mlai-code 0.3.2 → 0.3.4

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 (157) 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 +178 -31
  5. package/src/cloud.mjs +2 -0
  6. package/src/engine.mjs +1 -1
  7. package/src/headless.mjs +25 -9
  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 +20 -1
  14. package/src/relay.mjs +1 -1
  15. package/src/render.mjs +12 -3
  16. package/src/runtime.mjs +23 -1
  17. package/src/sandbox.mjs +19 -3
  18. package/src/server.mjs +46 -7
  19. package/src/session.mjs +53 -10
  20. package/src/shared-data.mjs +1 -1
  21. package/src/tui/acp.mjs +121 -19
  22. package/src/tui/app.mjs +24 -5
  23. package/src/tui/chat.mjs +43 -16
  24. package/src/tui/cmd-config.mjs +3 -1
  25. package/src/tui/cmd-core.mjs +11 -7
  26. package/src/tui/cmd-extensibility.mjs +13 -2
  27. package/src/tui/commands.mjs +2 -1
  28. package/src/tui/extensions-local.mjs +2 -0
  29. package/src/tui/mcp-config.mjs +22 -4
  30. package/src/tui/review-cli.mjs +15 -4
  31. package/src/tui/snapshots.mjs +57 -16
  32. package/src/tui/statusline.mjs +2 -1
  33. package/src/two-words.mjs +14 -8
  34. package/src/view-sync.mjs +48 -0
  35. package/src/wakeups.mjs +2 -2
  36. package/src/winenv.mjs +1 -1
  37. package/src/workspace-api.mjs +21 -4
  38. package/vendor/client-sdk/index.cjs +5 -4
  39. package/vendor/cloud-cli/mlai.mjs +184 -28
  40. package/worker/architecture_worker.py +33 -4
  41. package/worker/browser_tool.py +19 -1
  42. package/worker/build_planner.py +75 -7
  43. package/worker/chat_fork.py +4 -1
  44. package/worker/checkpoints.py +71 -8
  45. package/worker/cloud_pty.py +91 -2
  46. package/worker/command_safety.py +80 -3
  47. package/worker/compaction_brief.py +186 -0
  48. package/worker/connected_turn.py +30 -0
  49. package/worker/context_manager.py +6 -0
  50. package/worker/conversation.py +139 -12
  51. package/worker/deck_render.js +11 -0
  52. package/worker/deck_tools.py +825 -0
  53. package/worker/git_sync.py +42 -17
  54. package/worker/goals.py +4 -0
  55. package/worker/image_generation.py +102 -23
  56. package/worker/kimi_bridge.py +580 -52
  57. package/worker/kimi_gates.py +74 -6
  58. package/worker/kimi_hook.py +213 -19
  59. package/worker/media_tools.py +182 -0
  60. package/worker/memory.py +2 -2
  61. package/worker/mlai_hooks.py +314 -5
  62. package/worker/mlai_mcp.py +49 -11
  63. package/worker/mlai_review/README.md +35 -0
  64. package/worker/mlai_review/__init__.py +1 -0
  65. package/worker/mlai_review/__main__.py +261 -0
  66. package/worker/mlai_review/_compat.py +29 -0
  67. package/worker/mlai_review/analysis/__init__.py +3 -0
  68. package/worker/mlai_review/analysis/detect.py +50 -0
  69. package/worker/mlai_review/analysis/main.py +128 -0
  70. package/worker/mlai_review/analysis/messages.py +8 -0
  71. package/worker/mlai_review/analysis/tools.py +194 -0
  72. package/worker/mlai_review/application/__init__.py +1 -0
  73. package/worker/mlai_review/application/asking.py +53 -0
  74. package/worker/mlai_review/application/challengers.py +142 -0
  75. package/worker/mlai_review/application/context_builder.py +194 -0
  76. package/worker/mlai_review/application/errors.py +17 -0
  77. package/worker/mlai_review/application/feedback.py +52 -0
  78. package/worker/mlai_review/application/matching.py +46 -0
  79. package/worker/mlai_review/application/messages.py +41 -0
  80. package/worker/mlai_review/application/ports.py +222 -0
  81. package/worker/mlai_review/application/prompts.py +342 -0
