Re: [Groff] Letterspacing

[Peter Schaffter]

On Thu, Mar 27, 2014, Tadziu Hoffmann wrote:
> 
> > So there are two readily-available methods: varying
> > letter-spacing, or varying inter-word spacing.
> 

Not "or" -- "and".  Most times, I opt for word-spacing adjustments.  
The reason for my comparison was only to show that deft use of
letterspacing goes unnoticed.  If I'd used all the other tricks, I'm
sure I could have crafted a paragraph to make even the most die-hard
TeX fanatic weep.  As could we all. :)

> Overall, I think I'm with Werner on the issue of letterspacing.
> I usually find it visible (and irritating), but I can tolerate
> rather large amounts of word-spacing.

By letterspacing, I mean both tightening and loosening.  Sounds as
if you're talking about loosening only.  Either way, if you can see
it, it's because it's been done badly.  That's the point I was
demonstrating.

The other reason for the comparison was to get this very discussion
started.  How do we go about improving justified text?  Doug's
comments got me thinking.  We all agree groff's paragraph formatting
needs an overhaul, but is Knuth the right way to go?  Should it be
mentioned as a specific goal in the mission statement?

I have nothing but my gut to go on here, but my strongest feeling is
that there's a simpler way.

Groff uses a greedy, line-by-line approach to justifying text
(hereafter, I'll use "formatting lines" instead of "justifying
text"):
  - collect words until the line-length is reached/exceeded
  - look for legal breakpoints (spaces after words, hyphenation points)
  - choose the rightmost breakpoint that does not exceed line-length
  - expand spaces till the full measure is reached
  - break the line

When groff doesn't "get it right", we see lines with overlarge
wordspaces, which we fix by adjusting word- and letter-spacing
(and, very occasionally, glyph width).

KP posits that the solution to getting it right is to scan the
whole paragraph.  The number of words on each line is determined by
choosing a line arrangment for the whole paragraph that minimizes
the sum of the squares of the spaces at the ends of lines.  IOW,
an arrangement that strives to equalize wordspacing throughout the
whole paragraph.

Overall, the principle is sound.  If we could wave a magic wand and
get groff to do this, I doubt anyone would mind.  The problem is, as
Gunnar Ritter reminded me: "...the task of rewriting large portions
of an existing environment that assumes line based formatting."

KP works well with TeX because TeX was written from the start to
implement it.  Groff wasn't.  When I step back from pie-in-the-sky,
it's obvious that KP and groff aren't a mix, not without massive
rewrites of a key part of the code.  Like Doug, I smell an increase
in complexity that runs counter to simplicity, flexibility, and
portability.

I've been pondering this for a long, even raised the issue a few
times.  Groff's line-at-a-time formatting isn't awful, otherwise we
wouldn't be using the program.  It's fine for the majority of work
people do, and have done over the years.  Its limitations show up
in narrow-column work and fine typography, sure, but when they do,
we use tools groff already provides to massage individual lines
until our paragraphs look right, typically by adjustments to the
word-spacing, and, in my case, letter-spacing as well.  The end
result is often identical to a paragraph formatted by TeX.

In short, we work within the limitations of line-at-a-time
formatting by doing manual line-at-a-time adjusting.  Which begs
the question: is it really necessary to scan an entire paragraph
to determine optimal linebreaks if judiciously adjusted word-and
letter-spacing on a line-by-line basis can produce similar results?
Would it not make more sense to have groff, more or less as-is,
shoulder more of the burden of what we do manually, *using the same
strategies*, to achieve better *lines*, rather than focussing on
the whole paragraph?

I have some thoughts on how to approach this conceptually, which
I'll post in a day or two.  Been at this too long already.

-- 
Peter Schaffter
http://www.schaffter.ca