feat(syntax): detect languages from shebangs (#360)

* feat(syntax): detect languages from shebangs

* fix(syntax): accept equals form of env split-string

Author: fcoury-oai

default_config.toml

@@ -232,6 +232,7 @@ waiting = "steady_underscore"
 # extensions = ["build"]
 # filenames = ["Buildfile"]
 # aliases = ["build-script"]
+# shebangs = ["build-script"]
 # comment = "# %s"
 # indent_width = 2
 #

docs/LANGUAGES.md

@@ -1,7 +1,7 @@
 # Language extensions
 
 Red uses one language definition for syntax highlighting, exact filename and
-extension detection, comment templates, indentation, and language-server
+extension and shebang detection, comment templates, indentation, and language-server
 routing. Definitions can live in your configuration or in an installable
 external plugin package.
 
@@ -32,7 +32,32 @@ validate = true
 ```
 
 Extensions are case-insensitive and may start with a dot. Exact filenames are
-case-sensitive and take precedence over extensions. Aliases are accepted by
+case-sensitive and take precedence over extensions. When no filename or extension matches, Red reads up to 512 characters from the
+first line and checks its shebang. Bundled interpreters include `sh`, `bash`,
+`dash`, `ash`, `zsh` (using Bash syntax), `fish`, `pwsh`, `powershell`, `node`,
+`nodejs`, `lua`, `luajit`, and `husk`. No executable permission is required.
+
+Add interpreter basenames to any language definition or language-pack manifest:
+
+```toml
+[languages.python]
+shebangs = ["python", "python3", "pypy", "pypy3"]
+```
+
+This registers detection; Python highlighting still requires its language pack.
+Direct paths and common `env` forms are supported, including `env -S bash -eu`,
+`-i`, `-u NAME`, `-C DIR`, `--`, and environment assignments. Numeric interpreter
+versions such as `python3.12` fall back to the registered base name. Quoted
+commands, shell expansion, and arbitrary wrapper commands are not interpreted.
+An unknown or overlong shebang leaves the language undetected.
+
+Detection uses the current buffer, so editing the first line updates syntax even
+when it is offscreen. `:syntax <language>` overrides detection and `:syntax off`
+disables highlighting and structural operations. Shebang detection also applies
+to full-source previews. LSP routing still requires a filename or extension
+selector; this feature does not attach language servers to extensionless files.
+
+Aliases are accepted by
 `:syntax build-script`, syntax completion, and Markdown code fences.
 
 Every field is optional. A grammar-free language can still provide syntax

src/buffer.rs

@@ -119,7 +119,7 @@ pub struct SearchMatch {
 /// Buffer-local syntax-highlighting selection.
 #[derive(Debug, Clone, Default, PartialEq, Eq)]
 pub enum SyntaxSelection {
-    /// Detect syntax from the buffer's file name.
+    /// Detect syntax from the file name, then the first-line shebang.
     #[default]
     Auto,
     /// Disable syntax highlighting.

src/config.rs

@@ -1065,6 +1065,9 @@ pub struct LanguageConfig {
     /// Case-sensitive exact file names, such as `Dockerfile` or `Makefile`.
     #[serde(default)]
     pub filenames: Vec<String>,
+    /// Interpreter basenames recognized in a first-line shebang.
+    #[serde(default)]
+    pub shebangs: Vec<String>,
     /// Additional names accepted by syntax selection and injected fenced blocks.
     #[serde(default)]
     pub aliases: Vec<String>,
@@ -2285,6 +2288,7 @@ fn known_schema_path(path: &[String]) -> bool {
             "extensions"
                 | "filenames"
                 | "aliases"
+                | "shebangs"
                 | "comment"
                 | "text_width"
                 | "indent_width"

src/editor.rs

@@ -7278,7 +7278,10 @@ impl Editor {
         match buffer.syntax_selection() {
             SyntaxSelection::Auto => self
                 .highlighter
-                .language_id_for_file(buffer.file.as_deref())
+                .language_id_for_source(
+                    buffer.file.as_deref(),
+                    &buffer.line_prefix_contents(0, crate::highlighter::MAX_SHEBANG_CHARS + 1),