  82. package/worker/mlai_review/application/review_service.py +200 -0
  83. package/worker/mlai_review/application/reviewers.py +438 -0
  84. package/worker/mlai_review/application/streaming.py +91 -0
  85. package/worker/mlai_review/application/tools.py +138 -0
  86. package/worker/mlai_review/application/usage.py +145 -0
  87. package/worker/mlai_review/application/write_check.py +46 -0
  88. package/worker/mlai_review/contracts/__init__.py +5 -0
  89. package/worker/mlai_review/contracts/analysis.py +57 -0
  90. package/worker/mlai_review/contracts/review.py +42 -0
  91. package/worker/mlai_review/domain/__init__.py +4 -0
  92. package/worker/mlai_review/domain/changes.py +77 -0
  93. package/worker/mlai_review/domain/context.py +147 -0
  94. package/worker/mlai_review/domain/diagnostics.py +239 -0
  95. package/worker/mlai_review/domain/feedback.py +216 -0
  96. package/worker/mlai_review/domain/findings.py +505 -0
  97. package/worker/mlai_review/domain/lifecycle.py +85 -0
  98. package/worker/mlai_review/domain/partition.py +42 -0
  99. package/worker/mlai_review/domain/relations.py +166 -0
  100. package/worker/mlai_review/domain/relevance.py +202 -0
  101. package/worker/mlai_review/domain/report.py +99 -0
  102. package/worker/mlai_review/domain/structure.py +56 -0
  103. package/worker/mlai_review/domain/writes.py +86 -0
  104. package/worker/mlai_review/eval.py +221 -0
  105. package/worker/mlai_review/eval_cases/clean-rename.json +11 -0
  106. package/worker/mlai_review/eval_cases/missing-ownership-check.json +22 -0
  107. package/worker/mlai_review/eval_cases/off-by-one.json +22 -0
  108. package/worker/mlai_review/eval_cases/sql-injection.json +20 -0
  109. package/worker/mlai_review/feedback_report.py +60 -0
  110. package/worker/mlai_review/infrastructure/__init__.py +1 -0
  111. package/worker/mlai_review/infrastructure/analysis/__init__.py +1 -0
  112. package/worker/mlai_review/infrastructure/analysis/none.py +12 -0
  113. package/worker/mlai_review/infrastructure/analysis/translate.py +17 -0
  114. package/worker/mlai_review/infrastructure/github/__init__.py +0 -0
  115. package/worker/mlai_review/infrastructure/github/client.py +445 -0
  116. package/worker/mlai_review/infrastructure/github/feedback.py +212 -0
  117. package/worker/mlai_review/infrastructure/github/http.py +31 -0
  118. package/worker/mlai_review/infrastructure/github/rendering.py +398 -0
  119. package/worker/mlai_review/infrastructure/github/token.py +20 -0
  120. package/worker/mlai_review/infrastructure/local/__init__.py +1 -0
  121. package/worker/mlai_review/infrastructure/local/analyzer.py +48 -0
  122. package/worker/mlai_review/infrastructure/local/events.py +100 -0
  123. package/worker/mlai_review/infrastructure/local/git_reader.py +233 -0
  124. package/worker/mlai_review/infrastructure/local/publisher.py +138 -0
  125. package/worker/mlai_review/infrastructure/messages.py +54 -0
  126. package/worker/mlai_review/infrastructure/models/__init__.py +0 -0
  127. package/worker/mlai_review/infrastructure/models/openai_compatible.py +269 -0
  128. package/worker/mlai_review/infrastructure/structure/__init__.py +1 -0
  129. package/worker/mlai_review/infrastructure/structure/heuristic.py +186 -0
  130. package/worker/mlai_review/infrastructure/structure/tree_sitter.py +186 -0
  131. package/worker/mlai_review/infrastructure/workspace/__init__.py +1 -0
  132. package/worker/mlai_review/infrastructure/workspace/local.py +105 -0
