lll} old&new&\index{compat} key {\small (all but the last are in \texttt{-most})}\\\hline \keyname{node walk}&\index{for step=nodewalk}\footnote{Nodewalks are much improved in v2.0, so some syntax and keys are different than in v1.0!}&\indexdef{option of=compat>1.0-nodewalk}\\ \keyname{for}&\index{for step=\index{group}}&\indexdef{option of=compat>1.0-for}\\ \keyname{for all next}&\index{for step=\index{following siblings}}&\indexdef{option of=compat>1.0-forall}\\ \keyname{for all previous}&\index{for step=\index{preceding siblings}}&\indexdef{option of=compat>1.0-forall}\\ \keyname{for ancestors'}&\index{for step=\index{current and ancestors}}&\indexdef{option of=compat>1.0-ancestors}\\ (\keyname{for}) \keyname{linear next}&(\index{for step}) \index{next node}&\indexdef{option of=compat>1.0-linear}\\ (\keyname{for}) \keyname{linear previous}&(\index{for step}) \index{previous node}&\indexdef{option of=compat>1.0-linear}\\ \keyname{triangle}&\index{roof} (library \reflibrary{linguistics})&\indexdef{option of=compat>1.0-triangle}\\ \keyname{/tikz/fit to tree}&\index{fit to}\keyname{=}\index{tree}\footnote{The v1.0 key \keyname{/tikz/fit to tree} also set \keyname{inner sep=0}; the v2.0 key \index{fit to} does not do that.}&\indexdef{option of=compat>1.0-fittotree}\\ \keyname{begin forest}, \keyname{end forest}&none (use \index{style>stages})&\index{1.0-stages}\\ \keyname{end forest}, \keyname{end forest}&none (use \index{style>stages})&\index{1.0-stages}\\ \end{tabularx} \end{syntax} Good news: \begin{itemize} \item Added temporal propagators \index{before packing node} and \index{after packing node}. \item \emph{Much} improved nodewalks, see \S\ref{ref:nodewalks} and \S\ref{ref:spatial-propagators}. \item Implemented looping mechanisms and more conditionals, see \S\ref{ref:conditionals}. \item Implemented library support and started filling up the libraries: \begin{itemize} \item \reflibrary{linguistics}: \index{sn edges}, \index{nice empty nodes}, \index{draw brackets}, \index{c-commanded} and \index{c-commanders} \item \reflibrary{edges}: \index{forked edges} and \index{folder} \end{itemize} \item Implemented aggregate functions, see \S\ref{ref:aggregate-functions}. \item Added key \index{default preamble}. \item Implemented anchors \index{anchor>parent}, \index{anchor>children}, \index{anchor>first}, \index{anchor>last}, etc. \item Added key \index{split} and friends. \item Implemented sorting of children, see \S\ref{ref:dynamic}. \item Introduced registers, see \S\ref{sec:options-and-registers}. \item Implemented handlers \index{handler>option}, \index{handler>register} and \index{process args}. \item Implemented several friends to \index{process keylist}, introduced \index{processing order}s and \index{draw tree method}. \item Added the optional argument |(|\meta{stages}|)| to the \index{forest} environment and \index{Forest} macro. \item Implemented \index{autoforward}ing. \item Implemented flexible handling of unknown keys using \index{unknown to}. \item Implemented |pgfmath| functions \index{pgfmath>min_l}, \index{pgfmath>max_l}, \index{pgfmath>min_s}, \index{pgfmath>max_s}. \item Implemented augmented assignment operator \index{keylist-=\meta{keylist option}} for removing keys from keylists. \item Implemented a generalized \index{fit to} key. \item Implemented a very slow \foRest;-based indexing system (used to index this documentation) and included it in the gallery (\S\ref{sec:forest-index}). \item Added some minor keys: \index{edge path'}, \index{node format'}, \index{create'} and \index{plain content}. \item Added some developer keys: \index{copy command key}, \index{typeout}. \end{itemize} Bugfixes: \begin{itemize} \item In computation of numeric tree-structure info, when called for a non-root node. \item \TikZ;'s externalization internals (signature of |\tikzexternal@externalizefig@systemcall@uptodatecheck|) have changed: keep up to date, though only formally. \item \index{delay} was not behaving additively. \item \index{option>name}, \index{alias} and \index{baseline} didn't work properly when setting them for a non-current node. \item Augmented assignments