neuwirth
/
pmacs
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Merge pull request #146 from levineuwirth/web-grammars 

feat(web): HTML + CSS grammars + HTML script/style injections
This commit is contained in:
Levi Neuwirth 2026-07-23 20:23:33 +00:00 committed by GitHub
parent c49a8c71be 515c76f5ab
commit 47581f4034
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
5 changed files with 672 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
Cargo.lock generated
View File

@ -2566,8 +2566,10 @@ dependencies = [
"tree-sitter-cmake",
"tree-sitter-containerfile",
"tree-sitter-cpp",
"tree-sitter-css",
"tree-sitter-cuda",
"tree-sitter-go",
"tree-sitter-html",
"tree-sitter-javascript",
"tree-sitter-json",
"tree-sitter-lua",
@ -3794,6 +3796,16 @@ dependencies = [
"tree-sitter-language",
]
[[package]]
name = "tree-sitter-css"
version = "0.25.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "a5cbc5e18f29a2c6d6435891f42569525cf95435a3e01c2f1947abcde178686f"
dependencies = [
"cc",
"tree-sitter-language",
]
[[package]]
name = "tree-sitter-cuda"
version = "0.21.1"
@ -3814,6 +3826,16 @@ dependencies = [
"tree-sitter-language",
]
[[package]]
name = "tree-sitter-html"
version = "0.23.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "261b708e5d92061ede329babaaa427b819329a9d427a1d710abb0f67bbef63ee"
dependencies = [
"cc",
"tree-sitter-language",
]
[[package]]
name = "tree-sitter-javascript"
version = "0.25.0"

8
Cargo.toml
View File

@ -233,6 +233,14 @@ tree-sitter-md = "0.5"
# uses the in-repo overlay `builtin/queries/latex/highlights.scm` (the first
# such overlay; `include_str!`'d in `crate::syntax::BUILTIN_LANGUAGES`).
codebook-tree-sitter-latex = "0.6"
# Web grammars: HTML + CSS (`.html`/`.htm`/`.xhtml`, `.css`). The official
# tree-sitter-org grammars; both export `LANGUAGE` + query constants over
# `tree-sitter-language 0.1` (shared ABI crate, `tree-sitter` dev-only), so no
# overlay is needed. HTML's `INJECTIONS_QUERY` lights up `<script>` ->
# javascript (already registered) and `<style>` -> css via the #122 injection
# engine (see `crate::syntax::BUILTIN_LANGUAGES`).
tree-sitter-html = "0.23"
tree-sitter-css = "0.25"
# T M4.4 process supervisor: signal sending without `unsafe`. Keep
# the feature surface tight to keep build time low. `poll` feeds the
# compile-mode group readers (cancellable poll-based reads, Q#CM3).

319
docs/web-grammars-html-css-framing.md Normal file
View File

@ -0,0 +1,319 @@
# Web grammars (HTML + CSS) + HTML injections — framing
**Revision 4 — implemented on branch `web-grammars` (PR #146). Ground truth:
canonical `main` @ `4daa1b8` (after LaTeX #144 and inline-math-docs #145),
2026-07-23.** Rev 2 settled the capture set, injection scope, and LSP claim
(round 1); rev 3 corrected the `#match?` predicate analysis (round 2); rev 4
names the intended `@attribute` retro-paint on the already-bundled rust/lua/yaml
grammars — verified on a Rust buffer and pinned by a test (round 3). See §0.1.
## 0.1 Revision history
### Round 1 (rev 1 → rev 2)
- **F1 (high).** Q#WEB4 deferred the capture mapping. The upstream queries
settle it: HTML and CSS need exactly **`tag`** and **`attribute`**; `@tag.error`
prefix-walks to `tag`; there is **no** `@tag.delimiter` (rev-1 speculation).
Q#WEB4 now names both mappings and their styles, and acceptance 6 paints an
HTML **attribute** explicitly (a tag/property-only test could pass with
`attribute` unverified since `property` is already recognized).
- **F2 (medium).** The HTML injections query has **exactly two** patterns —
`script_element`→javascript, `style_element`→css. Q#WEB5 no longer claims
event-handler / `style=` coverage; inline-attribute injection is now a named
deferral.
- **F3 (medium).** No HTML/CSS language server is bundled: `lsp.lua`'s default
config list ends at YAML (`:247`) and `:521` returns `nil` without a
configured command. §2 and §5 corrected — the grammars give stable language
IDs so a *user-supplied* config attaches automatically, but nothing ships or
starts by default.