  133. package/worker/mlai_review/infrastructure/workspace/memory.py +77 -0
  134. package/worker/mlai_review/infrastructure/workspace/paths.py +17 -0
  135. package/worker/mlai_review/pipeline.py +89 -0
  136. package/worker/mlai_review/runtime.py +194 -0
  137. package/worker/mlai_review/settings.py +139 -0
  138. package/worker/model_stream.py +21 -7
  139. package/worker/offline_guard.py +1 -1
  140. package/worker/preview.py +62 -4
  141. package/worker/product_facts.py +20 -0
  142. package/worker/project_context.py +33 -7
  143. package/worker/reply_language.py +92 -10
  144. package/worker/repo_policy.py +5 -0
  145. package/worker/roles.json +199 -37
  146. package/worker/roles.py +271 -9
  147. package/worker/runner.py +366 -53
  148. package/worker/runtime_control.py +64 -3
  149. package/worker/secret_scan.py +3 -2
  150. package/worker/server_memory.py +122 -0
  151. package/worker/site_check.py +279 -0
  152. package/worker/skill_library.py +15 -3
  153. package/worker/skill_pack.py +170 -16
  154. package/worker/subagents.py +6 -2
  155. package/worker/tool_catalog.py +4 -3
  156. package/worker/verification.py +156 -19
  157. package/worker/wakeups.py +56 -4
@@ -0,0 +1,145 @@
1
+ """What one review costs in model tokens. A review opens a meter; every model call made while
2
+ it is open, from any reading, refuter or matcher task the review started, adds to it. Reviews
3
+ running side by side on one worker each have their own meter."""
4
+
5
+ from __future__ import annotations
6
+
7
+ import math
8
+ from collections.abc import Iterator
9
+ from contextlib import contextmanager
10
+ from contextvars import ContextVar
11
+ from dataclasses import dataclass, field
12
+ from typing import Any
13
+
14
+ from mlai_review.application.errors import PermanentReviewError
15
+
16
+ # Credit weights, kept in step with lib/plans.ts and worker/runner.py billable_tokens.
17
+ CACHED_INPUT_WEIGHT = 0.2
18
+ OUTPUT_WEIGHT = 3
19
+ RATE_KEYS = ("input", "cached", "cacheWrite", "output")
20
+
21
+
22
+ class CreditsExhausted(PermanentReviewError):
23
+ """The account's credits for this review are used up; no further model calls are made."""
24
+
25
+
26
+ @dataclass(slots=True)
27
+ class TokenUsage:
28
+ """Model calls and their tokens, as the provider reported them."""
29
+
30
+ calls: int = 0
31
+ prompt_tokens: int = 0
32
+ cached_tokens: int = 0 # of the prompt tokens, those served from the provider's cache
33
+ completion_tokens: int = 0
34
+ billable: int = 0 # credit-weighted tokens, as the app meters a run
35
+ rates: dict[str, float] | None = field(default=None, repr=False)
36
+ budget: int | None = None # billable tokens this review may use; None is no cap
37
+
38
+ def add(
39
+ self,
40
+ prompt_tokens: int,
41
+ cached_tokens: int,
42
+ completion_tokens: int,
43
+ raw: dict[str, Any] | None = None,
44
+ ) -> None:
45
+ self.calls += 1
46
+ self.prompt_tokens += prompt_tokens
47
+ self.cached_tokens += cached_tokens
48
+ self.completion_tokens += completion_tokens
49
+ reported = raw or {
50
+ "prompt_tokens": prompt_tokens,
51
+ "completion_tokens": completion_tokens,
52
+ "prompt_tokens_details": {"cached_tokens": cached_tokens},
53
+ }
54
+ self.billable += billable_tokens(reported, self.rates)
55
+
56
+ @property
57
+ def exhausted(self) -> bool:
58
+ return self.budget is not None and self.billable >= self.budget
59
+
60
+
61
+ def credit_rates(value: object) -> dict[str, float] | None:
62
+ """Per-model weights from lib/model-prices.ts, or None for the default weights."""