for count options were leaking `.0pt'. \item \index{create} didn't work properly in some cases. \item \keyname{triangle} (now \index{roof} in \reflibrary{linguistics}) didn't use |cycle| in the edge path \end{itemize} \end{description} \subsubsection{v1.0} \begin{description} \item[v1.0.10 (2015/07/22)] \mbox{} \begin{compactitem} \item Bugfix: a left-over debugging |\typeout| command was interfering with a |forest| within |tabular|, see \href{http://tex.stackexchange.com/questions/256509/odd-incompatibility-between-multi-line-forest-nodes-and-tabular}{this question on TeX.SE}. \item A somewhat changed versioning scheme \dots \end{compactitem} \item[v1.09 (2015/07/15)] \mbox{} \begin{compactitem} \item Bugfix: child alignment was not done in nodes with a single child, see \href{http://tex.stackexchange.com/questions/255309/elementary-forest-question-meaning-of-calign-parent-anchor-and-child-anchor}{this question on TeX.SE}. \end{compactitem} \item[v1.08 (2015/07/10)] \mbox{} \begin{compactitem} \item Fix externalization (compatibility with new |tikz| features). \end{compactitem} \item[v1.07 (2015/05/29)] \mbox{} \begin{compactitem} \item Require package |elocalloc| for local boxes, which were previously defined by package |etex|. \end{compactitem} \item[v1.06 (2015/05/04)] \mbox{} \begin{compactitem} \item Load |etex| package: since v2.1a, |etoolbox| doesn't do it anymore. \end{compactitem} \item[v1.05 (2014/03/07)] \mbox{} \begin{compactitem} \item Fix the node boundary code for rounded rectangle. (Patch contributed by Paul Gaborit.) \end{compactitem} \item[v1.04 (2013/10/17)] \mbox{} \begin{compactitem} \item Fixed an \href{http://tex.stackexchange.com/questions/138986/error-using-tikzexternalize-with-forest/139145}{externalization bug}. \end{compactitem} \item[v1.03 (2013/01/28)] \mbox{} \begin{compactitem} \item Bugfix: options of dynamically created nodes didn't get processed. \item Bugfix: the bracket parser was losing spaces before opening braces. \item Bugfix: a family of utility macros dealing with affixing token lists was not expanding content correctly. \item Added style \index{math content}. \item Replace key \keyname{tikz preamble} with more general \index{begin draw} and \index{end draw}. \item Add keys \keyname{begin forest} and \keyname{end forest}. \end{compactitem} \item[v1.02 (2013/01/20)] \mbox{} \begin{compactitem} \item Reworked style \index{style>stages}: it's easier to modify the processing flow now. \item Individual stages must now be explicitely called in the context of some (usually root) node. \item Added \index{delay n} and \index{if have delayed}. \item Added (experimental) \index{pack'}. \item Added reference to the \href{https://github.com/sasozivanovic/forest-styles}{style repository}. \end{compactitem} \item[v1.01 (2012/11/14)] \mbox{} \begin{compactitem} \item Compatibility with the |standalone| package: temporarily disable the effect of |standalone|'s package option |tikz| while typesetting nodes. \item Require at least the [2010/08/21] (v2.0) release of package |etoolbox|. \item Require version [2010/10/13] (v2.10, rcs-revision 1.76) of \PGF;/\TikZ;. Future compatibility: adjust to the change of the ``not yet positioned'' node name (2.10 |@| $\rightarrow$ 2.10-csv |PGFINTERNAL|). \item Add this changelog. \end{compactitem} \item[v1.0 (2012/10/31)] First public version \end{description} \paragraph{Acknowledgements} This package has turned out to be much more successful and widespread than I could have ever imagined and I want to thank all the users for the trust. Many of you have also contributed to the package in some way: by providing comments and ideas, sending patches, reporting bugs and so on. To you, I'm doubly grateful! I will not even try to list you all here, as the list is getting too long for me to maintain, but I do want to mention one person, a member of the friendly community at the excellent and indispensable \href{http://tex.stackexchange.com}{\TeX\ -- \LaTeX\ Stack Exchange} and the author of the very first \foRest;-based package, \href{http://ctan.org/pkg/prooftrees}{Prooftrees}: without |cfr|'s uncountable questions, answers, bug