### Round 2 (rev 2 → rev 3)
- **R2-1 (medium).** The rev-2 claim that pmacs *ignores* `#match?` was wrong.
`src/syntax.rs:1701` passes the source to `QueryCursor::captures`, so
tree-sitter evaluates the standard `#match?`/`#eq?`/`#any-of?` text predicates
natively (pmacs special-cases only `#is? local`, `:1703`). Corrected in §2 and
Q#WEB4: an ordinary CSS property has a single `@property` capture; only a
`--custom-property` also receives `@variable` (a benign double-capture).
Acceptance 6 no longer claims an ordinary property pins custom-property
precedence — it just verifies `color` → `@property` paints.
- **R2-2 (low).** The folding footprint in §0 was stale. At
`githubsucks/folding` @ `036a994` the branch touches `src/fold.rs`,
`editor.rs`, `editor_core.rs`, `lib.rs`, `lua_bindings/{fold.rs,mod.rs}`,
`semantic_render.rs`, `fold.lua`, and tests — not
`overlay.rs`/`daemon.rs`/`syntax.rs`. §0 now lists the real set; the
zero-overlap conclusion is unchanged (stronger, if anything).
### Round 3 (rev 3 → rev 4)
- **R3-1 (medium).** Q#WEB4 verified which captures pmacs *lacks* but not the
reverse: which already-bundled grammars *use* `@attribute`. Three do — rust
(`attribute_item`), lua (`<const>`), yaml (directives) — so the new
`("attribute", fg(3))` entry retro-paints their previously-unpainted spans
yellow. Named in Q#WEB4 as intended (verified on a Rust buffer: uniform
yellow, an improvement over unpainted), and pinned by
`rust_attribute_repaints_via_shared_attribute_capture`. `tag` is unaffected
(HTML/CSS only).
Add tree-sitter **HTML** and **CSS** grammars so `.html`/`.htm` and `.css`
buffers get lexical highlighting, and — the north-star payoff — light up
HTML's embedded-language **injections**: `<script>` → JavaScript (already
registered) and `<style>` → CSS (added here). This is the side-quest
backlog's "HTML/CSS grammars that light up more injection consumers."
## 0. Why this lane, why now (parallel-safety)
Board at this snapshot: LaTeX #144 and inline-math-docs #145 merged;
**folding Stage 1 (#142) is the only open PR** — a headless fold engine. Its
current diff (`main...githubsucks/folding` @ `036a994`) touches `src/fold.rs`
(new), `src/editor.rs`, `src/editor_core.rs`, `src/lib.rs`,
`src/lua_bindings/{fold.rs,mod.rs}`, `src/semantic_render.rs`,
`builtin/runtime/fold.lua`, and tests — **none of which this lane touches**.
gpu-invocation has landed. The GPU render path (`pmacs-gpu/src/main.rs`) remains
contended by folding's later Stage 3 — this lane never touches it.
This lane's footprint: two `Cargo.toml` grammar deps, two `BUILTIN_LANGUAGES`
entries, and — per Q#WEB4 — a small extension of `src/highlight.rs`'s recognized
capture set. **Zero file overlap with folding**; the three files it edits:
- `src/syntax.rs` — a localized append to the language table (after the `latex`
entry, currently ~`:1111`). Folding does not touch this file.
- `Cargo.toml` — two dependency lines in the grammar block (`:142`–`:220`).
- `src/highlight.rs` — appends to the capture-style `entries` table
(`:143-172`). Folding does not touch this file.
All are localized, low-conflict edits — the same pattern #144 and the
config-registry ∥ Vterm lanes landed conflict-free.
Unlike LaTeX (#144), **no in-repo query overlay is needed** — both crates
export their queries as constants. The overlay convention #144 established
stays available for future crates that don't (e.g. some of ruby/php).
## 1. What ships
- **HTML grammar** (`.html`/`.htm`/`.xhtml`) — tags, attributes, text,
comments, doctype.
- **CSS grammar** (`.css`) — selectors, properties, values, at-rules,
comments.
- **HTML injections** — `<script>…</script>` parses as JavaScript and
`<style>…</style>` parses as CSS, via HTML's crate-exported injections query
riding the #122 injection engine. CSS is registered here so the `<style>`
injection resolves.
