> Back in the day of standalone phototypesetters, eg. the mighty > Linotronic or CompuGraphic's 8400, everything was line-at-a-time yet > the results were almost always better than what can be achieved with > groff. The reason was that both word- and letter-spacing (track > kerning) were used to auto-justify lines. For each, min-opt-max > values could be given, which would then be used to determine optimal > breakpoints with the best overall grey to the line.
Letter-spacing is bad in general. A better solution is font expansion, as implemented e.g. in pdflatex. Unfortunately, MM fonts have been abandoned, since a width axis to expand or compress glyphs horizontally without modifying the counter width would be exactly the right thing for making the HZ algorithm work perfectly. Here is a short overview: http://www.tug.org/TUGboat/Articles/tb22-3/tb72thanh.pdf Note that implementing the HZ algorithm in pdftex was the subject of Hàn Thế Thành's thesis; you can read it here: http://www.pragma-ade.com/pdftex/thesis.pdf > I'm wondering if that older approach wouldn't be the better way to > go if, one day, groff gets an overhaul in this area. I don't know > the internals, but I'm thinking it might be easier to implement, > since it's still line-at-a-time. Practical experiments done by Thế Thanh have shown that font expansion or compression must not exceed approx. 2% of the natural width to be optically acceptable by high standards (MM fonts with a width axis could use slightly larger values, BTW). The same holds for letter spacing, which I consider a poor-man's alternative to font expansion. In other words, only a very small percentage of lines can be improved by playing around with `.tkf' IMHO. The final goal is to get a uniform grayness on a page, and for this you *must* use a paragraph-oriented algorithm. Given today's memory abundance and the high velocity of CPUs, the ideal route would be to implement a document-wide algorithm for typesetting a document (in contrast to TeX's page-wide approach). This could then handle widows and orphans gracefully, and produce optimized positioning of floats (i.e., images, tables, etc.). However, this is not a trivial task. For example, it should also gracefully handle the situation of two-column text, but with a centered image on the page so that the text of the two columns flow around it. Ditto for the generalized case of arbitrarily placed images in multi-column layout, or even non-rectangle image shapes. Werner