63
+ if not isinstance(value, dict):
64
+ return None
65
+ out: dict[str, float] = {}
66
+ for key in RATE_KEYS:
67
+ v = value.get(key)
68
+ if not isinstance(v, (int, float)) or isinstance(v, bool) or not math.isfinite(v) or v < 0:
69
+ return None
70
+ out[key] = float(v)
71
+ return out
72
+
73
+
74
+ def billable_tokens(usage: object, rates: dict[str, float] | None = None) -> int:
75
+ """One call's usage in billable tokens, exactly as lib/laptop-run.ts usageTokens and the
76
+ worker meter it: fresh input in full, cached a fifth, output three times, or per-model rates."""
77
+ if not isinstance(usage, dict):
78
+ return 0
79
+
80
+ def num(v: object) -> int:
81
+ return max(0, int(v)) if isinstance(v, (int, float)) and not isinstance(v, bool) else 0
82
+
83
+ read = num(usage.get("cache_read_input_tokens"))
84
+ if usage.get("prompt_tokens") is not None:
85
+ prompt = num(usage["prompt_tokens"])
86
+ else:
87
+ prompt = num(usage.get("input_tokens")) + read + num(usage.get("cache_creation_input_tokens"))
88
+ if usage.get("completion_tokens") is not None:
89
+ output = num(usage["completion_tokens"])
90
+ else:
91
+ output = num(usage.get("output_tokens"))
92
+ details = usage.get("prompt_tokens_details")
93
+ details = details if isinstance(details, dict) else {}
94
+ cached = min(prompt, num(details.get("cached_tokens")) or num(usage.get("cached_tokens")) or read)
95
+ rates = credit_rates(rates)
96
+ if not prompt and not output:
97
+ total = num(usage.get("total_tokens"))
98
+ return int(math.ceil(total * rates["output"])) if rates else total
99
+ if rates:
100
+ written = min(
101
+ prompt - cached,
102
+ num(usage.get("cache_creation_input_tokens")) or num(details.get("cache_write_tokens")),
103
+ )
104
+ return int(
105
+ math.ceil(
106
+ (prompt - cached - written) * rates["input"]
107
+ + cached * rates["cached"]
108
+ + written * rates["cacheWrite"]
109
+ + output * rates["output"]
110
+ )
111
+ )
112
+ return int(math.ceil(prompt - cached + cached * CACHED_INPUT_WEIGHT + output * OUTPUT_WEIGHT))
113
+
114
+
115
+ _meter: ContextVar[tuple[TokenUsage, ...]] = ContextVar("mlai_token_usage", default=())
116
+
117
+
118
+ @contextmanager
119
+ def metering(
120
+ rates: dict[str, float] | None = None, budget: int | None = None
121
+ ) -> Iterator[TokenUsage]:
122
+ """A fresh meter for the calls made inside the block (and the tasks it starts). Meters
123
+ nest: a call counts against every meter open around it, so a caller can meter a whole
124
+ run while the review service meters each step. Inner meters weigh tokens like the outer one."""
125
+ outer = _meter.get()
126
+ usage = TokenUsage(rates=credit_rates(rates) or (outer[-1].rates if outer else None), budget=budget)
127
+ token = _meter.set((*_meter.get(), usage))
128
+ try:
129
+ yield usage
130
+ finally:
131
+ _meter.reset(token)
132
+
133
+
134
+ def record(
135
+ prompt_tokens: int, cached_tokens: int, completion_tokens: int, raw: dict[str, Any] | None = None
136
+ ) -> None:
137
+ """Count one model call against the open meter; outside a review it is not counted."""
138
+ for usage in _meter.get():
139
+ usage.add(prompt_tokens, cached_tokens, completion_tokens, raw)
140
+
141
+
142
+ def check_budget() -> None:
143
+ """Before a model call: stop once any open meter has used its budget."""
144
+ if any(usage.exhausted for usage in _meter.get()):
145
+ raise CreditsExhausted("credits used up")
@@ -0,0 +1,46 @@
1
+ """A narrow look at every write a change adds. General readings judge a write by what it is
2
+ meant to do; asked about each write on its own, with its filter tested against example rows,
3
+ the model finds the ones that touch the wrong rows (see domain.writes)."""
