|
| 1 | +"""What the L2 judge sees for ONE rule: the upstream PRs it cites, with full diffs. |
| 2 | +
|
| 3 | +An intake event stores a merged PR with an 8 KB diff excerpt (enough to draft |
| 4 | +from), and a change set judges every rule against the evidence of its whole |
| 5 | +batch. The judge then reads ten PRs' worth of excerpts and still misses the |
| 6 | +code a rule describes: "the supplied diff ends before ..." was the most common |
| 7 | +reason for an unsure or failed verdict in the first live run. |
| 8 | +
|
| 9 | +``for_rule`` narrows the evidence to the PRs the rule cites (``^[PR #N]``, and |
| 10 | +the PRs its lifecycle footer names as evidence; all of them when it cites none |
| 11 | +we hold) and replaces each excerpt with the full |
| 12 | +per-file diffs of the PR as GitHub shows it (its head against its merge |
| 13 | +base with main), read from the service's own upstream |
| 14 | +mirror: files the rule names first, then source, tests, docs and configuration, within ``PER_FILE`` and |
| 15 | +``PER_RULE`` bytes. Only upstream merged-PR evidence is expanded; anything the |
| 16 | +mirror cannot answer keeps its excerpt (the judge says "unsure" rather than |
| 17 | +guessing), so expansion never makes a verdict easier than today. |
| 18 | +""" |
| 19 | + |
| 20 | +from __future__ import annotations |
| 21 | + |
| 22 | +import re |
| 23 | + |
| 24 | +from ..knowledge_service.facts import CODE_SPAN, EVIDENCE_PR, PR_CITATION, FactsError |
| 25 | +from ..knowledge_service.lifecycle import FOOTER |
| 26 | + |
| 27 | +PER_FILE = 24 * 1024 |
| 28 | +PER_RULE = 48 * 1024 |
| 29 | +NOTE = ("full per-file diffs of the merged commit (files the rule names first); " |
| 30 | + "a file cut at the byte limit says so") |
| 31 | + |
| 32 | + |
| 33 | +def _named(text: str) -> list[str]: |
| 34 | + out = [] |
| 35 | + for m in CODE_SPAN.finditer(text): |
| 36 | + token = m.group("tok").strip().partition("::")[0].strip("/") |
| 37 | + if "/" in token or re.search(r"\.[A-Za-z0-9]{1,5}$", token): |
| 38 | + out.append(token) |
| 39 | + return out |
| 40 | + |
| 41 | + |
| 42 | +def _rank(path: str) -> int: |
| 43 | + """Source before tests before docs and configuration (the code a rule describes first).""" |
| 44 | + parts = path.lower().split("/") |
| 45 | + if any(p in ("docs", "doc", "recipes", "examples") for p in parts[:-1]) or \ |
| 46 | + parts[-1].endswith((".md", ".rst", ".txt", ".yaml", ".yml", ".json", ".toml")): |
| 47 | + return 2 |
| 48 | + if any(p in ("tests", "test") for p in parts[:-1]) or parts[-1].startswith("test_"): |
| 49 | + return 1 |
| 50 | + return 0 |
| 51 | + |
| 52 | + |
| 53 | +def _matches(path: str, names: list[str]) -> bool: |
| 54 | + return any(path == n or path.endswith("/" + n) or n.endswith("/" + path) for n in names) |
| 55 | + |
| 56 | + |
| 57 | +def _cut(text: str, limit: int) -> str: |
| 58 | + data = text.encode("utf-8") |
| 59 | + if len(data) <= limit: |
| 60 | + return text |
| 61 | + return data[:limit].decode("utf-8", "ignore") + f"\n[... cut at {limit} bytes of {len(data)}]\n" |
| 62 | + |
| 63 | + |
| 64 | +def for_rule(text: str, evidence: list[dict], observer) -> list[dict]: |
| 65 | + """The evidence for one rule (``text``: the rule as proposed, or as it was).""" |
| 66 | + cited = {f"PR #{m.group('n')}" for m in PR_CITATION.finditer(text)} |
| 67 | + # a retirement's justification lives in its footer (evidence="PR #N"), not in a citation |
| 68 | + for line in text.splitlines(): |
| 69 | + footer = FOOTER.match(line.strip()) |
| 70 | + if footer: |
| 71 | + cited |= {f"PR #{m.group('n')}" for m in EVIDENCE_PR.finditer(footer.group("attrs"))} |
| 72 | + items = [e for e in evidence if e.get("source_reference") in cited] or list(evidence) |
| 73 | + if observer is None: |
| 74 | + return items |
| 75 | + names, budget, out = _named(text), PER_RULE, [] |
| 76 | + for item in items: |
| 77 | + number = re.fullmatch(r"PR #(\d+)", str(item.get("source_reference") or "")) |
| 78 | + if not number or "merged_at" not in item or budget <= 0: |
| 79 | + out.append(item) # not an upstream merged PR (or no room left): as recorded |
| 80 | + continue |
| 81 | + try: |
| 82 | + sha = str(item.get("merge_commit_sha") or observer.pull(int(number.group(1))).get("merge_commit_sha") or "") |
| 83 | + if not sha: |
| 84 | + raise FactsError("no merge commit") |
| 85 | + files = list(item.get("changed_files") or []) |
| 86 | + ordered = sorted(files, key=lambda f: (not _matches(f, names), _rank(f))) |
| 87 | + diffs: dict[str, str] = {} |
| 88 | + for path in ordered: |
| 89 | + if budget <= 0: |
| 90 | + break |
| 91 | + diff = _cut(observer.pr_diff(int(number.group(1)), sha, path), min(PER_FILE, budget)) |
| 92 | + if diff: |
| 93 | + diffs[path] = diff |
| 94 | + budget -= len(diff.encode("utf-8")) |
| 95 | + except FactsError: |
| 96 | + out.append(item) |
| 97 | + continue |
| 98 | + if not diffs: |
| 99 | + out.append(item) |
| 100 | + continue |
| 101 | + expanded = {k: v for k, v in item.items() if k != "diff_excerpt"} |
| 102 | + out.append({**expanded, "diffs": diffs, "diffs_note": NOTE, |
| 103 | + "diffs_omitted": [f for f in ordered if f not in diffs]}) |
| 104 | + return out |
0 commit comments