Merge pull request #25 from domenicquirl/v0.6

Cargo.toml

@@ -1,33 +1,44 @@
 [package]
-edition = "2018"
+edition   = "2018"
-name = "cstree"
+name      = "cstree"
-version = "0.5.0"
+version   = "0.6.0"
-authors = ["Domenic Quirl <DomenicQuirl@pm.me>", "Aleksey Kladov <aleksey.kladov@gmail.com>"]
+authors   = [
+    "Domenic Quirl <DomenicQuirl@pm.me>",
+    "Aleksey Kladov <aleksey.kladov@gmail.com>",
+]
 description = "Library for generic lossless syntax trees"
-license = "MIT OR Apache-2.0"
+license   = "MIT OR Apache-2.0"
 repository = "https://github.com/domenicquirl/cstree"
-readme = "README.md"
+readme    = "README.md"
+
+[profile.release]
+debug = true
 
 [dependencies]
-lasso = "0.5"
+lasso     = { version="0.5", features=["inline-more"] }
 text-size = "1.0.0"
-fxhash= "0.2.1"
+fxhash    = "0.2.1"
-parking_lot= "0.11.1"
+parking_lot = "0.11.1"
 
 # Arc
 triomphe = "0.1.2"
 
 [dependencies.serde]
-version = "1.0"
+version        = "1.0"
-optional = true
+optional       = true
 default-features = false
-features = ["derive"]
+features       = ["derive"]
 
 [dev-dependencies]
-m_lexer = "0.0.4"
+m_lexer       = "0.0.4"
-serde_json = "1.0.61"
+serde_json    = "1.0.61"
-serde_test = "1.0.119"
+serde_test    = "1.0.119"
 crossbeam-utils = "0.8"
+criterion     = "0.3"
+
+[[bench]]
+name  = "main"
+harness = false
 
 [features]
 default = []

benches/main.rs

@@ -0,0 +1,84 @@
+use criterion::{criterion_group, criterion_main, Criterion, Throughput};
+use cstree::*;
+use lasso::{Interner, Rodeo};
+
+#[derive(Debug)]
+pub enum Element<'s> {
+    Node(Vec<Element<'s>>),
+    Token(&'s str),
+}
+
+#[derive(Debug, Copy, Clone, Eq, PartialEq, Ord, PartialOrd, Hash)]
+pub enum TestLang {}
+impl Language for TestLang {
+    type Kind = SyntaxKind;
+
+    fn kind_from_raw(raw: SyntaxKind) -> Self::Kind {
+        raw
+    }
+
+    fn kind_to_raw(kind: Self::Kind) -> SyntaxKind {
+        kind
+    }
+}
+
+pub fn build_tree_with_cache<'c, 'i, I>(root: &Element<'_>, cache: &'c mut NodeCache<'i, I>) -> GreenNode
+where
+    I: Interner,
+{
+    let mut builder = GreenNodeBuilder::with_cache(cache);
+    build_recursive(root, &mut builder, 0);
+    let (node, interner) = builder.finish();
+    assert!(interner.is_none());
+    node
+}
+
+pub fn build_recursive<'c, 'i, I>(root: &Element<'_>, builder: &mut GreenNodeBuilder<'c, 'i, I>, mut from: u16) -> u16
+where
+    I: Interner,
+{
+    match root {
+        Element::Node(children) => {
+            builder.start_node(SyntaxKind(from));
+            for child in children {
+                from = build_recursive(child, builder, from + 1);
+            }
+            builder.finish_node();
+        }
+        Element::Token(text) => {
+            builder.token(SyntaxKind(from), *text);
+        }
+    }
+    from
+}
+
+fn two_level_tree() -> Element<'static> {
+    use Element::*;
+    Node(vec![
+        Node(vec![Token("0.0"), Token("0.1")]),
+        Node(vec![Token("1.0")]),
+        Node(vec![Token("2.0"), Token("2.1"), Token("2.2")]),
+    ])
+}
+
+pub fn create(c: &mut Criterion) {
+    let mut group = c.benchmark_group("qualification");
+    group.throughput(Throughput::Elements(1));
+
+    let mut interner = Rodeo::new();
+    let mut cache = NodeCache::with_interner(&mut interner);
+    let tree = two_level_tree();
+
+    group.bench_function("two-level tree", |b| {
+        b.iter(|| {
+            for _ in 0..100_000 {
+                let _tree = build_tree_with_cache(&tree, &mut cache);
+            }
+        })
+    });
+
+    group.finish();
+}
+
+criterion_group!(benches, create);
+criterion_main!(benches);

src/green/builder.rs

@@ -1,6 +1,9 @@
-use std::convert::TryFrom;
+use std::{
+    convert::TryFrom,
+    hash::{Hash, Hasher},
+};
 
