a07663cbdb
- build.py：从「招标文件档案」和读标清单生成规矩投标文件稿；技术参数按 op / value / unit 算满足的值， 分项报价按分项上限比例分摊，落款日期取截止日前 3 天，偏离表按格式表头决定列（含设备名称列）， 读标漏抽的技术参数也补上响应；routes 支持「+」组合和 prose 散文响应。 - mutate.py：按词库注入要求项问题（单点 / 连续块，带不带标志词）和规则问题，写响应清单； 符合性条件响应表的问题按表级记一个。 - score.py：按单元去重、类型族比对、读标漏抽单列、端到端召回、无法对照单列。 - run_check.py：加只检索不调模型的基线；render.py：扫描件按最后一次出现定位。 - vocab-dev.json：开发集注入词库（留出集的另行封存，不进 git）。 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
184 lines
7.6 KiB
Python
184 lines
7.6 KiB
Python
"""把投标文件稿（spec.json）渲染成 PDF，并回填响应清单里的位置（evals/check/GUIDELINES.md §5、§6）。
|
||
|
||
cd server && ../.venv/bin/python ../evals/check/render.py <投标文件编号> [...] # 需要 Gotenberg（:3109）
|
||
|
||
- 每个分册一份 PDF：evals/out/check/<bid>/<分册编号>.pdf；
|
||
- `scan` 块单独成页，渲染后把那一页换成图片（没有文字层），用来检验扫描件的处理（C10）；
|
||
- 页眉写 volume.header（默认是本项目名称；注入 P06 时改成别的项目名称）；
|
||
- 渲染后用读标同一套切句找到每个响应块所在的页，写进 manifest.json 的 where（页码 + 片段）。
|
||
"""
|
||
|
||
from __future__ import annotations
|
||
|
||
import html
|
||
import io
|
||
import json
|
||
import pathlib
|
||
import sys
|
||
from typing import Any
|
||
|
||
import httpx
|
||
import pypdfium2 as pdfium
|
||
|
||
from bid_doctor.document.pipeline import text_layer
|
||
|
||
ROOT = pathlib.Path(__file__).resolve().parent
|
||
OUT = ROOT.parent / "out" / "check"
|
||
GOTENBERG = "http://127.0.0.1:3109/forms/chromium/convert/html"
|
||
SNIPPET = 24 # 回填位置时取块文字的前多少个字
|
||
|
||
CSS = """
|
||
@page { size: A4; margin: 22mm 18mm 20mm; }
|
||
body { font-family: "Noto Serif CJK SC", "Noto Sans CJK SC", serif; font-size: 11pt; line-height: 1.75; margin: 0; }
|
||
h1 { font-size: 17pt; text-align: center; margin: 0 0 12pt; }
|
||
h2 { font-size: 14pt; margin: 14pt 0 6pt; }
|
||
h3 { font-size: 12pt; margin: 10pt 0 4pt; }
|
||
p { margin: 0 0 6pt; text-align: justify; }
|
||
table { width: 100%; border-collapse: collapse; margin: 6pt 0 10pt; font-size: 10pt; }
|
||
th, td { border: 0.6pt solid #333; padding: 3pt 4pt; vertical-align: top; }
|
||
th { background: #f2f2f2; }
|
||
.page { page-break-before: always; }
|
||
.scan { page-break-before: always; page-break-after: always; padding: 24mm 10mm; border: 1pt solid #999; min-height: 150mm; }
|
||
.scan h3 { text-align: center; }
|
||
.sign { margin-top: 18pt; text-align: right; }
|
||
"""
|
||
|
||
|
||
def esc(s: Any) -> str:
|
||
return html.escape(str(s if s is not None else ""))
|
||
|
||
|
||
def block_html(b: dict[str, Any]) -> str:
|
||
t = b.get("type")
|
||
if t == "title":
|
||
return f"<h1>{esc(b['text'])}</h1>"
|
||
if t == "heading":
|
||
level = int(b.get("level", 2))
|
||
cls = ' class="page"' if b.get("new_page") else ""
|
||
return f"<h{level}{cls}>{esc(b['text'])}</h{level}>"
|
||
if t == "para":
|
||
return f"<p>{esc(b['text'])}</p>"
|
||
if t == "sign":
|
||
return "<div class='sign'>" + "".join(f"<p>{esc(x)}</p>" for x in b["lines"]) + "</div>"
|
||
if t == "table":
|
||
head = "".join(f"<th>{esc(h)}</th>" for h in b.get("header") or [])
|
||
rows = "".join("<tr>" + "".join(f"<td>{esc(c)}</td>" for c in r["cells"]) + "</tr>" for r in b["rows"])
|
||
title = f"<h3>{esc(b['title'])}</h3>" if b.get("title") else ""
|
||
return f"{title}<table><thead><tr>{head}</tr></thead><tbody>{rows}</tbody></table>"
|
||
if t == "scan":
|
||
lines = "".join(f"<p>{esc(x)}</p>" for x in b.get("lines") or [])
|
||
return f"<div class='scan'><h3>{esc(b['text'])}</h3>{lines}</div>"
|
||
raise ValueError(f"未知的块类型：{t}")
|
||
|
||
|
||
def volume_html(spec: dict[str, Any], vol: dict[str, Any]) -> tuple[str, str]:
|
||
body = "".join(block_html(b) for b in vol["blocks"])
|
||
header = esc(vol.get("header") or spec.get("project_name") or "")
|
||
# 页眉用 Gotenberg 的 header.html 传，每页都有；读标切句会把各页重复的行认成页眉页脚带，
|
||
# 查标的「别的项目名称」规则要专门扫这里（m3-plan §2.4）
|
||
return (
|
||
f"<!doctype html><html><head><meta charset='utf-8'><style>{CSS}</style></head><body>{body}</body></html>",