reports and ideas, \foRest; would be a much poorer package indeed. \subsection{Known bugs} \label{sec:bugs} If you find a bug (there are bound to be some \dots), please contact me at \href{mailto:saso.zivanovic@guest.arnes.si}{saso.zivanovic@guest.arnes.si}. \subsubsection*{System requirements} This package requires \LaTeX\ and e\TeX. If you use something else: sorry. The requirement for \LaTeX\ might be dropped in the future, when I get some time and energy for a code-cleanup (read: to remedy the consequences of my bad programming practices and general disorganization). The requirement for e\TeX\ will probably stay. If nothing else, \foRest; is heavy on boxes: every node requires its own \dots\ and consequently, I have freely used e\TeX\ constructs in the code \dots \subsubsection*{\PGF; internals} \FoRest; relies on some details of \PGF; implementation, like the name of the ``not yet positioned'' nodes. Thus, a new bug might appear with the development of \PGF;. If you notice one, please let me know. \subsubsection*{Edges cutting through sibling nodes} \label{sec:cutting-edge} In the following example, the R--B edge crosses the AAA node, although \index{ignore edge} is set to the default |false|. \begin{forestexample}[index={calign,{value=calign>first},align,{value=align>center},base,{value=base>bottom}}] \begin{forest} calign=first [R[AAAAAAAAAA\\AAAAAAAAAA\\AAAAAAAAAA,align=center,base=bottom][B]] \end{forest} \end{forestexample} This happens because s-distances between the adjacent children are computed before child alignment (which is obviously the correct order in the general case), but child alignment non-linearly influences the edges. Observe that the with a different value of \index{calign}, the problem does not arise. \begin{forestexample}[index={calign,{value=calign>last},align,{value=align>center},base,{value=base>bottom}}] \begin{forest} calign=last [R[AAAAAAAAAA\\AAAAAAAAAA\\AAAAAAAAAA,align=center,base=bottom][B]] \end{forest} \end{forestexample} While it would be possible to fix the situation after child alignment (at least for some child alignment methods), I have decided against that, since the distances between siblings would soon become too large. If the AAA node in the example above was large enough, B could easily be pushed off the paper. The bottomline is, please use manual adjustment to fix such situations. \subsubsection*{Orphans} \label{sec:orphans} If the \index{option>l} coordinates of adjacent children are too different (as a result of manual adjustment or tier alignment), the packing algorithm might have nothing so say about the desired distance between them: in this sense, node C below is an ``orphan.'' \begin{forestexample}[index={for step,tree,s sep,option>l,dimen*}] \begin{forest} for tree={s sep=0,draw}, [R[A][B][C,l*=2][D][E]] \end{forest} \end{forestexample} To prevent orphans from ending up just anywhere, I have decided to vertically align them with their preceding sibling --- although I'm not certain that's really the best solution. In other words, you can rely that the sequence of s-coordinates of siblings is non-decreasing. The decision also influences a similar situation illustrated below. The packing algorithm puts node E immediately next to B (i.e.\ under C): however, the monotonicity-retaining mechanism then vertically aligns it with its preceding sibling, D. \begin{forestexample}[index={for step,tree,s sep,tier}] \begin{forest} for tree={s sep=0,draw}, [R[A[B,tier=bottom]][C][D][E,tier=bottom]] \end{forest} \end{forestexample} Obviously, both examples also create the situation of an edge crossing some sibling node(s). Again, I don't think anything sensible can be done about this, in general. \addcontentsline{toc}{section}{References} \bibliography{tex} \bibliographystyle{plain} \newpage \phantomsection \addcontentsline{toc}{section}{\indexname} \addtocontents{toc}{\protect\setbox0=\protect\vbox{\protect\iffalse}\protect\fi} \PrintIndex \addtocontents{toc}{\protect\iffalse{\protect\fi}} \end{document} %%% Local Variables: %%% mode: latex %%% fill-column: 100 %%% TeX-command-default: "Make PDF" %%% TeX-master: t %%% End: