|
1
|
""" |
|
2
|
AST optimization utilities for navegador ingestion. |
|
3
|
|
|
4
|
Provides four independent classes that can be composed to accelerate |
|
5
|
re-ingestion of large repositories: |
|
6
|
|
|
7
|
TreeCache — LRU cache for parsed tree-sitter trees (#42) |
|
8
|
IncrementalParser — Wraps tree-sitter parse() with old_tree support (#43) |
|
9
|
GraphDiffer — Node-level diffing to skip unchanged writes (#44) |
|
10
|
ParallelIngester — ThreadPoolExecutor wrapper around RepoIngester (#45) |
|
11
|
""" |
|
12
|
|
|
13
|
from __future__ import annotations |
|
14
|
|
|
15
|
import concurrent.futures |
|
16
|
import logging |
|
17
|
import threading |
|
18
|
from collections import OrderedDict |
|
19
|
from dataclasses import dataclass, field |
|
20
|
from pathlib import Path |
|
21
|
from typing import Any |
|
22
|
|
|
23
|
logger = logging.getLogger(__name__) |
|
24
|
|
|
25
|
|
|
26
|
# ── #42 — LRU cache for parsed trees ───────────────────────────────────────── |
|
27
|
|
|
28
|
|
|
29
|
class TreeCache: |
|
30
|
""" |
|
31
|
Thread-safe LRU cache that maps ``(path, content_hash)`` to a parsed |
|
32
|
tree-sitter tree. |
|
33
|
|
|
34
|
Args: |
|
35
|
max_size: Maximum number of trees to hold in memory. When the cache |
|
36
|
is full the least-recently-used entry is evicted. |
|
37
|
""" |
|
38
|
|
|
39
|
def __init__(self, max_size: int = 256) -> None: |
|
40
|
if max_size < 1: |
|
41
|
raise ValueError("max_size must be >= 1") |
|
42
|
self._max_size = max_size |
|
43
|
# OrderedDict used as an LRU store: most-recently used is at the end. |
|
44
|
self._cache: OrderedDict[tuple[str, str], Any] = OrderedDict() |
|
45
|
self._hits = 0 |
|
46
|
self._misses = 0 |
|
47
|
self._lock = threading.Lock() |
|
48
|
|
|
49
|
# ── public API ──────────────────────────────────────────────────────────── |
|
50
|
|
|
51
|
def get(self, path: str, content_hash: str) -> Any | None: |
|
52
|
"""Return the cached tree or ``None`` on a cache miss.""" |
|
53
|
key = (path, content_hash) |
|
54
|
with self._lock: |
|
55
|
if key in self._cache: |
|
56
|
self._hits += 1 |
|
57
|
# Move to the end (most recently used). |
|
58
|
self._cache.move_to_end(key) |
|
59
|
return self._cache[key] |
|
60
|
self._misses += 1 |
|
61
|
return None |
|
62
|
|
|
63
|
def put(self, path: str, content_hash: str, tree: Any) -> None: |
|
64
|
"""Insert (or update) a tree in the cache, evicting LRU entry if full.""" |
|
65
|
key = (path, content_hash) |
|
66
|
with self._lock: |
|
67
|
if key in self._cache: |
|
68
|
self._cache.move_to_end(key) |
|
69
|
self._cache[key] = tree |
|
70
|
else: |
|
71
|
if len(self._cache) >= self._max_size: |
|
72
|
# Evict oldest entry (front of the OrderedDict). |
|
73
|
self._cache.popitem(last=False) |
|
74
|
self._cache[key] = tree |
|
75
|
|
|
76
|
def clear(self) -> None: |
|
77
|
"""Remove all cached trees and reset statistics.""" |
|
78
|
with self._lock: |
|
79
|
self._cache.clear() |
|
80
|
self._hits = 0 |
|
81
|
self._misses = 0 |
|
82
|
|
|
83
|
def stats(self) -> dict[str, int]: |
|
84
|
"""Return a snapshot of cache statistics.""" |
|
85
|
with self._lock: |
|
86
|
return { |
|
87
|
"hits": self._hits, |
|
88
|
"misses": self._misses, |
