\documentclass[margin=1cm]{standalone}

\usepackage{tikz}
\pgfdeclareshape{what}{
     \anchor{center}{
        \pgfpointorigin
    }
    \foregroundpath{
        \pgfscope
            \pgfpathmoveto{\pgfpoint{-4pt}{0pt}}
            \pgfpathlineto{\pgfpoint{4pt}{0pt}}
            \pgfpathlineto{\pgfpoint{0pt}{4pt}}
            \pgfpathlineto{\pgfpoint{-4pt}{0pt}}
            % \pgfpathclose
            \pgfusepath{draw,fill}
        \endpgfscope
    }
}

\begin{document}
\begin{tikzpicture}
    \draw[] (0,0) -- node[what] {} ++(0,-2);
    \draw[red, shorten <=.5cm] (1,0) -- node[what] {} ++(0,-2);
    \draw[blue,shorten <=.5cm] (2,0) -- node[draw] {} ++(0,-2);
\end{tikzpicture}

\end{document}

In reply to @r_mano:matrix.org

\documentclass[margin=1cm]{standalone}

\usepackage{tikz}
\pgfdeclareshape{what}{
     \anchor{center}{
        \pgfpointorigin
    }
    \foregroundpath{
        \pgfscope
            \pgfpathmoveto{\pgfpoint{-4pt}{0pt}}
            \pgfpathlineto{\pgfpoint{4pt}{0pt}}
            \pgfpathlineto{\pgfpoint{0pt}{4pt}}
            \pgfpathlineto{\pgfpoint{-4pt}{0pt}}
            % \pgfpathclose
            \pgfusepath{draw,fill}
        \endpgfscope
    }
}

\begin{document}
\begin{tikzpicture}
    \draw[] (0,0) -- node[what] {} ++(0,-2);
    \draw[red, shorten <=.5cm] (1,0) -- node[what] {} ++(0,-2);
    \draw[blue,shorten <=.5cm] (2,0) -- node[draw] {} ++(0,-2);
\end{tikzpicture}

\end{document}

shorten <=.5cm is applied when the node shape is drawn. Yes this option (and shorten >) takes effect even when the path has no arrows. Without \pgfpathclose, you can also use \pgfsetshortenstart{0pt} \pgfsetshortenend{0pt} to do the resetting in pgf level.

PS: Commenting out the second \pgfpathlineto{\pgfpoint{-4pt}{0pt}} and using node[what, draw=black] help me figure out what's really wrong.

In reply to @r_mano:matrix.org

\documentclass[margin=1cm]{standalone}

\usepackage{tikz}
\pgfdeclareshape{what}{
     \anchor{center}{
        \pgfpointorigin
    }
    \foregroundpath{
        \pgfscope
            \pgfpathmoveto{\pgfpoint{-4pt}{0pt}}
            \pgfpathlineto{\pgfpoint{4pt}{0pt}}
            \pgfpathlineto{\pgfpoint{0pt}{4pt}}
            \pgfpathlineto{\pgfpoint{-4pt}{0pt}}
            % \pgfpathclose
            \pgfusepath{draw,fill}
        \endpgfscope
    }
}

\begin{document}
\begin{tikzpicture}
    \draw[] (0,0) -- node[what] {} ++(0,-2);
    \draw[red, shorten <=.5cm] (1,0) -- node[what] {} ++(0,-2);
    \draw[blue,shorten <=.5cm] (2,0) -- node[draw] {} ++(0,-2);
\end{tikzpicture}

\end{document}

shorten <=.5cm is applied when the node shape is drawn. Yes this option (and shorten >) takes effect even when the path has no arrows. Without \pgfpathclose, you can also use \pgfsetshortenstart{+0pt} \pgfsetshortenend{+0pt} to do the resetting in pgf level. The plus sign is a pgfmath trick so it knows the math-expr is only a simple value and the parsing will be quicker.

PS: Commenting out the second \pgfpathlineto{\pgfpoint{-4pt}{0pt}} and using node[what, draw=black] help me figure out what's really wrong.

If the outer path has non-empty arrows, \pgfsetarrows{-} is also required.

It's a pity that there's no lists of options which are inherited, used, and reset inside a node/pic/matrix/graph.

If the outer path has non-empty arrows, \pgfsetarrows{-} is also required.

It's a pity that there's no documented lists of options which are inherited, used, and reset inside a node/pic/matrix/graph.

If the outer path has non-empty arrows, \pgfsetarrows{-} is also required.

It's a pity that there's no documented lists of options which are inherited, used, and reset inside a node/pic/matrix/graph.

muzimuzhi thanks! So I should \pgfscope \pgfsetshortenstart{+0pt} \pgfsetshortenend{+0pt} \pgfsetarrows{-} ... \endpgfscope at the start of every shape backgroundpath which is supposed to be drawn independently from the outer path? Is the scoping sufficient to guarantee that the outer path is not disturbed?

