mlai-code 0.3.3 → 0.3.5

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (136) hide show
  1. package/README.md +1 -1
  2. package/package.json +1 -1
  3. package/src/checkpoints.mjs +217 -0
  4. package/src/cli.mjs +153 -26
  5. package/src/cloud.mjs +2 -0
  6. package/src/engine.mjs +1 -1
  7. package/src/headless.mjs +39 -11
  8. package/src/home.mjs +3 -2
  9. package/src/host.mjs +141 -13
  10. package/src/ide-quick.mjs +147 -0
  11. package/src/ide.mjs +182 -0
  12. package/src/netbox.mjs +1 -1
  13. package/src/parity.mjs +17 -0
  14. package/src/relay.mjs +1 -1
  15. package/src/render.mjs +16 -5
  16. package/src/runtime.mjs +2 -0
  17. package/src/server.mjs +46 -7
  18. package/src/session.mjs +31 -8
  19. package/src/tui/acp.mjs +121 -19
  20. package/src/tui/app.mjs +2 -1
  21. package/src/tui/chat.mjs +35 -10
  22. package/src/tui/cmd-config.mjs +3 -1
  23. package/src/tui/cmd-core.mjs +9 -5
  24. package/src/tui/cmd-extensibility.mjs +5 -0
  25. package/src/tui/extensions-local.mjs +2 -0
  26. package/src/tui/mcp-config.mjs +22 -4
  27. package/src/tui/review-cli.mjs +15 -4
  28. package/src/tui/snapshots.mjs +57 -16
  29. package/src/tui/transcript.mjs +13 -2
  30. package/src/two-words.mjs +14 -8
  31. package/src/view-sync.mjs +48 -0
  32. package/src/winenv.mjs +1 -1
  33. package/src/workspace-api.mjs +21 -4
  34. package/vendor/client-sdk/index.cjs +2 -1
  35. package/vendor/cloud-cli/mlai.mjs +23 -11
  36. package/worker/architecture_worker.py +33 -4
  37. package/worker/browser_tool.py +19 -1
  38. package/worker/chat_fork.py +4 -1
  39. package/worker/checkpoints.py +71 -8
  40. package/worker/command_safety.py +56 -3
  41. package/worker/connected_turn.py +30 -0
  42. package/worker/conversation.py +125 -13
  43. package/worker/git_sync.py +4 -4
  44. package/worker/image_generation.py +2 -2
  45. package/worker/kimi_bridge.py +251 -31
  46. package/worker/kimi_gates.py +25 -6
  47. package/worker/kimi_hook.py +154 -13
  48. package/worker/media_tools.py +4 -1
  49. package/worker/mlai_hooks.py +222 -3
  50. package/worker/mlai_mcp.py +26 -6
  51. package/worker/mlai_review/README.md +35 -0
  52. package/worker/mlai_review/__init__.py +1 -0
  53. package/worker/mlai_review/__main__.py +261 -0
  54. package/worker/mlai_review/_compat.py +29 -0
  55. package/worker/mlai_review/analysis/__init__.py +3 -0
  56. package/worker/mlai_review/analysis/detect.py +50 -0
  57. package/worker/mlai_review/analysis/main.py +128 -0
  58. package/worker/mlai_review/analysis/messages.py +8 -0
  59. package/worker/mlai_review/analysis/tools.py +194 -0
  60. package/worker/mlai_review/application/__init__.py +1 -0
  61. package/worker/mlai_review/application/asking.py +53 -0
  62. package/worker/mlai_review/application/challengers.py +142 -0
  63. package/worker/mlai_review/application/context_builder.py +194 -0
  64. package/worker/mlai_review/application/errors.py +17 -0
  65. package/worker/mlai_review/application/feedback.py +52 -0
  66. package/worker/mlai_review/application/matching.py +46 -0
  67. package/worker/mlai_review/application/messages.py +41 -0
  68. package/worker/mlai_review/application/ports.py +222 -0
  69. package/worker/mlai_review/application/prompts.py +342 -0
  70. package/worker/mlai_review/application/review_service.py +200 -0
  71. package/worker/mlai_review/application/reviewers.py +438 -0
  72. package/worker/mlai_review/application/streaming.py +91 -0
  73. package/worker/mlai_review/application/tools.py +138 -0
  74. package/worker/mlai_review/application/usage.py +145 -0
  75. package/worker/mlai_review/application/write_check.py +46 -0
  76. package/worker/mlai_review/contracts/__init__.py +5 -0
  77. package/worker/mlai_review/contracts/analysis.py +57 -0
  78. package/worker/mlai_review/contracts/review.py +42 -0
  79. package/worker/mlai_review/domain/__init__.py +4 -0
  80. package/worker/mlai_review/domain/changes.py +77 -0
  81. package/worker/mlai_review/domain/context.py +147 -0
  82. package/worker/mlai_review/domain/diagnostics.py +239 -0
  83. package/worker/mlai_review/domain/feedback.py +216 -0
  84. package/worker/mlai_review/domain/findings.py +505 -0
  85. package/worker/mlai_review/domain/lifecycle.py +85 -0
  86. package/worker/mlai_review/domain/partition.py +42 -0
  87. package/worker/mlai_review/domain/relations.py +166 -0