4
+
5
+ from __future__ import annotations
6
+
7
+ from abc import ABC, abstractmethod
8
+ from typing import Any
9
+
10
+ from mlai_review.application import prompts
11
+ from mlai_review.application.asking import ask
12
+ from mlai_review.application.ports import LanguageModel
13
+ from mlai_review.domain.context import ReviewContext
14
+ from mlai_review.domain.writes import writes_in
15
+
16
+
17
+ class WriteCheck(ABC):
18
+ @abstractmethod
19
+ async def answer(self, context: ReviewContext) -> dict[str, Any] | None:
20
+ """Findings about the change's writes, in the readings' format; None when there was
21
+ nothing to check."""
22
+
23
+
24
+ class NoWriteCheck(WriteCheck):
25
+ async def answer(self, context: ReviewContext) -> dict[str, Any] | None:
26
+ return None
27
+
28
+
29
+ class LlmWriteCheck(WriteCheck):
30
+ """One model call per change (or part), listing every write it adds."""
31
+
32
+ def __init__(self, model: LanguageModel) -> None:
33
+ self._model = model
34
+
35
+ async def answer(self, context: ReviewContext) -> dict[str, Any] | None:
36
+ sites = writes_in(context.changes)
37
+ if not sites:
38
+ return None
39
+ return await ask(
40
+ self._model,
41
+ system=prompts.SHARED,
42
+ user=prompts.user_message(context) + prompts.write_check_task(sites),
43
+ tools=None,
44
+ budget=None,
45
+ who="writes",
46
+ )
@@ -0,0 +1,5 @@
1
+ """Messages exchanged between ML AI services.
2
+
3
+ Services depend on this package, never on each other. Types here are plain, serialisable data;
4
+ changing a field is a cross-service change and needs both sides deployed compatibly.
5
+ """
@@ -0,0 +1,57 @@
1
+ """The contract between the orchestrator and the static-analysis sandbox: what the sandbox is
2
+ asked to analyse, and the diagnostics it reports back."""
3
+
4
+ from __future__ import annotations
5
+
6
+ from dataclasses import asdict, dataclass, field
7
+ from typing import Any
8
+
9
+ ANALYSIS_RESULT_VERSION = 1
10
+
11
+
12
+ @dataclass(frozen=True, slots=True)
13
+ class Diagnostic:
14
+ """One message from one tool, on one line of one file."""
15
+
16
+ tool: str # ruff, tsc, eslint, ...
17
+ path: str # repository-relative, POSIX style
18
+ line: int # 1-based
19
+ rule: str # the tool's code for the message, e.g. F821, TS2322, no-unused-vars
20
+ message: str
21
+ level: str # "error" or "warning"
22
+
23
+
24
+ @dataclass(frozen=True, slots=True)
25
+ class AnalysisResult:
26
+ version: int = ANALYSIS_RESULT_VERSION
27
+ tools_run: tuple[str, ...] = ()
28
+ diagnostics: tuple[Diagnostic, ...] = ()
29
+ # Tools that could not run, with the reason: shown to operators, never to the model.
30
+ errors: dict[str, str] = field(default_factory=dict)
31
+
32
+ def as_dict(self) -> dict[str, Any]:
33
+ return {
34
+ "version": self.version,
35
+ "tools_run": list(self.tools_run),
36
+ "diagnostics": [asdict(d) for d in self.diagnostics],
37
+ "errors": dict(self.errors),
38
+ }
39
+
40
+ @classmethod
41
+ def from_dict(cls, data: dict[str, Any]) -> AnalysisResult:
42
+ return cls(
43
+ version=int(data.get("version", ANALYSIS_RESULT_VERSION)),
44
+ tools_run=tuple(str(t) for t in data.get("tools_run", [])),
45
+ diagnostics=tuple(
46
+ Diagnostic(
47
+ tool=str(d["tool"]),
48
+ path=str(d["path"]),
49
+ line=int(d["line"]),
50
+ rule=str(d.get("rule", "")),
51
+ message=str(d.get("message", "")),
52
+ level=str(d.get("level", "error")),
53
+ )
54
+ for d in data.get("diagnostics", [])
55
+ ),
56
+ errors={str(k): str(v) for k, v in (data.get("errors") or {}).items()},
57
+ )
@@ -0,0 +1,42 @@
1
+ """The contract for asking the orchestrator to review a pull request."""
