---
id: wiki-2026-0508-concrete-syntax-tree-cst
title: Concrete Syntax Tree (CST)
category: 10_Wiki/Topics
status: verified
canonical_id: self
aliases: [Parse Tree, CST, Lossless Syntax Tree]
duplicate_of: none
source_trust_level: A
confidence_score: 0.9
verification_status: applied
tags: [parser, ast, tooling, language-engineering]
raw_sources: []
last_reinforced: 2026-05-10
github_commit: pending
tech_stack:
  language: Rust/JS
  framework: tree-sitter/rowan
---

# Concrete Syntax Tree (CST)

## 매 한 줄
> **"매 source 의 every token (whitespace, comment 포함) 까지 보존하는 lossless tree"**. AST 가 semantic-only 인 반면 CST 는 매 source 의 round-trip 가능. 매 modern tooling — tree-sitter, rowan (rust-analyzer), Roslyn — 매 CST 위에서 매 IDE feature, refactoring, formatter 를 매 build.

## 매 핵심

### 매 vs AST
- **AST**: semantic node 만 (`if`, `BinaryOp`, etc) — comment, whitespace 버림.
- **CST**: 모든 token + trivia 보존 — `source == reprint(cst)`.
- CST → AST 의 lowering 가능, 역은 매 lossy.

### 매 핵심 properties
- **Lossless**: print 시 원본 byte-for-byte 복구.
- **Error-tolerant**: incomplete/invalid code 도 partial tree.
- **Incremental**: edit 시 affected subtree 만 reparse (tree-sitter).
- **Untyped or weakly-typed**: 모든 node 가 동질 — typed wrapper 로 navigate.

### 매 응용
1. IDE: syntax highlight, fold, outline, indent.
2. Refactoring: rename, extract method (preserve formatting).
3. Formatter: prettier, rustfmt — CST 로 layout 결정.
4. Linter: tree-sitter queries.

## 💻 패턴

### tree-sitter parsing
```javascript
const Parser = require('tree-sitter');
const TS = require('tree-sitter-typescript').typescript;
const parser = new Parser();
parser.setLanguage(TS);
const tree = parser.parse('const x: number = 1;');
// Walk
const cursor = tree.walk();
do { console.log(cursor.nodeType, cursor.startIndex, cursor.endIndex); }
while (cursor.gotoNextSibling() || cursor.gotoFirstChild());
```

### tree-sitter query (S-expression)
```scheme
; Find all function declarations
(function_declaration
  name: (identifier) @func.name
  parameters: (formal_parameters) @func.params)

; Find unused imports
(import_statement
  source: (string) @import.source) @import
```

### Incremental edit
```javascript
const oldTree = parser.parse(oldSrc);
const newSrc = oldSrc.slice(0, 10) + 'INSERTED' + oldSrc.slice(10);
oldTree.edit({
  startIndex: 10, oldEndIndex: 10, newEndIndex: 18,
  startPosition: {row: 0, column: 10},
  oldEndPosition: {row: 0, column: 10},
  newEndPosition: {row: 0, column: 18},
});
const newTree = parser.parse(newSrc, oldTree);  // reuses unchanged subtrees
```

### rowan (rust-analyzer) typed wrapper
```rust
// Untyped GreenNode + typed SyntaxNode wrapper
use rowan::{GreenNode, SyntaxNode};

#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
#[repr(u16)]
enum SyntaxKind { L_PAREN, R_PAREN, IDENT, FN_KW, FN_DEF, ROOT, /* ... */ }

struct FnDef(SyntaxNode<MyLang>);
impl FnDef {
    fn name(&self) -> Option<String> {
        self.0.children().find(|n| n.kind() == SyntaxKind::IDENT)
              .map(|n| n.text().to_string())
    }
}
```

### Lossless rewrite (rename)
```rust
// Replace token in CST and reprint — preserves comments/whitespace.
fn rename(node: &SyntaxNode, old: &str, new: &str) -> String {
    let mut out = String::new();
    for tok in node.descendants_with_tokens() {
        if let Some(t) = tok.as_token() {
            if t.kind() == SyntaxKind::IDENT && t.text() == old { out.push_str(new); }
            else { out.push_str(t.text()); }
        }
    }
    out
}
```

## 매 결정 기준
| 상황 | Approach |
|---|---|
| Compiler / type checker | AST (semantic 충분) |
| IDE / LSP / formatter | CST (lossless 필수) |
| Refactoring tool | CST + typed wrapper |
| Quick analysis script | tree-sitter query |
| New language design | rowan or tree-sitter base |

**기본값**: User-facing tool 이면 매 CST. Compiler internal pass 면 매 AST.

## 🔗 Graph
- 부모: [[Parser]]
- 변형: [[AST]]
- 응용: [[Prettier]]

## 🤖 LLM 활용
**언제**: Source-to-source transformation, IDE-grade tooling, codemods.
**언제 X**: Pure semantic analysis (AST 만 충분).

## ❌ 안티패턴
- **Regex on source**: 매 fragile — CST query 사용.
- **AST 로 formatter**: comment 손실 — CST 필수.
- **Hand-rolled parser**: error recovery 빠짐 — tree-sitter/lark 사용.
- **Full reparse on every keystroke**: incremental edit API 사용.

## 🧪 검증 / 중복
- Verified (tree-sitter docs, rust-analyzer rowan, Roslyn architecture).
- 신뢰도 A.

## 🕓 Changelog
| 날짜 | 변경 |
|---|---|
| 2026-05-08 | Phase 1 |
| 2026-05-10 | Manual cleanup — CST vs AST + tree-sitter/rowan patterns |