## 2. Ground truth (scouted 2026-07-23, `main` @ `4daa1b8`)
### Crate facts (verified)
- **`tree-sitter-html` 0.23.2** (2026-06). Exports `LANGUAGE: LanguageFn`,
**`HIGHLIGHTS_QUERY`**, **`INJECTIONS_QUERY`**, `NODE_TYPES`. Deps:
`tree-sitter-language ^0.1` (the shim every pmacs grammar uses),
`tree-sitter ^0.24` **dev-only** → ABI-compatible with our `tree-sitter 0.26`
via `.into()`. Its `INJECTIONS_QUERY` is the load-bearing piece: it names
`javascript` and `css` via `#set! injection.language`.
- **`tree-sitter-css` 0.25.0** (2026-05). Exports `LANGUAGE: LanguageFn`,
**`HIGHLIGHTS_QUERY`**, `NODE_TYPES`. Deps: `tree-sitter-language ^0.1`,
`tree-sitter ^0.25` **dev-only** → shim-ABI-fine. No injections query (CSS
injects nothing).
Both export their highlights query as a constant, so **no overlay/vendoring**
(the LaTeX complication) applies.
### Codebase
- **`BUILTIN_LANGUAGES`** (`src/syntax.rs:816`…`];`) currently ends at the
`latex` entry (~`:1104-1111`, added by #144). No `html`/`css` entry. Append
the two new entries before `];`.
- **`javascript` is registered** (`src/syntax.rs:1010`, `extensions: js/mjs/cjs`,
`injections_query: &[]`), so HTML's `<script>` → `javascript` resolves today.
`css` does not exist → HTML's `<style>` → `css` resolves only once this lane
adds it.
- **Injection engine** — `collect_injection_matches` (`src/syntax.rs:375-427`)
reads `@injection.content` (`:381`) + `injection.language` (dynamic node text
or `#set!`, `:387-402`) + `injection.include-children` (`:402`).
`resolve_injected_language` (`:304`) resolves the name against
`BUILTIN_LANGUAGES` + `default_injection_aliases` (`:234`). Working
precedents: rust `INJECTIONS_QUERY` (`:823`), markdown block/inline
(`:848`/`:862`). HTML's `#set! injection.language "javascript"|"css"` is the
same shape.
- **Recognized capture set** — `src/highlight.rs:143-172`, resolved by a
dotted-prefix walk (`:189`). Confirmed against the two upstream queries
(HTML v0.23.2 `highlights.scm`, CSS v0.25.0 `highlights.scm`): the **only**
captures not already recognized are **`@tag`** and **`@attribute`** — used by
BOTH grammars (`@tag.error` prefix-walks to `tag`; there is no `@tag.delimiter`).
Everything else maps: HTML's `@constant` (doctype), `@string` (attribute
value), `@comment`, `@punctuation.bracket`; CSS's `@operator`/`@property`/
`@function`/`@keyword`/`@number`/`@type`/`@string.special`/`@punctuation.*`.
CSS also has two `#match?`-gated `@variable` patterns for `--custom-props`.
pmacs passes the buffer text to `QueryCursor::captures` (`src/syntax.rs:1701`),
so tree-sitter **evaluates** the standard `#match?`/`#eq?`/`#any-of?` text
predicates natively; pmacs adds handling only for the `#is? local` property
predicate (the `property_predicates … "local"` filter at `:1703`). So an
ordinary property (`color`) has a single `@property` capture, and only a custom
property (`--brand`) additionally receives `@variable` — see Q#WEB4.
- **Detection** — the `extensions` field drives `language_name_for_path`
(`:1223`) ahead of the LSP filetype map in the Lua chain
(`builtin/runtime/syntax.lua:452-466`); no Lua edit. **No HTML/CSS language
server is bundled**: `builtin/runtime/lsp.lua`'s default config list ends at
YAML (`:247`), and `:521` returns `nil` without a configured command. The
grammars provide stable language IDs (`html`/`css`), so a **user-supplied**
`pmacs.lsp.config.html`/`.css` attaches automatically, but nothing ships or
starts by default.
## 3. Decisions
### Q#WEB1 — Bundle `tree-sitter-html` 0.23 + `tree-sitter-css` 0.25
Add both to the `Cargo.toml` grammar block; loaders
`|| tree_sitter_html::LANGUAGE.into()` and `|| tree_sitter_css::LANGUAGE.into()`.