  88. package/worker/mlai_review/domain/relevance.py +202 -0
  89. package/worker/mlai_review/domain/report.py +99 -0
  90. package/worker/mlai_review/domain/structure.py +56 -0
  91. package/worker/mlai_review/domain/writes.py +86 -0
  92. package/worker/mlai_review/eval.py +221 -0
  93. package/worker/mlai_review/eval_cases/clean-rename.json +11 -0
  94. package/worker/mlai_review/eval_cases/missing-ownership-check.json +22 -0
  95. package/worker/mlai_review/eval_cases/off-by-one.json +22 -0
  96. package/worker/mlai_review/eval_cases/sql-injection.json +20 -0
  97. package/worker/mlai_review/feedback_report.py +60 -0
  98. package/worker/mlai_review/infrastructure/__init__.py +1 -0
  99. package/worker/mlai_review/infrastructure/analysis/__init__.py +1 -0
  100. package/worker/mlai_review/infrastructure/analysis/none.py +12 -0
  101. package/worker/mlai_review/infrastructure/analysis/translate.py +17 -0
  102. package/worker/mlai_review/infrastructure/github/__init__.py +0 -0
  103. package/worker/mlai_review/infrastructure/github/client.py +445 -0
  104. package/worker/mlai_review/infrastructure/github/feedback.py +212 -0
  105. package/worker/mlai_review/infrastructure/github/http.py +31 -0
  106. package/worker/mlai_review/infrastructure/github/rendering.py +398 -0
  107. package/worker/mlai_review/infrastructure/github/token.py +20 -0
  108. package/worker/mlai_review/infrastructure/local/__init__.py +1 -0
  109. package/worker/mlai_review/infrastructure/local/analyzer.py +48 -0
  110. package/worker/mlai_review/infrastructure/local/events.py +100 -0
  111. package/worker/mlai_review/infrastructure/local/git_reader.py +233 -0
  112. package/worker/mlai_review/infrastructure/local/publisher.py +138 -0
  113. package/worker/mlai_review/infrastructure/messages.py +54 -0
  114. package/worker/mlai_review/infrastructure/models/__init__.py +0 -0
  115. package/worker/mlai_review/infrastructure/models/openai_compatible.py +269 -0
  116. package/worker/mlai_review/infrastructure/structure/__init__.py +1 -0
  117. package/worker/mlai_review/infrastructure/structure/heuristic.py +186 -0
  118. package/worker/mlai_review/infrastructure/structure/tree_sitter.py +186 -0
  119. package/worker/mlai_review/infrastructure/workspace/__init__.py +1 -0
  120. package/worker/mlai_review/infrastructure/workspace/local.py +105 -0
  121. package/worker/mlai_review/infrastructure/workspace/memory.py +77 -0
  122. package/worker/mlai_review/infrastructure/workspace/paths.py +17 -0
  123. package/worker/mlai_review/pipeline.py +89 -0
  124. package/worker/mlai_review/runtime.py +194 -0
  125. package/worker/mlai_review/settings.py +139 -0
  126. package/worker/preview.py +15 -0
  127. package/worker/product_facts.py +3 -1
  128. package/worker/repo_policy.py +5 -0
  129. package/worker/roles.py +29 -2
  130. package/worker/runner.py +164 -23
  131. package/worker/runtime_control.py +95 -9
  132. package/worker/secret_scan.py +3 -2
  133. package/worker/server_memory.py +122 -0
  134. package/worker/subagents.py +6 -2
  135. package/worker/tool_catalog.py +5 -2
  136. package/worker/verification.py +113 -8
@@ -0,0 +1,445 @@
1
+ """GitHub adapters for the orchestrator's ports."""
2
+
3
+ from __future__ import annotations
4
+
5
+ import logging
6
+ from typing import Any, ClassVar
7
+
8
+ import httpx
9
+
10
+ from mlai_review.contracts.review import ReviewPullRequestInput
11
+ from mlai_review.application.errors import ReviewError
12
+ from mlai_review.application.ports import PullRequestReader, ReviewPublisher
13
+ from mlai_review.domain.changes import ChangeSet, FileChange
14
+ from mlai_review.domain.context import PullRequestInfo
15
+ from mlai_review.domain.findings import Finding
16
+ from mlai_review.domain.lifecycle import PriorFinding
17
+ from mlai_review.domain.report import ReviewReport, Verdict
18
+ from mlai_review.infrastructure import messages
19
+ from mlai_review.infrastructure.github import rendering
20
+ from mlai_review.infrastructure.github.token import StaticTokenProvider
21
+ from mlai_review.infrastructure.github.http import github_headers, raise_for_github_status
22
+
23
+ _log = logging.getLogger("mlai.orchestrator.github")
24
+ _PAGE_SIZE = 100
25
+ _MAX_PAGES = 30 # GitHub returns at most 3,000 files for a pull request.
