@godcreator/word-report 0.4.1 → 0.5.1

package/dist/site.d.mts

@@ -1,5 +1,5 @@
-import { r as AssemblyResult, v as ReportDocument } from "./types-Bcl48xQN.mjs";
-import { a as ReportTocItem, i as ReportSummary, n as ReportPageComments, o as reportAssetHrefOf, r as ReportPageData, s as reportHrefOf, t as ReportListData } from "./page-data-CJ6nCBnW.mjs";
+import { r as AssemblyResult, y as ReportDocument } from "./types-Ea2OT6Ac.mjs";
+import { a as ReportTocItem, i as ReportSummary, n as ReportPageComments, o as reportAssetHrefOf, r as ReportPageData, s as reportHrefOf, t as ReportListData } from "./page-data-C1veAE55.mjs";
 import { CommentAvatar, CommentBodyComponent } from "@godcreator/comments/site";
 //#region src/site/report-pages.d.ts
 declare function ReportListPage(): import("react").JSX.Element;

package/dist/site.mjs

@@ -1,4 +1,4 @@
-import { t as assembleDocument } from "./assemble-BpllQNVY.mjs";
+import { t as assembleDocument } from "./assemble-BEt9_xsB.mjs";
 import { n as reportHrefOf, t as reportAssetHrefOf } from "./page-data-DLPoVgXy.mjs";
 import { Link, useLoaderData } from "react-router";
 import { DocsBody, DocsPage, DocsTitle } from "@fumadocs/base-ui/layouts/docs/page";

package/dist/{types-Bcl48xQN.d.mts → types-Ea2OT6Ac.d.mts}

@@ -7,6 +7,7 @@ type ReportInline = {
   type: 'math';
   id: string;
   latex: string;
+  plainText?: PlainMathRun[];
 } | {
   type: 'ref';
   id: string;
@@ -29,6 +30,11 @@ type ReportInline = {
 } | {
   type: 'break';
 };
+/** 行内公式在 Word 里写成文字时的一段：网页仍按 latex 显示，Word 照这些文字与标记写，不插 MathType */
+interface PlainMathRun {
+  text: string;
+  marks?: ('em' | 'sub' | 'sup')[];
+}
 interface Paragraph {
   type: 'paragraph';
   id: string;
@@ -194,4 +200,4 @@ declare class ReportContentError extends Error {
   constructor(message: string, file?: string, pointer?: string);
 }
 //#endregion
-export { SingleImageFigure as C, TableCell as E, Section as S, Table as T, ReportContentError as _, CellBlock as a, ReportInline as b, ContentWarning as c, FigurePanel as d, Include as f, ReportBlock as g, Paragraph as h, Bibliography as i, Equation as l, MultiImageFigure as m, AssemblyOptions as n, Code as o, IndexEntry as p, AssemblyResult as r, CompiledReport as s, Abstract as t, Figure as u, ReportDocument as v, SourceLocation as w, ReportList as x, ReportIndex as y };
+export { Section as C, TableCell as D, Table as E, ReportList as S, SourceLocation as T, ReportBlock as _, CellBlock as a, ReportIndex as b, ContentWarning as c, FigurePanel as d, Include as f, PlainMathRun as g, Paragraph as h, Bibliography as i, Equation as l, MultiImageFigure as m, AssemblyOptions as n, Code as o, IndexEntry as p, AssemblyResult as r, CompiledReport as s, Abstract as t, Figure as u, ReportContentError as v, SingleImageFigure as w, ReportInline as x, ReportDocument as y };

package/engine/README.md

@@ -1,6 +1,6 @@
 # Word report engine
 
-正式 Word COM / MathType 引擎迁入本仓。来源、提交与每份原始文件的 SHA-256 见 `migration-provenance.json`。仅迁入源仓已提交基线，未迁入源仓未提交代码、probe、旧运行输出或虚拟环境。Python 包版本仍为 0.0.1，依赖版本保持来源固定值。
+正式 Word COM / MathType 引擎迁入本仓。来源、提交与每份原始文件的 SHA-256 见仓根 `probes/2610131513_word-report-migration-baseline/migration-provenance.json`。仅迁入源仓已提交基线，未迁入源仓未提交代码、probe、旧运行输出或虚拟环境。Python 包版本仍为 0.0.1，依赖版本保持来源固定值。
 
 ## 环境
 
@@ -34,8 +34,8 @@
 
 ## 迁入回归
 
-2026-10-13 的证据摘要见 `migration-verification.json`：23 个 Python 单测与依赖检查通过；原 demo 经迁入代码重新生成 12 页 DOCX/PDF，6 个 MathType 保存重开、5 张表、2 张源图片、正文、域引用和临时编号增删检查通过，模板哈希未变。页面显示仍待用户检查。
+2026-10-13 的证据在仓根 `probes/2610131513_word-report-migration-baseline/`，摘要是其中的 `migration-verification.json`：23 个 Python 单测与依赖检查通过；原 demo 经迁入代码重新生成 12 页 DOCX/PDF，6 个 MathType 保存重开、5 张表、2 张源图片、正文、域引用和临时编号增删检查通过，模板哈希未变。页面显示仍待用户检查。
 