+                )
                 .map(ToString::to_string),
             SyntaxSelection::Off => None,
             SyntaxSelection::Language(language) => self
@@ -19424,7 +19427,12 @@ impl Editor {
         }
 
         self.highlighter
-            .language_id_for_file(self.current_buffer().file.as_deref())
+            .language_id_for_source(
+                self.current_buffer().file.as_deref(),
+                &self
+                    .current_buffer()
+                    .line_prefix_contents(0, crate::highlighter::MAX_SHEBANG_CHARS + 1),
+            )
             .map(str::to_string)
             .or_else(|| self.current_buffer().file_type())
     }
@@ -34479,6 +34487,33 @@ builtin = "rust"
         assert_eq!(editor.config.commenting.languages["buildspec"], "# %s");
     }
 
+    #[tokio::test]
+    async fn shebang_reload_updates_open_unsaved_buffer() {
+        let directory = tempfile::tempdir().unwrap();
+        let path = directory.path().join("config.toml");
+        std::fs::write(
+            &path,
+            "[languages.custom]\nshebangs = ['custom']\ncomment = '# %s'\nindent_width = 2\n",
+        )
+        .unwrap();
+        let mut editor = test_editor(80, 12);
+        editor.buffer_manager[0] = Buffer::new(None, "#!/bin/custom\nhello\n".into());
+        editor.current_buffer_mut().insert_str(0, 1, "changed ");
+        let contents = editor.current_buffer().contents();
+        let revision = editor.current_buffer().revision();
+        editor.set_language_reload_source(path.clone(), Vec::new());
+        editor.reload_languages().await.unwrap();
+        assert_eq!(editor.current_language_id().as_deref(), Some("custom"));
+        assert_eq!(editor.indentation().shift_width, 2);
+        assert!(editor.configured_comment_syntax().is_some());
+        assert_eq!(editor.current_buffer().contents(), contents);
+        assert_eq!(editor.current_buffer().revision(), revision);
+        assert!(editor.current_buffer().is_dirty());
+        std::fs::write(&path, "[languages.custom]\nshebangs = ['other']\n").unwrap();
+        editor.reload_languages().await.unwrap();
+        assert_eq!(editor.current_language_id(), None);
+    }
+
     #[tokio::test]
     async fn rejected_language_reload_keeps_previous_registry_and_configuration() {
         let directory = tempfile::tempdir().unwrap();
@@ -38119,6 +38154,46 @@ builtin = "rust"
         assert_eq!(span_shape(&reopened), span_shape(&expected));
     }
 
+    #[test]
+    fn shebang_follows_edits_offscreen_and_manual_selection() {
+        let mut editor = rust_test_editor(100, 120, 22);
+        let text = format!(
+            "#!/bin/bash\n{}",
+            "if true; then echo hello; fi\n".repeat(100)
+        );
+        editor.buffer_manager[0] = Buffer::new(Some("script".into()), text);
+        assert!(!editor
+            .viewport_highlight_spans(0, 60, 20)
+            .unwrap()
+            .is_empty());
+        assert_eq!(
+            editor.highlight_cache[&0].language_id.as_deref(),
+            Some("bash")
+        );
+        assert_eq!(editor.current_language_id().as_deref(), Some("bash"));
+        assert!(editor.configured_comment_syntax().is_some());
+        editor.current_buffer_mut().insert_str(0, 0, "#");
+        assert!(editor
+            .viewport_highlight_spans(0, 60, 20)
+            .unwrap()
+            .is_empty());
+        editor
+            .current_buffer_mut()
+            .set_syntax_selection(SyntaxSelection::Language("fish".into()));
+        editor.viewport_highlight_spans(0, 60, 20).unwrap();
+        assert_eq!(
+            editor.highlight_cache[&0].language_id.as_deref(),
+            Some("fish")
+        );
+        editor
+            .current_buffer_mut()
+            .set_syntax_selection(SyntaxSelection::Off);
+        assert!(editor
+            .viewport_highlight_spans(0, 60, 20)
+            .unwrap()
+            .is_empty());
+    }
+
     #[test]
     fn viewport_highlight_cache_follows_buffer_syntax_selection() {
         let mut editor = rust_test_editor(100, 120, 22);