ABI is fine via the shared `tree-sitter-language` shim, as every current
grammar. No provenance saga (these are the official tree-sitter-org grammars,
not squatted republishes).
### Q#WEB2 — Register both; CSS before HTML's injection can resolve
- `html`: `highlights_query: &[tree_sitter_html::HIGHLIGHTS_QUERY]`,
`injections_query: &[tree_sitter_html::INJECTIONS_QUERY]`, `locals_query: &[]`.
- `css`: `highlights_query: &[tree_sitter_css::HIGHLIGHTS_QUERY]`, injections and
locals empty.
Both live in the same `BUILTIN_LANGUAGES`, so `resolve_injected_language`
finds `css` (and the existing `javascript`) when HTML's injections query fires.
### Q#WEB3 — Extensions
`html`: `["html", "htm", "xhtml"]`. `css`: `["css"]`. SCSS/LESS/Sass are
distinct grammars (`scss`/`less` node sets) and are deferred (§5).
### Q#WEB4 — Add exactly two capture entries: `tag` and `attribute`
The LaTeX lane could rename captures because it owned an editable overlay.
Here the highlights queries are **crate constants** — not editable — so the
reconciliation is to **extend `src/highlight.rs`'s `entries` table**
(`:143-172`). The upstream queries settle the exact set: the **only** captures
neither grammar's query already resolves are `tag` and `attribute`. Add exactly
two entries:
- `("tag", fg(5))` — HTML `(tag_name)` and CSS element/nesting/universal
selectors, in the keyword hue (magenta) but non-bold to stay light in dense
markup. `@tag.error` (HTML erroneous end tags) prefix-walks to this entry, so
it needs no separate mapping.
- `("attribute", fg(3))` — HTML `(attribute_name)` and CSS
pseudo-/attribute-selector names, in the type hue (yellow) — distinct from
`tag`, from `property`/`operator` (cyan), and from `string` values (green).
There is **no** `tag.delimiter` (rev-1 speculation, removed); HTML's `<`/`>`/
`</`/`/>` are `@punctuation.bracket`, already handled. This extension is:
- **low-conflict** — folding does not touch `highlight.rs`;
- **general** — `tag`/`attribute` are standard tree-sitter web captures, so it
also serves future html-ish grammars (vue/svelte/astro).
**Retro-paint on already-bundled languages (intended, added rev 4).** The
capture table is global, so adding `attribute` also colours the `@attribute`
capture that three bundled grammars already emit but which was previously
unrecognized (and so unpainted): **rust** (`attribute_item`/`inner_attribute_item`
— every `#[derive(…)]` / `#![…]`), **lua** (the `<const>`/`<close>` variable
attribute), and **yaml** (`%YAML`/`%TAG` directives). On merge these begin
painting `attribute` yellow (`fg 3`). Verified on a Rust buffer: `#[derive(Debug)]`
paints uniformly yellow (the `attribute_item` span carries no narrower overriding
captures) — a distinct-attribute convention most editors follow, and an
improvement over unpainted. This is **chosen, not incidental**: it is pinned by
`rust_attribute_repaints_via_shared_attribute_capture` (`src/highlight.rs`).
`tag` is clean — only HTML/CSS use it, so it has no retro-effect.
**On CSS custom properties (corrected in rev 3):** pmacs passes the buffer text
to `QueryCursor::captures` (`src/syntax.rs:1701`), so tree-sitter evaluates the
standard `#match?`/`#eq?`/`#any-of?` predicates natively — pmacs special-cases
only `#is? local`. So an ordinary property (`color`) matches only the
unconditional `(property_name) @property` and paints cleanly; a custom property
(`--brand`) additionally satisfies `#match? "^--"` and also receives `@variable`
— a benign double-capture whose winner is a within-layer precedence detail, out
of scope for v0 and not relied upon. Rejected: a shadow overlay re-capturing the
same nodes with recognized names (fragile, precedence-dependent, duplicative).
### Q#WEB5 — Injection scope: script + style only
HTML's `INJECTIONS_QUERY` (v0.23.2) has **exactly two** patterns:
`(script_element (raw_text) @injection.content) (#set! injection.language "javascript")`
and the same for `(style_element …)` → `"css"`. There is **no** event-handler
(`onclick=…`) or inline `style=` attribute injection — those are a named
deferral (§5). Each element is its own subtree, so `injection.combined` (many
matches → one shared parse, the PHP-in-HTML case) is not needed and stays
deferred. Acceptance pins script + style.
