bug in fontinst 1.801: rounding problems (was: duplicated ker nings)

Lars Hellström Lars.Hellstrom@math.umu.se
Mon, 5 Oct 1998 13:19:28 +0100 (MET)

Alan Jeffrey wrote:
>Gosh, what a remarkable bit of detective work, well done!
Actually, it was pure luck. I happend to be looking at that part of
fontinst, when a sudden "Waaaiiit a minute! What if ..." struck me. I might
have found the murder weapon, but the murderer is still to be caught.

>I guess the question is now what the fix is: peraps all generations of
>MAPFONT should go through the `divide by 8 multiply by 8 again' routine to
>make sure they're all rounded.  This won't make any difference to dimens
>that have gone through the process, and will round others appropriately.
Actually, I would prefer it if scalings with a factor of 1000 were not made
at all. Not scaling uses less of TeX's memory and is most likely faster.

Vladimir Volovich wrote:
>btw, why scaling with factor 1000 occurs in the first place? :-) it
>was not requested in the test file which demonstrates a bug (and in ae
>fonts too).

Many of the metric commands requires a scaling factor. If you want to make
a glyph punctdash that is the same thing as emdash (latin.mtx does this),
you have to write


This is where the scaling sneaks in.

Rowland wrote:
>Not all of us have access to present-day processors, you know.

As "present-day processors" was used, I believe the even the 68LC040 would
qualify with a margin. Its ancestor 68000 probably would not.

Finally, I would like to point out that we might also consider doing
scaling with some kind of multiple word arthimetic. To me, it seems like

   % #1 is a \count register whose contents should be scaled.
   % #2 is the scaling factor. I assume it is a valid TeX <number>,
   %    rather than a fontinst <integer expression>.
   \divide \a_count 10000
   \multiply \c_count 10000
   \advance \b_count -\c_count
   \multiply \a_count #2
   \multiply \b_count #2
   \multiply \a_count 10
   \divide \b_count 1000 % One could do something fancier here, to
                         % achieve proper rounding rather than mere
                         % truncation.
   \advance \a_count \b_count

would pretty much do the trick.

Lars Hellström