grid fitting

[Laurent.SIEBENMANN@math.u-psud.fr (Laurent SIEBENMANN)]

grid fitting (was Problem with dvips and Acrobat)


BKPH: > 
 > One of the great features of PS is that it enables one to
 > create resolution-independent output.  Grid fitting is a
 > function that the rasterizer should perform, not the
 > application producing the PS.

Knuth bet in the late 70's that the conceptual model for
printing would long be the integer planar grid  Z x Z.  He
soon lost that poker game when in the mid 80's Postscript and
established as model the real cartesian plane R x R.

Then Rokicki boobed bigger and better in using the Knuth
model in a program that serves Postscript.

BKPH:
 > It [dvips] rounds the actual character advance widths to some
 > discrete grid and then replaces the true advance widths with
 > the vector of coarsely quantized values it has constructed.
 > Resulting in a PS file that is not truly
 > resolution-independent --- a file that is optimized for a
 > specific resolution and that will print *worse* than if no
 > adjustments had been made at a different resolution.

Uniformly worse! It is a good bet (and an empirical
observation) that AT ***ALL*** RESOLUTIONS dvips output will
print *worse* than if no res dependant grid adjustments had been
made by dvips.  An uncoordinated doubles pair plays worse tennis
than a single player.

Melissa:
 > About the only way around this problem is to tell DVIPS that
 > you have a 2400dpi printer or somesuch so that the roundoff
 > errors (1/4800th of a point) are unlikely to add up to much.

Right!  The only remedy I know is to set the resolution to
maximum. Did you know that gentle reader?  If not, then according
to Rowland, you do not qualify as moderately intelligent.
Since the remedy is squarely counterintuitive and undocumented
to boot, would it not be better to say that there is a tiny
minority centered on this list that has some clue.

Strangely, max res still leaves dvips as low man on my empirical
quality scale.  My guess is that the drift fitting (alias
drift control) jazz is so complex that the code is undebuggable
and the performance slightly capricious.

Melissa:
 > Maybe someday someone will become impassioned enough about such
 > minutiae to fix DVIPS.

Hmm.  At 1200 dpi and above optimal grid fitting is not a big
issue (in spite of Melissa's wild arithmetic).  BUT on our
computer screens -- that are just beginning to offer classical
typographic quality -- it is going to be a huge issue for the
forseeable future.

      Cheers

             Larry Siebenmann



PS.  Gratified to hear more or less coherent lamantations on the
grid fitting issue, I wonder if anyone is up to debating the related
issues in virtual font design. (We all know it is NOT Berthold's cup
of tea!) See my unanswered posting on "invisible rules in vfs".  The
grid fitting problem is most acute there since the placement
accuracy required is several times higher than in interletter
spacing.