perf(linter): batch plugins into single visitor with anchor-kind dispatch

.changeset/batch-plugin-visitor.md

@@ -0,0 +1,5 @@
+---
+"@biomejs/biome": patch
+---
+
+Improved plugin performance by batching all plugins into a single syntax visitor with a kind-to-plugin lookup map, reducing per-node dispatch overhead from O(N) to O(1) where N is the number of plugins.

.github/workflows/benchmark_js.yml

@@ -1,4 +1,4 @@
-name: Benchmarks JS
+name: Benchmarks JS/GritQL
 
 on:
   workflow_dispatch:
@@ -17,6 +17,7 @@ on:
       - 'crates/biome_formatter/**/*.rs'
       - 'crates/biome_rowan/**/*.rs'
       - 'crates/biome_parser/**/*.rs'
+      - 'crates/biome_grit_patterns/**/*.rs'
   push:
     branches:
       - main
@@ -30,6 +31,7 @@ on:
       - 'crates/biome_formatter/**/*.rs'
       - 'crates/biome_rowan/**/*.rs'
       - 'crates/biome_parser/**/*.rs'
+      - 'crates/biome_grit_patterns/**/*.rs'
 
 env:
   RUST_LOG: info
@@ -47,6 +49,7 @@ jobs:
           - biome_js_parser
           - biome_js_formatter
           - biome_js_analyze
+          - biome_grit_patterns
 
     steps:
 

crates/biome_analyze/src/analyzer_plugin.rs

@@ -1,5 +1,5 @@
-use camino::Utf8PathBuf;
-use rustc_hash::FxHashSet;
+use camino::{Utf8Path, Utf8PathBuf};
+use rustc_hash::{FxHashMap, FxHashSet};
 use std::hash::Hash;
 use std::{fmt::Debug, sync::Arc};
 
@@ -23,6 +23,14 @@ pub trait AnalyzerPlugin: Debug + Send + Sync {
     fn query(&self) -> Vec<RawSyntaxKind>;
 
     fn evaluate(&self, node: AnySyntaxNode, path: Arc<Utf8PathBuf>) -> Vec<RuleDiagnostic>;
+
+    /// Returns true if this plugin should run on the given file path.
+    ///
+    /// Stub that always returns `true` — file-scoping will be implemented
+    /// in a companion PR (#9171) via the `includes` plugin option.
+    fn applies_to_file(&self, _path: &Utf8Path) -> bool {
+        true
+    }
 }
 
 #[derive(Copy, Clone, Debug, Eq, PartialEq, Hash)]
@@ -37,6 +45,10 @@ pub enum PluginTargetLanguage {
 pub struct PluginVisitor<L: Language> {
     query: FxHashSet<L::Kind>,
     plugin: Arc<Box<dyn AnalyzerPlugin>>,
+
+    /// When set, all nodes in this subtree are skipped until we leave it.
+    /// Used to skip subtrees that fall entirely outside the analysis range
+    /// (see the `ctx.range` check in `visit`).
     skip_subtree: Option<SyntaxNode<L>>,
 }
 
