Merge pull request #115 from levineuwirth/shell-grammar

feat(highlight): shell/bash tree-sitter grammar for the shell family

Cargo.lock

@@ -2550,6 +2550,7 @@ dependencies = [
  "thiserror 2.0.18",
  "toml",
  "tree-sitter",
+ "tree-sitter-bash",
  "tree-sitter-c",
  "tree-sitter-cpp",
  "tree-sitter-cuda",
@@ -3716,6 +3717,16 @@ dependencies = [
  "tree-sitter-language",
 ]
 
+[[package]]
+name = "tree-sitter-bash"
+version = "0.25.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9e5ec769279cc91b561d3df0d8a5deb26b0ad40d183127f409494d6d8fc53062"
+dependencies = [
+ "cc",
+ "tree-sitter-language",
+]
+
 [[package]]
 name = "tree-sitter-c"
 version = "0.24.2"

Cargo.toml

@@ -154,6 +154,19 @@ tree-sitter-cpp = "0.23"
 # it shares the ABI crate with the other grammars — no second
 # `tree-sitter` in the graph (its own `tree-sitter` dep is dev-only).
 tree-sitter-cuda = "0.21"
+# Shell / bash. Lexical highlighting for `.sh`/`.bash` and the wider
+# shell family (`.zsh`/`.ksh`/`.ash`/`.bats`) — the LSP half
+# (bash-language-server + shellcheck/shfmt) was already wired in
+# `builtin/runtime/lsp.lua`; this fills in the missing grammar. bash
+# does not inherit another grammar, so `HIGHLIGHT_QUERY` (singular, the
+# c/cpp/md idiom) is self-contained — no base-query composition like
+# `cuda`. Exports `LANGUAGE` over `tree-sitter-language 0.1` (shared ABI
+# crate, no second `tree-sitter` in the graph). Note: because the
+# grammar's language name (`bash`) matches the `pmacs.lsp.config.bash`
+# key, every extension claimed here also auto-attaches
+# bash-language-server (shellcheck refuses zsh, so `.zsh` diagnostics
+# may be sparse; highlighting is unaffected).
+tree-sitter-bash = "0.25"
 # T M9.7: markdown grammar so prompt result buffers with
 # `_meta.format = "markdown"` get structured highlighting through the
 # same M4 path as rust/lua — no special-case painter in Lua.

builtin/runtime/lsp.lua

@@ -215,9 +215,15 @@ pmacs.lsp.filetypes.jsx = pmacs.lsp.filetypes.jsx or "javascriptreact"
 -- `language_for_path` resolves `.lua` first; this entry is the LSP
 -- fallback and keeps the language id stable if that ever changes.
 pmacs.lsp.filetypes.lua = pmacs.lsp.filetypes.lua or "lua"
--- Bash / shell (bash-language-server).
-pmacs.lsp.filetypes.sh = pmacs.lsp.filetypes.sh or "bash"
-pmacs.lsp.filetypes.bash = pmacs.lsp.filetypes.bash or "bash"
+-- Bash / shell (bash-language-server). pmacs bundles a bash grammar, so
+-- `language_for_path` already resolves these to `bash`; this map is the
+-- LSP-only fallback that keeps the language id stable if that grammar is
+-- ever dropped (same role as the `lua`/`cuda` entries). The wider shell
+-- family (`.zsh`/`.ksh`/`.ash`/`.bats`) rides the same server; shellcheck
+-- declines zsh, so `.zsh` diagnostics may be sparse.
+for _, ext in ipairs({ "sh", "bash", "zsh", "ksh", "ash", "bats" }) do
+  pmacs.lsp.filetypes[ext] = pmacs.lsp.filetypes[ext] or "bash"
+end
 -- TOML (taplo).
 pmacs.lsp.filetypes.toml = pmacs.lsp.filetypes.toml or "toml"
 -- Zig (zls). `.zon` is Zig Object Notation, handled by the same server.

src/syntax.rs

@@ -443,6 +443,24 @@ pub const BUILTIN_LANGUAGES: &[LanguageEntry] = &[
             tree_sitter_cuda::HIGHLIGHTS_QUERY,
         ],
     },
+    // Shell / bash. Lexical highlighting for the shell family; the LSP
+    // half (bash-language-server) was already wired in `lsp.lua`. Unlike
+    // `cuda`, bash's `highlights.scm` is self-contained (no `; inherits:`
+    // delta), so a single fragment suffices. The extension set is wider
+    // than the `.sh`/`.bash` the LSP filetype map covered: `.zsh`/`.ksh`/
+    // `.ash` are close-enough dialects and `.bats` is bash. None collide
+    // with an entry above. Because the language name is `bash` — matching
+    // the `pmacs.lsp.config.bash` key — opening any of these also
+    // auto-attaches bash-language-server. Extensionless shebang scripts
+    // (`#!/bin/sh`) and rc dotfiles (`.bashrc`) are NOT covered: detection
+    // is extension-keyed, and shebang/filename sniffing is a separate
+    // (deferred) feature.
+    LanguageEntry {
+        name: "bash",
+        extensions: &["sh", "bash", "zsh", "ksh", "ash", "bats"],
+        loader: || tree_sitter_bash::LANGUAGE.into(),
+        highlights_query: &[tree_sitter_bash::HIGHLIGHT_QUERY],
+    },
 ];
 
 /// Registry that the Lua surface ([`crate::lua_bindings::install_parse`])
@@ -953,6 +971,95 @@ mod tests {
         );
     }
 