### Q#WEB6 — No protocol, frontend, or GPU change
Pure instance-side: two grammar entries, one injections query, a capture-table
extension. No wire type, no TUI/GPU edit.
## 4. Categorical bets
1. **The shim makes ABI a non-issue.** Both crates ride `tree-sitter-language
0.1`, exactly like every bundled grammar.
2. **HTML's value is the injection, not the tags.** Highlighting a web page
*with* its embedded JS and CSS is the north-star injection consumer;
registering CSS is the prerequisite, which is why the two grammars ship
together as one lane.
3. **Extending the core capture table is the correct reconciliation for
crate-exported queries.** You cannot rename a `const` query's captures;
teaching the highlighter the standard web captures is the general fix and
costs one localized, uncontended edit.
## 5. Deferred (named)
- **SCSS / LESS / Sass** grammars (distinct node sets; own extensions).
- **`injection.combined`** (PHP-in-HTML and other many→one-parse schemes) — the
existing side-quest deferral; not needed for script/style.
- **CSS-in-JS / HTML-in-JS** via `tree-sitter-javascript`'s `INJECTIONS_QUERY`
(the `javascript` entry is `injections_query: &[]` today) — a clean
follow-up: add the JS injections query so tagged template literals
(`` css`…` ``, `` html`…` ``) parse. Separate consumer, separate PR.
- **Inline HTML attribute injection** — event handlers (`onclick=…`) and
`style=` attributes are not in the upstream injections query (Q#WEB5); adding
them would need a pmacs-side injections overlay.
- **Vue / Svelte / Astro** single-file-component grammars.
- **Bundled HTML/CSS LSP configs** — no server ships today (§2). Shipping
default `pmacs.lsp.config.html`/`.css` entries (vscode-langservers-extracted)
is a separate follow-up; until then the grammar's stable language ID lets a
user config attach automatically.
## 6. Acceptance
Mirrors the #144 / CUDA conventions (`src/syntax.rs` table guards + smokes;
`src/highlight.rs` paint template `grid_paints_injected_child_keyword`).
1. **Table guards** `builtin_languages_include_html` / `_css`: entries exist,
claim their extensions, carry non-empty highlights; `html` also carries
`INJECTIONS_QUERY`.
2. **Load-and-parse smokes**: a minimal HTML document and a CSS rule parse with
the expected root node kind and `!has_error()` (exact kinds — likely
`document` / `stylesheet` — confirmed at implementation).
3. **Highlights resolve**: `reg.highlights_query("html")` / `("css")` compile
against their grammars (the node-name compatibility gate).
4. **Extension resolution**: `.html`/`.htm`/`.xhtml` → `html`; `.css` → `css`.
5. **Injection (the payoff)**: an HTML buffer with
`<style>a{color:red}</style>` and `<script>let x=1</script>` produces a
`ParseTreeBundle` whose child layers resolve to `css` and `javascript`, and
a grid-paint asserts an injected CSS property and JS keyword paint **inside**
the embedded regions (mirroring `grid_paints_injected_child_keyword`).
6. **Paint (non-vacuous highlighting)**: parse `<a href="x">` and assert **both**
the `<a>` tag (`@tag`) **and** the `href` attribute name (`@attribute`) paint
the two new non-default styles — the attribute assertion is load-bearing, since
a tag-only test could pass with `@attribute` unverified. Also assert a CSS
selector (`@tag`) paints the new `tag` style and an ordinary CSS property
(`color` → `@property`) paints non-default (an ordinary property has a single
unconditional capture, so no precedence subtlety — Q#WEB4).
7. **Full gate suite** per `CLAUDE.md`.
## 7. Prior art in pmacs
- **LaTeX lane #144** (`docs/latex-grammar-math-substrate-framing.md`) — the
grammar-add mechanics, the table-guard/smoke/paint test conventions, and the
compile-gate-plus-paint-test discipline. (Its overlay convention is not
needed here.)
- **Multi-language injections #122**
(`docs/multi-language-injections-framing.md`) — the `ParseTreeBundle` +
`Layer` engine HTML's injections ride.
- **markdown → rust fenced code** (`src/syntax.rs:848`, tested by
`src/highlight.rs`'s `grid_paints_injected_child_keyword`) — the working
injection precedent this lane's acceptance 5 mirrors.