26
+
27
+
28
+ class GitHubPullRequestReader(PullRequestReader):
29
+ _PR: ClassVar[dict[str, str]] = {"pull_requests": "read"}
30
+ _CONTENTS: ClassVar[dict[str, str]] = {"contents": "read"}
31
+
32
+ def __init__(self, *, http: httpx.AsyncClient, tokens: StaticTokenProvider) -> None:
33
+ self._http = http
34
+ self._tokens = tokens
35
+
36
+ async def pull_request(self, target: ReviewPullRequestInput) -> PullRequestInfo:
37
+ response = await self._http.get(
38
+ f"/repos/{target.owner}/{target.repo}/pulls/{target.pull_number}",
39
+ headers=await _auth(self._tokens, target, self._PR),
40
+ )
41
+ raise_for_github_status(
42
+ response, messages.OP_READ_PULL_REQUEST.format(number=target.pull_number)
43
+ )
44
+ body = response.json()
45
+ return PullRequestInfo(title=body.get("title") or "", body=body.get("body") or "")
46
+
47
+ async def head(self, target: ReviewPullRequestInput) -> tuple[str, str]:
48
+ response = await self._http.get(
49
+ f"/repos/{target.owner}/{target.repo}/pulls/{target.pull_number}",
50
+ headers=await _auth(self._tokens, target, self._PR),
51
+ )
52
+ raise_for_github_status(
53
+ response, messages.OP_READ_PULL_REQUEST.format(number=target.pull_number)
54
+ )
55
+ body = response.json()
56
+ return str(body["head"]["sha"]), str(body["base"]["sha"])
57
+
58
+ async def changed_files(
59
+ self, target: ReviewPullRequestInput, *, since: str | None = None
60
+ ) -> ChangeSet:
61
+ if since is not None:
62
+ incremental = await self._changed_since(target, since)
63
+ if incremental is not None:
64
+ return incremental
65
+ headers = await _auth(self._tokens, target, self._PR)
66
+ operation = messages.OP_LIST_FILES.format(repository=target.repository_full_name)
67
+ files: list[FileChange] = []
68
+ for page in range(1, _MAX_PAGES + 1):
69
+ response = await self._http.get(
70
+ f"/repos/{target.owner}/{target.repo}/pulls/{target.pull_number}/files",
71
+ params={"per_page": _PAGE_SIZE, "page": page},
72
+ headers=headers,
73
+ )
74
+ raise_for_github_status(response, operation)
75
+ batch = response.json()
76
+ files += _file_changes(batch)
77
+ if len(batch) < _PAGE_SIZE:
78
+ break
79
+ return ChangeSet(files=tuple(files))
80
+
81
+ async def _changed_since(self, target: ReviewPullRequestInput, since: str) -> ChangeSet | None:
82
+ """The files changed between a commit and the head, or None when the commit is not
83
+ an ancestor of the head (a force push), in which case the whole change is needed."""
84
+ headers = await _auth(self._tokens, target, self._CONTENTS)
85
+ operation = messages.OP_COMPARE.format(since=since[:12], head=target.head_sha[:12])
86
+ files: list[FileChange] = []
87
+ for page in range(1, _MAX_PAGES + 1):
88
+ response = await self._http.get(
89
+ f"/repos/{target.owner}/{target.repo}/compare/{since}...{target.head_sha}",
90
+ params={"per_page": _PAGE_SIZE, "page": page},
91
+ headers=headers,
92
+ )
93
+ if response.status_code == 404:
94
+ return None # the commit is gone from the branch
95
+ raise_for_github_status(response, operation)
96
+ body = response.json()
97
+ if body.get("status") == "diverged":
98
+ return None
99
+ batch = body.get("files") or []
100
+ files += _file_changes(batch)
101
+ if len(batch) < _PAGE_SIZE:
102
+ break
103
+ return ChangeSet(files=tuple(files), since=since)
104
+
105
+
106
+ def _file_changes(batch: list[dict[str, Any]]) -> list[FileChange]:
107
+ return [
108
+ FileChange(
109
+ path=f["filename"],
110
+ status=f["status"],
111
+ additions=f["additions"],
112
+ deletions=f["deletions"],
113
+ patch=f.get("patch"),
114
+ )
115
+ for f in batch
116
+ ]
117
+
118
+
119
+ # The publication id of a review shown as a commit status rather than a check run.
120
+ STATUS_PUBLICATION = 0
121
+ # Neither a check run nor a commit status could be written (a personal token without the Commit
122
+ # statuses permission): the review still runs and posts its comments and summary.
123
+ NO_PUBLICATION = -1
124
+
125
+
126
+ class GitHubReviewPublisher(ReviewPublisher):
127
+ """Publishes to two GitHub surfaces: the check run (always) and a pull request review with
128
+ inline comments (for findings on lines the diff shows)."""
