ps2pk vs. gsftopk

[bkph@ai.mit.edu (Berthold K.P. Horn)]

Hi:

> I don't have compared the quality of the rendering of ghostscript,
> ps2pk and ATM (MSDOS). I would appreciate if somebody does this in such
> a way (TUGboat article?) that we eventually can work on serious
> problems like distorted stems.

I do not know how such a comparison can be done other than in a 
subjective manner.  Do you have some ideas?  Subjectively, the 
ATM rendering is better than the other rasterizers.  

Which is not at all surprising.  It is the top font rasterizer, designed
by the people that designed Type 1 hinting.  The internal details are
trade secret and all clone attempts so far, have fallen short.  This
includes IBM's attempt, and the Sun MicroSystems attempt. It even includes
the rasterizer that Sun was silly enough to buy from Adobe recently
(after killing their own F3 and T1 efforts...).

The Type 1 rasterizing algorithm has not been described or published.  One
can only conjecture how parts of it work through tedious experimentation.
For example, the Type 1 `fill' operator is totally different from the
PostScript `fill' (or `eofill'), which --- amongst other things --- on
average adds 1/2 pixel too much black all the way around a shape (another 
reason mot to call Type 1 fonts `PostScript' fonts :=)

A comparison is particularly hard, since in some cases the rasterization
is distinctly different --- but it is not clear which is `better.'  I am
thinking, for example, of cases where a font at a particular point size is
more `open' (one more pixel in x height) when done using ATM than using
the `standard' Adobe printer PostScript interpreter.  The `open' version
looks easier to read, but suffers from neighboring letters coming a bit close.

Even if you overlay bitmaps created by ATM or the printer PS rasterizer,
how can you tell which is better?  (Assuming their are no obvious errors
like drop-outs, counters filling in, equal stems not rendering equal,
overshoots not being suppressed etc).