yeast: Use SmallVec<[Id; 1]> for rule output

Every rule firing returned `Vec<Id>` even though the overwhelming
majority of rules produce a single replacement node. Switch the
Transform return type, try_rule, apply_rules, and apply_rules_inner
to `RuleOutput = SmallVec<[Id; 1]>`. Re-export `smallvec` and
`SmallVec` from the yeast crate so generated rule! macro code can
refer to them by short paths.

For the rule! macro, generate `yeast::smallvec![__id]` for the
shorthand form and `yeast::RuleOutput = yeast::SmallVec::new()` for
the full template form, so rule outputs stay inline for the common
single-Id case.

Per-firing memory: a Vec allocation per rule firing turns into zero
when the rule produces 0 or 1 Ids. Rules that produce multiple Ids
still allocate (SmallVec spills to heap when the inline buffer is
exceeded), unchanged from before.

Cargo.lock

@@ -3415,6 +3415,7 @@ dependencies = [
  "serde",
  "serde_json",
  "serde_yaml",
+ "smallvec",
  "tree-sitter",
  "tree-sitter-python",
  "tree-sitter-ruby",

MODULE.bazel

@@ -142,6 +142,7 @@ use_repo(
     "vendor_ts__serde_json-1.0.145",
     "vendor_ts__serde_with-3.14.1",
     "vendor_ts__serde_yaml-0.9.34-deprecated",
+    "vendor_ts__smallvec-1.15.1",
     "vendor_ts__syn-2.0.106",
     "vendor_ts__toml-0.9.7",
     "vendor_ts__tracing-0.1.41",

misc/bazel/3rdparty/tree_sitter_extractors_deps/BUILD.bazel

@@ -541,6 +541,18 @@ alias(
     tags = ["manual"],
 )
 
+alias(
+    name = "smallvec-1.15.1",
+    actual = "@vendor_ts__smallvec-1.15.1//:smallvec",
+    tags = ["manual"],
+)
+
+alias(
+    name = "smallvec",
+    actual = "@vendor_ts__smallvec-1.15.1//:smallvec",
+    tags = ["manual"],
+)
+
 alias(
     name = "syn-2.0.106",
     actual = "@vendor_ts__syn-2.0.106//:syn",

misc/bazel/3rdparty/tree_sitter_extractors_deps/defs.bzl

@@ -391,6 +391,7 @@ _NORMAL_DEPENDENCIES = {
             "serde": Label("@vendor_ts__serde-1.0.228//:serde"),
             "serde_json": Label("@vendor_ts__serde_json-1.0.145//:serde_json"),
             "serde_yaml": Label("@vendor_ts__serde_yaml-0.9.34-deprecated//:serde_yaml"),
+            "smallvec": Label("@vendor_ts__smallvec-1.15.1//:smallvec"),
             "tree-sitter": Label("@vendor_ts__tree-sitter-0.26.8//:tree_sitter"),
             "tree-sitter-python": Label("@vendor_ts__tree-sitter-python-0.23.6//:tree_sitter_python"),
             "tree-sitter-ruby": Label("@vendor_ts__tree-sitter-ruby-0.23.1//:tree_sitter_ruby"),
@@ -4278,6 +4279,7 @@ def crate_repositories():
         struct(repo = "vendor_ts__serde_json-1.0.145", is_dev_dep = False),
         struct(repo = "vendor_ts__serde_with-3.14.1", is_dev_dep = False),
         struct(repo = "vendor_ts__serde_yaml-0.9.34-deprecated", is_dev_dep = False),
+        struct(repo = "vendor_ts__smallvec-1.15.1", is_dev_dep = False),
         struct(repo = "vendor_ts__syn-2.0.106", is_dev_dep = False),
         struct(repo = "vendor_ts__toml-0.9.7", is_dev_dep = False),
         struct(repo = "vendor_ts__tracing-0.1.41", is_dev_dep = False),

shared/yeast-macros/src/parse.rs

@@ -643,7 +643,7 @@ pub fn parse_rule_top(input: TokenStream) -> Result<TokenStream> {
                 true,
                 __source_range,
             );
-            vec![__id]
+            yeast::smallvec![__id]
         }
     } else {
         // Full template form
@@ -657,7 +657,7 @@ pub fn parse_rule_top(input: TokenStream) -> Result<TokenStream> {
         }
 
