Building a Human68k Font Converter for TrueType
Converting the fonts used by the Sharp X68000 into TrueType can seem like a small cosmetic task, yet it opens a valuable route into modern workflows. Human68k bitmap fonts carry the visual identity of Japanese home computing, and many exist only in old FNT resources or disk images. A useful converter should preserve the original glyph shapes, character mapping and spacing while producing a file that current editors, browsers and design tools can handle.
The project is more than a routine file export. A Human68k tool must interpret platform-specific font structures, translate source codes into Unicode and decide exactly how a fixed pixel raster becomes scalable outlines. That makes the result a preservation asset rather than a loose approximation. It also lets an Australian collector use a carefully reconstructed face in a word processor, document or web project without abandoning the X68000 context that gives the font its meaning.
Reading the Source Font
Human68k font files normally use the .FNT extension and contain packed monochrome glyph rasters. A collection often separates its Latin and Japanese resources, with 8×16 and 16×16 pixel faces serving different screens and applications. The binary layout is specific to the Human68k ecosystem, so the converter needs its own parser rather than a generic BMP loader. Dimensions, pixel row stride, offsets and spacing data must be read according to the format revision being processed.
Character identity needs separate attention. Human68k text commonly uses ASCII, JIS X 0208 kana and kanji, and half-width katakana, but source codes are not Unicode code points. A conventional Shift-JIS mapping places 0x8140 at U+3000 and 0x82A0 at U+3042; half-width katakana occupies another contiguous block. Put these relationships in a versioned mapping table, record how duplicate and unmapped glyphs are handled, and emit a Unicode cmap that applications can query without guessing.
A dependable parser should reject truncated files, invalid offsets and inconsistent glyph counts instead of following bad arithmetic into a crash. Confirm the byte order against known files rather than assuming every inherited DOS-era font behaves like native Motorola data. Keep each original .FNT untouched, record a checksum and retain notes about the source. The SMB research archive is a useful reference when cross-checking specifications, older research and machine-specific behaviour.
Designing the Conversion Pipeline
Separate the source parser, character mapper, outline builder and TrueType writer into distinct components. That separation keeps difficult bitmap decoding away from the font assembly stage and makes broken inputs easy to diagnose. Python is a practical choice: fontTools supplies a reliable object model for font tables, while the TTGlyphPen class can receive the contours produced from each raster. A command interface might accept --input, --map, --units-per-em and --output. The result can run on the Windows x86-64, Linux and Apple silicon desktops common in the Australian market.
Represent every decoded glyph with its source code, bitmap, dimensions, baseline, side bearings, advance width and Unicode destination. A JSON mapping file works well for the first release, with explicit rules for private-use code points when no agreed Unicode value exists. Use the source pixels as the design grid, convert metrics into font units and set the em square to a documented value such as 1000. Do not stretch an 8×16 face into a 16×16 grid and describe the result as historically exact.
Generate TrueType contours that match the original lit pixels, then let the writer build glyf, hmtx, cmap, head, hhea, maxp, name, post and OS/2 data as appropriate. The hmtx advances should reflect source spacing, including side bearings, rather than the width of the dark pixels alone. Keep Unicode normalisation and glyph ordering under explicit control. For Japanese text, check kana, kanji and punctuation separately, since one visually similar character can carry different code-point semantics.
Choosing the Outlining Method
The central design choice is whether the TrueType file should imitate the raster or reinterpret it as a clean vector face. Retain the source pixel grid for a faithful archival edition, or redraw the shapes for a smoother publishing edition. A converter can produce either, yet mixing the two without warning makes an archive difficult to trust.
| Consideration | Human68k source | Bitmap-faithful TrueType | Redrawn vector TrueType |
|---|---|---|---|
| Glyph data | Packed monochrome pixels | Contours following pixel edges | Smooth quadratic Bézier curves |
| Size handling | Fixed raster dimensions | Scalable, but enlarged edges remain stepped | Scalable with smooth interpolation |
| Character mapping | Platform-specific code tables | Unicode cmap and explicit source mapping |
Unicode cmap and explicit source mapping |
| Native-size match | Exact reference | Very high when aligned to the grid | Depends on redraw accuracy |
| Editing | Awkward outside original software | Simple but preserves the raster character | Clean, flexible outlines |
| Main effort | Preserving source files | Automated outlining and quality checks | Manual or assisted vector reconstruction |
The bitmap-faithful approach follows horizontal and vertical pixel edges, retaining the corners, counters and blank areas seen on the X68000. It is well suited to reconstruction, teaching material and close visual comparison. Scaling it up reveals stair-step edges, but that is a truthful property of the source, not a failed outline conversion.