129
+
130
+ _CHECKS: ClassVar[dict[str, str]] = {"checks": "write"}
131
+ _REVIEWS: ClassVar[dict[str, str]] = {"pull_requests": "write"}
132
+
133
+ def __init__(
134
+ self,
135
+ *,
136
+ http: httpx.AsyncClient,
137
+ tokens: StaticTokenProvider,
138
+ web_base_url: str = "https://github.com",
139
+ details_url: str | None = None,
140
+ ) -> None:
141
+ self._http = http
142
+ self._tokens = tokens
143
+ self._web = web_base_url.rstrip("/")
144
+ self._details_url = details_url
145
+ self._login: str | None = None
146
+ self._login_read = False
147
+
148
+ async def _mine(self, target: ReviewPullRequestInput, comment: dict[str, Any]) -> bool:
149
+ """Whether a comment was written by the account this review posts as. Markers alone
150
+ can be copied by anyone who may comment, so a pasted summary or finding marker must
151
+ not steer the review. When the account cannot be read, markers are trusted as before."""
152
+ if not self._login_read:
153
+ self._login_read = True
154
+ response = await self._http.get("/user", headers=await _auth(self._tokens, target, {}))
155
+ if response.is_success:
156
+ login = (response.json() or {}).get("login")
157
+ self._login = str(login).lower() if login else None
158
+ if self._login is None:
159
+ return True
160
+ return str((comment.get("user") or {}).get("login") or "").lower() == self._login
161
+
162
+ async def start(self, target: ReviewPullRequestInput) -> int:
163
+ """A check run when the token may create one (a GitHub App's); otherwise a commit status
164
+ under the same name, which is what an account's own token can write."""
165
+ response = await self._http.post(
166
+ f"/repos/{target.owner}/{target.repo}/check-runs",
167
+ headers=await _auth(self._tokens, target, self._CHECKS),
168
+ json={
169
+ "name": rendering.CHECK_NAME,
170
+ "head_sha": target.head_sha,
171
+ "status": "in_progress",
172
+ "external_id": target.triggered_by_delivery,
173
+ "output": {
174
+ "title": rendering.IN_PROGRESS_TITLE,
175
+ "summary": rendering.IN_PROGRESS_SUMMARY,
176
+ },
177
+ },
178
+ )
179
+ if response.status_code in (403, 404):
180
+ if not await self._status(target, "pending", rendering.IN_PROGRESS_SUMMARY):
181
+ _log.warning(messages.STATUS_SKIPPED, target.pull_number)
182
+ return NO_PUBLICATION
183
+ return STATUS_PUBLICATION
184
+ raise_for_github_status(
185
+ response, messages.OP_CREATE_CHECK.format(repository=target.repository_full_name)
186
+ )
187
+ check_id: int = response.json()["id"]
188
+ return check_id
189
+
190
+ async def _status(self, target: ReviewPullRequestInput, state: str, description: str) -> bool:
191
+ """Write the commit status; False when the token may not (403 or 404)."""
192
+ body: dict[str, Any] = {
193
+ "state": state,
194
+ "context": rendering.CHECK_NAME,
195
+ "description": description[: rendering.MAX_STATUS_CHARS],
196
+ }
197
+ if self._details_url:
198
+ body["target_url"] = self._details_url
199
+ response = await self._http.post(
200
+ f"/repos/{target.owner}/{target.repo}/statuses/{target.head_sha}",
201
+ headers=await _auth(self._tokens, target, self._CHECKS),
202
+ json=body,
203
+ )
204
+ if response.status_code in (403, 404) and not _rate_limited(response):
205
+ return False
206
+ raise_for_github_status(
207
+ response, messages.OP_CREATE_CHECK.format(repository=target.repository_full_name)
208
+ )
209
+ return True
210
+
211
+ async def complete(
212
+ self, target: ReviewPullRequestInput, publication_id: int, report: ReviewReport
213
+ ) -> None:
214
+ for prior in report.resolved:
215
+ await self._resolve(target, prior)
216
+ if publication_id == STATUS_PUBLICATION:
217
+ await self._status(target, rendering.status_state(report), rendering.title(report))
218
+ elif publication_id != NO_PUBLICATION:
219
+ await self._complete_check(target, publication_id, report)
220
+ try:
221
+ await self._write_overview(target, report)
222
+ except ReviewError as exc:
223
+ # Best effort: the check run already carries the same list.
224
+ _log.warning(messages.OVERVIEW_FAILED, target.pull_number, exc)
225
+
226
+ async def _complete_check(
227
+ self, target: ReviewPullRequestInput, publication_id: int, report: ReviewReport
228
+ ) -> None:
229
+ response = await self._http.patch(
230
+ f"/repos/{target.owner}/{target.repo}/check-runs/{publication_id}",
231
+ headers=await _auth(self._tokens, target, self._CHECKS),
232
+ json={
233
+ "status": "completed",
234
+ "conclusion": rendering.conclusion(report),
235
+ "output": {"title": rendering.title(report), "summary": rendering.summary(report)},
236
+ },
237
+ )
238
+ raise_for_github_status(
239
+ response, messages.OP_COMPLETE_CHECK.format(check_id=publication_id)
240
+ )
241
+
242
+ async def _write_overview(self, target: ReviewPullRequestInput, report: ReviewReport) -> None:
243
+ """One conversation comment per pull request, rewritten each review: created when
244
+ there is something worth a look, updated (to say so) when there no longer is."""