180
src/highlight.rs
View File

@ -169,6 +169,11 @@ impl Theme {
("decorator", fg(13)),
("regexp", fg(2)),
("typeParameter", fg_italic(11)),
// Web grammars (HTML/CSS, framing Q#WEB4): the only captures their
// crate-exported queries use that the set above lacks. `@tag.error`
// prefix-walks to `tag`.
("tag", fg(5)),
("attribute", fg(3)),
];
let by_capture = entries
.iter()
@ -1410,4 +1415,179 @@ mod tests {
"the \\section title text is painted @keyword.control (bold)"
);
}
#[test]
fn rust_attribute_repaints_via_shared_attribute_capture() {
// Intended side effect, named in the framing (Q#WEB4): the
// `("attribute", fg(3))` entry added for HTML/CSS also colours the
// `@attribute` capture that tree-sitter-rust (attribute_item), -lua
// (`<const>`), and -yaml (directives) already emit. A Rust
// `#[derive(Debug)]` — previously unpainted, since `@attribute` was
// unrecognized — now paints the attribute style throughout. Pinned so
// the retro-paint on this repo's primary language is a chosen effect,
// not an incidental one.
use crate::buffer::{Buffer, BufferId, EditOp};
use crate::cell::{Cell, CellSize};
use crate::syntax::{ParseView, SyntaxRegistry};
let reg = SyntaxRegistry::new();
let language = reg.language("rust").expect("rust grammar");
let src = b"#[derive(Debug)]\nstruct S;\n";
let mut buf = Buffer::new(BufferId::next(), "a.rs");
buf.apply_edit(EditOp::Insert { pos: 0, bytes: src })
.unwrap();
let view = ParseView::new(&buf, language, "rust".to_owned());
let handle = view.handle();
let _vid = buf.attach_view(Box::new(view));
let mut req = handle.make_request();
req.injection_aliases = reg.injection_alias_snapshot();
let bundle = crate::syntax::run_parse(req).expect("rust parse");
handle.install(reg.resolve_layer_queries(&bundle));
let mut hv = SyntaxHighlightView::new(handle, reg.theme());
let (rows, cols) = (1usize, 20usize);
let mut backing: Vec<Cell> = vec![Cell::default(); rows * cols];
let mut grid = CellGrid {
cells: &mut backing,
stride: cols as u32,
size: CellSize::new(rows as u32, cols as u32),
};
let viewport = Viewport {
buffer_start: 0,
buffer_end: u64::MAX,
cell_origin: CellCoord::new(0, 0),
cell_size: CellSize::new(rows as u32, cols as u32),
gutter_w: 0,
};
let registry = buf;
hv.render(&registry, viewport, &mut grid);
// `derive` (col 2) sits inside the `attribute_item` and paints the
// shared @attribute style (fg 3) — the intended retro-paint.
assert_eq!(
grid.get(CellCoord::new(0, 2)).style.fg,
pmacs_protocol::cell::Color::Indexed(3),
"a Rust #[derive] attribute paints the shared @attribute style (fg 3)"
);
}
#[test]
fn web_grid_paints_html_tag_and_attribute() {
// Q#WEB4 acceptance: the two capture entries this lane adds (`tag`,
// `attribute`) actually reach painted cells. The attribute assertion is
// load-bearing — a tag-only test could pass with `@attribute` unverified.
use crate::buffer::{Buffer, BufferId, EditOp};
use crate::cell::{Cell, CellSize};
use crate::syntax::{ParseView, SyntaxRegistry};
let reg = SyntaxRegistry::new();
let language = reg.language("html").expect("html grammar");
// Line 0: <a href="x">Hi</a> — `a` (tag_name) at col 1, `href`
// (attribute_name) at col 3.