|
89
|
"size": len(self._cache), |
|
90
|
"max_size": self._max_size, |
|
91
|
} |
|
92
|
|
|
93
|
# ── dunder helpers ──────────────────────────────────────────────────────── |
|
94
|
|
|
95
|
def __len__(self) -> int: |
|
96
|
with self._lock: |
|
97
|
return len(self._cache) |
|
98
|
|
|
99
|
|
|
100
|
# ── #43 — Incremental re-parsing via tree-sitter old_tree API ──────────────── |
|
101
|
|
|
102
|
|
|
103
|
class IncrementalParser: |
|
104
|
""" |
|
105
|
Wraps tree-sitter's ``parser.parse()`` to pass ``old_tree`` when a |
|
106
|
previously-parsed tree for the same path is available. |
|
107
|
|
|
108
|
Parsed trees are stored in a :class:`TreeCache` so subsequent calls for |
|
109
|
an unchanged file return instantly without hitting the tree-sitter C |
|
110
|
extension. |
|
111
|
|
|
112
|
Args: |
|
113
|
cache: A :class:`TreeCache` instance. A default cache is created if |
|
114
|
none is provided. |
|
115
|
""" |
|
116
|
|
|
117
|
def __init__(self, cache: TreeCache | None = None) -> None: |
|
118
|
self._cache = cache if cache is not None else TreeCache() |
|
119
|
|
|
120
|
def parse( |
|
121
|
self, |
|
122
|
source_bytes: bytes, |
|
123
|
language: Any, |
|
124
|
path: str, |
|
125
|
content_hash: str, |
|
126
|
) -> Any: |
|
127
|
""" |
|
128
|
Parse *source_bytes* using *language* and return the tree. |
|
129
|
|
|
130
|
If a cached tree exists for *(path, content_hash)* it is returned |
|
131
|
immediately. Otherwise the most recent tree for *path* (with a |
|
132
|
different hash, if any) is retrieved from the cache and passed as |
|
133
|
``old_tree`` to ``language.parser.parse()`` to enable incremental |
|
134
|
parsing, then the new tree is stored. |
|
135
|
|
|
136
|
Args: |
|
137
|
source_bytes: Raw UTF-8 source code. |
|
138
|
language: A tree-sitter ``Language`` object (or mock in tests). |
|
139
|
path: Repository-relative path, used as cache key. |
|
140
|
content_hash: Content hash, used as cache key. |
|
141
|
|
|
142
|
Returns: |
|
143
|
A parsed tree-sitter ``Tree``. |
|
144
|
""" |
|
145
|
# Fast path: tree for this exact content is already cached. |
|
146
|
cached = self._cache.get(path, content_hash) |
|
147
|
if cached is not None: |
|
148
|
return cached |
|
149
|
|
|
150
|
# Look for a stale tree for this path to use as old_tree. |
|
151
|
old_tree = self._get_stale_tree(path) |
|
152
|
|
|
153
|
# Build a parser using the language object. tree-sitter parsers are |
|
154
|
# instantiated differently depending on version; we support both the |
|
155
|
# legacy API (tree_sitter.Parser) and the new Language.parser attribute. |
|
156
|
try: |
|
157
|
import tree_sitter # type: ignore[import] |
|
158
|
|
|
159
|
parser = tree_sitter.Parser() |
|
160
|
parser.set_language(language) |
|
161
|
except Exception: |
|
162
|
# Fallback: language might already be a parser-like object. |
|
163
|
parser = language |
|
164
|
|
|
165
|
if old_tree is not None: |
|
166
|
tree = parser.parse(source_bytes, old_tree) |
|
167
|
else: |
|
168
|
tree = parser.parse(source_bytes) |
|
169
|
|
|
170
|
self._cache.put(path, content_hash, tree) |
|
171
|
return tree |
|
172
|
|
|
173
|
# ── internal helpers ────────────────────────────────────────────────────── |
|