2
+
3
+ from __future__ import annotations
4
+
5
+ from dataclasses import dataclass
6
+
7
+ REVIEW_WORKFLOW = "ReviewPullRequest"
8
+ REVIEW_TASK_QUEUE = "reviews"
9
+ # The name of the check the orchestrator posts; the gateway recognises re-run requests by it.
10
+ CHECK_NAME = "ML AI Review"
11
+
12
+
13
+ @dataclass(frozen=True, slots=True)
14
+ class ReviewPullRequestInput:
15
+ """Everything the orchestrator needs to review one pull request at one head commit."""
16
+
17
+ installation_id: int
18
+ repository_id: int
19
+ repository_full_name: str
20
+ pull_number: int
21
+ head_sha: str
22
+ base_sha: str
23
+ triggered_by_delivery: str
24
+ # The head commit before this push (GitHub's `before` on synchronize), when known. Lets
25
+ # the orchestrator review only what changed since, if that commit was reviewed.
26
+ previous_head_sha: str | None = None
27
+ # Set when someone asked for the review in a comment ("@ml-ai-review review"): the
28
+ # orchestrator acknowledges that comment. Such requests carry no commit (head_sha and
29
+ # base_sha are empty): the orchestrator reviews the pull request's head as it is then.
30
+ requested_in_comment: int | None = None
31
+
32
+ @property
33
+ def needs_head(self) -> bool:
34
+ return not self.head_sha
35
+
36
+ @property
37
+ def owner(self) -> str:
38
+ return self.repository_full_name.split("/", 1)[0]
39
+
40
+ @property
41
+ def repo(self) -> str:
42
+ return self.repository_full_name.split("/", 1)[1]
@@ -0,0 +1,4 @@
1
+ """Review domain: changes, reports and the reviewers that turn one into the other.
2
+
3
+ Standard library only.
4
+ """
@@ -0,0 +1,77 @@
1
+ """What a pull request changed, and which lines of the new files the diff shows."""
2
+
3
+ from __future__ import annotations
4
+
5
+ import re
6
+ from dataclasses import dataclass, field
7
+
8
+ _HUNK = re.compile(r"^@@ -\d+(?:,\d+)? \+(\d+)(?:,(\d+))? @@")
9
+
10
+
11
+ @dataclass(frozen=True, slots=True)
12
+ class FileChange:
13
+ path: str
14
+ status: str # added, modified, removed, renamed, ...
15
+ additions: int
16
+ deletions: int
17
+ # Unified diff for this file; absent for binary files or very large diffs.
18
+ patch: str | None = None
19
+
20
+
21
+ @dataclass(frozen=True, slots=True)
22
+ class ChangeSet:
23
+ files: tuple[FileChange, ...] = field(default_factory=tuple)
24
+ # When set, the files are only what changed since this commit, not the whole pull request.
25
+ since: str | None = None
26
+
27
+ @property
28
+ def additions(self) -> int:
29
+ return sum(f.additions for f in self.files)
30
+
31
+ @property
32
+ def deletions(self) -> int:
33
+ return sum(f.deletions for f in self.files)
34
+
35
+ def file(self, path: str) -> FileChange | None:
36
+ return next((f for f in self.files if f.path == path), None)
37
+
38
+
39
+ @dataclass(frozen=True, slots=True)
40
+ class DiffLine:
41
+ kind: str # "+" added, "-" removed, " " unchanged context
42
+ # For added and context lines, their number in the new file. For removed lines, the number
43
+ # of the new-file line they now precede (so the removal can be shown in place).
