Re: FontMatrix revisited (CFF & T1)

[Just van Rossum]

At 5:44 AM -0400 01-08-2000, Tom Kacvinsky wrote:
>Ever look at Fontographer generated fonts (of course, silly me, you
>worked on RoboFog , right?)  There are spurious floats all over the
>place in the output (usually in the global hint structures).

Even so, I don't see how a few floats in a global array has much impact on
the parsing time of an entire font. It just doesn't seem a worthwile
component to optimize for speed.

>Floats in CFF just aren't implemented as strings that we can pass to
>atof().  The parsing routine (start with 30, end with a nibble set
>to 0xF, etc...) is a necessary evil.  Or did you mean, construct a
>string of the float using the parsing loop, and then pass this to
>atof() ?

That's exactly what I meant. The thing is, those font matrices contain very
small values. Converting these first to 16.16 fixed point numbers and
_then_ scale them by a factor of a thousand or so simply loses too much
precision. To me it seems easier to first convert to a float (or double),
do the math, and convert to fixed as the very last step.

Is atof() used anywhere else in the library? If not, would using it have an
important impact of the code size? I know David likes to keep things as
small as possible...

Just