[tex-k] dvips Z option and Zandt's alt-rule.pro revisited

Tomas G. Rokicki rokicki at CS.Stanford.EDU
Fri Mar 5 23:59:01 CET 2004

I've never yet spent the time to really review the change.

The dvips rule rendering code evolved over many years to deal with 
many many problems:

   - Off-by-one problems in the rendering on some interpreters
     when fill and image were used together;

   - Outright rendering bugs when image is used for some
     sets of parameters (causing major portions of rules to be

   - Speed issues with image

   - Screen display dropping rules altogether, because the
     rendering resolution didn't match the generating resolution.

Essentially, there was no portable way to mix image and fill
back then and get perfect pixel alignment due to a large set of
bugs in the interpreters.  Yet, using image for rules didn't
work correctly in a number of environments, or took way too
long to render in others.  (I'm not talking a few seconds, but
rather *minutes*.)

Things are probably better now, but I'm still a bit concerned
that the change be pixel-precise; I'm a little bothered by a
div 2 I see there and I'll have to convince myself that it will
do the correct thing in all cases.  It probably will, but I want
to make sure.

At 1200 dpi, a single pixel is probably less of an issue than it
was at 300 dpi (where dvips was developed) but it's still important
to get it right.

And just because the bugs are in old interpreters doesn't mean we
can ignore them; there are still an awful lot of old LaserJets
around, with ancient interpreters, that need to generate the
correct output.


More information about the tex-k mailing list