fontinst and reals (was: fontinst with 8y.etx)

[Melissa O'Neill]

Hilmar Schlegel wrote:
> You can either rewrite fontinst to use floating point metrics instead of
> AFM-unit integers or use "raw" TFMs generated by fontinst with exactly
> identical rounded data. There is no real precision loss since
> differences are below 1 AFM unit (1/1000 em) except for CM fonts which
> use advance width of non-integer dimensions.

Multiple master font instances also have potentially non-integer advance
widths, so it isn't just an issue for CM fonts.

Also, is it really the case that fontinst rounds to the nearest integer?
I was under the impression that it truncates down. But then I stopped
using fontinst for my multiple-master fonts several months back, when
it couldn't handle real advance widths at all.

    Melissa.

P.S. I use FontKit, my unreleased set of Perl tools, for TeX font
installation -- mostly for on-the-fly generation of metrics for
multiple master fonts. FontKit does not yet do glyph fakery,
autonaming, or automatic FD files, but does do just about everything
else (i.e. scaling, oldstyle digits, characters from expert fonts,
etc.). I don't see FontKit as becoming a competitor for fontinst any
time soon (although it is the only tool I know that knows how to use
SKIP entries in LIGTABLES). I wrote it partly because of the rounding
issues with fontinst and because fontinst used to cause my elderly
version of TeX to blow up. Mostly though I wrote it both because it
was fun, and because that way I'd have a font installation kit whose
internals I could easily understand.