let src = b"<a href=\"x\">Hi</a>\n";
let mut buf = Buffer::new(BufferId::next(), "index.html");
buf.apply_edit(EditOp::Insert { pos: 0, bytes: src })
.unwrap();
let view = ParseView::new(&buf, language, "html".to_owned());
let handle = view.handle();
let _vid = buf.attach_view(Box::new(view));
let mut req = handle.make_request();
req.injection_aliases = reg.injection_alias_snapshot();
let bundle = crate::syntax::run_parse(req).expect("html parse");
handle.install(reg.resolve_layer_queries(&bundle));
let mut hv = SyntaxHighlightView::new(handle, reg.theme());
let (rows, cols) = (1usize, 40usize);
let mut backing: Vec<Cell> = vec![Cell::default(); rows * cols];
let mut grid = CellGrid {
cells: &mut backing,
stride: cols as u32,
size: CellSize::new(rows as u32, cols as u32),
};
let viewport = Viewport {
buffer_start: 0,
buffer_end: u64::MAX,
cell_origin: CellCoord::new(0, 0),
cell_size: CellSize::new(rows as u32, cols as u32),
gutter_w: 0,
};
let registry = buf;
hv.render(&registry, viewport, &mut grid);
let tag_cell = grid.get(CellCoord::new(0, 1));
let attr_cell = grid.get(CellCoord::new(0, 3));
assert_ne!(
tag_cell.style,
Cell::default().style,
"the <a> tag_name paints @tag (non-default)"
);
assert_ne!(
attr_cell.style,
Cell::default().style,
"the href attribute_name paints @attribute (non-default)"
);
assert_ne!(
tag_cell.style, attr_cell.style,
"@tag and @attribute use the two distinct new styles"
);
}
#[test]
fn html_injects_css_and_js() {
// The payoff (Q#WEB acceptance 5): HTML's INJECTIONS_QUERY parses
// <style> as CSS and <script> as JavaScript, and the child layers paint
// INSIDE the embedded regions — styling only the injected grammars can
// produce. `css` is resolvable only because this lane registered it.
use crate::buffer::{Buffer, BufferId, EditOp};
use crate::cell::{Cell, CellSize};
use crate::syntax::{ParseView, SyntaxRegistry};
let reg = SyntaxRegistry::new();
let language = reg.language("html").expect("html grammar");
// Line 0: <style>a{color:red}</style> — `color` (CSS property) at col 9.
// Line 1: <script>let x=1</script> — `let` (JS keyword) at col 8.
let src = b"<style>a{color:red}</style>\n<script>let x=1</script>\n";
let mut buf = Buffer::new(BufferId::next(), "page.html");
buf.apply_edit(EditOp::Insert { pos: 0, bytes: src })
.unwrap();
let view = ParseView::new(&buf, language, "html".to_owned());
let handle = view.handle();
let _vid = buf.attach_view(Box::new(view));
let mut req = handle.make_request();
req.injection_aliases = reg.injection_alias_snapshot();
let bundle = crate::syntax::run_parse(req).expect("html parse");
handle.install(reg.resolve_layer_queries(&bundle));
let mut hv = SyntaxHighlightView::new(handle, reg.theme());
let (rows, cols) = (2usize, 60usize);
let mut backing: Vec<Cell> = vec![Cell::default(); rows * cols];
let mut grid = CellGrid {
cells: &mut backing,
stride: cols as u32,
size: CellSize::new(rows as u32, cols as u32),
};
let viewport = Viewport {
buffer_start: 0,
buffer_end: u64::MAX,
cell_origin: CellCoord::new(0, 0),
cell_size: CellSize::new(rows as u32, cols as u32),
gutter_w: 0,
};
let registry = buf;
hv.render(&registry, viewport, &mut grid);
// Inside <style>: the CSS `color` property paints (non-default) —
// proves the <style> -> css injection resolved and parsed.
assert_ne!(
grid.get(CellCoord::new(0, 9)).style,
Cell::default().style,
"the CSS `color` property paints inside the <style> injection"
);
// Inside <script>: the JS `let` keyword paints bold — proves the
// <script> -> javascript injection resolved and parsed.
assert!(
grid.get(CellCoord::new(1, 8)).style.bold,
"the JS `let` keyword paints (bold) inside the <script> injection"
);
}
}

143
src/syntax.rs
View File

@ -1108,6 +1108,28 @@ pub const BUILTIN_LANGUAGES: &[LanguageEntry] = &[
locals_query: &[],
injections_query: &[],
},
// HTML + CSS (framing `docs/web-grammars-html-css-framing.md`). Both crates
// export their query constants (no overlay). HTML's `INJECTIONS_QUERY`
// wires `<script>` -> javascript (already registered) and `<style>` -> css
// (below), riding the #122 injection engine; `css` must be registered here
// for that injection to resolve. The `tag`/`attribute` captures these
// queries use are taught to the highlighter in `crate::highlight` (Q#WEB4).
