Making hyphens in text slightly longer and thinner

Question

Even though I would think that this is a duplicate, I was not able to find any answer to a rather simple question. I am concerned about the appearance of various hyphens or dashes. In this discussion, it is suggested that one has to type, say, Robertson-Walker metric as Robertson--Walker, i.e. with an en-dash, whereas semi-simple group would come with a hyphen, i.e. just semi-simple. But what worries me is that, apart from the fact that the hyphen is very short, it's also thicker than the en-dash.

I'd like to have the hyphen being typeset like an en-dash, just a bit shorter, i.e. as in the middle but not bottom example of this list (the top entry is just the en-dash):

Here is the MWE to play with.

\documentclass{article}
\pagestyle{empty}
\begin{document}
Robertson--Walker metric\dots looks fine but IMHO semi-simple Lie group doen't look
too good. Is there any way to make \verb|semi-simple| being typeset as 
semi\raisebox{2pt}{\rule{4pt}{0.2pt}}simple?

Comparison: \begin{tabular}{l}
A--B \\ A\raisebox{2pt}{\rule{4pt}{0.2pt}}B \\ A-B
\end{tabular}
\end{document}

QUESTION: Is there a (simple or semi-simple) way of achieving this?

NOTE: I am not interested in proposals that would mess up the minus signs in equations or do other harm, and I'd also like to avoid having to type some LaTeX commands instead of the hyphens. In the ideal case, only the appearance of - would change.

Answer 1

You can use the pre_linebreak_filter to walk the hlist, search for disc nodes and replace their pre, post, and replace fields with rules if they contain the hyphen. This leaves the hyphen invariant in math mode and in non-discretionary contexts.

It is true that as of today arXiv does not accept LuaTeX but there are ways around that: https://www.monperrus.net/martin/how-to-use-lualatex-arxiv

\documentclass{article}
\usepackage{luacode}
\begin{luacode*}

local pt = 2^16

local r = node.new("rule")
r.width  = 4*pt
r.height = 2.2*pt
r.depth  = -2*pt

local function is_hyphen(head)
    return head and head.id == node.id("glyph") and head.char == 45
end

local function longhyphen(head)
    for n in node.traverse(head) do
        if n.id == node.id("disc") then
            for _,v in pairs({"pre","post","replace"}) do
                if is_hyphen(n[v]) then
                    n[v] = node.copy(r)
                end
            end
        end
    end
    return head
end

luatexbase.add_to_callback("pre_linebreak_filter",
                           longhyphen,
                           "longhyphen")

\end{luacode*}

\begin{document}

Robertson--Walker

semi-simple

$a - b$

no disc-

-disc

\end{document}

Answer 2

Another option, apart from the robust and preferable LuaTeX-solution, is to mess with the catcodes.

The hyphen is now an active character. In math mode it simply expands to a minus, but in text mode it swallows the next two tokens, examines them whether they are hyphens and either inserts the rule to replace a single hyphen or it expands to the appropriate number of original hyphens to form the dash ligatures.

This fails horribly in case of e.g.

-\begin{environement}

One could maybe make it brace-group-aware, but then, why would I do that?

No refunds!

\documentclass{article}

\chardef\textminus=`\-
\mathchardef\mathminus=\mathcode`\-

\begingroup
\makeatletter
\catcode`-=13
\protected\long\gdef\longhyphen#1#2{%
  \ifx#1-\relax
    \textminus\textminus
    \ifx#2-\relax
      \textminus
    \else
      #2%
    \fi
  \else
    \raisebox{2pt}{\rule{4pt}{0.2pt}}#1#2%
  \fi
}
\protected\gdef-{%
  \ifmmode
  \expandafter\mathminus
  \else
  \expandafter\longhyphen
  \fi
}
\endgroup

\begin{document}

Robertson---Walker

Robertson--Walker

semi-simple

$a - b$

\catcode`-=13

Robertson---Walker

Robertson--Walker

semi-simple

$a - b$

\end{document}

Answer 3

Another possibility, that has not been explored yet it using a virtual font. To this end we follow the How to create a virtual font? tutorial to create our instance.

First copy the base font (I use cmr10) to the current directory and convert it to a property list.

cp `kpsewhich cmr10.tfm` .
tftopl cmr10.tfm > cmlh10.vpl

I chose the filename cmlh10 where lh stands for “long hyphen”. Open the file cmlh10.vpl and add the following line at the very top.

(MAPFONT D 0 (FONTNAME cmr10))

Then find the line starting with (CHARACTER O 55 and replace the character entry with the following

(CHARACTER O 55
   (CHARWD R 0.500002)
   (CHARHT R 0.430555)
   (CHARIC R 0.027779)
   (MAP
      (SELECTFONT D 0)
      (SETCHAR O 173)
      )
   )

The slots 55 and 173 are both in octal. The slot 55 holds the hyphen, slot 173 the endash, i.e. we are mapping the hyphen to endash. Another alternative would be to draw a custom rule instead of just mapping endash to hyphen.

(CHARACTER O 55
   (CHARWD R 0.4)
   (CHARHT R 0.28)
   (MAP
      (MOVEUP R 0.255)
      (SETRULE R 0.025 R 0.4)
      )
   )

Assemble the virtual font using

vptovf cmlh10.vpl

In you TeX document replace cmr10 with cmlh10 in the NFSS font loading.

\DeclareFontShape{OT1}{cmr}{m}{n}{
        <-6>    cmr5
        <6-7>   cmr6
        <7-8>   cmr7
        <8-9>   cmr8
        <9-10>  cmr9
        <10-12> cmlh10
        <12-17> cmr12
        <17->   cmr17
      }{}
\documentclass{article}
\pagestyle{empty}
\begin{document}

Robertson--Walker

semi-simple

$a - b$

no disc-

-disc

\end{document}

In the screenshot I mapped the endash to hyphen.