Older: fontina 0.1.0, the first release,  Up: News

0.2.0, and a browser you can work in

2026-09-14

fontina 0.2.0 is out, ten days after 0.1.1: forty-two features, thirty-eight fixes and five performance changes. Most of it is the browser, a much larger set of questions you can ask of your library, and a long tail of hardening that came out of reviewing the last two milestones rather than out of bug reports.

The browser

fontina ui was a way of looking at the index. It is now a way of working on it.

Mark many faces and act on them once. Take back the last thing that changed the index, whatever it was. Open a command palette over every command the program has, listed where you are rather than in a manual. Build a filter while watching the count move, using the same flags the command line takes — so the row at the bottom of the browser is the command that would have done the same thing, and copying it is how you script what you just did by hand.

Three of the additions answer questions rather than run commands. What else in this library is nearly this font ranks the rest of your faces against the one you are looking at and shows the numbers behind the ranking. Who can set this text takes a paste and tells you which faces cover it. What pairs with this reads what a font says it is — its classification, not a guess from the name — and ranks on that.

The layout follows the terminal now: as many panes as the window can carry and no more, one palette resolved against what the terminal can actually show, one box, one bar, one left edge.

It is also measurably quicker. A pane builds the rows it draws rather than the rows it holds, the listing queries left the thread that draws, and there is a repaint budget in scripts/bench that fails the build if a keystroke stops being instant on ten thousand faces. The preview cache now says what it costs as well as what it saves.

Two of the four browser milestones are finished and two are not: what it looks like is three quarters done, and the comparison work — pinning faces side by side — is still ahead.

Better questions

The index has held all of this since M0. The filters had not caught up.

• --lang and a language facet, with a font’s two different claims about a language kept apart rather than merged.
• --mono and --proportional, and a spacing facet. On a developer’s library this is the most useful single division there is, and it was read from every font and then thrown away.
• --script repeats, so you can ask for two scripts at once, and --script-min asks how much of one a face actually covers.
• A variable font is found at every weight it spans, not only at its default. A face whose weight axis runs 200 to 800 now answers a search for 400.

Handing fonts over

collection export --bundle produces a collection you can give to somebody else, fonts and all. fontina tag sync reads and writes the operating system’s own file tags, in one direction at a time so it cannot quietly lose anyone’s work. Every command that takes faces reads them from standard input too, so programs can pipe in. A specimen now says how many licensed fonts it is carrying, and is laid out so the type can be judged rather than merely displayed.

Hardening

Most of this was found by looking rather than by breaking.

The WOFF2 decoder panicked on a file it was handed; it is contained now. An empty cmap no longer hides the rest of a font’s report. Two processes creating the same index at the same instant no longer race. The watcher keeps up with what people actually do to a font directory. The rasteriser clipped a preview instead of smearing it, and stopped overflowing. A scan says which files it walked past and why, rather than silently ignoring them.

The login agent that puts your activations back after a reboot touches only files fontina wrote, which is the sort of thing worth checking before it ships rather than after.

Packaging

Manifests for Homebrew, Scoop, winget and the AUR are in the repository. They are not published yet — that needs credentials only a person can create — but the release now also builds an arm64 .deb and an aarch64 .rpm alongside the x86_64 ones, and every package is installed for real with apt and dnf in a clean Debian, Ubuntu and Fedora before it goes out.

Download has the archives and how to check them: a SHA-256 for every file, a SLSA build provenance attestation that proves the archive came from the release workflow rather than somebody’s laptop, and an SPDX bill of materials.

Why 0.2.0 and not 0.1.2

Because forty-two features is a minor release. release-please was configured to move the patch number for a feature until 1.0, which is allowed before 1.0 and still wrong: somebody running 0.1.1 who saw 0.1.2 would have read it as bug fixes. That setting is off, here and from now on.

Pre-1.0 still means what it said: command output, the JSON schemas and the health-check identifiers may change, and the changelog says when one does. Releases lists every version.