44
+ new_line: int
45
+ text: str
46
+
47
+
48
+ def parse_patch(patch: str | None) -> tuple[DiffLine, ...]:
49
+ if not patch:
50
+ return ()
51
+ lines: list[DiffLine] = []
52
+ current = 0
53
+ for raw in patch.splitlines():
54
+ header = _HUNK.match(raw)
55
+ if header:
56
+ current = int(header.group(1))
57
+ elif raw.startswith("\\") or not current:
58
+ continue # "", or text before the first hunk
59
+ elif raw.startswith("-"):
60
+ lines.append(DiffLine("-", current, raw[1:]))
61
+ else:
62
+ kind = "+" if raw.startswith("+") else " "
63
+ lines.append(DiffLine(kind, current, raw[1:] if raw else ""))
64
+ current += 1
65
+ return tuple(lines)
66
+
67
+
68
+ def lines_in_diff(patch: str | None) -> frozenset[int]:
69
+ """Line numbers of the new file that the diff shows: added lines and the unchanged context
70
+ around them. A finding on one of these lines can be pointed at in the diff itself."""
71
+ return frozenset(d.new_line for d in parse_patch(patch) if d.kind != "-")
72
+
73
+
74
+ def changed_lines(patch: str | None) -> frozenset[int]:
75
+ """New-file line numbers where something changed: added lines, and the line each removal
76
+ now precedes. The smallest set of places a reviewer must look at."""
77
+ return frozenset(d.new_line for d in parse_patch(patch) if d.kind != " ")
@@ -0,0 +1,147 @@
1
+ """Everything a reviewer is shown about one pull request."""
2
+
3
+ from __future__ import annotations
4
+
5
+ from dataclasses import dataclass, field, replace
6
+
7
+ from mlai_review.domain.changes import ChangeSet, changed_lines, lines_in_diff
8
+ from mlai_review.domain.diagnostics import Diagnostic
9
+ from mlai_review.domain.feedback import Dismissal
10
+ from mlai_review.domain.relations import Relation
11
+ from mlai_review.domain.structure import Outline
12
+
13
+ # How a changed file is shown. A change inside a function shows the whole function; a function
14
+ # longer than this, or a change outside any function, shows a window around the change instead.
15
+ MAX_FUNCTION_LINES = 160
16
+ WINDOW = 8
17
+ # Regions this close together are shown as one.
18
+ MERGE_GAP = 4
19
+ # At most this many outline lines (imports, definition headers) per file.
20
+ MAX_SKELETON_LINES = 60
21
+
22
+
23
+ @dataclass(frozen=True, slots=True)
24
+ class PullRequestInfo:
25
+ title: str
26
+ body: str
27
+
28
+
29
+ @dataclass(frozen=True, slots=True)
30
+ class LineLocation:
31
+ path: str
32
+ line: int
33
+ # True when the diff shows this line, so a finding can be pointed at in the diff itself.
34
+ in_diff: bool
35
+
36
+
37
+ @dataclass(frozen=True, slots=True)
38
+ class Region:
39
+ start: int
40
+ end: int # inclusive
41
+
42
+
43
+ @dataclass(frozen=True, slots=True)
44
+ class FileView:
45
+ """What the reviewer sees of one changed file: its outline, and the regions around the
46
+ change, instead of the whole text. The text itself stays here for rendering only."""
47
+
48
+ text: str
49
+ outline: Outline | None
50
+ regions: tuple[Region, ...]
51
+ skeleton: tuple[int, ...]
52
+
53
+ @property
54
+ def line_count(self) -> int:
55
+ return len(self.text.splitlines())
56
+
57
+ @property
58
+ def chars(self) -> int:
59
+ """Size of what will be shown, for budgeting."""
