Persian digits are not a font feature

The usual hope is that a font will turn a Latin digit into a Persian one. This site does not do that. The character is chosen before paint, with fa-IR. Vazirmatn then has a glyph for it, because Persian digits sit in the Arabic block and the variable file includes that block. Inter, in the global stylesheet, is Latin and Latin-extended only. Nothing in fonts.css swaps the figure.

The short note Put right-to-left in on day one is the cluster piece. Two routes, not a flipped page walks through lang and the calendar. This one stays with digits and type.

A digit is a different character

toLocaleString('fa-IR') and Intl.NumberFormat('fa-IR') write Persian digits. That is not a CSS substitution. The code turns a number into a string of Persian characters, and the string is what enters the page.

Reading time on an article does this. So do the home-page stats, the language percentages on the projects page, and the counter that climbs in motion.ts. That counter uses fa-IR only when document.documentElement.lang starts with fa. A Persian document is fa-IR, so the check passes. An English document is en-US, and the same function builds Latin digits.

Post cards format on their own, with useGrouping: false, so a small count does not grow a thousands separator. The rail beside the home page does one more thing. padStart pads with the Persian character ۰, not with a Latin 0. A one-digit Persian string is still one character. A Latin pad would sit a foreign zero next to a Persian digit.

The date on that same page becomes Persian by another call, but the figure and the month name both come from the locale. formatDate lands in the time element. The minute count beside it is formatted on its own. One is the calendar day. The other is a reading count. On a Persian page both use Persian digits, because both received fa-IR.

Two families, two ranges

On html the default stack is Inter, then Vazirmatn, then the system UI. html:lang(fa) reverses that order. The selector is :lang(fa), which also matches fa-IR. An exact html[lang='fa'] does not match a document whose language is fa-IR, so it never puts Vazirmatn first.

Vazirmatn comes from @fontsource-variable/vazirmatn. One of its ranges is the Arabic block, from U+0600. Persian digits, U+06F0 through U+06F9, live in that block. Inter in fonts.css is two files, Latin and Latin-extended. Neither reaches the Arabic block. Inter cannot draw a Persian digit, even if one showed up on an English page. The glyph falls through to Vazirmatn or to the system font.

Latin digits, U+0030 through U+0039, sit in the Latin range of both families. On a Persian page Vazirmatn is first, so a Latin digit inside a repository name usually comes from Vazirmatn’s Latin, not from Inter. A more specific family wins where one is set. The command palette starts with ui-monospace and keeps Vazirmatn behind it. The intro canvas asks for Vazirmatn first in both languages. Share images use the separate file Vazirmatn-Bold.ttf, not the variable face on the page.

Weight, line-height, and motion are in Type, space, and the motion of a page. This page does not repeat them.

Not every number is Persian

Stars and forks on a project card are raw JavaScript numbers, inside dir="ltr". They do not pass through toLocaleString. Repository names, languages, topics, and the ⌘K button are left-to-right islands too. A code block in an article sets direction: ltr.

Direction and digit shape are two decisions. A left-to-right island can still hold a Persian digit if the string was formatted first. Month labels on the home activity strip are built with fa-IR-u-ca-persian and then placed in a grid with dir="ltr". The month label is Persian. The week columns do not mirror.

Language percentages go through num, so on a Persian page they are Persian. The same page shows a star count in Latin digits beside them. That split is not one global rule applied halfway. Each place that builds a number decides the format there.

The calendar and the figure are not the same choice

formatDate formats a post date with fa-IR. In current Intl that locale is both the Solar Hijri calendar and Persian digits. formatMonth and the home activity strip are more explicit: fa-IR-u-ca-persian. The busiest day on the projects page passes calendar: 'persian' to Intl.

The clock on the about page is an Intl.DateTimeFormat in the page locale, with timeZone: 'Asia/Tehran'. The label is “my time in Tehran.” The clock asks for font-variant-numeric: tabular-nums, and the home stat numbers ask for the same. That is a request for equal-width figures. The characters themselves still come from Intl. The global stylesheet has no OpenType feature that swaps Latin and Persian digits.

A digit on the canvas, apart from the Latin grid

The commit city writes the busiest day’s count with Intl.NumberFormat in the page locale and puts it on a sprite. When the page is Persian, dates in that scene also pass calendar: 'persian'. The activity cells are a different object, and they stay dir="ltr". One view can carry a Persian digit on a label and still keep week columns from mirroring.

Home stats and language percentages call toLocaleString with default grouping. Post cards turn grouping off with useGrouping: false. A large contribution total on the home page gets a separator. The reading time on a card does not. Both are Persian digits when the page is Persian. The difference is the separator, not the font.

Share images do not build these figures. They load Vazirmatn Bold and set direction on the SVG by themselves. If a post title already contains a digit, that is the character written in the markdown, not a fresh fa-IR format. The page’s stats never reach that PNG.

Page direction does not make this choice. dir says which edge a line starts from. It does not say which character 4 is. That difference is in Direction does not finish the page.

A separate route keeps a separate digit

Persian and English do not share the HTML file. The Persian home page formats stats with fa-IR. The English home page uses en-US. If both languages were one document and only dir changed, the same number string would reach both. Here the formatter receives the language while the page is built. Addresses and hreflang are in The translation is a second file.

The collaboration page says what kind of work is welcome. The only public contact is Telegram, @MH759.

If one thing stays from this page, it is this: the font does not invent a Persian digit. fa-IR builds the character, Vazirmatn draws the Arabic block, and a number that never passed through that formatter stays Latin.