bug at paths with a node as starting point

[sistlind]

There is a bug, if a node is used as a starting point of a path without defining an explicit anchor:

\begin{circuitikz}
	\draw 
	
	%npns rechte Diffstufe
	(3,0) node[npn, anchor=C] (npn5) {}
	(npn5.B)++(1,0) node[] {$T_5$}% Beschriftung
	(4,0) node[npn, xscale=-1, anchor=C] (npn6) {}
	(npn6.B)++(-1,0) node[] {$T_6$}% Beschriftung
	(npn5.E)++(0.5,-0.5) node[npn,xscale=-1,anchor=C](npn4) {} 
	(npn4.B)++(-1,0) node[] {$T_4$}% Beschriftung
	
	%npns linke Diffstufe 
	(0,0) node[npn, anchor=C] (npn1) {}
	(npn1.B)++(1,0) node[] {$T_1$}% Beschriftung
	(1,0) node[npn, xscale=-1, anchor=C] (npn2) {}
	(npn2.B)++(-1,0) node[] {$T_2$}% Beschriftung
	(npn1.E)++(0.5,-0.5) node[npn, anchor=C](npn3) {}
	(npn3.B)++(1,0) node[] {$T_3$} % Beschriftung
	
	(npn1.C)++(0,1) node[](P3){}
	(npn6.C)++(0,1) node[](P4){}
	
	(npn1.C) to[short] (P3)
	(npn6.C) to[short] (P4)
	% weiter linke Diffstufe
	(npn1.E) -- (npn2.E) 
	(P3) to[R,l^=$R_1$,-] ++(0,2) node[vcc](){}
	(npn1.B) to[short,l_=LO-,-o] ++(-0.5,0) 
	(npn6.B) to[short,l=LO-,-o] ++(0.5,0) 
	(P4) to[R,l_=$R_2$,-] ++(0,2) node[vcc]{}
	(npn1.E)++(0.5,0) to[short,*-] (npn3.C)
	(npn3.B) to[short,l_=RF+,-o] ++(-0.5,0)
	(npn4.B) to[short,l=RF-,-o] ++(0.5,0)
	(npn5.B) -- (npn2.B)

	% weiter rechte Diffstufe
	(npn5.E) -- (npn6.E) 
	(npn5.E)++(0.5,0) to[short,*-] (npn4.C)
	
	%Verbindung zw. rechter und linker Diffstufe
	(npn3.E) -- (npn4.E)
	
	
	(npn5.C) to[short,-*] (P3)
	(npn2.C) to[short,-*] (P4)
	
	(npn2.B)++(0.2,0)to[short,-o]++(0,-0.5) node[](P5){}
	(P5)++(0,-0.3)  node[](P6){LO+}
	
	(npn3.E) ++(1.5,0) to[dcisource=$I1$,*-] ++(0,-2) node[ground](){}
	%------------------------------------------
	;
\end{circuitikz}

There appear anwanted lines(between emitter of T2 and R1 as well as from LO- to R2):

[Rmano]

The new path fixes this. Notice that now the original example compiles like this:

Which is correct --- P3 and P4 are nodes and so they have a non-zero sizes. You can see what's happening if you add [draw=red] to them:

The behavior of forcing the start point at the center of the node is not standard and creates a lot of problems.

[sistlind]

Hi @Rmano ,
if your last screenshot is the result of the new paths, in my opinion this is a huge drawback of this code, as it will break nearly all the drawings I have done for my lectures. I think it is not convenient to always specify the center anchor. Furthermore, nodes do not need to have a border. See difference between "coordinate" and "node".

EDIT: Did a closer look to this example. In this case it is ok, because P3 should really be declared as coordinate, not as path. Nevertheless, I am not sure if it does not break some other stuff.

Here I posted some other example: https://tex.stackexchange.com/questions/360264/correct-linejoins-at-tikz-to-path

[Rmano]

Hi, I understand what you say, but the correct behavior in tikz is to connect the border of the nodes (even if invisible):

\documentclass[border=10pt]{standalone}
\usepackage{tikz}
\usetikzlibrary{arrows.meta,positioning,calc}
\begin{document}
\begin{tikzpicture}
    \node [](A) at (0,.2) {};
    \node [](B) at (2,.2) {};
    \node [](C) at (0,0) {};
    \node [](D) at (2,0) {};
    \draw (A) -- (B);
    \draw (C.center) -- (D.center);
\end{tikzpicture}
\end{document}

For this coordinates exists --- there are tens of post in TeX.SE about this, for example this,

Forcing the behavior of substitute the start and end node with the center to the center of nodes has as the effect of

1. that you can't use to[R] (for example) between the border of two nodes, and that you have to manually select the appropriate anchor border;
2. that you can't start the path replacement code with -- (\tikztostart), which will force the ugly hack for the line joins (that sometimes fails as in the case at #417)

So I think it's a good thing to do (I created an embedded rollback point at 1.1.2 for backward compatibility on purpose); the original code is now correctly rendered by TikZ standard (and it's easily fixable by changing node to coordinate or adding inner sep=0pt to the two nodes).

To explain point 1 here: this code

\begin{tikzpicture}[every node/.style={inner sep=1cm}]
    \node [draw](A) at (0,0) {A};
    \node [draw](B) at (4,2) {B};
    \draw (A) to[R] (B);
\end{tikzpicture}

renders as

in 1.1.2 and previous
and as

now --- you can easily have the old behavior (if you really wanted it) with using A.center and B.center explicitly, but it's really complex to obtain the new one --- which in the rest of tikz is considered the correct one.

[Rmano]

Here I posted some other example: https://tex.stackexchange.com/questions/360264/correct-linejoins-at-tikz-to-path

Thanks, I will study this.

[Rmano]

Notice that in the example posted above now it works as should (in my opinion):

\begin{tikzpicture}[very thick]
    \draw (2,2) node[draw=red](A){};
    \draw (2,2) coordinate(B);
    \draw (0,0) to[esource](0,2) to[esource](A) to[esource] ++(2,-2);
    \begin{scope}[xshift=4.5cm]
        \draw (2,2) coordinate(B);
        \draw (0,0) to[esource](0,2) to[esource] (B) to[esource] ++(2,-2);
    \end{scope}
\end{tikzpicture}

The line joins are correct and the landing/starting positions from the node are the expected ones.