174
|
|
|
175
|
def _get_stale_tree(self, path: str) -> Any | None: |
|
176
|
"""Return any cached tree for *path* regardless of hash, or ``None``.""" |
|
177
|
with self._cache._lock: |
|
178
|
for (cached_path, _), tree in self._cache._cache.items(): |
|
179
|
if cached_path == path: |
|
180
|
return tree |
|
181
|
return None |
|
182
|
|
|
183
|
@property |
|
184
|
def cache(self) -> TreeCache: |
|
185
|
return self._cache |
|
186
|
|
|
187
|
|
|
188
|
# ── #44 — Graph node diffing ────────────────────────────────────────────────── |
|
189
|
|
|
190
|
|
|
191
|
@dataclass |
|
192
|
class DiffResult: |
|
193
|
"""Summary of a node-level diff for one file.""" |
|
194
|
|
|
195
|
added: int = 0 |
|
196
|
modified: int = 0 |
|
197
|
unchanged: int = 0 |
|
198
|
removed: int = 0 |
|
199
|
|
|
200
|
@property |
|
201
|
def total_changes(self) -> int: |
|
202
|
return self.added + self.modified + self.removed |
|
203
|
|
|
204
|
|
|
205
|
@dataclass |
|
206
|
class NodeDescriptor: |
|
207
|
"""Minimal, hashable description of a graph node used for comparison.""" |
|
208
|
|
|
209
|
label: str |
|
210
|
name: str |
|
211
|
line_start: int |
|
212
|
# Extra properties that contribute to the "modified" check. |
|
213
|
extra: dict[str, Any] = field(default_factory=dict) |
|
214
|
|
|
215
|
def identity_key(self) -> tuple[str, str, int]: |
|
216
|
return (self.label, self.name, self.line_start) |
|
217
|
|
|
218
|
def __eq__(self, other: object) -> bool: |
|
219
|
if not isinstance(other, NodeDescriptor): |
|
220
|
return NotImplemented |
|
221
|
return self.identity_key() == other.identity_key() and self.extra == other.extra |
|
222
|
|
|
223
|
|
|
224
|
class GraphDiffer: |
|
225
|
""" |
|
226
|
Compares newly-parsed nodes against what is already stored in the graph |
|
227
|
so that only genuinely changed nodes need to be written. |
|
228
|
|
|
229
|
Args: |
|
230
|
store: A :class:`~navegador.graph.store.GraphStore` instance. |
|
231
|
""" |
|
232
|
|
|
233
|
def __init__(self, store: Any) -> None: |
|
234
|
self._store = store |
|
235
|
|
|
236
|
# ── public API ──────────────────────────────────────────────────────────── |
|
237
|
|
|
238
|
def diff_file( |
|
239
|
self, |
|
240
|
file_path: str, |
|
241
|
new_nodes: list[NodeDescriptor], |
|
242
|
) -> DiffResult: |
|
243
|
""" |
|
244
|
Compare *new_nodes* against graph nodes currently stored for |
|
245
|
*file_path*. |
|
246
|
|
|
247
|
Args: |
|
248
|
file_path: Repository-relative path of the file being re-parsed. |
|
249
|
new_nodes: Nodes produced by the latest parse pass. |
|
250
|
|
|
251
|
Returns: |
|
252
|
A :class:`DiffResult` with counts of added / modified / unchanged / |
|
253
|
removed nodes. |
|
254
|
""" |
|
255
|
existing_nodes = self._fetch_existing_nodes(file_path) |
|
256
|
|
|
257
|
existing_by_key: dict[tuple[str, str, int], NodeDescriptor] = { |
|
258
|
n.identity_key(): n for n in existing_nodes |
|
259
|
} |
|
260
|
new_by_key: dict[tuple[str, str, int], NodeDescriptor] = { |
|
261
|
n.identity_key(): n for n in new_nodes |
|
262
|
} |
|
263
|
|
|
264
|
result = DiffResult() |
|
265
|
|
|
266
|
for key, new_node in new_by_key.items(): |
|
267
|
if key not in existing_by_key: |