60
+ lines = self.text.splitlines()
61
+ shown = {n for r in self.regions for n in range(r.start, r.end + 1)} | set(self.skeleton)
62
+ return sum(len(lines[n - 1]) + 8 for n in shown if n <= len(lines))
63
+
64
+ @classmethod
65
+ def focus(cls, text: str, patch: str | None, outline: Outline | None) -> FileView:
66
+ line_count = len(text.splitlines())
67
+ ranges: list[tuple[int, int]] = []
68
+ for line in sorted(changed_lines(patch)):
69
+ line = min(max(line, 1), max(line_count, 1))
70
+ function = outline.enclosing_function(line) if outline else None
71
+ if function is not None and function.lines <= MAX_FUNCTION_LINES:
72
+ ranges.append((function.start, function.end))
73
+ else:
74
+ ranges.append((max(1, line - WINDOW), min(line_count, line + WINDOW)))
75
+ skeleton = outline.skeleton(max_lines=MAX_SKELETON_LINES) if outline else ()
76
+ return cls(text, outline, _merge(ranges), skeleton)
77
+
78
+
79
+ def _merge(ranges: list[tuple[int, int]]) -> tuple[Region, ...]:
80
+ merged: list[Region] = []
81
+ for start, end in sorted(ranges):
82
+ if merged and start <= merged[-1].end + MERGE_GAP:
83
+ merged[-1] = Region(merged[-1].start, max(merged[-1].end, end))
84
+ else:
85
+ merged.append(Region(start, end))
86
+ return tuple(merged)
87
+
88
+
89
+ @dataclass(frozen=True, slots=True)
90
+ class ReviewContext:
91
+ pull_request: PullRequestInfo
92
+ # The files that get reviewed: the change set without noise.
93
+ changes: ChangeSet
94
+ # The reviewer's view of each changed file, keyed by path. May be partial: removed files,
95
+ # unreadable files and anything beyond the size budget are reviewed from their diff.
96
+ views: dict[str, FileView] = field(default_factory=dict)
97
+ # False when file contents could not be read at all (e.g. no Contents permission).
98
+ contents_available: bool = True
99
+ # Changed files left out as noise, for the record.
100
+ skipped: tuple[str, ...] = field(default_factory=tuple)
101
+ # Lines elsewhere in the repository that define what the change calls, or use what it
102
+ # changes. Found automatically; the reviewer can read further with its tools.
103
+ related: tuple[Relation, ...] = field(default_factory=tuple)
104
+ # What static-analysis tools reported on lines the change touches.
105
+ diagnostics: tuple[Diagnostic, ...] = field(default_factory=tuple)
106
+ # Earlier findings in this repository the team marked as not useful.
107
+ dismissed: tuple[Dismissal, ...] = field(default_factory=tuple)
108
+ # For one part of a change reviewed separately: the pull request's other changed files.
109
+ elsewhere: tuple[str, ...] = field(default_factory=tuple)
110
+ # The team's own guidance for reviews of this repository (its settings or .mlai/review.yml).
111
+ instructions: str = ""
112
+
113
+ def part(self, paths: tuple[str, ...]) -> ReviewContext:
114
+ """The same review narrowed to some of its files: their changes, views and tool
115
+ messages, the related code that mentions what they mention, and the names of the
116
+ files left out."""
117
+ keep = set(paths)
118
+ files = tuple(f for f in self.changes.files if f.path in keep)
119
+ texts = [f.patch or "" for f in files] + [
120
+ self.views[f.path].text for f in files if f.path in self.views
121
+ ]
122
+ return replace(
123
+ self,
124
+ changes=replace(self.changes, files=files),
125
+ views={p: v for p, v in self.views.items() if p in keep},
126
+ related=tuple(r for r in self.related if any(r.symbol in t for t in texts)),
127
+ diagnostics=tuple(d for d in self.diagnostics if d.path in keep),
128
+ elsewhere=tuple(f.path for f in self.changes.files if f.path not in keep),
129
+ )
130
+
131
+ def locate(self, path: str, line: int) -> LineLocation | None:
132
+ """Where a claimed finding really is, or None if the claim points nowhere real.
133
+
134
+ A finding must name a file this change adds or modifies, and a line that exists in it.
135
+ When the file's text is unknown, only lines the diff shows can be checked, so only those
136
+ are accepted: an unverifiable location is never published.
137
+ """
138
+ change = self.changes.file(path)
139
+ if change is None or change.status == "removed" or line < 1:
140
+ return None
141
+ in_diff = line in lines_in_diff(change.patch)
142
+ view = self.views.get(path)
143
+ if view is not None and line > view.line_count:
144
+ return None
145
+ if view is None and not in_diff:
146
+ return None
147
+ return LineLocation(path=path, line=line, in_diff=in_diff)