245
+ if report.verdict not in (Verdict.ISSUES_FOUND, Verdict.NO_ISSUES, Verdict.TOO_LARGE):
246
+ return
247
+ existing = await self._overview_id(target)
248
+ body = rendering.overview(
249
+ report,
250
+ head_sha=target.head_sha,
251
+ blob_base=f"{self._web}/{target.owner}/{target.repo}/blob",
252
+ )
253
+ headers = await _auth(self._tokens, target, self._REVIEWS)
254
+ base = f"/repos/{target.owner}/{target.repo}/issues"
255
+ response = (
256
+ await self._http.patch(
257
+ f"{base}/comments/{existing}", headers=headers, json={"body": body}
258
+ )
259
+ if existing is not None
260
+ else await self._http.post(
261
+ f"{base}/{target.pull_number}/comments", headers=headers, json={"body": body}
262
+ )
263
+ )
264
+ raise_for_github_status(
265
+ response, messages.OP_WRITE_OVERVIEW.format(number=target.pull_number)
266
+ )
267
+
268
+ async def _overview_id(self, target: ReviewPullRequestInput) -> int | None:
269
+ found = await self._overview(target)
270
+ return int(found["id"]) if found is not None else None
271
+
272
+ async def last_reviewed(self, target: ReviewPullRequestInput) -> str | None:
273
+ """The commit the summary comment says was last reviewed. Best effort: without it the
274
+ whole change is reviewed, which is never wrong."""
275
+ try:
276
+ found = await self._overview(target)
277
+ except ReviewError:
278
+ return None
279
+ return rendering.reviewed_in(found.get("body") or "") if found is not None else None
280
+
281
+ async def _overview(self, target: ReviewPullRequestInput) -> dict[str, Any] | None:
282
+ headers = await _auth(self._tokens, target, self._REVIEWS)
283
+ for page in range(1, _MAX_PAGES + 1):
284
+ response = await self._http.get(
285
+ f"/repos/{target.owner}/{target.repo}/issues/{target.pull_number}/comments",
286
+ params={"per_page": _PAGE_SIZE, "page": page},
287
+ headers=headers,
288
+ )
289
+ raise_for_github_status(
290
+ response, messages.OP_LIST_ISSUE_COMMENTS.format(number=target.pull_number)
291
+ )
292
+ batch = response.json()
293
+ for comment in batch:
294
+ # Posted as the connected account, so the marker alone says it is ours.
295
+ if rendering.is_overview(comment.get("body") or "") and await self._mine(
296
+ target, comment
297
+ ):
298
+ return dict(comment)
299
+ if len(batch) < _PAGE_SIZE:
300
+ return None
301
+ return None
302
+
303
+ async def reviewed(self, target: ReviewPullRequestInput, sha: str) -> bool:
304
+ response = await self._http.get(
305
+ f"/repos/{target.owner}/{target.repo}/commits/{sha}/check-runs",
306
+ params={"check_name": rendering.CHECK_NAME, "per_page": 10},
307
+ headers=await _auth(self._tokens, target, self._CHECKS),
308
+ )
309
+ if response.status_code in (404, 422):
310
+ return False # unknown commit, or one git no longer has
311
+ if response.status_code == 403 and not _rate_limited(response):
312
+ runs = [] # a personal token without the Checks permission: look at statuses only
313
+ else:
314
+ raise_for_github_status(response, messages.OP_LIST_CHECKS.format(sha=sha[:12]))
315
+ runs = response.json().get("check_runs") or []
316
+ if any(
317
+ r.get("status") == "completed" and r.get("conclusion") in rendering.REVIEWED
318
+ for r in runs
319
+ ):
320
+ return True
321
+ statuses = await self._http.get(
322
+ f"/repos/{target.owner}/{target.repo}/commits/{sha}/statuses",
323
+ params={"per_page": 100},
324
+ headers=await _auth(self._tokens, target, self._CHECKS),
325
+ )
326
+ if not statuses.is_success:
327
+ return False
328
+ return any(
329
+ s.get("context") == rendering.CHECK_NAME
330
+ and rendering.status_reviewed(s.get("state") or "", s.get("description") or "")
331
+ for s in statuses.json()
332
+ )
333
+
334
+ async def acknowledge(self, target: ReviewPullRequestInput) -> None:
335
+ """An eyes reaction on the comment that asked for the review. Best effort: a reaction that cannot be
336
+ added must not stop the review."""
