Field Notes: Windows 3.1 Edition

2024-06-04

by drj

Can we create new fonts for 20th century operating systems? Short version: Yes! Use fontmake.

Reminder on installing fonts in Windows 3.1

Use Fonts control panel to install fonts.

This is a bit annoying because i haven't worked out how to sync folders in DOSBox, so if i copy new files into an already mounted folder DOSBox won't see it. I have to quit and restart.

Observations

Ranalina can be installed. (it is called Ranalint in the screenshots because i am using the development version which has a different name from the production version).

Ubuntu Regular can be installed.

In Character Map Ranalina only displays the ASCII range, but Ubuntu Regular seems fine.

A screenshot showing a sample of Ranalina font and its
Character Map; it is clear in the Character Map that the
entire bottom portion (after «~», at ASCII position 0x7E) is
not available and is showing the `.notdef` glyph. — Note how the `.notdef` glyph occupies the bottom half

A screenshot showing a sample of Ubuntu Regular font and
its Character Map; most of the Character Map shows proper
glyphs. — Note how most characters are available in Ubuntu Regular

Slightly to my surprise the OS/2 table version seems not to matter. Ubuntu Regular has a version 3 OS/2 table, and Windows 3.1 copes with that, despite the fact version 3 was specified well after Windows 3.1.

The OS/2 table is somewhat forwards and backwards compatible in the sense that newer versions generally only append fields, and therefore Windows 3.1 code that "just grabs the bytes from their known locations" will work even when there are more fields (which will be ignored).

Is it the cmap?

(yes)

I rebuilt ranalina.ttf by converting it to UFO with tx (from AFKDO) and compiling the UFO to TTF with fontmake (from googlefonts):

tx -ufo -o ranalina.ufo -f ranalina.ttf
fontmake -o ttf -f ranalina.ufo
# outputs the font file master_ttf/ranalina.ttf

This version of Ranalina has a different Character Map:

A screenshot from an emulated Windows 3.1 showing a
sample of Ranalina font and its Character Map. — Note how that in the bottom half, several characters are present

the lowercase is missing, because Ranalina has no true lowercase: duplicate entries in the cmap map the lowercase letters to the same glyphs as the uppercase, effectively double mapping the glyphs and giving them two unicodes. tx assigns each glyph only one unicode value when converting to UFO (despite that fact that both TTF and UFO support multiple unicodes mapping to the same glyph).
various accented letters are missing, because Ranalina currently supports these by using combining accents (U+0300 COMBINING GRAVE ACCENT and so on), and does not encode the pre-composed forms.
however, the remaining "bottom half" characters are shown where Ranalina has them, which is an improvement over the original Ranalina. For example Œ ± £.

What cmap?

My (somewhat ad hoc) seefont cmap tool can give some more insight into the cmap tables. In the output below, RANALINA.TTF is the original Ranaline font file; UBUREG.TTF is Ubuntu Regular; RANA-FM.TTF is Ranalina rebuilt by tx and fontmake above.

; for a in ~/WIN31/RANALINA.TTF ~/WIN31/WINDOWS/SYSTEM/{UBUREG,RANA-FM}.TTF
do
echo ==== $a ====
seefont cmap "$a" | grep '^[^ ]'
done
==== /Users/drj/WIN31/RANALINA.TTF ====
cmap,length, 2020
cmap,sub,0,4,828,0/4 Unicode/BMP+non-BMP
cmap,sub,3,1,28,3/1 Microsoft/Unicode-BMP
cmap,sub,3,10,828,3/10 Microsoft/Unicode-UCS-4
cmap@28 format 4
cmap@828 format 12
==== /Users/drj/WIN31/WINDOWS/SYSTEM/UBUREG.TTF ====
cmap,length, 1514
cmap,sub,0,3,28,0/3 Unicode/BMP
cmap,sub,1,0,992,1/0 Macintosh/Roman
cmap,sub,3,1,28,3/1 Microsoft/Unicode-BMP
cmap@28 format 4
cmap@992 format 6
==== /Users/drj/WIN31/WINDOWS/SYSTEM/RANA-FM.TTF ====
cmap,length, 430
cmap,sub,0,3,20,0/3 Unicode/BMP
cmap,sub,3,1,20,3/1 Microsoft/Unicode-BMP
cmap@20 format 4

The TrueType cmap table contains a number of subtables each in a particular format, which correspond to binary file structures. The subtables can be in different encodings (for example Unicode/MacRoman/Shift-JIS), and different subtables can share the same data as long as it's compatible. The last example in the previous console output shows the simplest case of this: the RANA-FM.TTF file has two subtables that in fact share the same data.

The output of the tool is a bit cryptic, but an example should clarify. From RANALINA.TTF:

cmap,sub,0,4,828,0/4 Unicode/BMP+non-BMP

This (CSV formatted) row tells us that the cmap table has a subtable of Platform Code 0 and Platform Specific Encoding 4, starting at byte offset 828 (from the beginning of the cmap table); the final part of the row summarises that as Unicode platform with BMP+non-BMP encoding. More details in the TrueType Reference Manual.

Discussion

The fonts which display "bottom half" Character Map characters are UBUREG and RANA-FM and the common cmap subtables between them are 0/3 and 3/1. Which one is Windows using? RANALINA.TTF also has a 3/1 subtable, so why doesn't it work? Is Windows 3.1 using the "Unicode platform" subtable? It is confused by the presence of a Format 12 subtable?