Santhosh Thottingal is a practitioner of Malayalam computing and type design, an engineer at Wikimedia, based in Kerala, India. Since 2020, he has been doing something almost nobody else does: he uses MetaPost — a 1980s drawing language, the successor to Donald Knuth's MetaFont — to design modern variable fonts. MetaPost describes every curve, every glyph component, with mathematical formulas. No GUI. No dragging Bézier handles with a mouse. Input is code; output is letterform. With this method he created Nupuram, a Malayalam typeface superfamily with four variable axes: weight, width, slant, and softness.

Why is this the beauty of the remainder? Because there is a triple chisel-construct cycle at work. First layer: MetaFont/MetaPost was Knuth's construct for parametric type design — define letter shapes programmatically, change parameters to change the entire font. This construct was chiseled away by GUI font editors (Fontographer, FontLab, Glyphs) in the 1990s. The type design world nearly forgot programmatic approaches. MetaPost became a fossil, written in WEB literate programming language, compiled to Pascal, its toolchain nearly impossible to extend. It is the remainder of type engineering history — an abandoned path. Second layer: variable font technology (OpenType 1.8, 2016) is the modern construct of web typography. But its design assumption is clear: designers draw multiple "masters" (thin, regular, bold) by hand in GUI tools, then the system interpolates between them. Nobody planned for a 1980s programming language to directly generate variable fonts. When Thottingal uses MetaPost's parametric output to produce variable fonts, he finds a crevice between two eras of construct — neither system made room for this use. Third layer: Malayalam script itself. The toolchain of Latin type design carries an unspoken assumption: letters are composed of straight lines and simple curves. Malayalam's arcs are extraordinarily complex, its fluid conjuncts demand curvature continuity (G2 or even G3), with very few right angles. This writing system is the remainder of Latin-centric font technology — it relentlessly exposes the tools' insufficiency. Thottingal's spiral splines (used in the Manjari typeface, based on Raph Levien's research) and now his two-parameter curve fitting are inventions driven by this remainder.

In February 2026, he went further. He began building his own curve fitting system in Rust, based on Raph Levien's two-parameter curve family — a more modern smooth curve fitting approach than Hobby's algorithm. He used this system to produce variable-width stroke outlines that remain interpolatable for variable fonts. The problem is exquisitely precise: variable fonts require all masters to have exactly the same number of points and paths, but generating parallel curves from variable-width strokes naturally produces an unpredictable number of points. His solution is linear perturbation — offset the skeleton curve by a direction field multiplied by a width function, guaranteeing constant point count. G1 continuity failures (tangent jumps) at segment joins are handled by a multi-stage error correction system: extract corners from the outline, correct misclassifications using skeleton information, detect and repair tangent discontinuities. The entire process is an engineering approximation — he says so himself. But it is precisely in the gap between exact mathematics and engineering compromise that the curves acquire their aesthetic quality. Beauty is the remainder of the optimization process.

The naming gap here is enormous. The type design community does not recognize MetaPost ("that's not our tool"). The MetaPost community is nearly extinct. Curve mathematics researchers (Levien, Hobby) do not make fonts. The Rust/Kurbo ecosystem is a general-purpose graphics library, not a font tool. The Malayalam language community needs fonts but does not know parametric design theory. No existing field can fully name what Thottingal is doing: using the mathematical kernel of a dead 1980s language, rewritten in Rust, to generate modern variable fonts for a complex non-Latin writing system. He calls it "non-conventional type design." The name is provisional, because the logic is still growing — he explicitly states that the next step is building a dedicated domain-specific language (DSL), no longer MetaPost itself, but a new system inheriting MetaPost's spirit while focusing on type design. When a practitioner is actively chiseling away the construct of their own previous stage, advancing toward a next stage that cannot yet be named, what we are looking at is the beauty of the remainder.

Han Qin (秦汉) · @hqinjarsy · SAE Judgment & Aesthetics