Wrapping long lines with no spaces while respecting kerning

Question

There were many posts asking about breaking long string of text/digits (with no spaces) into justified lines. This post is one example. The mostly-agreed-upon solution seems to be adding a zero-width glue with a little stretch between every pair of characters within that string; maybe throw in an extra discretionary to insert/customize hyphens at line ends.

But none of the solutions I saw addressed kerning. Here’re two scenarios where I’d like to preserve kerning:

1. Long string of digits (say, 100 digits of pi), set in proportional old-style figures. You see, font designers often kern between 7 (seven.osf) and 4 (four.osf), among many other pairs. Example font: Source Serif 4.
2. Long string of kanas, set in a Japanese font with palt and kern OpenType features turned on. You see, font designers often specify proportional alternate widths for フ and ォ, and also kern between them, among many other kanas and kana pairs. Example font: Source Han Serif. (Yes, I’m aware that LuaTeX-ja already implements this.)

Can kerning be preserved somehow? eTeX/XeTeX solutions preferred, but LuaTeX solutions also welcome.

% run with XeLaTeX or LuaLaTeX
\documentclass{article}
\usepackage{fontspec}

\setmainfont{Source Serif 4}[Numbers={Proportional,OldStyle}]

\newcommand*\+{\hskip 0pt plus 1pt \relax}

\begin{document}

3.14159265358979323846264338327950288419716, figures kerned, cannot be wrapped.

3\+.\+1\+4\+1\+5\+9\+2\+6\+5\+3\+5\+8\+9\+7\+9\+3\+2\+3\+8\+4\+6\+2\+6\+4\+3\+3\+8\+3\+2\+7\+9\+5\+0\+2\+8\+8\+4\+1\+9\+7\+1\+6, can be wrapped, figures no longer kerned.

\end{document}

Answer 1

You can save the kerns after you have split your data in a sequence.

\documentclass{article}
\usepackage{fontspec}

\setmainfont{Source Serif Pro}[Numbers={Proportional,OldStyle}]

\ExplSyntaxOn

\seq_new:N \l__ruixi_split_items_seq
\seq_new:N \l__ruixi_split_kerns_seq

\NewDocumentCommand{\Split}{m}
 {
  \seq_set_split:Nnn \l__ruixi_split_items_seq {} { #1 }
  \seq_clear:N \l__ruixi_split_kerns_seq
  \int_step_inline:nnn { 2 } { \seq_count:N \l__ruixi_split_items_seq }
   {
    \hbox_set:Nn \l_tmpa_box
     {
      \seq_item:Nn \l__ruixi_split_items_seq { ##1-1 }
      \seq_item:Nn \l__ruixi_split_items_seq { ##1 }
     }% with kern
    \hbox_set:Nn \l_tmpb_box
     {
      \seq_item:Nn \l__ruixi_split_items_seq { ##1-1 }
      \kern0pt
      \seq_item:Nn \l__ruixi_split_items_seq { ##1 }
     }% without kern
    \seq_put_right:Ne \l__ruixi_split_kerns_seq
     {
      \dim_eval:n { \box_wd:N \l_tmpa_box - \box_wd:N \l_tmpb_box }
     }
   }
  \seq_put_right:Nn \l__ruixi_split_kerns_seq { 0pt }
  \seq_map_indexed_inline:Nn \l__ruixi_split_items_seq
   {
    ##2 \hspace{\seq_item:Nn \l__ruixi_split_kerns_seq { ##1 } plus 0.3pt}
   }
  \unskip
 }
\ExplSyntaxOff

\begin{document}

3.14159265358979323846264338327950288419716

\Split{3.14159265358979323846264338327950288419716}

\parbox[t]{3cm}{\Split{3.14159265358979323846264338327950288419716}}

\end{document}

Of course, if you use this inside a paragraph, the glue can participate to the stretching of the line.

If I set a box containing \Split{3.14159}} I get

\hbox(5.39+1.4)x31.42, direction TLT
.\TU/SourceSerifPro(0)/m/n/10 󰀞
.\glue -0.19 plus 0.3
.\TU/SourceSerifPro(0)/m/n/10 .
.\glue -0.38 plus 0.3
.\TU/SourceSerifPro(0)/m/n/10 󰀜
.\glue 0.0 plus 0.3
.\TU/SourceSerifPro(0)/m/n/10 󰀟
.\glue 0.0 plus 0.3
.\TU/SourceSerifPro(0)/m/n/10 󰀜
.\glue 0.0 plus 0.3
.\TU/SourceSerifPro(0)/m/n/10 󰀠
.\glue 0.0 plus 0.3
.\TU/SourceSerifPro(0)/m/n/10 󰀤

The same without \Split

\hbox(5.39+1.4)x31.42, direction TLT
.\TU/SourceSerifPro(0)/m/n/10 󰀞
.\kern-0.19 (font)
.\TU/SourceSerifPro(0)/m/n/10 .
.\kern-0.38 (font)
.\TU/SourceSerifPro(0)/m/n/10 󰀜
.\TU/SourceSerifPro(0)/m/n/10 󰀟
.\TU/SourceSerifPro(0)/m/n/10 󰀜
.\TU/SourceSerifPro(0)/m/n/10 󰀠
.\TU/SourceSerifPro(0)/m/n/10 󰀤

Answer 2

This is a solution for digits, but there is no reason it shouldn't work with characters. The secret sauce is that I used mod so the space is only inserted after groups of digits, so pairs are preserved withing the group and hence the font kerning. With cjk and other complex utf cases a Lua solution would be better to do more checks. Presentation, I used common practice in number theory books to move the 3.4 a bit out to the left.

% run with XeLaTeX or LuaLaTeX
\documentclass{article}
\usepackage{fontspec}

\setmainfont{Source Serif 4}[{Numbers=Proportional,OldStyle}]

\newcommand*\+{\hskip 0pt plus 1pt \relax}

\begin{document}

3.14159265358979323846264338327950288419716, figures kerned, cannot be wrapped.

3\+.\+1\+4\+1\+5\+9\+2\+6\+5\+3\+5\+8\+9\+7\+9\+3\+2\+3\+8\+4\+6\+2\+6\+4\+3\+3\+8\+3\+2\+7\+9\+5\+0\+2\+8\+8\+4\+1\+9\+7\+1\+6, can be wrapped, figures no longer kerned.

\ExplSyntaxOn
\cs_set:Npn\spaceone{\hspace{5pt}}
%\cs_set:Npn\spacetwo #1{\fbox{#1}\hskip1sp}
\tl_put_left:Nn\l_tmpa_tl{14159265358979 323846264338327950288419716}
\fboxsep=0pt\fboxrule=0sp
\int_set:Nn\l_tmpa_int{0}

\def\insertskip #1 
{#1
 \int_incr:N\l_tmpa_int
 \int_set:Nn\l_tmpb_int{ \int_mod:nn{\int_use:N\l_tmpa_int}{4} } 
 \int_compare:nNnTF{\int_use:N\l_tmpb_int}=0{\int_gset:Nn\l_tmpa_int{0}\spaceone}{}
}

Output:\par
\DeclareDocumentCommand\Out{}{
\parindent0pt
\begin{minipage}[t]{1em}
3.
\end{minipage}
\begin{minipage}[t]{4cm}\large
    \tl_map_function:NN\l_tmpa_tl\insertskip
\end{minipage}  
}
\ExplSyntaxOff
\Out
\end{document}

Top figure without Proportional, bottom with Proportional.