LanguageEntry {
name: "html",
extensions: &["html", "htm", "xhtml"],
loader: || tree_sitter_html::LANGUAGE.into(),
highlights_query: &[tree_sitter_html::HIGHLIGHTS_QUERY],
locals_query: &[],
injections_query: &[tree_sitter_html::INJECTIONS_QUERY],
},
LanguageEntry {
name: "css",
extensions: &["css"],
loader: || tree_sitter_css::LANGUAGE.into(),
highlights_query: &[tree_sitter_css::HIGHLIGHTS_QUERY],
locals_query: &[],
injections_query: &[],
},
];
/// LaTeX highlights overlay (framing Q#LX2). The chosen grammar crate
@ -2372,6 +2394,127 @@ mod tests {
}
}
#[test]
fn builtin_languages_include_html_and_css() {
// Both crate grammars export their query constants (no overlay). HTML
// additionally carries an injections query (script/style); CSS does not.
let html = BUILTIN_LANGUAGES
.iter()
.find(|l| l.name == "html")
.expect("`html` language entry must be present");
for ext in ["html", "htm", "xhtml"] {
assert!(html.extensions.contains(&ext), "`html` claims `.{ext}`");
}
assert!(
!html.highlights_query.is_empty(),
"`html` ships a highlights query"
);
assert!(
!html.injections_query.is_empty(),
"`html` ships an injections query (script/style)"
);
let css = BUILTIN_LANGUAGES
.iter()
.find(|l| l.name == "css")
.expect("`css` language entry must be present");
assert!(css.extensions.contains(&"css"), "`css` claims `.css`");
assert!(
!css.highlights_query.is_empty(),
"`css` ships a highlights query"
);
}
#[test]
fn html_grammar_loads_and_parses() {
// ABI acceptance: `tree-sitter-html` (LanguageFn over
// `tree-sitter-language 0.1`) is accepted by our `tree-sitter` 0.26 core.
let reg = SyntaxRegistry::new();
let language = reg
.language("html")
.expect("`html` language loads from BUILTIN_LANGUAGES");
let mut buf = fresh_buffer("index.html");
buf.apply_edit(EditOp::Insert {
pos: 0,
bytes: b"<!DOCTYPE html>\n<html><body><a href=\"x\">Hi</a></body></html>\n",
})
.unwrap();
let view = ParseView::new(&buf, language, "html".to_owned());
let handle = view.handle();
let _vid = buf.attach_view(Box::new(view));
let bundle = parse_synchronously(&handle);
assert_eq!(
bundle.root_tree().root_node().kind(),
"document",
"HTML grammar roots at document"
);
assert!(
!bundle.root_tree().root_node().has_error(),
"HTML grammar parses a document without error"
);
}
#[test]
fn css_grammar_loads_and_parses() {
let reg = SyntaxRegistry::new();
let language = reg
.language("css")
.expect("`css` language loads from BUILTIN_LANGUAGES");
let mut buf = fresh_buffer("style.css");
buf.apply_edit(EditOp::Insert {
pos: 0,
bytes: b"a { color: red; }\n",
})
.unwrap();
let view = ParseView::new(&buf, language, "css".to_owned());
let handle = view.handle();
let _vid = buf.attach_view(Box::new(view));
let bundle = parse_synchronously(&handle);
assert_eq!(
bundle.root_tree().root_node().kind(),
"stylesheet",
"CSS grammar roots at stylesheet"
);
assert!(
!bundle.root_tree().root_node().has_error(),
"CSS grammar parses a rule without error"
);
}
#[test]
fn html_and_css_highlights_resolve() {
// The crate-exported queries compile against their grammars (node-name
// compatibility gate), and both use the `@tag` capture this lane teaches
// the highlighter (Q#WEB4).
let reg = SyntaxRegistry::new();
for lang in ["html", "css"] {
let query = reg
.highlights_query(lang)
.unwrap_or_else(|| panic!("{lang} highlights compile against the grammar"));
let names = query.capture_names();
assert!(
names.contains(&"tag"),
"{lang} highlights use the @tag capture; got {names:?}"
);
}
}
#[test]
fn language_for_path_resolves_web_extensions() {
let reg = SyntaxRegistry::new();
for (path, lang) in [
("index.html", "html"),
("page.htm", "html"),
("doc.xhtml", "html"),
("style.css", "css"),
] {
assert_eq!(
reg.language_name_for_path(path).as_deref(),
Some(lang),
"{path} resolves to {lang}"
);
}
}
#[test]
fn builtin_languages_include_bash() {
// Regression guard: the bash entry claims the wider shell family