l0ng-ai
22e1ab1694
tty7 is split into two Rust processes: a persistent daemon that owns the shells and a GPU-rendered client that talks to it over a local socket. Because the shells live in the daemon, quitting and reopening the app leaves the session intact — detach and reattach, no tmux required. - Persistent sessions — the daemon holds the PTYs and child processes, so closing a window or swapping in a new build never takes a shell down. - Performance — an 11 MB `cat` completes in 95 ms and DOOM-fire renders at 888 fps; the daemon drains the PTY at device speed off the render path. - Shell-aware — new tabs and splits open in the current working directory; zsh, bash, fish, and PowerShell are set up automatically. - Enhanced prompt — inline completion, syntax highlighting, history, and in-terminal search, with rich flag/subcommand signatures for common tools. - Tabs, resizable splits, a command palette, click-to-open links, desktop notifications, eight themes, and CJK/IME input. Native builds for macOS, Windows, and Linux. Built on Zed's gpui and Alacritty's VT core.
fig-convert — Fig autocomplete specs → tty7 completion signatures
tty7's per-command completion (flags/subcommands/args with descriptions) is
driven by signature data generated from Fig's autocomplete spec corpus
(MIT-licensed, community-maintained, hundreds of commands). This is the
build-time converter; the runtime consumer is src/terminal/signature.rs.
Why a converter (and not the JS engine)
Fig specs are authored in TypeScript and shipped pre-compiled as ESM on npm.
Each spec export defaults an object whose static shape (subcommands, options,
args, descriptions, static generator scripts) is exactly what a completion
menu needs. The only truly dynamic parts are functions — generator postProcess,
custom, filterTerm, generateSpec — which need a JS runtime to evaluate.
We convert ahead of time: run each spec once, snapshot the
static shape to JSON, and drop the functions. No JS engine at runtime. (An
embedded QuickJS evaluating the functions live would be the alternative; we don't need it.) Dynamic
value completion (git checkout <branch>) is recovered later by running a
generator's static script and splitting stdout on newlines — the scripts are
already captured; only the executor is future work.
Regenerate
# 1. Fetch the compiled spec corpus once (into this dir, gitignored).
npm pack @withfig/autocomplete
tar xzf withfig-autocomplete-*.tgz # -> package/build/<cmd>.js
# 2. Convert the commands tty7 embeds into assets/completions/.
node convert.mjs --build-dir package/build --out ../../assets/completions git docker
convert.mjs executes each <cmd>.js, walks the object graph keeping only
static fields, resolves loadSpec references (e.g. docker compose grafts in
the docker-compose spec), and writes minified assets/completions/<cmd>.json.
Each command is converted in isolation: one spec that throws is reported (✗)
and skipped, never aborting the batch — of Fig's ~716 specs, 715 convert and
only a non-command index barrel is skipped. Pass as many command names as you
like in one run.
Output schema
Mirrors the serde model in src/terminal/signature.rs one-to-one:
- root:
{ name, description, options[], args[], subcommands[] } - subcommand:
{ names[], description, hidden, options[], args[], subcommands[] } - option:
{ names[], description, args[], required, repeatable, hidden } - arg:
{ name, description, optional, variadic, template[], suggestions[], generators[] } - suggestion:
{ names[], description } - generator:
{ script[] }(static shell command only;postProcessdropped)
Scope & rollout
Signatures are read from disk, not embedded: signature::spec_source resolves
a completions/ directory (inside the macOS .app bundle's Resources, beside
the executable on Linux/Windows, or the in-tree assets/completions for
cargo run/tests) and lazily loads <cmd>.json by command name. So adding a
command is just dropping its JSON into assets/completions/ — the packaging
scripts copy the whole directory into each bundle, and no recompile or binary
bloat is involved. A big spec is ~300–500 KiB; the full corpus is ~700 commands,
so ship whatever subset you want rather than all of it if bundle size matters.
The in-tree set is ~95 everyday commands (version control, container/k8s,
language package managers, cloud CLIs, shell/sysadmin tools) — ~5.5 MB, curated
from the corpus. The two outliers aws (~51 MB) and gcloud (~18 MB) are
deliberately excluded; add them only if you also trim them. To add a command,
run convert.mjs with its name (step 2 above) and commit the resulting JSON.