RE: [ft] otf autohint/nohint problem

[Roger Flores]

These questions are a follow up after tinkering with the various hinters
and trying to find which one looks the best for the set of fonts used.
I don't need answers, but I thought people might want to discuss them.

1. My understanding of hints is that they are needed when the font is
small to help the appearance.  Is there some threshold above which they
can be ignored?  Could this lead to performance improvements (because
the hints can be trivially skipped or the module can avoid loading)?

2. Are there any tests showing that fonts at a large point size are
identical regardless of the hint module used?

3. I was thinking about the error from the hint modules and how to
measure and reduce it.   We obviously can't have an ideal hinter, but I
was wondering if we could still calculate an ideal hinter even if we
can't make one?  For instance, one view of a perfect hinter could be
that it shows all the edge transitions.  If we calculate the number of
transitions possible on a pixel row, and then compare to what a hinter
produces, we can get a rating for each row of the character.  Rotate for
the vertical dimension.  Apply to all characters.  Now we have an
automatic test for how well a hinter is working.  Look at the worst
characters and improve them.  Make sure none of the other characters go
down, over time.  You could confirm that adding a hint reduced the hint
error.  You could check at which point sizes a hint reduces error (maybe
the hint isn't implemented correctly).  There are obviously
complications (what is the result for a one pixel tall capital 'H') but
even a simple version would probably reveal a lot.

4. I was thinking about trying to figure out which hinter to use.  The
answer seems to vary depending on which fonts are installed, meaning you
can't predict at compile time.  If we had a hint error measurer we could
automatically select the best hinter for the font at runtime.  We could
probably guess well with a trick.  For instance, we probably don't need
to measure every single character.  There might be two or three
characters that give us the right decision 90% of the time.
 

-Roger Flores