|
||
header,
|
||
)
|
||
|
||
|
||
def render_pdf(page_html: str, header: str) -> bytes:
|
||
head = (
|
||
"<html><head><style>body{font-family:'Noto Sans CJK SC',sans-serif;font-size:8pt;color:#555;"
|
||
"width:100%;margin:0 18mm;}</style></head><body>" + esc(header) + "</body></html>"
|
||
)
|
||
files = [
|
||
("files", ("index.html", page_html.encode(), "text/html")),
|
||
("files", ("header.html", head.encode(), "text/html")),
|
||
]
|
||
data = {"marginTop": "0.9", "marginBottom": "0.8", "marginLeft": "0.7", "marginRight": "0.7"}
|
||
r = httpx.post(GOTENBERG, files=files, data=data, timeout=120)
|
||
r.raise_for_status()
|
||
return r.content
|
||
|
||
|
||
def rasterize(pdf: bytes, pages: set[int]) -> bytes:
|
||
"""把指定页（从 1 起）换成图片页：没有文字层，模拟扫描件。"""
|
||
if not pages:
|
||
return pdf
|
||
src = pdfium.PdfDocument(pdf)
|
||
out = pdfium.PdfDocument.new()
|
||
for i in range(len(src)):
|
||
if i + 1 in pages:
|
||
img = src[i].render(scale=1.6).to_pil().convert("L")
|
||
buf = io.BytesIO()
|
||
img.save(buf, format="PDF", resolution=115)
|
||
out.import_pages(pdfium.PdfDocument(buf.getvalue()))
|
||
else:
|
||
out.import_pages(src, [i])
|
||
buf = io.BytesIO()
|
||
out.save(buf)
|
||
return buf.getvalue()
|
||
|
||
|
||
def page_texts(pdf: bytes) -> dict[int, str]:
|
||
pages = text_layer.extract("B", pdf)
|
||
return {p.page_no: text_layer.norm("".join(s.text for s in p.sentences)) for p in pages}
|
||
|
||
|
||
def block_text(b: dict[str, Any], row: dict[str, Any] | None = None) -> str:
|
||
if row is not None:
|
||
# 定位用投标人写的那一格（anchor 指定；默认第 3 格「投标文件中的内容」），不用照抄的要求原文
|
||
cells = [str(c) for c in row["cells"]]
|
||
at = row.get("anchor", 2 if len(cells) >= 3 else len(cells) - 1)
|
||
return cells[at]
|
||
if b.get("type") == "scan":
|
||
return str(b["text"])
|
||
return str(b.get("text") or " ".join(b.get("lines") or []))
|
||
|
||
|
||
def locate(texts: dict[int, str], snippet: str, last: bool = False) -> int | None:
|
||
"""last：取最后一次出现的页（附件在文件末尾，名称常和前面格式文件的标题相同）。"""
|
||
q = text_layer.norm(snippet)[:SNIPPET]
|
||
hits = [p for p, t in texts.items() if q and q in t]
|
||
return (hits[-1] if last else hits[0]) if hits else None
|
||
|
||
|
||
def render(bid: str) -> None:
|
||
d = OUT / bid
|
||
spec = json.loads((d / "spec.json").read_text())
|
||
manifest_path = d / "manifest.json"
|
||
manifest = json.loads(manifest_path.read_text()) if manifest_path.exists() else None
|
||
where: dict[str, list[dict[str, Any]]] = {}
|
||
problems: list[str] = []
|
||
for vol in spec["volumes"]:
|
||
page_html, header = volume_html(spec, vol)
|
||
pdf = render_pdf(page_html, header)
|
||
texts = page_texts(pdf)
|
||
scans = {p for b in vol["blocks"] if b.get("type") == "scan" and (p := locate(texts, b["text"], last=True))}
|
||
located: list[tuple[list[str], str]] = []
|
||
for b in vol["blocks"]:
|
||
if b.get("type") == "table":
|
||
located += [(r.get("responds") or [], block_text(b, r)) for r in b["rows"]]
|
||
else:
|
||
located.append((b.get("responds") or [], block_text(b)))
|
||
for units, snippet in located:
|
||
if not units:
|
||
continue
|
||
page = locate(texts, snippet)
|
||
if page is None:
|
||
problems.append(f"{vol['id']}：找不到「{snippet[:SNIPPET]}」")
|
||
continue
|
||
for u in units:
|
||
where.setdefault(u, []).append(
|
||
{"volume": vol["id"], "p": page, "t": text_layer.norm(snippet)[:SNIPPET]}
|
||
)
|
||
(d / f"{vol['id']}.pdf").write_bytes(rasterize(pdf, scans))
|
||
print(f"{bid}/{vol['id']}.pdf {len(texts)} 页，扫描页 {sorted(scans)}")
|
||
if manifest is not None:
|
||
for uid, entry in manifest["units"].items():
|
||
entry["where"] = where.get(uid, [])
|
||
manifest_path.write_text(json.dumps(manifest, ensure_ascii=False, indent=1))
|
||
for p in problems:
|
||
print(" ! " + p)
|
||
|
||
|
||
if __name__ == "__main__":
|
||
for b in sys.argv[1:]:
|
||
render(b)
|