|
268
|
result.added += 1 |
|
269
|
elif new_node != existing_by_key[key]: |
|
270
|
result.modified += 1 |
|
271
|
else: |
|
272
|
result.unchanged += 1 |
|
273
|
|
|
274
|
for key in existing_by_key: |
|
275
|
if key not in new_by_key: |
|
276
|
result.removed += 1 |
|
277
|
|
|
278
|
return result |
|
279
|
|
|
280
|
# ── internal helpers ────────────────────────────────────────────────────── |
|
281
|
|
|
282
|
def _fetch_existing_nodes(self, file_path: str) -> list[NodeDescriptor]: |
|
283
|
""" |
|
284
|
Query the graph for all nodes associated with *file_path* and return |
|
285
|
them as :class:`NodeDescriptor` objects. |
|
286
|
|
|
287
|
The query returns rows of ``[label, name, line_start]``. Any row |
|
288
|
where *name* or *line_start* is ``None`` is skipped. |
|
289
|
""" |
|
290
|
cypher = ( |
|
291
|
"MATCH (n {file_path: $file_path}) " |
|
292
|
"RETURN labels(n)[0] AS label, n.name AS name, n.line_start AS line_start" |
|
293
|
) |
|
294
|
result = self._store.query(cypher, {"file_path": file_path}) |
|
295
|
rows = result.result_set or [] |
|
296
|
nodes: list[NodeDescriptor] = [] |
|
297
|
for row in rows: |
|
298
|
label, name, line_start = row[0], row[1], row[2] |
|
299
|
if name is None or line_start is None: |
|
300
|
continue |
|
301
|
nodes.append( |
|
302
|
NodeDescriptor(label=str(label), name=str(name), line_start=int(line_start)) |
|
303
|
) |
|
304
|
return nodes |
|
305
|
|
|
306
|
|
|
307
|
# ── #45 — Parallel ingestion with worker pool ───────────────────────────────── |
|
308
|
|
|
309
|
|
|
310
|
class ParallelIngester: |
|
311
|
""" |
|
312
|
Processes repository files concurrently using a |
|
313
|
:class:`concurrent.futures.ThreadPoolExecutor`. |
|
314
|
|
|
315
|
The class wraps a :class:`~navegador.ingestion.parser.RepoIngester` and |
|
316
|
overrides its sequential file-iteration loop with a parallel one while |
|
317
|
reusing all of its parsing, hashing and graph-writing logic. |
|
318
|
|
|
319
|
Args: |
|
320
|
store: Graph store passed through to the underlying |
|
321
|
:class:`~navegador.ingestion.parser.RepoIngester`. |
|
322
|
redact: Enable sensitive-content redaction. |
|
323
|
""" |
|
324
|
|
|
325
|
def __init__(self, store: Any, redact: bool = False) -> None: |
|
326
|
from navegador.ingestion.parser import RepoIngester |
|
327
|
|
|
328
|
self._store = store |
|
329
|
self._redact = redact |
|
330
|
self._ingester = RepoIngester(store, redact=redact) |
|
331
|
|
|
332
|
def ingest_parallel( |
|
333
|
self, |
|
334
|
repo_path: str | Path, |
|
335
|
max_workers: int | None = None, |
|
336
|
clear: bool = False, |
|
337
|
incremental: bool = False, |
|
338
|
) -> dict[str, int]: |
|
339
|
""" |
|
340
|
Ingest a repository, parsing files concurrently. |
|
341
|
|
|
342
|
Args: |
|
343
|
repo_path: Path to the repository root. |
|
344
|
max_workers: Number of worker threads. Defaults to |
|
345
|
``min(32, cpu_count + 4)`` via |
|
346
|
:class:`~concurrent.futures.ThreadPoolExecutor`. |
|
347
|
clear: Wipe the graph before ingesting. |
|
348
|
incremental: Skip files whose content hash hasn't changed. |
|
349
|
|
|
350
|
Returns: |
|
351
|
Aggregated stats dict with keys: files, functions, classes, |
