Re: [ft-devel] New `slight' auto-hinting mode

[Werner LEMBERG]

>>> Currently there is a need to adjust pen position when hinting
>>> places two adjacent glyph edges too near or too far.  This concept
>>> does not go away with subpixel pen positioning.
>>
>> Why do you think so?  Where is the difference to the integer case?
> 
> Let's trace all the steps backwards.  Let's consider usual glyphs
> with two well defined edge stems: a, b, d, e, g, h, n, o, p, q, s,
> u. Let's pick a size when these edge stems have width not exactly
> but close to one pixel.  Hinting regardless of shift should produce
> consistent images of the same width with the edge stems covering
> exactly one pixel.

Define `consistent'.  For example, I could live with, say, four
different versions of glyph `a', dependent on the (fractional) start
position.

> Now think of a word "unabandoned".  In order to look good every
> hinted stem should have very predictable position.  What is also
> true is that the pen position is also quite prescribed exactly in
> the middle between letters, which because of hinting happens to be
> either at the pixel boundary or exactly in the middle of the pixel.
> As a consequence, the advance width must also be integer distance
> either between pixel boundaries or between the middles.

Well, ...

> I honestly do not see how you can achieve predictably periodic
> regular hinted look of "unabandoned" with fractional advance.  Any
> accumulating errors of fractional advance will ruin the hinted look.
> I think it is obvious that hinting simply contradicts to fractional
> advance and subpixel positioning.

... this doesn't really convince me.  We should try to make the warper
work with fractional values to get more insight.


    Werner