Files
aigc-home/docs/hel-892/refonly-document-evidence/finalize.py
T

177 lines
7.2 KiB
Python

"""Bind actual GET, reversible text, screenshots and completed checks."""
import hashlib
import json
import re
import shutil
from build_definition import OUT, REF, REPO
from rebuild import BASE
BROWSER = REPO / ".local-evidence/refonly-document/browser"
def sha(raw):
return hashlib.sha256(raw).hexdigest()
def write(name, data):
(OUT / name).write_text(json.dumps(data, ensure_ascii=False, indent=2) + "\n")
def main():
actual = json.loads((BROWSER / "actual-get.json").read_bytes())
export = OUT / ("actual-" + actual["name"] + ".json")
doc = json.loads(export.read_bytes())
assert actual["doc_json"] == doc
shutil.copyfile(BROWSER / "actual-get.json", OUT / "actual-get.json")
write(
"get-binding.json",
dict(
workflow_id=actual["id"],
get_sha256=sha((BROWSER / "actual-get.json").read_bytes()),
export_sha256=sha(export.read_bytes()),
doc_json_equals_isolated_export=True,
comparison="Parsed JSON content equality, not original HTTP serialization bytes",
pins=json.loads((OUT / "pins.json").read_bytes()),
),
)
for p in (REPO / ".local-evidence").glob("refonly-*.log"):
shutil.copyfile(p, OUT / (p.stem + ".txt"))
for name in [
"configure-refonly.py",
"prose-refonly.py",
"checks-refonly.py",
"browser-refonly.py",
"finish-refonly.py",
]:
shutil.copyfile(REPO / ".local-evidence" / name, OUT / (name + ".txt"))
for p in (REPO / ".local-evidence").glob("refonly-*.txt"):
shutil.copyfile(p, OUT / p.name)
b = doc["fixed_version"]
browser = json.loads((OUT / "browser-results.json").read_bytes())
backend = json.loads((OUT / "backend-request-counts.json").read_bytes())
reports = [OUT.parent / "多参生视频无审核版原保存与未知身份固定参考中文说明.md"]
assert len(reports) == 1
report = reports[0].read_text()
texts = []
for n in b["comments"]:
raw = n["text"]
literal = json.dumps(raw, ensure_ascii=False, indent=2)
assert "\n```json\n" + literal + "\n```" in report
assert json.loads(literal) == raw
renders = [r for r in browser["text_renderings"] if r["kind"] == "Note" + n["node_id"]]
assert len(renders) == 2 and all(r["decoded_sha256"] == sha(raw.encode()) for r in renders)
texts.append(
dict(
node_id=n["node_id"],
raw_codepoints=len(raw),
raw_sha256=sha(raw.encode()),
report_json_sha256=sha(literal.encode()),
decoded_sha256=sha(json.loads(literal).encode()),
CR=raw.count("\r"),
LF=raw.count("\n"),
literal_backslash_n=raw.count("\\n"),
tail_LF=raw.endswith("\n"),
browser_renderings=renders,
)
)
assert not b["comments"] and not doc.get("source_notes")
source_texts = []
for f in b["fields"]:
if f["node_id"] != "1" or f["field"] != "position_0":
continue
raw = f["value"]
renders = [r for r in browser["text_renderings"] if r["kind"] == "prompt1"]
assert len(renders) == 2 and all(
r["rendered_utf8_sha256"] == sha(raw.encode()) for r in renders
)
source_texts.append(
dict(
node_id="1",
characters=len(raw),
LF=raw.count("\n"),
CR=raw.count("\r"),
tail_LF=raw.endswith("\n"),
raw_sha256=sha(raw.encode()),
report_json_sha256=sha(json.dumps(raw, ensure_ascii=False).encode()),
decoded_sha256=sha(raw.encode()),
browser_renderings=renders,
)
)
write(
"rendered-text-sha256.json",
dict(raw_json_decoded_separate=True, notes=texts, source_texts=source_texts),
)
fields = []
original = json.loads((OUT / "definition-0.json").read_bytes())
for f, source_field in zip(b["fields"], original["fields"], strict=True):
assert f["pointer"] == source_field["pointer"]
assert f["value"] == source_field["value"]
assert type(f["value"]).__name__ == f["json_type"] == type(source_field["value"]).__name__
assert "named_pointer" not in f and "named_value" not in f
literal = json.dumps(source_field["value"], ensure_ascii=False, indent=2)
assert "\n```json\n" + literal + "\n```" in report
assert json.loads(literal) == f["value"]
fields.append(
dict(
pointer=f["pointer"],
named_state=f["named_state"],
json_type=f["json_type"],
source_order_report_json_sha256=sha(literal.encode()),
get_json_sha256=sha(json.dumps(f["value"], ensure_ascii=False, indent=2).encode()),
get_value_type=type(f["value"]).__name__,
)
)
write("saved-value-evidence.json", dict(saved_entries=40, fields=fields))
viewed = json.loads((OUT / "viewed-screenshots.json").read_bytes())
write(
"screenshot-review.json",
dict(
viewed=[dict(path=n, sha256=sha((BROWSER / n).read_bytes())) for n in viewed],
total=len(list(BROWSER.glob("*.png"))),
findings="代表默认用途、22未知身份、共享MODEL/独立Math/普通output链、保存边界/完整prompt/字符串64实际查看;长文DOM/SHA,不称逐字目验,无横溢,未改CSS/字体。",
),
)
tests = int(re.search(r"(\d+) passed", (OUT / "refonly-tests-final.txt").read_text())[1])
frozen = json.loads((OUT / "baseline-frozen-files.json").read_bytes())
closure = json.loads((OUT / "source-closure.json").read_bytes())
rows = json.loads((REF / "sources.json").read_bytes())["sources"]
write(
"validation-results.json",
dict(
base=BASE,
backend_tests=tests,
browser_assertions=browser["assertion_count"],
browser_groups=browser["assertionGroups"],
source_rows=len(rows),
source_closure_files=closure["count"],
accurate_nodes=0,
unresolved_implementation_nodes=0,
unknown_nodes=22,
saved_entries=40,
notes=0,
screenshots=len(list(BROWSER.glob("*.png"))),
viewed=len(viewed),
backend_fixed_gets=backend["fixed_count"],
browser_fixed_gets=browser["request_accounting"]["fixed_count"],
backend_polling_gets=backend["polling_count"],
browser_polling_gets=browser["request_accounting"]["polling_count"],
fixture_interceptions=len(browser["interceptions"]),
readiness_gets=1,
forbidden_requests=0,
frozen_baseline_files=frozen["count"],
old_fixed_blocks=133,
other_reviews=155,
frontend_build="rerun this batch",
full_browser="rerun this batch; no accumulation",
upstream_executed=False,
target_connected=False,
),
)
print("PASS actual GET/export,40 un-named original positions and full prompt raw/DOM SHA")
if __name__ == "__main__":
main()