A while back I got frustrated with struggling to read absurdly large numbers in terminal windows, and set about thinking how I might apply some logic in the terminal to subtly bunch together groups of three digits as a form of thousand separator. Eventually it occurred to me to try doing it in the font with ligature rules (initially as a joke, but then I looked into it) and it turns out Numderline is a thing which does that.

Well, it does it in terminals which support font ligature rendering, with caveats about efficiency, cursor placement, screen edges, and digits which dance as you type… but I would suggest these compromises have been worth it.

You could discover that for yourself, of course, so what am I adding?

I didn’t really get along with underlines for digit grouping. I would prefer to stick with spaces, because they’re a more common convention. So I hacked around a bit, tweaking up the spacing support and adding hexadecimal support (that is, things starting with 0x grouping by fours).

But also, since I found that most (not all) terminals support it, and CSS also supports it, I decided to rely on so-called “font features” to make the digit grouping configurable.

The common way of stuffing extra ligature rules into a font is to put them under the calt feature, which is on by default. But in various applications in various places where you choose your font, you should also have the option of specifying a bunch of other features by FourCC or by common names. The Iosevka font, for example, offers extensive customisation in this space.

So to test this I used the codes dgsp (“digit spaces”, I guess), dgco, and dgdo (for commas and dots).

In CSS, for example, if using the patched font then spaces could be enabled with:

font-feature-settings: "dgsp";

Then I extended on that with dgcd and dgdd if you want such marks to the right of the decimal point as well. Personally I find that a bit disorienting (especially when the dot has to be changed to a comma, which just is a big fat lie), so I like to stick with spaces throughout.

But there’s so much more that I did not do. I did not implement Chinese spacings (except in hexadecimal, I guess). I did not do Indian spacing. I didn’t do apostrophes as separators. I did not add the option to group digits after the decimal by fives like Wikipedia does. I didn’t do anything at all for any numeral system outside of US-ASCII.

I also didn’t do anything about things like git hashes, which have no prefix to signal that they’re hexadecimal. They just come out messy. What I probably should do, there, is to disable decimal digit grouping for anything that abuts a letter (though this falls down for, eg., currency, like CAD1000000).

TODO: all those things

And also I haven’t figured out how to do all this without breaking the existing ligature rules, because of the specific combination of tools and bug workarounds needed to make forward progress (see for example the original [Numbderline][Numbderline 2] mention of such). Somebody helpfully posted a comment on my own bug on that matter, but I have not yet followed up.

TODO: that thing, too

Before going any further with extensive configurability, I stopped to ask myself what the point of it all would be if I did do that, when I already have what I want.

In the abstract I think the principal benefit is not having to handle text in quite such a locale-specific way (with an outstanding caveat about the decimal comma versus dot, etc.). It becomes a presentation-layer problem. This means you can cut and paste it elsewhere and you won’t be foiled by extra glyphs causing confusion in other software. Application settings or CSS can make their best effort at internationalised font configuration, but if you copy-paste the text it’s still in its original bare-bones format.

So I looked into more documentation on how and when to use font features, and what to do with my made-up codes.

I found a lot of verbiage about language and script support, which seemed sufficient to auto-detect the appropriate configuration. Thankfully other documentation said specifically to not do that (TODO: find that reference). The application is meant to make all such decisions and signal what it wants by configuring the font appropriately.

After all, this would be complicated by an international userbase who are used to dealing with English software which doesn’t support their native systems properly (and, little by little, a userbase which doesn’t even know their own native systems any more), meaning that sometimes localisation efforts are seen as unhelpful and confusing. Certainly a strong case for the font not making its own deductions.

I think configuring the font among many different conventions means reserving a bunch of different “features” identifying different conventions, for which I can (and already did) just go ahead and make up my own. But if a serious effort was worthwhile then things would need to be more structured and complete.

And then I stopped, because I don’t know of anybody who actually cares about all that.