Draw a curve as an outline to another curve

Question

How can I offset a hobby curve, so that the resulting curve serves as a shifted and parallel outline to the original curve? Here is what I got so far:

\documentclass[border=5pt,tikz]{standalone}

\usepackage{tikz}
\usetikzlibrary{calc,hobby}

\tikzset{
    point/.style={circle,inner sep=0pt,minimum size=3pt,fill=red},
}

\begin{document}
\begin{tikzpicture}
\coordinate [label=below right:$A$] (A) at (0,0);
\coordinate [label=above right:$B$] (B) at (0,-5);
\coordinate [label=above left:$C$] (C) at (2.5,-3);
\coordinate [label=above left:$D$] (D) at (5,-2);
\coordinate [label=below left:$E$] (E) at (4.5,0);

\draw (A) -- (B);
\draw ([out angle=0]B) to [curve through ={(C)}] ([in angle=-135]D);
\draw (D) -- (E);
\draw (E) -- (A);

\draw[blue,dashed] (A) -- ([xshift=-.3cm]A);
\draw[blue,dashed] ([xshift=-.3cm]A) -- ([xshift=-.3cm]B);
\draw[blue,dashed] ([xshift=-.3cm]B) to[out=-90,in=180] ([yshift=-.3cm]B);
\draw[blue,dashed] ([out angle=0,yshift=-.3cm]B) to [curve through ={([yshift=-.3cm]C)}] ([in angle=-135,yshift=-.3cm]D);

\node[point] at (A) {};
\node[point] at (B) {};
\node[point] at (C) {};
\node[point] at (D) {};
\node[point] at (E) {};
\end{tikzpicture}
\end{document}

Which produces the following image:

As you can see, the blue-dashed line from A to B is 0.3cm apart from the original line. However, the blue-dashed curve does not maintain a constant distance from the original curve.

I tried the solution from JLDiaz in this post and that works perfectly, as long as you don't use hobby curves, because they don't seem to support the coordinate[pos=xxx] option that is used in the solution. Using regular curves instead of hobby curves is not an option for me, since the hobby curves produce a much nicer "flow".

I figure that shifting something around in parallel is such a common task, that there must be an easier way of doing it, right? Can anyone point me into the right direction?

Answer 1

Without a fairly sizeable amount of work, the best TikZ can do by default is something like this. It's almost there, but frustratingly tricky to get the kinks ironed out.

\documentclass[tikz,border=5]{standalone}
\usetikzlibrary{calc,hobby,decorations}
\tikzset{
   point/.style={circle,inner sep=0pt,minimum size=3pt,fill=red},
}
\begin{document}
\begin{tikzpicture}
\coordinate [label=below right:$A$] (A) at (0,0);
\coordinate [label=above right:$B$] (B) at (0,-5);
\coordinate [label=above left:$C$] (C) at (2.5,-3);
\coordinate [label=above left:$D$] (D) at (5,-2);
\coordinate [label=below left:$E$] (E) at (4.5,0);

\draw
  (A) -- (B) to [out angle=0, in angle=-135, curve through=(C)]
  (D) -- (E) -- cycle;

\draw [decoration={curveto, raise=-0.3cm}, decorate, blue, dashed]
  (A) -- (B) to [out angle=0, in angle=-135, curve through=(C)]
  (D) -- (E) -- cycle;

\foreach \n in {A,...,E}
  \node[point] at (\n) {};
\end{tikzpicture}
\end{document}

However, bearing in mind the caveats from this answer, here is a decoration for out-setting which not really very efficient and will probably fail for paths where the exterior angle between consecutive paths is acute or the interior angle is very small. But it is at least passable for the current example. The basic idea is to extend each path at the corners with a straight line similar to the approach taken in mitering. It might give the metapost-guys some ideas ;)

Answer 2

This is an oddly fiddly task. I've tried to produce a robust solution to this problem a couple of times without ever getting a very satisfactory or general result. But here is my latest effort. This is in Metapost but it might give the tikz-guys some ideas.

Compile with mpost...

prologues := 3;
outputtemplate := "%j%c.eps";

beginfig(1);

numeric u;  u = 1cm;
path shape;  
shape = ( origin -- (0,-5) { right } .. (5/2,-3) .. { dir 45 } (5,-2) -- (9/2,0) -- cycle ) scaled u;

draw shape; 
for t=0 upto 4: 
  fill fullcircle scaled 3 shifted point t of shape withcolor red;
  label(char (65+t), 6 up rotated angle direction t+1/16 of shape shifted point t+1/16 of shape);
endfor

path C; C = fullcircle scaled 20; 
numeric a, b, s; s=1/4; 
draw for t=s step s until length shape:
    hide(
        a := directiontime (point t of shape - precontrol t of shape) of C;
        b := directiontime (postcontrol t of shape - point t of shape) of C;
        if b<a: a:= a-8; fi
        % uncomment this draw command to see the contruction circles...
        % draw C shifted point t of shape withcolor .8[red,white];
    )
    subpath (a,b) of C shifted point t of shape {direction t+eps of shape} ..
endfor cycle dashed evenly;

endfig;
end.

The idea here is to use the pre-control and post-control points of each step along the path to define an appropriate arc of a circle and then connect up all of these arcs. Where the three control points are in a line this arc will be a single point (which is what we want). Reducing the step size improves the shape of the surrounding curve, but if you go too small you get some extra loop artefacts. This is what I mean about it not being a very robust approach. But I hope it provides at least some inspiration.

And while I'm at it, there is another technique you could try: drawing with a fat pen, and then erasing the interior.

beginfig(2);
  draw shape withpen pencircle scaled 12 withcolor blue;
  undraw shape withpen pencircle scaled 11;
  unfill shape;
  draw shape;
endfig;

with shape defined as in the first figure this produces:

but don't try adding a dash pattern with such a fat pen, it will look very strange. The unfill shape gets rid of the line inside the shape.