🚀 go-pugleaf

Tbl totally botches the line-length setting for text blocks which are
to span more than one column, if the starting column has an explicit
width specification.  It takes the width to be that of the starting
column.  Taking various other parameters from the starting column is
not unreasonable, but picking up the width from there defeats the
whole notion of spanning.  I have a tentative fix;  quite possibly
it misses some fine points, but it seems a good first approximation
to what's wanted, and it runs my test cases well.  The policy is that
the formula given in the tbl manual is used unchanged if no explicit
widths are present (it gives reasonable results), and otherwise the
width is the sum of
	- the widths of columns with explicitly-specified widths
	- (linelength/ncolumns) for each defaulted column
	- all intervening column-separation spaces

Here's a diff on src/cmd/tbl/tg.c;  beware that line numbers may be off,
so don't apply it blindly.

9a13
> int colno, plus, ndefault, septotal;
33,36c37,62
< if (cll[icol][0])
< 	fprintf(tabout, ".ll %sn\n", cll[icol]);
< else
< 	fprintf(tabout, ".ll \\n(%du*%du/%du\n",SL,ctspan(ilin,icol),ncol+1);
---
> plus = ' ';
> septotal = 0;
> ndefault = 0;
> fprintf(tabout, ".ll");
> for (colno = icol; colno < ncol && (colno == icol || ctype(ilin, colno) == 's'); colno++)
> 	{
> 	if (cll[colno][0])
> 		{
> 		fprintf(tabout, "%c%sn", plus, cll[colno]);
> 		plus = '+';
> 		}
> 	else
> 		ndefault++;
> 	if (colno != icol)
> 		septotal += sep[colno-1];
> 	}
> if (plus == '+')
> 	{
> 	if (septotal != 0)
> 		fprintf(tabout, "+%dn", septotal);
> 	if (ndefault != 0)
> 		fprintf(tabout, "+\\n(%du*%du/%du", SL, ndefault, ncol);
> 	fprintf(tabout, "\n");
> 	}
> else
> 	fprintf(tabout, " \\n(%du*%du/%du\n",SL,ctspan(ilin,icol),ncol+1);