From 8e00cef24ee366a079a5efc8844cf6410f2c26bf Mon Sep 17 00:00:00 2001
From: mizzlelover <mizzlelover@users.noreply.github.com>
Date: Thu, 10 Sep 2026 22:37:19 +0800
Subject: [PATCH] =?UTF-8?q?=E4=BF=AE=E5=A4=8D=E7=89=88=E5=A4=B4=E5=9D=90?=
 =?UTF-8?q?=E6=A0=87=E9=87=8F=E6=B5=8B=E7=9A=84=E4=B8=AD=E6=96=87=E5=88=86?=
 =?UTF-8?q?=E8=AF=8D=E5=85=BC=E5=AE=B9?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 tests/measure-header-coordinates.mjs | 45 +++++++++++++++++++++-------
 1 file changed, 35 insertions(+), 10 deletions(-)

diff --git a/tests/measure-header-coordinates.mjs b/tests/measure-header-coordinates.mjs
index a92da94..b07b87d 100644
--- a/tests/measure-header-coordinates.mjs
+++ b/tests/measure-header-coordinates.mjs
@@ -20,16 +20,41 @@ function decodeHtml(value) {
 
 function words(pdf) {
   const html = execFileSync("pdftotext", ["-f", "1", "-l", "1", "-bbox-layout", pdf, "-"], { encoding: "utf8" });
-  return [...html.matchAll(/<word\s+[^>]*yMin="([0-9.]+)"[^>]*>([\s\S]*?)<\/word>/g)].map((match) => ({
-    yMinPt: Number(match[1]),
-    text: decodeHtml(match[2]),
+  return [...html.matchAll(/<word\s+[^>]*xMin="([0-9.]+)"[^>]*yMin="([0-9.]+)"[^>]*xMax="([0-9.]+)"[^>]*>([\s\S]*?)<\/word>/g)].map((match) => ({
+    xMinPt: Number(match[1]),
+    yMinPt: Number(match[2]),
+    xMaxPt: Number(match[3]),
+    text: decodeHtml(match[4]),
   }));
 }
 
-function findWord(pdfWords, expected) {
-  const word = pdfWords.find(({ text }) => text.includes(expected));
-  if (!word) throw new Error(`${expected} was not found on the first page`);
-  return word;
+function findText(pdfWords, expected) {
+  // Poppler can split a Chinese phrase into adjacent <word> elements depending
+  // on the installed font. Rebuild visual rows before matching so the
+  // measurement is independent of that tokenization choice.
+  const rows = [];
+  for (const word of [...pdfWords].sort((left, right) => left.yMinPt - right.yMinPt || left.xMinPt - right.xMinPt)) {
+    const row = rows.find(({ yMinPt }) => Math.abs(yMinPt - word.yMinPt) <= 1.5);
+    if (row) row.words.push(word);
+    else rows.push({ yMinPt: word.yMinPt, words: [word] });
+  }
+  for (const row of rows) {
+    const ordered = row.words.sort((left, right) => left.xMinPt - right.xMinPt);
+    const text = ordered.map(({ text }) => text).join("");
+    const start = text.indexOf(expected);
+    if (start < 0) continue;
+    const matchingWords = [];
+    let offset = 0;
+    for (const word of ordered) {
+      const end = offset + word.text.length;
+      if (end > start && offset < start + expected.length) matchingWords.push(word);
+      offset = end;
+    }
+    return {
+      yMinPt: Math.min(...matchingWords.map(({ yMinPt }) => yMinPt)),
+    };
+  }
+  throw new Error(`${expected} was not found on the first page`);
 }
 
 function mm(word) {
@@ -38,9 +63,9 @@ function mm(word) {
 
 const withFields = words(withFieldsPdf);
 const withoutFields = words(withoutFieldsPdf);
-const copy = findWord(withFields, copyNo);
-const agencyWithFields = findWord(withFields, agencyName);
-const agencyWithoutFields = findWord(withoutFields, agencyName);
+const copy = findText(withFields, copyNo);
+const agencyWithFields = findText(withFields, agencyName);
+const agencyWithoutFields = findText(withoutFields, agencyName);
 const copyMm = mm(copy);
 const agencyWithFieldsMm = mm(agencyWithFields);
 const agencyWithoutFieldsMm = mm(agencyWithoutFields);