Possible bug with arrowhead positioning & orientation

[alexreg]

Here's a MWE.

\documentclass{standalone}

\usepackage{tikz}

\usetikzlibrary{snakes, arrows.meta, shapes}

\begin{document}

\tikzset{
	node/.style = {draw, shape=rectangle, minimum width=3cm},
	edge/.style = {draw, >=Stealth},
}

\begin{tikzpicture}[line join=bevel, scale=1.0]
	\node (foo) at (1714.7bp,355.8bp) [node] {foo};
	\node (bar) at (1018.7bp,180.8bp) [node] {bar};
	\draw [->,edge] (foo) ..controls (1698.9bp,331.68bp) and (1662.1bp,282.02bp)  .. (1617.7bp,260.55bp) .. controls (1394.9bp,152.86bp) and (1310.8bp,222.14bp)  .. (1065.7bp,188.55bp) .. controls (1061.7bp,188.01bp) and (1057.6bp,187.43bp)  .. (bar);
\end{tikzpicture}

\end{document}

As you can see, the arrow head not only has the wrong orientation (by 180 degrees) but also it extends beyond the endpoint of the edge! What's going on here?

N.B. This output is part of a file produced by dot2tex, and looks perfectly reasonable, as far as I can tell.

[muzimuzhi]

A fuzzy guess is that, the last curve is too short to fit in an arrow, thus you got a degenerated result.

% the last curve
(1065.7bp,188.55bp) .. controls (1061.7bp,188.01bp) and (1057.6bp,187.43bp)  .. (bar)

As far as the last subpath (whether it's a curve or line-to) is short enough, one got a degenerated result. Perhaps pgf/tikz should raise an arrow in such cases.

\documentclass[tikz, margin=5pt]{standalone}
% \usepackage{tikz}
\usetikzlibrary{arrows.meta}

\begin{document}

\tikzset{
  >=Stealth
}

\begin{tikzpicture}
  % original coordinates
  % (1069.69637pt,189.25706pt) .. controls (1065.68137pt,188.71503pt) and (1061.566pt,188.13284pt) .. (1063.61786pt,188.48318pt)
  \draw[->]
    (0, 0) .. controls (4pt, .5pt) and (8pt, 1pt) .. (6pt, 0.8pt);

\end{tikzpicture}

\begin{tikzpicture}[>=Stealth]
  \foreach \i [count=\cnt] in {0, .5, 1, 1.5, 2, 3, 4, 5} {
    \draw[->, xshift={0.5cm*\cnt}, , yshift=-.5cm, x=1pt*\i, y=1pt*\i]
      (0, 0) .. controls (4, .5) and (8, 1) .. (6, .8);
    \draw[->, xshift={0.5cm*\cnt}, x=1pt*\i, y=1pt*\i]
      (0, 0) -- (6, .8);
  }
\end{tikzpicture}

\end{document}

[alexreg]

I see. It sounds like a pretty straightforward thing to implement. Just removing the special casing? Or is there some nuance I’m not aware of? I could submit a PR myself in any case.

[muzimuzhi]

Entrance of the rabbit hole: \pgf@parse@end

• May need to merge the last n (n >= 1) subpaths to get a new, long-enough subpath. (But the distance between subpath[n].start to subpath[-1].end is not necessarily larger than that of subpath[n-1].start to subpath[-1].end.)
• Compatibility with the bending library?

[alexreg]

Thank you. The architecture of the package looks overwhelming, but I'll give it a go!

Is there a reason you wouldn't take the approach of measuring the actual total path length, rather than this one?

[muzimuzhi]

I don't know why. When I started contributing to this repo, most of its functionalities had been stable and the original author, Till Tantau had been inactive. Probably @hmenke know more.

[alexreg]

@muzimuzhi Good to know. Perhaps he can reply here when he has a moment, let's see.