337
+ if target.requested_in_comment is None:
338
+ return
339
+ response = await self._http.post(
340
+ f"/repos/{target.owner}/{target.repo}/issues/comments/"
341
+ f"{target.requested_in_comment}/reactions",
342
+ headers=await _auth(self._tokens, target, self._REVIEWS),
343
+ json={"content": "eyes"},
344
+ )
345
+ if not response.is_success:
346
+ _log.warning(
347
+ messages.ACKNOWLEDGE_SKIPPED, target.requested_in_comment, response.status_code
348
+ )
349
+
350
+ async def post_finding(self, target: ReviewPullRequestInput, finding: Finding) -> None:
351
+ """One comment per finding, posted as soon as it is confirmed: on its line when the
352
+ diff shows it (GitHub allows nothing else there), otherwise on the file."""
353
+ body: dict[str, Any] = {"commit_id": target.head_sha, **_comment(finding)}
354
+ if not finding.in_diff:
355
+ body = {
356
+ "commit_id": target.head_sha,
357
+ "path": finding.path,
358
+ "subject_type": "file",
359
+ "body": rendering.file_comment(finding),
360
+ }
361
+ response = await self._http.post(
362
+ f"/repos/{target.owner}/{target.repo}/pulls/{target.pull_number}/comments",
363
+ headers=await _auth(self._tokens, target, self._REVIEWS),
364
+ json=body,
365
+ )
366
+ raise_for_github_status(response, messages.OP_POST_COMMENT.format(path=finding.path))
367
+
368
+ async def prior_findings(self, target: ReviewPullRequestInput) -> tuple[PriorFinding, ...]:
369
+ headers = await _auth(self._tokens, target, self._REVIEWS)
370
+ found: list[PriorFinding] = []
371
+ for page in range(1, _MAX_PAGES + 1):
372
+ response = await self._http.get(
373
+ f"/repos/{target.owner}/{target.repo}/pulls/{target.pull_number}/comments",
374
+ params={"per_page": _PAGE_SIZE, "page": page},
375
+ headers=headers,
376
+ )
377
+ raise_for_github_status(
378
+ response, messages.OP_LIST_COMMENTS.format(number=target.pull_number)
379
+ )
380
+ batch = response.json()
381
+ for comment in batch:
382
+ fp = rendering.fingerprint_in(comment.get("body") or "")
383
+ if (
384
+ fp is not None
385
+ and comment.get("in_reply_to_id") is None
386
+ and await self._mine(target, comment)
387
+ ):
388
+ found.append(
389
+ PriorFinding(
390
+ fingerprint=fp,
391
+ path=str(comment["path"]),
392
+ comment_id=int(comment["id"]),
393
+ line=comment.get("line"),
394
+ file_level=comment.get("subject_type") == "file",
395
+ title=rendering.title_in(comment.get("body") or ""),
396
+ explanation=rendering.explanation_in(comment.get("body") or ""),
397
+ )
398
+ )
399
+ if len(batch) < _PAGE_SIZE:
400
+ break
401
+ return tuple(found)
402
+
403
+ async def _resolve(self, target: ReviewPullRequestInput, prior: PriorFinding) -> None:
404
+ """Reply on the earlier comment so the author can resolve the thread. GitHub does not
405
+ let an App resolve review threads ("Resource not accessible by integration"), so the
406
+ reply is the whole of what we can do. Best effort: a comment that is gone must not fail
407
+ the review."""
408
+ headers = await _auth(self._tokens, target, self._REVIEWS)
409
+ base = f"/repos/{target.owner}/{target.repo}/pulls/{target.pull_number}"
410
+ reply = await self._http.post(
411
+ f"{base}/comments/{prior.comment_id}/replies",
412
+ headers=headers,
413
+ json={"body": rendering.resolved_reply(target.head_sha)},
414
+ )
415
+ if not reply.is_success:
416
+ _log.warning(messages.RESOLVE_SKIPPED, prior.comment_id, reply.status_code)
417
+
418
+
419
+ def _rate_limited(response: httpx.Response) -> bool:
420
+ return response.headers.get("x-ratelimit-remaining") == "0"
421
+
422
+
423
+ def _comment(finding: Finding) -> dict[str, Any]:
424
+ """A suggestion must sit on exactly the lines it replaces; other comments sit on the
425
+ finding's line."""
426
+ fix = finding.fix
427
+ comment: dict[str, Any] = {
428
+ "path": finding.path,
429
+ "line": finding.line,
430
+ "side": "RIGHT",
431
+ "body": rendering.inline_comment(finding),
432
+ }
433
+ if fix is not None and fix.suggestable:
434
+ comment["line"] = fix.end
435
+ if fix.start < fix.end:
436
+ comment["start_line"] = fix.start
437
+ comment["start_side"] = "RIGHT"
438
+ return comment
439
+
440
+
441
+ async def _auth(
442
+ tokens: StaticTokenProvider, target: ReviewPullRequestInput, permissions: dict[str, str]
443
+ ) -> dict[str, str]:
444
+ token = await tokens.token(target.installation_id, target.repository_id, permissions)
445
+ return {"Authorization": f"Bearer {token}", **github_headers()}
@@ -0,0 +1,212 @@
1
+ """Feedback on ML AI's comments, read back from GitHub: reactions and replies come with the
2
+ review comments; thread resolution needs GraphQL; review times come from the check runs."""