|
352
|
edges, skipped, errors. |
|
353
|
""" |
|
354
|
from navegador.graph.schema import NodeLabel |
|
355
|
from navegador.ingestion.parser import LANGUAGE_MAP, _file_hash |
|
356
|
|
|
357
|
repo_path = Path(repo_path).resolve() |
|
358
|
if not repo_path.exists(): |
|
359
|
raise FileNotFoundError(f"Repository not found: {repo_path}") |
|
360
|
|
|
361
|
if clear: |
|
362
|
self._store.clear() |
|
363
|
|
|
364
|
# Repository node (same as RepoIngester.ingest). |
|
365
|
self._store.create_node( |
|
366
|
NodeLabel.Repository, |
|
367
|
{ |
|
368
|
"name": repo_path.name, |
|
369
|
"path": str(repo_path), |
|
370
|
}, |
|
371
|
) |
|
372
|
|
|
373
|
# Collect all candidate files up-front (fast, single-threaded). |
|
374
|
candidate_files = [ |
|
375
|
f for f in self._ingester._iter_source_files(repo_path) if LANGUAGE_MAP.get(f.suffix) |
|
376
|
] |
|
377
|
|
|
378
|
aggregated: dict[str, int] = { |
|
379
|
"files": 0, |
|
380
|
"functions": 0, |
|
381
|
"classes": 0, |
|
382
|
"edges": 0, |
|
383
|
"skipped": 0, |
|
384
|
"errors": 0, |
|
385
|
} |
|
386
|
lock = threading.Lock() |
|
387
|
|
|
388
|
def _process_file(source_file: Path) -> None: |
|
389
|
language = LANGUAGE_MAP[source_file.suffix] |
|
390
|
rel_path = str(source_file.relative_to(repo_path)) |
|
391
|
content_hash = _file_hash(source_file) |
|
392
|
|
|
393
|
if incremental and self._ingester._file_unchanged(rel_path, content_hash): |
|
394
|
with lock: |
|
395
|
aggregated["skipped"] += 1 |
|
396
|
return |
|
397
|
|
|
398
|
if incremental: |
|
399
|
self._ingester._clear_file_subgraph(rel_path) |
|
400
|
|
|
401
|
parse_path, effective_root = self._ingester._maybe_redact_to_tmp(source_file, repo_path) |
|
402
|
try: |
|
403
|
parser = self._ingester._get_parser(language) |
|
404
|
file_stats = parser.parse_file(parse_path, effective_root, self._store) |
|
405
|
self._ingester._store_file_hash(rel_path, content_hash) |
|
406
|
with lock: |
|
407
|
aggregated["files"] += 1 |
|
408
|
aggregated["functions"] += file_stats.get("functions", 0) |
|
409
|
aggregated["classes"] += file_stats.get("classes", 0) |
|
410
|
aggregated["edges"] += file_stats.get("edges", 0) |
|
411
|
except Exception: |
|
412
|
logger.exception("Failed to parse %s", source_file) |
|
413
|
with lock: |
|
414
|
aggregated["errors"] += 1 |
|
415
|
finally: |
|
416
|
import shutil |
|
417
|
|
|
418
|
if effective_root is not repo_path: |
|
419
|
shutil.rmtree(effective_root, ignore_errors=True) |
|
420
|
|
|
421
|
with concurrent.futures.ThreadPoolExecutor(max_workers=max_workers) as executor: |
|
422
|
futures = {executor.submit(_process_file, f): f for f in candidate_files} |
|
423
|
for future in concurrent.futures.as_completed(futures): |
|
424
|
# Exceptions are already caught inside _process_file; this |
|
425
|
# re-raises any unexpected ones that slipped through. |
|
426
|
future.result() |
|
427
|
|
|
428
|
logger.info( |
|
429
|
"ParallelIngester finished %s: %d files, %d functions, %d skipped, %d errors", |
|
430
|
repo_path.name, |
|
431
|
aggregated["files"], |
|
432
|
aggregated["functions"], |
|
433
|
aggregated["skipped"], |
|
434
|
aggregated["errors"], |
|
435
|
) |
|
436
|
return aggregated |
|
437
|
|