Numderline

Numderline is a font patcher that uses OpenType font shaping trickery to make it easier to visually parse large numbers. It has multiple variants for different preferences, fonts and contexts.

I (Tristan Hume) was inspired to build this by my job at Jane Street involving staring at a lot of large numbers, mostly latencies in nanoseconds, and wanting to pick out larger quantities like milliseconds and microseconds.

See the code on Github to contribute or patch fonts of your choice.

Underlining

The flagship "Numderline" variant underlines alternating groups of 3 digits. It's what I use at work.

Grouping

This variant squishes digits together into groups of 3. It's nice looking and works for most fonts but is annoying for editable text since it jumps around while typing.

Bolding

For fonts with lots of weights, making alternating groups slightly bolder is an alternative to underlines.

Commas

Inserting fake commas is a good option in proportional contexts where changing the width doesn't matter.

Monospace Commas

By combining the grouping variant with tiny commas this variant adds commas while remaining monospaced.

Debug

The debug version shows how the font uses shaping tricks to replace digits with alternate versions that track the position modulo 7.

Notes

• These fonts only work in applications with proper font shaping support. This includes browsers, many native apps and the Kitty terminal. Unfortunately some apps that support basic ligatures don't support the fancy shaping features I use, including Sublime Text 3 and iTerm2. Other apps like Emacs and most terminals don't support shaping at all.
• The flagship underlining variant is my favorite, but it only works on fonts with a sufficiently wide, low and thin underscore like DejaVu Sans Mono. This could be fixed by synthesizing an underline contour for a font instead of using the underscore character, contributions welcome!
• "SauceCode" variants are based on the Source Code Pro font, but the SIL Open Font License prohibits derived fonts from using the same name.
• Unfortunately I couldn't get patching of fonts with existing ligatures to preserve the original ligatures, although it's probably possible with some debugging of arcane font stuff.
• All data in the example table is based on a random generation script, not any real data.