3
+
4
+ from __future__ import annotations
5
+
6
+ from collections import defaultdict
7
+ from datetime import datetime
8
+ from typing import Any, ClassVar
9
+
10
+ import httpx
11
+
12
+ from mlai_review.application.ports import FeedbackSource, Repository
13
+ from mlai_review.domain.feedback import PostedComment, ReviewRun
14
+ from mlai_review.infrastructure import messages
15
+ from mlai_review.infrastructure.github import rendering
16
+ from mlai_review.infrastructure.github.token import StaticTokenProvider
17
+ from mlai_review.infrastructure.github.http import github_headers, raise_for_github_status
18
+
19
+ _PAGE_SIZE = 100
20
+ # Repository-wide comments are listed newest first; a review only needs recent ones and a
21
+ # report rarely needs more than a few thousand.
22
+ _MAX_COMMENT_PAGES = 30
23
+ _MAX_PULLS = 100
24
+ _MAX_COMMITS = 50
25
+ _THREADS = """
26
+ query($owner: String!, $name: String!, $number: Int!) {
27
+ repository(owner: $owner, name: $name) {
28
+ pullRequest(number: $number) {
29
+ reviewThreads(first: 100) {
30
+ nodes { isResolved comments(first: 1) { nodes { databaseId } } }
31
+ }
32
+ }
33
+ }
34
+ }"""
35
+
36
+
37
+ class GitHubFeedbackSource(FeedbackSource):
38
+ _READ: ClassVar[dict[str, str]] = {"pull_requests": "read", "checks": "read"}
39
+
40
+ def __init__(self, *, http: httpx.AsyncClient, tokens: StaticTokenProvider) -> None:
41
+ self._http = http
42
+ self._tokens = tokens
43
+
44
+ async def locate(self, full_name: str) -> Repository:
45
+ owner, name = full_name.split("/", 1)
46
+ installation_id, repository_id = await self._tokens.locate(owner, name)
47
+ return Repository(full_name, installation_id, repository_id)
48
+
49
+ async def posted_comments(
50
+ self, repository: Repository, since: datetime, *, with_threads: bool = True
51
+ ) -> tuple[PostedComment, ...]:
52
+ raw = await self._comments_since(repository, since)
53
+ ours = [
54
+ c
55
+ for c in raw
56
+ if c.get("in_reply_to_id") is None
57
+ and rendering.fingerprint_in(c.get("body") or "") is not None
58
+ and _created(c) >= since
59
+ ]
60
+ replies: dict[int, list[dict[str, Any]]] = defaultdict(list)
61
+ for c in raw:
62
+ if c.get("in_reply_to_id") is not None:
63
+ replies[int(c["in_reply_to_id"])].append(c)
64
+ resolved: dict[int, bool] = {}
65
+ if with_threads:
66
+ for number in sorted({_pull_number(c) for c in ours}):
67
+ resolved |= await self._thread_states(repository, number)
68
+ return tuple(_posted(c, replies[int(c["id"])], resolved.get(int(c["id"]))) for c in ours)
69
+
70
+ async def review_runs(self, repository: Repository, since: datetime) -> tuple[ReviewRun, ...]:
71
+ headers = await self._headers(repository)
72
+ base = f"/repos/{repository.owner}/{repository.name}"
73
+ pulls = await self._http.get(
74
+ f"{base}/pulls",
75
+ params={"state": "all", "sort": "updated", "direction": "desc", "per_page": _MAX_PULLS},
76
+ headers=headers,
77
+ )
78
+ raise_for_github_status(
79
+ pulls, messages.OP_LIST_PULLS.format(repository=repository.full_name)
80
+ )
81
+ runs: list[ReviewRun] = []
82
+ for pull in pulls.json():
83
+ if _parse(pull["updated_at"]) < since:
84
+ break
85
+ commits = await self._http.get(
86
+ f"{base}/pulls/{pull['number']}/commits",
87
+ params={"per_page": _MAX_COMMITS},
88
+ headers=headers,
89
+ )
90
+ raise_for_github_status(commits, messages.OP_LIST_COMMITS.format(number=pull["number"]))
91
+ for commit in commits.json():
92
+ runs += await self._runs_on(base, int(pull["number"]), commit["sha"], headers)
93
+ return tuple(r for r in runs if r.started_at >= since)
94
+
95
+ async def _runs_on(
96
+ self, base: str, number: int, sha: str, headers: dict[str, str]
97
+ ) -> list[ReviewRun]:
98
+ response = await self._http.get(
99
+ f"{base}/commits/{sha}/check-runs",
100
+ # filter=all: by default GitHub lists only the latest run per check name.