@@ -102,6 +114,10 @@ where
             return;
         }
 
+        if !self.plugin.applies_to_file(&ctx.options.file_path) {
+            return;
+        }
+
         let rule_timer = profiling::start_plugin_rule("plugin");
         let diagnostics = self
             .plugin
@@ -126,3 +142,139 @@ where
         ctx.signal_queue.extend(signals);
     }
 }
+
+/// A batched syntax visitor that evaluates multiple plugins in a single visitor.
+///
+/// Instead of registering N separate `PluginVisitor` instances (one per plugin),
+/// this holds all plugins together and dispatches using a kind-to-plugin lookup
+/// map. This reduces visitor-dispatch overhead and enables O(1) kind matching
+/// per node instead of iterating all plugins.
+pub struct BatchPluginVisitor<L: Language> {
+    plugins: Vec<Arc<Box<dyn AnalyzerPlugin>>>,
+
+    /// Maps each syntax kind to the indices of plugins that query for it.
+    kind_to_plugins: FxHashMap<L::Kind, Vec<usize>>,
+
+    /// When set, all nodes in this subtree are skipped until we leave it.
+    /// Used to skip subtrees that fall entirely outside the analysis range
+    /// (see the `ctx.range` check in `visit`).
+    skip_subtree: Option<SyntaxNode<L>>,
+
+    /// Cached per-plugin results of `applies_to_file`. Populated lazily on
+    /// first `WalkEvent::Enter` — the file path is constant for the entire walk.
+    applicable: Option<Vec<bool>>,
+}
+
+impl<L> BatchPluginVisitor<L>
+where
+    L: Language + 'static,
+    L::Kind: Eq + Hash,
+{
+    /// Creates a batched plugin visitor from a slice of plugins.
+    ///
+    /// # Safety
+    /// Caller must ensure all plugins target language `L`. The `RawSyntaxKind`
+    /// values returned by each plugin's `query()` are converted to `L::Kind`
+    /// via `from_raw` without validation.
+    pub unsafe fn new_unchecked(plugins: AnalyzerPluginSlice) -> Self {
+        let mut all_plugins = Vec::with_capacity(plugins.len());
+        let mut kind_to_plugins: FxHashMap<L::Kind, Vec<usize>> = FxHashMap::default();
+
+        for (idx, plugin) in plugins.iter().enumerate() {
+            all_plugins.push(Arc::clone(plugin));
+            let mut seen_kinds = FxHashSet::default();
+            for raw_kind in plugin.query() {
+                let kind = L::Kind::from_raw(raw_kind);
+                if seen_kinds.insert(kind) {
+                    kind_to_plugins.entry(kind).or_default().push(idx);
+                }
+            }
+        }
+
+        Self {
+            plugins: all_plugins,
+            kind_to_plugins,
+            skip_subtree: None,
+            applicable: None,
+        }
+    }
+}
+
+impl<L> Visitor for BatchPluginVisitor<L>
+where
+    L: Language + 'static,
+    L::Kind: Eq + Hash,
+{
+    type Language = L;
+
+    fn visit(
+        &mut self,
+        event: &WalkEvent<SyntaxNode<Self::Language>>,
+        ctx: VisitorContext<Self::Language>,
+    ) {
+        let node = match event {
+            WalkEvent::Enter(node) => node,
+            WalkEvent::Leave(node) => {
+                if let Some(skip_subtree) = &self.skip_subtree
+                    && skip_subtree == node
+                {
+                    self.skip_subtree = None;
+                }
+
+                return;
+            }
+        };
+
+        if self.skip_subtree.is_some() {
+            return;
+        }
+
+        if let Some(range) = ctx.range
+            && node.text_range_with_trivia().ordering(range).is_ne()
+        {
+            self.skip_subtree = Some(node.clone());
+            return;
+        }
+
+        let kind = node.kind();
+
+        let Some(plugin_indices) = self.kind_to_plugins.get(&kind) else {
+            return;
+        };
+
+        let applicable = self.applicable.get_or_insert_with(|| {
+            self.plugins
+                .iter()
+                .map(|p| p.applies_to_file(&ctx.options.file_path))
+                .collect()
+        });
+
+        for &idx in plugin_indices {
+            if !applicable[idx] {
+                continue;
+            }
+
+            let plugin = &self.plugins[idx];
+            let rule_timer = profiling::start_plugin_rule("plugin");
+            let diagnostics = plugin.evaluate(node.clone().into(), ctx.options.file_path.clone());
+            rule_timer.stop();
+
+            let signals = diagnostics.into_iter().map(|diagnostic| {
+                let name = diagnostic
+                    .subcategory
+                    .clone()
+                    .unwrap_or_else(|| "anonymous".into());
+
+                SignalEntry {
+                    text_range: diagnostic.span().unwrap_or_default(),
+                    signal: Box::new(PluginSignal::<L>::new(diagnostic)),
+                    rule: SignalRuleKey::Plugin(name.into()),
+                    category: RuleCategory::Lint,
+                    instances: Default::default(),
+                }
+            });
+
+            ctx.signal_queue.extend(signals);
+        }
+    }
+}

