Spaces:
Sleeping
Sleeping
File size: 9,049 Bytes
98aa8c9 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 | """Chunk ingested content into retrieval units.
Reads data/manifest.json + files produced by ingest.py.
Writes data/chunks.jsonl — one JSON chunk per line.
Chunk schema (same for all source types):
id stable unique id
text what gets embedded (includes a context header)
source_type code | doc | issue
path file path in the repo (or issues/N.json)
symbol function/class name, doc heading, or issue title
url link to the exact GitHub location
start_line / end_line (code and docs only)
Usage:
python src/chunk.py
"""
import json
import subprocess
from pathlib import Path
import tree_sitter_python as tspython
from tree_sitter import Language, Parser
DATA_DIR = Path("data")
REPO_DIR = DATA_DIR / "repo"
MAX_CLASS_LINES = 150 # classes bigger than this get split into methods
MIN_CHUNK_CHARS = 50 # drop trivial fragments
MAX_ISSUE_BODY = 3000
MAX_COMMENT = 1000
PY_LANGUAGE = Language(tspython.language())
PARSER = Parser(PY_LANGUAGE)
# ---------------------------------------------------------------- helpers
def default_branch() -> str:
out = subprocess.run(["git", "-C", str(REPO_DIR), "rev-parse",
"--abbrev-ref", "HEAD"],
capture_output=True, text=True)
return out.stdout.strip() or "master"
def github_url(repo: str, branch: str, path: str,
start: int | None = None, end: int | None = None) -> str:
url = f"https://github.com/{repo}/blob/{branch}/{path}"
if start is not None:
url += f"#L{start}-L{end}"
return url
# ---------------------------------------------------------------- code
def chunk_code_file(source: bytes, rel_path: str, repo: str,
branch: str) -> list[dict]:
"""AST-aware chunking: one chunk per top-level function/class.
Large classes are split into per-method chunks. Module-level leftovers
(imports, constants) become one '<module>' chunk.
"""
tree = PARSER.parse(source)
text_lines = source.decode(errors="replace").splitlines()
chunks: list[dict] = []
covered: set[int] = set() # 1-based line numbers already in a chunk
def node_text(node) -> str:
return source[node.start_byte:node.end_byte].decode(errors="replace")
def emit(node, symbol: str, context: str = "") -> None:
start, end = node.start_point[0] + 1, node.end_point[0] + 1
header = f"# File: {rel_path}"
if context:
header += f" | {context}"
header += f" | Symbol: {symbol}\n"
body = node_text(node)
if len(body) < MIN_CHUNK_CHARS:
covered.update(range(start, end + 1))
return
chunks.append({
"id": f"{rel_path}::{symbol}::L{start}",
"text": header + body,
"source_type": "code",
"path": rel_path,
"symbol": symbol,
"url": github_url(repo, branch, rel_path, start, end),
"start_line": start,
"end_line": end,
})
covered.update(range(start, end + 1))
def unwrap(node):
"""decorated_definition wraps the real def; we want the decorator
text included but the inner node's name."""
if node.type == "decorated_definition":
for child in node.children:
if child.type in ("function_definition", "class_definition"):
return child, node
return node, node
def symbol_name(def_node) -> str:
name = def_node.child_by_field_name("name")
return name.text.decode() if name else "?"
for node in tree.root_node.children:
inner, outer = unwrap(node)
if inner.type == "function_definition":
emit(outer, symbol_name(inner))
elif inner.type == "class_definition":
cls_name = symbol_name(inner)
n_lines = inner.end_point[0] - inner.start_point[0] + 1
if n_lines <= MAX_CLASS_LINES:
emit(outer, cls_name)
else:
# Big class: one chunk per method
body = inner.child_by_field_name("body")
for child in (body.children if body else []):
m_inner, m_outer = unwrap(child)
if m_inner.type == "function_definition":
emit(m_outer, f"{cls_name}.{symbol_name(m_inner)}",
context=f"Class: {cls_name}")
# Module-level leftovers: imports, constants, top-level statements
leftover = [i for i in range(1, len(text_lines) + 1)
if i not in covered and text_lines[i - 1].strip()]
if leftover:
body = "\n".join(text_lines[i - 1] for i in leftover)
if len(body) >= MIN_CHUNK_CHARS:
start, end = leftover[0], leftover[-1]
chunks.append({
"id": f"{rel_path}::<module>",
"text": f"# File: {rel_path} | Symbol: <module>\n{body}",
"source_type": "code",
"path": rel_path,
"symbol": "<module>",
"url": github_url(repo, branch, rel_path, start, end),
"start_line": start,
"end_line": end,
})
return chunks
# ---------------------------------------------------------------- docs
def chunk_markdown_file(text: str, rel_path: str, repo: str,
branch: str) -> list[dict]:
"""Split on '## ' headings; each section keeps the page title as context."""