-use fxhash::FxHashMap;
+use fxhash::{FxHashMap, FxHasher32};
 use text_size::TextSize;
 
 use crate::{
@@ -132,11 +135,11 @@ where
         }
 
         impl ChildrenIter {
-            fn new(data: [Option<GreenElement>; CHILDREN_CACHE_THRESHOLD], count: usize) -> Self {
+            fn new() -> Self {
                 ChildrenIter {
-                    data,
+                    data: [None, None, None],
-                    idx: 0,
+                    idx:  0,
-                    len: count,
+                    len:  0,
                 }
             }
         }
@@ -157,19 +160,24 @@ where
             }
         }
 
-        let mut data: [Option<GreenElement>; CHILDREN_CACHE_THRESHOLD] = [None, None, None];
+        let mut new_children = ChildrenIter::new();
-        let mut count = 0;
+        let mut hasher = FxHasher32::default();
-
+        let mut text_len: TextSize = 0.into();
-        for child in children {
+        for (i, child) in children.into_iter().enumerate() {
-            data[count] = Some(child);
+            text_len += child.text_len();
-            count += 1;
+            child.hash(&mut hasher);
+            new_children.data[i] = Some(child);
+            new_children.len += 1;
         }
-        let children = ChildrenIter::new(data, count);
 
-        let head = GreenNodeHead::from_child_iter(kind, children.clone());
+        let head = GreenNodeHead {
+            kind,
+            text_len,
+            child_hash: hasher.finish() as u32,
+        };
         self.nodes
-            .entry(head.clone())
+            .entry(head)
-            .or_insert_with(|| GreenNode::from_head_and_children(head, children))
+            .or_insert_with_key(|head| GreenNode::from_head_and_children(head.clone(), new_children))
             .clone()
     }
 
@@ -177,7 +185,10 @@ where
         let text_len = TextSize::try_from(text.len()).unwrap();
         let text = self.interner.get_or_intern(text);
         let data = GreenTokenData { kind, text, text_len };
-        self.tokens.entry(data).or_insert_with(|| GreenToken::new(data)).clone()
+        self.tokens
+            .entry(data)
+            .or_insert_with_key(|data| GreenToken::new(*data))
+            .clone()
     }
 }
 

src/green/node.rs

@@ -15,29 +15,9 @@ use triomphe::{Arc, HeaderWithLength, ThinArc};
 #[repr(align(2))] //to use 1 bit for pointer tagging. NB: this is an at-least annotation
 #[derive(Debug, Clone, PartialEq, Eq, Hash)]
 pub(super) struct GreenNodeHead {
-    kind:       SyntaxKind,
+    pub(super) kind:       SyntaxKind,
-    text_len:   TextSize,
+    pub(super) text_len:   TextSize,
-    child_hash: u32,
+    pub(super) child_hash: u32,
-}
-
-impl GreenNodeHead {
-    #[inline]
-    pub(super) fn from_child_iter<I>(kind: SyntaxKind, children: I) -> Self
-    where
-        I: Iterator<Item = GreenElement>,
-    {
-        let mut hasher = FxHasher32::default();
-        let mut text_len: TextSize = 0.into();
-        for child in children {
-            text_len += child.text_len();
-            child.hash(&mut hasher);
-        }
-        Self {
-            kind,
-            text_len,
-            child_hash: hasher.finish() as u32,
-        }
-    }
 }
 
 /// Internal node in the immutable "green" tree.

src/syntax/resolved.rs

@@ -22,7 +22,6 @@ use crate::{
 /// # See also
 /// [`SyntaxNode`]
 /// [`SyntaxNode::new_root_with_resolver`]
-#[derive(Clone)]
 #[repr(transparent)]
 pub struct ResolvedNode<L: Language, D: 'static = ()> {
     pub(super) syntax: SyntaxNode<L, D>,
@@ -41,6 +40,14 @@ impl<L: Language, D> ResolvedNode<L, D> {
     }
 }
 
+impl<L: Language, D> Clone for ResolvedNode<L, D> {
+    fn clone(&self) -> Self {
+        Self {
+            syntax: self.syntax.clone(),
+        }
+    }
+}
+
 impl<L: Language, D> Deref for ResolvedNode<L, D> {
     type Target = SyntaxNode<L, D>;
 
@@ -77,6 +84,14 @@ impl<L: Language, D> ResolvedToken<L, D> {
     }
 }
 
+impl<L: Language, D> Clone for ResolvedToken<L, D> {
+    fn clone(&self) -> Self {
+        Self {
+            syntax: self.syntax.clone(),
+        }
+    }
+}
+
 impl<L: Language, D> Deref for ResolvedToken<L, D> {
     type Target = SyntaxToken<L, D>;