fix: apply cached GSMArena confirmations to source files

Author: Seungpyo1007

app/verify/gsmarena_backfill.py

@@ -794,12 +794,16 @@ def write_source_url_if_unchanged(path: Path, record: dict[str, Any], url: str)
         current = json.loads(raw)
     except (OSError, UnicodeDecodeError, json.JSONDecodeError):
         return False
-    if not isinstance(current, dict) or content_hash(current) != content_hash(record):
+    if not isinstance(current, dict):
+        return False
+    source_urls = current.get("source_urls")
+    if isinstance(source_urls, list) and url in source_urls:
+        return True
+    if content_hash(current) != content_hash(record):
         return False
     updated = add_source_url_text(raw, url)
     if updated is None:
-        source_urls = current.get("source_urls")
-        return isinstance(source_urls, list) and url in source_urls
+        return False
     tmp = path.with_suffix(path.suffix + ".tmp")
     try:
         tmp.write_bytes(updated.encode("utf-8"))
@@ -1165,7 +1169,15 @@ def fetch(url: str) -> tuple[int | None, str, str]:
         digest = content_hash(record)
         cached = cache.get(rel)
         if cached and cached.get("hash") == digest and cached.get("decision") in DECISIONS:
-            result.rows.append(_row_from_cache(cached))
+            row = _row_from_cache(cached)
+            if not dry_run and cached["decision"] == CONFIRM:
+                url = cached.get("proposed_url")
+                if not isinstance(url, str) or not write_source_url_if_unchanged(
+                    repo / rel, record, url
+                ):
+                    row["decision"] = "write-failed"
+                    row["reason"] = "source-urls-write-failed-or-record-changed"
+            result.rows.append(row)
             result.cached += 1
             continue
         outcome = evaluate_record(record, fetcher, fetch)
@@ -1190,8 +1202,8 @@ def fetch(url: str) -> tuple[int | None, str, str]:
                     "source-urls-write-failed-or-record-changed",
                     outcome.suffix_only,
                 )
-        # Transient blocks stay uncached so a later run can retry them. A real
-        # CONFIRM is cached only after the guarded source_urls write succeeded.
+        # Transient blocks stay uncached so a later run can retry them. Cached
+        # dry-run CONFIRMs are written through the same guard on an apply run.
         if outcome.liveness not in RETRY_LIVENESS and outcome.reason != "network-error":
             append_cache(
                 cache_entry(rel_path=rel, record=record, result=outcome, ts=_now_iso()),

tests/verify/test_gsmarena_backfill.py

@@ -21,7 +21,6 @@
     add_source_url_text,
     backfill,
     compare_specs,
-    content_hash,
     gate_page,
     parse_cat_file_batch,
     parse_page,
@@ -370,7 +369,65 @@ def test_dry_run_does_not_write_the_record_and_caches(tmp_path: Path):
     )
     assert client.requests == 0
     assert again.cached == 1
-    assert content_hash(DX650_RECORD) == again.rows[0]["decision"] or again.cached == 1
+    assert again.rows[0]["decision"] == CONFIRM
+
+    # The cached confirmation still needs to be applied to the source file.
+    applied = backfill(
+        repo=repo,
+        cache_path=state / "gsmarena_backfill_cache.jsonl",
+        index_path=state / "phone_index.json",
+        summary_path=state / "summary.md",
+        limit=5,
+        sleep_s=0,
+        dry_run=False,
+        refresh_index=False,
+        client=client,
+        paths=[rel],
+        records={rel: DX650_RECORD},
+    )
+    url = "https://www.gsmarena.com/acer_dx650-2888.php"
+    written = target.read_text(encoding="utf-8")
+    assert json.loads(written)["source_urls"].count(url) == 1
+    assert applied.cached == 1
+    assert applied.rows[0]["decision"] == CONFIRM
+    assert client.requests == 0
+
+    # Applying again must leave the file byte-for-byte unchanged.
+    reapplied = backfill(
+        repo=repo,
+        cache_path=state / "gsmarena_backfill_cache.jsonl",
+        index_path=state / "phone_index.json",
+        summary_path=state / "summary.md",
+        limit=5,
+        sleep_s=0,
+        dry_run=False,
+        refresh_index=False,
+        client=client,
+        paths=[rel],
+        records={rel: DX650_RECORD},
+    )
+    assert target.read_text(encoding="utf-8") == written
+    assert reapplied.rows[0]["decision"] == CONFIRM
+    assert client.requests == 0
+
+    changed = {**DX650_RECORD, "battery_mah": 9999}
+    target.write_text(json.dumps(changed, indent=2) + "\n", encoding="utf-8")
+    stale = backfill(
+        repo=repo,
+        cache_path=state / "gsmarena_backfill_cache.jsonl",
+        index_path=state / "phone_index.json",
+        summary_path=state / "summary.md",
+        limit=5,
+        sleep_s=0,
+        dry_run=False,
+        refresh_index=False,
+        client=client,
+        paths=[rel],
+        records={rel: DX650_RECORD},
+    )
+    assert stale.rows[0]["decision"] == "write-failed"
+    assert url not in target.read_text(encoding="utf-8")
+    assert client.requests == 0
 
 
 def test_retry_after_is_parsed_and_honored_without_a_fixed_delay(tmp_path: Path, monkeypatch):