Pixel tracing can automate a smoother-looking alternative, yet diagonal strokes may acquire rounded joints or shifted corners. Hand redrawing gives the best curves and point economy, but demands far more time and subjective judgement. TrueType hinting can improve small-size alignment; it cannot repair inaccurate geometry. Where both editions are useful, publish them under distinct names and document the outlining policy.
Testing the Converter Against the X68000
Testing should cover structure and appearance. Use fontTools’ ttx utility to list and decode the generated tables, then check that the Unicode lookups, advance widths and vertical metrics agree with the source records. Automated tests can assert expected mappings for representative ASCII, kana, kanji and punctuation, while also ensuring that no source glyph was silently discarded. Golden images add a useful regression check whenever the outline or rasterisation code changes.
Visual comparison needs care because TrueType is not automatically a Human68k-compatible font on the X68000. Capture reference images from the original machine or a suitable emulator, then render the converted file at matching integer pixel sizes with antialiasing disabled. A monochrome FreeType raster is a convenient cross-check. Small differences can arise from grid fitting, so compare stroke placement and counters rather than treating every changed edge as an error.
Collectors in Sydney and Melbourne often share software, spare parts and reference images at local retro-computing gatherings, which makes side-by-side testing easier than working with an isolated download. Brisbane’s subtropical humidity deserves a little extra care with boxed 3.5-inch floppy disks: keep archived media in a dry, stable cupboard and inspect it before use. Hardware purchases from Japanese sellers also fit an Australian budget differently, with AUD exchange rates, GST, freight and bank charges included in the real cost.
Australian mains power adds another practical constraint. The X68000 was designed for Japanese 100 V supply, while Australian mains is nominally 230 V at 50 Hz, and Type I plugs do not fit Japanese Type A sockets. Use a properly rated isolation transformer and an RCD-protected circuit rather than a travel adaptor, and consult a qualified electrician if the setup is unfamiliar. For a working bench, an expansion board such as Nereid-X can preserve a usable test system, while an expansion-slot case modification can make connector access less awkward.
Preserving Provenance and Sharing the Tool
A release should preserve provenance as carefully as it preserves pixels. Include the converter source, dependency versions, mapping tables, outlining mode and commands used to generate each output. Store a SHA-256 checksum for the original .FNT and its converted counterpart. Name the font family and style accurately, and state whether it is a pixel-faithful reconstruction or a redrawn edition. That documentation lets another hobbyist repeat the process and distinguish an original resource from a later interpretation.
Font licences need separate attention from software licences. The right to read the Human68k format does not grant permission to redistribute a commercial game font, and the ownership of a vector outline can be just as sensitive as the ownership of its source bitmap. Bundle only faces whose redistribution is permitted, or ask the rights holder for permission. A small, clearly licensed sample is enough to exercise the test suite and demonstrate the workflow without publishing an entire copyrighted collection.
Document the workflow for people who are new to the X68000 while still giving experienced developers enough detail to extend it. Explain how to select a .FNT file, supply a Unicode mapping, run a conversion and inspect the result on a modern operating system. Include rendering comparisons for both the original raster and the new font, noting any glyphs that remain unmapped. Convert a legally shareable sample, compare it against the original display, and publish the verified font and mapping notes on X68K.NET.
Nereid-X Expansion Board
A personally-produced LAN+USB+Memory expansion board for Sharp X68000 series computers. Multiple production runs were offered, including a final batch and a later revival reproduction run.
Power Supply Repair
X68 power supply repair and modification services were offered by the site owner, with documentation shared through diary entries spanning 2001–2006.
Server & Networking
Notes on FreeBSD administration, ISP changes, server migration, and networking topics. The site itself ran on FreeBSD with the hns diary system and Namazu search integration.
Get in touch
X68K.NET connects Sharp X68000 enthusiasts through community links and shared projects. Reach out with questions about the Nereid project or X68 resources.