In reply to @muzzi:matrix.org
hmenke: Since Conventional Commits has been applied for a while with a tacit agreement, maybe it's time to mention and require it in PR template. This could save us from reminding new contributors.

In reply to @hmenke:matrix.org
samcarter: That's a good idea. When and where will the next DANTE meeting take place? The website doesn't list anything yet.

In reply to @r_mano:matrix.org

If the outer path has non-empty arrows, \pgfsetarrows{-} is also required.

It's a pity that there's no documented lists of options which are inherited, used, and reset inside a node/pic/matrix/graph.

muzimuzhi thanks! So I should \pgfscope \pgfsetshortenstart{+0pt} \pgfsetshortenend{+0pt} \pgfsetarrows{-} ... \endpgfscope at the start of every shape backgroundpath which is supposed to be drawn independently from the outer path? Is the scoping sufficient to guarantee that the outer path is not disturbed?

There're more to be reset. What I can immediately think of is using \pgfsetcornersarced{\pgfpointorigin} to reset outer non-zero rounded corners.

Is the scoping sufficient to guarantee that the outer path is not disturbed?

Yes I think so.

In reply to @r_mano:matrix.org

If the outer path has non-empty arrows, \pgfsetarrows{-} is also required.

It's a pity that there's no documented lists of options which are inherited, used, and reset inside a node/pic/matrix/graph.

muzimuzhi thanks! So I should \pgfscope \pgfsetshortenstart{+0pt} \pgfsetshortenend{+0pt} \pgfsetarrows{-} ... \endpgfscope at the start of every shape backgroundpath which is supposed to be drawn independently from the outer path? Is the scoping sufficient to guarantee that the outer path is not disturbed?

Depending on the design goal, there may be more to be reset. What I can immediately think of is using \pgfsetcornersarced{\pgfpointorigin} to reset outer non-zero rounded corners.

Is the scoping sufficient to guarantee that the outer path is not disturbed?

Yes I think so.

In reply to @muzzi:matrix.org

If the outer path has non-empty arrows, \pgfsetarrows{-} is also required.

It's a pity that there's no documented lists of options which are inherited, used, and reset inside a node/pic/matrix/graph.

In reply to @hmenke:matrix.org
Well, generally everything should be assumed to be inherited, because even if we can reset certain things in the graphics state, these changes will not propagate back into pgfkeys. To be sure that everything is in sync, all the (un)desired options should be reset manually.

\documentclass{article}
\usepackage{tikz}

\makeatletter
\pgfutil@g@addto@macro\tikz@node@reset@hook
  {\tikzset{shorten <=+0pt, shorten >=+0pt, arrows={-}, rounded corners=+0pt}}
\makeatother

\pgfdeclareshape{what}{
  \anchor{center}{
    \pgfpointorigin
  }
  \foregroundpath{
    \pgfpathmoveto{\pgfpoint{-4pt}{0pt}}
    \pgfpathlineto{\pgfpoint{4pt}{0pt}}
    \pgfpathlineto{\pgfpoint{0pt}{4pt}}
    \pgfpathlineto{\pgfpoint{-4pt}{0pt}}
    \pgfusepath{draw,fill}
  }
}

\begin{document}
\begin{tikzpicture}[back]
    \draw[] (0,0) -- node[what] {} ++(0,-2);
    
    \draw[fill=red, shorten <=.5cm] (1,0) -- node[what] {} ++(0,-2);
    \draw[fill=red, shorten <=.5cm] (2,0) -- node[what, shorten <=5pt] {} ++(0,-2);
    
    \draw[fill=blue,rounded corners=10pt] (3,0) -- node[what] {} ++(0,-2);
    \draw[fill=blue,rounded corners=10pt] (4,0) -- node[what, rounded corners=2pt] {} ++(0,-2);
    
    \draw[fill=green, <->] (5,0) -- node[what] {} ++(0,-2);
    \draw[fill=green, <->] (6,0) -- node[what, shorten <=3pt, <-] {} ++(0,-2);
\end{tikzpicture}
\end{document}

In reply to @r_mano:matrix.org
muzimuzhi: I see it is difficult... sometimes you want them to propagate, and sometimes no. So I think I will go for the minimal fix; really the user that noticed the problem was trying to do a very strange thing with the shape. (See https://tex.stackexchange.com/questions/632051/tikz-shorten-with-circuitikz)

sometimes you want them to propagate, and sometimes no

true. Having every relation between any of tikz options and node/pic/matrix/graph documented is such a hard task, and it's even impossible to resort the inconsistencies when tikz's main features has been populated for such a long time.

Oh I've opened that tex-sx question, but didn't read through.