crates/biome_analyze/src/lib.rs

@@ -30,7 +30,8 @@ mod visitor;
 pub use biome_diagnostics::category_concat;
 
 pub use crate::analyzer_plugin::{
-    AnalyzerPlugin, AnalyzerPluginSlice, AnalyzerPluginVec, PluginTargetLanguage, PluginVisitor,
+    AnalyzerPlugin, AnalyzerPluginSlice, AnalyzerPluginVec, BatchPluginVisitor,
+    PluginTargetLanguage, PluginVisitor,
 };
 pub use crate::categories::{
     ActionCategory, OtherActionCategory, RefactorKind, RuleCategories, RuleCategoriesBuilder,

crates/biome_css_analyze/src/lib.rs

@@ -14,8 +14,8 @@ pub use crate::registry::visit_registry;
 use crate::suppression_action::CssSuppressionAction;
 use biome_analyze::{
     AnalysisFilter, AnalyzerOptions, AnalyzerPluginSlice, AnalyzerSignal, AnalyzerSuppression,
-    ControlFlow, LanguageRoot, MatchQueryParams, MetadataRegistry, Phases, PluginTargetLanguage,
-    PluginVisitor, RuleAction, RuleRegistry, to_analyzer_suppressions,
+    BatchPluginVisitor, ControlFlow, LanguageRoot, MatchQueryParams, MetadataRegistry, Phases,
+    PluginTargetLanguage, RuleAction, RuleRegistry, to_analyzer_suppressions,
 };
 use biome_css_syntax::{CssFileSource, CssLanguage, TextRange};
 use biome_diagnostics::Error;
@@ -151,15 +151,19 @@ where
         analyzer.add_visitor(phase, visitor);
     }
 
-    for plugin in plugins {
-        // SAFETY: The plugin target language is correctly checked here.
+    let css_plugins: Vec<_> = plugins
+        .iter()
+        .filter(|p| p.language() == PluginTargetLanguage::Css)
+        .cloned()
+        .collect();
+
+    if !css_plugins.is_empty() {
+        // SAFETY: All plugins have been verified to target CSS above.
         unsafe {
-            if plugin.language() == PluginTargetLanguage::Css {
-                analyzer.add_visitor(
-                    Phases::Syntax,
-                    Box::new(PluginVisitor::new_unchecked(plugin.clone())),
-                )
-            }
+            analyzer.add_visitor(
+                Phases::Syntax,
+                Box::new(BatchPluginVisitor::new_unchecked(&css_plugins)),
+            );
         }
     }
 

crates/biome_grit_patterns/Cargo.toml

@@ -11,6 +11,10 @@ keywords.workspace   = true
 categories.workspace = true
 publish              = false
 
+[[bench]]
+harness = false
+name    = "grit_query"
+
 [dependencies]
 biome_analyze        = { workspace = true }
 biome_console        = { workspace = true }
@@ -39,9 +43,16 @@ serde_json           = { workspace = true, optional = true }
 
 [dev-dependencies]
 biome_test_utils = { path = "../biome_test_utils" }
+criterion        = { package = "codspeed-criterion-compat", version = "*" }
 insta            = { workspace = true }
 tests_macros     = { path = "../tests_macros" }
 
+[target.'cfg(all(target_family="unix", not(all(target_arch = "aarch64", target_env = "musl"))))'.dev-dependencies]
+tikv-jemallocator = { workspace = true }
+
+[target.'cfg(target_os = "windows")'.dev-dependencies]
+mimalloc = { workspace = true }
+
 [features]
 schema = ["biome_js_parser/schema", "dep:schemars", "serde"]
 serde  = ["dep:serde", "dep:serde_json"]