         quote! {
-            let mut __nodes: Vec<usize> = Vec::new();
+            let mut __nodes: yeast::RuleOutput = yeast::SmallVec::new();
             #(#transform_items)*
             __nodes
         }

shared/yeast/Cargo.toml

@@ -8,6 +8,7 @@ clap = { version = "4.4.10", features = ["derive"] }
 serde = { version = "1.0.193", features = ["derive"] }
 serde_json = "1.0.108"
 serde_yaml = "0.9"
+smallvec = "1.15"
 tree-sitter = ">= 0.23.0"
 yeast-macros = { path = "../yeast-macros" }
 

shared/yeast/src/lib.rs

@@ -4,6 +4,7 @@ extern crate self as yeast;
 
 use serde::Serialize;
 use serde_json::{json, Value};
+pub use smallvec::{smallvec, SmallVec};
 
 pub mod build;
 pub mod captures;
@@ -29,6 +30,10 @@ type Id = usize;
 type FieldId = u16;
 type KindId = u16;
 
+/// The output of one rule firing: a small list of replacement node Ids,
+/// inline for the common "single replacement" case.
+pub type RuleOutput = SmallVec<[Id; 1]>;
+
 pub const CHILD_FIELD: u16 = u16::MAX;
 
 #[derive(Debug)]
@@ -452,9 +457,10 @@ impl From<tree_sitter::Range> for NodeContent {
 
 /// The transform function for a rule: takes the AST, captured variables, a
 /// fresh-name scope, and the source range of the matched node, and returns
-/// the IDs of the replacement nodes.
+/// the IDs of the replacement nodes (typically a single Id; uses
+/// `SmallVec` to keep the common case inline).
 pub type Transform = Box<
-    dyn Fn(&mut Ast, Captures, &tree_builder::FreshScope, Option<tree_sitter::Range>) -> Vec<Id>
+    dyn Fn(&mut Ast, Captures, &tree_builder::FreshScope, Option<tree_sitter::Range>) -> RuleOutput
         + Send
         + Sync,
 >;
@@ -492,7 +498,7 @@ impl Rule {
         ast: &mut Ast,
         node: Id,
         fresh: &tree_builder::FreshScope,
-    ) -> Result<Option<Vec<Id>>, String> {
+    ) -> Result<Option<RuleOutput>, String> {
         let mut captures = Captures::new();
         if self.query.do_match(ast, node, &mut captures)? {
             fresh.next_scope();
@@ -544,7 +550,7 @@ fn apply_rules(
     ast: &mut Ast,
     id: Id,
     fresh: &tree_builder::FreshScope,
-) -> Result<Vec<Id>, String> {
+) -> Result<RuleOutput, String> {
     let index = RuleIndex::new(rules);
     apply_rules_inner(&index, ast, id, fresh, 0, None)
 }
@@ -556,7 +562,7 @@ fn apply_rules_inner(
     fresh: &tree_builder::FreshScope,
     rewrite_depth: usize,
     skip_rule: Option<*const Rule>,
-) -> Result<Vec<Id>, String> {
+) -> Result<RuleOutput, String> {
     if rewrite_depth > MAX_REWRITE_DEPTH {
         return Err(format!(
             "Desugaring exceeded maximum rewrite depth ({MAX_REWRITE_DEPTH}). \
@@ -576,7 +582,7 @@ fn apply_rules_inner(
             // query doesn't loop. Other rules and child traversal are
             // unaffected.
             let next_skip = if rule.repeated { None } else { Some(rule_ptr) };
-            let mut results = Vec::new();
+            let mut results = RuleOutput::new();
             for node in result_node {
                 results.extend(apply_rules_inner(
                     index,
@@ -625,7 +631,7 @@ fn apply_rules_inner(
         }
     }
     ast.nodes[id].fields = fields;
-    Ok(vec![id])
+    Ok(smallvec![id])
 }
 
 /// One phase of a desugaring pass: a named bundle of rules that runs to