+    #[test]
+    fn builtin_languages_include_bash() {
+        // Regression guard: the bash entry claims the wider shell family
+        // and ships a highlights query, so shell scripts get lexical color
+        // (they had LSP via bash-language-server but no grammar before).
+        let bash = BUILTIN_LANGUAGES
+            .iter()
+            .find(|l| l.name == "bash")
+            .expect("`bash` language entry must be present");
+        for ext in ["sh", "bash", "zsh", "ksh", "ash", "bats"] {
+            assert!(bash.extensions.contains(&ext), "`bash` claims `.{ext}`");
+        }
+        assert!(
+            !bash.highlights_query.is_empty(),
+            "`bash` ships a highlights query"
+        );
+    }
+
+    #[test]
+    fn bash_grammar_loads_and_parses_script() {
+        // ABI acceptance: the `tree-sitter-bash` 0.25 grammar must be
+        // accepted by our `tree-sitter` 0.26 core — `set_language`
+        // succeeds and a tree is produced. A representative script
+        // (shebang, `set`, parameter expansion, function, `if`) parses
+        // without error, so the entry wired a working grammar.
+        let reg = SyntaxRegistry::new();
+        let language = reg
+            .language("bash")
+            .expect("`bash` language loads from BUILTIN_LANGUAGES");
+        let mut buf = fresh_buffer("deploy.sh");
+        buf.apply_edit(EditOp::Insert {
+            pos: 0,
+            bytes: b"#!/usr/bin/env bash\nset -euo pipefail\nname=${1:-world}\n\
+                     greet() { echo \"hello, $name\"; }\nif [ -n \"$name\" ]; then greet; fi\n",
+        })
+        .unwrap();
+        let view = ParseView::new(&buf, language, "bash".to_owned());
+        let handle = view.handle();
+        let _vid = buf.attach_view(Box::new(view));
+        let bundle = parse_synchronously(&handle);
+        assert_eq!(
+            bundle.tree.root_node().kind(),
+            "program",
+            "bash grammar roots at `program`"
+        );
+        assert!(
+            !bundle.tree.root_node().has_error(),
+            "bash grammar parses a representative script without error"
+        );
+    }
+
+    #[test]
+    fn bash_highlights_compile_with_captures() {
+        // The self-contained bash `highlights.scm` must compile against
+        // the grammar and yield real capture classes (a crate bump that
+        // drifted query and grammar apart would surface here).
+        let reg = SyntaxRegistry::new();
+        let query = reg
+            .highlights_query("bash")
+            .expect("bash highlights compile");
+        assert!(
+            query.capture_names().len() >= 5,
+            "bash highlights resolve several capture classes; got {}",
+            query.capture_names().len()
+        );
+    }
+
+    #[test]
+    fn language_for_path_resolves_bash_extensions() {
+        // The whole shell family resolves to the `bash` grammar via
+        // extension detection; `.zsh`/`.ksh`/`.ash`/`.bats` are new here
+        // (only `.sh`/`.bash` were covered by the LSP filetype map before).
+        let reg = SyntaxRegistry::new();
+        for path in [
+            "deploy.sh",
+            "lib.bash",
+            "prompt.zsh",
+            "script.ksh",
+            "init.ash",
+            "test_cli.bats",
+        ] {
+            assert_eq!(
+                reg.language_name_for_path(path).as_deref(),
+                Some("bash"),
+                "{path} resolves to bash"
+            );
+        }
+    }
+
     #[test]
     fn byte_to_point_handles_first_line() {
         let src = b"hello world";

tests/m4_acceptance.rs

@@ -5804,6 +5804,51 @@ fn m4_12_default_bundle_wires_cuda() {
     assert_eq!(probe.get::<String>("grammar_cuh").unwrap(), "cuda");
 }
 
+/// The default bundle wires the shell family: `pmacs.lsp.config.bash`
+/// targets bash-language-server (pre-existing), the bundled bash grammar
+/// resolves the wider extension set (`.sh`/`.zsh`/`.bats`) to `bash`
+/// through `pmacs.parse.language_for_path`, and the LSP filetype fallback
+/// maps the new extensions too (belt-and-suspenders if the grammar is
+/// dropped).
+#[test]
+fn m4_12_default_bundle_wires_bash() {
+    use pmacs::editor::EditorState;
+
+    let s = EditorState::new();
+    let probe: mlua::Table = s
+        .lua_host
+        .lua()
+        .load(
+            r"
+            local out = {}
+            out.cfg_cmd = pmacs.lsp.config.bash and pmacs.lsp.config.bash.command
+            out.ft_zsh = pmacs.lsp.filetypes.zsh
+            out.ft_bats = pmacs.lsp.filetypes.bats
+            -- Grammar-backed detection resolves the wider set directly.
+            out.grammar_sh = pmacs.parse.language_for_path('deploy.sh')
+            out.grammar_zsh = pmacs.parse.language_for_path('prompt.zsh')
+            out.grammar_bats = pmacs.parse.language_for_path('test_cli.bats')
+            return out
+            ",
+        )
+        .eval()
+        .expect("probe bash wiring");
+    assert_eq!(
+        probe.get::<String>("cfg_cmd").unwrap(),
+        "bash-language-server",
+        "config.bash targets bash-language-server"
+    );
+    assert_eq!(probe.get::<String>("ft_zsh").unwrap(), "bash");
+    assert_eq!(probe.get::<String>("ft_bats").unwrap(), "bash");
+    assert_eq!(
+        probe.get::<String>("grammar_sh").unwrap(),
+        "bash",
+        "bundled grammar resolves `.sh` to bash"
+    );
+    assert_eq!(probe.get::<String>("grammar_zsh").unwrap(), "bash");
+    assert_eq!(probe.get::<String>("grammar_bats").unwrap(), "bash");
+}
+
 /// Typing-perf: the default bundle coalesces full-document
 /// `didChange` notifications instead of sending one per keystroke
 /// (each send copies the whole buffer several times and writes