name prefix and name suffix vs. edge: starting point of edge paths are incorrect

[ghost]

It seems that there is some problem using nodes with name prefix (or name suffix for that matter) and the edge path construction. When one switches on either of them, the edge will connect to the node center of the starting node.

\documentclass[tikz,border=3mm]{standalone}
\begin{document}
\begin{tikzpicture}[nodes={circle,draw},>=stealth]
 \begin{scope}[name prefix=L-]
  \path node (A) {A} (2,0) node (B){B};
  \path (A) edge (B);
 \end{scope}
 \begin{scope}[xshift=4cm]
  \path node (A) {A} (2,0) node(B){B};
  \path (A) edge (B);
 \end{scope}
\end{tikzpicture}
\end{document}

Left is a scope with a name prefix, and the edge extends to the center of the A node, where as in the right scope, where there is no node prefix, the result is as expected. If we use a name suffix, the same problem arises.

[QJLc]

I may get the point of this issue.

See tikz.code.tex.

After line 5427, I add lines:

    \def\tikz@parse@node@temp{#2}%
    \ifx\tikz@parse@node@temp\tikztostart%
      \tikz@shapebordertrue%
    \fi%

this will set the (\tikztostart) on shape border. But I do not know whether this will cause errors.

I guess, the key point about this issue is , command \tikz@scan@one@point sets \tikz@shapeborderfalse, the full name L-A is saved in \tikztostart, so, \tikz@shapeborderfalse is always valid, and then, command \tikz@@movetowill set

\let\tikz@moveto@waiting=\relax

[hmenke]

Hm, I'm not sure whether this is the right way to approach this. It seems like a very symptomatic treatment. This patch also solves the problem and is maybe a bit more idiomatic. I'm not sure which way to go.

--- a/tex/generic/pgf/frontendlayer/tikz/tikz.code.tex
+++ b/tex/generic/pgf/frontendlayer/tikz/tikz.code.tex
@@ -5431,7 +5431,13 @@
         \tikz@shapebordertrue%
         \def\tikz@shapeborder@name{\tikz@pp@name{#2}}%
       \fi%
-    \fi
+    \else\ifcsname pgf@sh@ns@#2\endcsname
+      \expandafter\ifx\csname pgf@sh@ns@#2\endcsname\tikz@coordinate@text%
+      \else
+        \tikz@shapebordertrue%
+        \def\tikz@shapeborder@name{#2}%
+      \fi%
+    \fi\fi
   \fi%
   \edef\tikz@marshal{\noexpand#1{\noexpand\pgfqpoint{\the\pgf@x}{\the\pgf@y}}}%
   \tikz@marshal%

[QJLc]

Yes, I have learned it, your patch is very reasonable.
The command
\tikz@scan@one@point{\tikz@@moveto}(A)
will define
\def\tikz@shapeborder@name{L-A}
so, the added line
+ \def\tikz@shapeborder@name{#2}%
could repeat the defination, is this repetition needed?