101
+ params={"check_name": rendering.CHECK_NAME, "filter": "all", "per_page": _PAGE_SIZE},
102
+ headers=headers,
103
+ )
104
+ if response.status_code in (404, 422):
105
+ return []
106
+ raise_for_github_status(response, messages.OP_LIST_CHECKS.format(sha=sha[:12]))
107
+ incomplete = rendering.INCOMPLETE_TITLE
108
+ return [
109
+ ReviewRun(
110
+ pull_number=number,
111
+ head_sha=sha,
112
+ started_at=_parse(run["started_at"]),
113
+ completed_at=_parse(run["completed_at"]),
114
+ completed=(run.get("output") or {}).get("title") != incomplete,
115
+ )
116
+ for run in response.json().get("check_runs") or []
117
+ if run.get("status") == "completed" and run.get("completed_at")
118
+ ]
119
+
120
+ async def _comments_since(
121
+ self, repository: Repository, since: datetime
122
+ ) -> list[dict[str, Any]]:
123
+ headers = await self._headers(repository)
124
+ url = f"/repos/{repository.owner}/{repository.name}/pulls/comments"
125
+ found: list[dict[str, Any]] = []
126
+ for page in range(1, _MAX_COMMENT_PAGES + 1):
127
+ response = await self._http.get(
128
+ url,
129
+ params={
130
+ "sort": "created",
131
+ "direction": "desc",
132
+ "since": since.isoformat(),
133
+ "per_page": _PAGE_SIZE,
134
+ "page": page,
135
+ },
136
+ headers=headers,
137
+ )
138
+ raise_for_github_status(
139
+ response, messages.OP_LIST_REPO_COMMENTS.format(repository=repository.full_name)
140
+ )
141
+ batch = response.json()
142
+ found += batch
143
+ if len(batch) < _PAGE_SIZE:
144
+ break
145
+ return found
146
+
147
+ async def _thread_states(self, repository: Repository, number: int) -> dict[int, bool]:
148
+ response = await self._http.post(
149
+ "/graphql",
150
+ headers=await self._headers(repository),
151
+ json={
152
+ "query": _THREADS,
153
+ "variables": {"owner": repository.owner, "name": repository.name, "number": number},
154
+ },
155
+ )
156
+ raise_for_github_status(response, messages.OP_READ_THREADS.format(number=number))
157
+ pull = ((response.json().get("data") or {}).get("repository") or {}).get("pullRequest")
158
+ states: dict[int, bool] = {}
159
+ for thread in ((pull or {}).get("reviewThreads") or {}).get("nodes") or []:
160
+ first = ((thread.get("comments") or {}).get("nodes") or [{}])[0]
161
+ if first.get("databaseId") is not None:
162
+ states[int(first["databaseId"])] = bool(thread.get("isResolved"))
163
+ return states
164
+
165
+ async def _headers(self, repository: Repository) -> dict[str, str]:
166
+ token = await self._tokens.token(
167
+ repository.installation_id, repository.repository_id, self._READ
168
+ )
169
+ return {"Authorization": f"Bearer {token}", **github_headers()}
170
+
171
+
172
+ def _posted(
173
+ comment: dict[str, Any], replies: list[dict[str, Any]], resolved: bool | None
174
+ ) -> PostedComment:
175
+ reactions = comment.get("reactions") or {}
176
+ people = [r for r in replies if not _by_bot(r)]
177
+ ours = [r for r in replies if _by_bot(r)]
178
+ body = comment.get("body") or ""
179
+ return PostedComment(
180
+ comment_id=int(comment["id"]),
181
+ pull_number=_pull_number(comment),
182
+ path=str(comment["path"]),
183
+ fingerprint=rendering.fingerprint_in(body) or "",
184
+ title=rendering.title_in(body),
185
+ created_at=_created(comment),
186
+ thumbs_up=int(reactions.get("+1", 0)),
187
+ thumbs_down=int(reactions.get("-1", 0)),
188
+ replies=tuple(str(r.get("body") or "") for r in sorted(people, key=_created)),
189
+ outdated=comment.get("line") is None and comment.get("subject_type") != "file",
190
+ fixed_by_push=any(rendering.is_resolved_reply(r.get("body") or "") for r in ours),
191
+ thread_resolved=resolved,
192
+ tentative=rendering.is_tentative(body),
193
+ )
194
+
195
+
196
+ def _by_bot(comment: dict[str, Any]) -> bool:
197
+ """Ours: posted by an App, or, when posted as the connected account, carrying our wording."""
198
+ if (comment.get("user") or {}).get("type") == "Bot":
199
+ return True
200
+ return rendering.is_resolved_reply(comment.get("body") or "")
201
+
202
+
203
+ def _pull_number(comment: dict[str, Any]) -> int:
204
+ return int(str(comment["pull_request_url"]).rstrip("/").rsplit("/", 1)[1])
205
+
206
+
207
+ def _created(comment: dict[str, Any]) -> datetime:
208
+ return _parse(comment["created_at"])
209
+
210
+
211
+ def _parse(value: str) -> datetime:
212
+ return datetime.fromisoformat(value.replace("Z", "+00:00"))