lines = text.splitlines()
page_title = next((l.lstrip("# ").strip() for l in lines
if l.startswith("# ")), rel_path)
sections: list[tuple[str, int]] = [] # (heading, start_line 1-based)
for i, line in enumerate(lines):
if line.startswith("## "):
sections.append((line.lstrip("# ").strip(), i + 1))
if not sections or sections[0][1] > 1:
sections.insert(0, (page_title, 1))
chunks = []
for idx, (heading, start) in enumerate(sections):
end = sections[idx + 1][1] - 1 if idx + 1 < len(sections) else len(lines)
body = "\n".join(lines[start - 1:end]).strip()
if len(body) < MIN_CHUNK_CHARS:
continue
header = f"# Doc: {rel_path} | Page: {page_title} | Section: {heading}\n"
chunks.append({
"id": f"{rel_path}::{heading}::L{start}",
"text": header + body,
"source_type": "doc",
"path": rel_path,
"symbol": heading,
"url": github_url(repo, branch, rel_path, start, end),
"start_line": start,
"end_line": end,
})
return chunks
# ---------------------------------------------------------------- issues
def chunk_issue(record: dict) -> dict | None:
"""One chunk per issue: title + body + top comments."""
parts = [f"# Issue #{record['number']}: {record['title']}"]
if record.get("labels"):
parts.append(f"Labels: {', '.join(record['labels'])}")
if record.get("body"):
parts.append(record["body"][:MAX_ISSUE_BODY])
for c in record.get("comments", []):
parts.append(f"---\nComment: {c[:MAX_COMMENT]}")
text = "\n".join(parts)
if len(text) < MIN_CHUNK_CHARS:
return None
return {
"id": f"issue::{record['number']}",
"text": text,
"source_type": "issue",
"path": f"issues/{record['number']}.json",
"symbol": record["title"],
"url": record["url"],
"start_line": None,
"end_line": None,
}
# ---------------------------------------------------------------- main
def main() -> None:
manifest = json.loads((DATA_DIR / "manifest.json").read_text())
repo = manifest["repo"]
branch = default_branch()
all_chunks: list[dict] = []
for rec in manifest["files"]:
path = REPO_DIR / rec["path"]
if rec["source_type"] == "code":
all_chunks += chunk_code_file(path.read_bytes(), rec["path"],
repo, branch)
elif rec["source_type"] == "doc":
all_chunks += chunk_markdown_file(
path.read_text(errors="replace"), rec["path"], repo, branch)
for issue_file in sorted((DATA_DIR / "issues").glob("*.json")):
chunk = chunk_issue(json.loads(issue_file.read_text()))
if chunk:
all_chunks.append(chunk)
out = DATA_DIR / "chunks.jsonl"
with out.open("w", encoding="utf-8") as f:
for c in all_chunks:
f.write(json.dumps(c, ensure_ascii=False) + "\n")
counts: dict[str, int] = {}
sizes = []
for c in all_chunks:
counts[c["source_type"]] = counts.get(c["source_type"], 0) + 1
sizes.append(len(c["text"]))
print(f"[done] {len(all_chunks)} chunks -> {out}")
print(f"[done] by type: {counts}")
print(f"[done] chars/chunk: min={min(sizes)} "
f"avg={sum(sizes)//len(sizes)} max={max(sizes)}")
if __name__ == "__main__":
main()
|