rux vocab

Print the elements, attributes, directives and honored CSS the runtime understands, as JSON, for editors.

rux vocab

JSON on stdout: elements, their attributes, the directives, the script globals, the tags that never nest, and every CSS property the runtime honors.

It exists for editors. The VS Code extension offers its completions from this.

Why it is a command and not a document

The lists that a crate already owns are read from that crate rather than copied into the output:

  • The CSS properties are the same slice the unhonored-property warning consults.
  • The tags that never nest are the same list the formatter indents by.

So the guarantee an editor can make is that if it offers something, it works. A completion list built from a second, hand-kept copy would eventually suggest a property the runtime then warns does nothing, which is worse than offering no completions at all.

It also means a property honored in a later release reaches completions without anyone remembering to update a second list.

The drift this prevents already happened

The extension used to carry its own copy of the tags that never nest. It inherited HTML's set, which has img and not Rux's <image>, and over-indented everything after an <image src="…"> for two releases before anyone noticed.

The extension ships a generated copy of this output so completions work before cargo install ruxlang has finished, and prefers whatever the installed binary says when there is one. A release gate regenerates the committed copy and fails if it differs.

Shape

{
  "version": "0.7.0-dev",
  "elements":         [{ "name": "view", "detail": "…", "doc": "…" }],
  "globalAttributes": [{ "name": "class", "detail": "…", "doc": "…" }],
  "directives":       [{ "name": "r-for", "detail": "…", "doc": "…" }],
  "elementAttributes": { "image": [{ "name": "src", "detail": "…", "doc": "…" }] },
  "scriptGlobals":    [{ "name": "signal", "detail": "…", "doc": "…" }],
  "voidTags":         ["image", "input", "…"],
  "cssProperties":    ["display", "width", "…"]
}

The element and attribute tables are declared by the command itself, because no single crate owns them today: tags are strings all the way through the parser. A test pins them against the reference, so a tag offered as a completion has to be one the documentation describes.