-首次运行在 `out/migration-baseline-20261013-01` 留下失败记录：已提交基线的链接锚点保留 `@LINK0@`。迁入包显式替换锚点文字后，在全新目录 `out/migration-baseline-20261013-02` 重新生成并通过自动验收；没有使用源仓未提交的修补代码。
+首次运行（同一目录的 `runs/migration-baseline-20261013-01`）留下失败记录：已提交基线的链接占位文字 `@LINK0@` 没被替换。迁入包显式替换占位文字后，在全新目录（`runs/migration-baseline-20261013-02`）重新生成并通过自动验收；没有使用源仓未提交的修补代码。
 
 独立 CLI 回归时，MathType 将读回文本的外层 `$...$` 换成了 `\[...\]`，旧校验错误地判定公式变化。当前只规范化一对外层数学分隔符，内部 TeX 仍逐字比较；不会改写公式、改 MathType 全局设置或忽略内容差异。对应回归同时检查错误公式与不匹配分隔符仍被拒绝。

package/engine/schemas/report-content.schema.json

@@ -143,6 +143,48 @@
         "latex": {
           "type": "string",
           "minLength": 1
+        },
+        "plainText": {
+          "type": "array",
+          "minItems": 1,
+          "description": "Word writes these runs as text instead of a MathType object; the web view still renders latex.",
+          "items": {
+            "type": "object",
+            "properties": {
+              "text": {
+                "$ref": "#/$defs/line"
+              },
+              "marks": {
+                "type": "array",
+                "items": {
+                  "enum": [
+                    "em",
+                    "sub",
+                    "sup"
+                  ]
+                },
+                "uniqueItems": true,
+                "not": {
+                  "allOf": [
+                    {
+                      "contains": {
+                        "const": "sub"
+                      }
+                    },
+                    {
+                      "contains": {
+                        "const": "sup"
+                      }
+                    }
+                  ]
+                }
+              }
+            },
+            "required": [
+              "text"
+            ],
+            "additionalProperties": false
+          }
         }
       },
       "required": [

package/engine/src/docreport/__main__.py

@@ -16,6 +16,7 @@ from types import SimpleNamespace
 from .content import read_json, validate_content, validate_layout
 from .compose import ContentComposer
 from .mathtype import DEFAULT_DLL, fill_and_reopen, validate_pending
+from .plain_math import clear_east_asian_hints
 from .verify import document_state, package_state, pdf_state, sha, content_state, abstract_state
 from .word import opened_document, refresh, word_application
 
@@ -114,9 +115,18 @@ def build(directory, output, dll=DEFAULT_DLL, *, data_only=False, layout_path=No
     evidence = dict(status="running", startedAt=datetime.now(timezone.utc).isoformat(),
                     report=str(directory), visualReview="pending", events=[])
 
+    # Rewriting the whole record per event is quadratic: a report with 858 equations logs ~3500 events.
+    last_written = [0.0]
+
+    def persist():
+        write_json(output / "verification.json", evidence)
+        last_written[0] = time.perf_counter()
+
     def record(event, **values):
+        values.setdefault('elapsedSeconds', round(time.perf_counter() - started, 3))
         evidence["events"].append(dict(event=event, **values))
-        write_json(output / "verification.json", evidence)
+        if time.perf_counter() - last_written[0] >= 2:
+            persist()
         print(json.dumps(dict(event=event, **values), ensure_ascii=True), flush=True)
 
     manifest = dict(report=str(directory), inputs=inputs,
@@ -142,10 +152,12 @@ def build(directory, output, dll=DEFAULT_DLL, *, data_only=False, layout_path=No
             manifest["environment"]["word"] = dict(version=word.Version, build=word.Build)
             doc = word.Documents.Add(Template=str(template), NewTemplate=False)
             doc.UpdateStylesOnOpen = False
+            record("compose-start")
             pending = module.build(doc, content, output)
             validate_pending(pending)
             doc.SaveAs2(str(draft), 16, AddToRecentFiles=False)
             record("draft-saved", equations=len(pending))
+        record("inline-math-hints-cleared", runs=clear_east_asian_hints(draft))
 
         if pending:
             fill_and_reopen(draft, destination, pending, record, dll.resolve())
@@ -156,16 +168,19 @@ def build(directory, output, dll=DEFAULT_DLL, *, data_only=False, layout_path=No
             with opened_document(word, destination) as doc:
                 refresh(doc)
                 doc.Save()
+            # Checks and the PDF read the field results as saved; refreshing again here would hide a stale save.
             with opened_document(word, destination, True) as doc:
-                refresh(doc)
                 generic = document_state(doc, len(pending))
                 record("document-check", **generic)
                 if content.get('format') == 'docreport':
                     abstracts = abstract_state(doc, content)
                     if abstracts:
                         record('abstract-check', abstracts=abstracts)
+                record("report-check-start")
                 record("report-check", result=module.validate(doc, content))
+                record("pdf-export-start")
                 doc.ExportAsFixedFormat(str(destination.with_suffix(".pdf")), 17)
+                record("pdf-export-finished")
             saved_hash = sha(destination)
             if callable(getattr(module, "validate_editing", None)):
                 with opened_document(word, destination, True) as doc:
@@ -195,6 +210,7 @@ def build(directory, output, dll=DEFAULT_DLL, *, data_only=False, layout_path=No
         record("exception", detail=traceback.format_exc())
         raise
     finally:
+        persist()
         write_json(output / "manifest.json", manifest)
 
 

package/engine/src/docreport/compose.py

@@ -7,7 +7,8 @@ Word objects and deferred MathType/field placeholders.
 from pathlib import Path
 import re
 
-from docreport.word import utf16_length, find_text
+from docreport.plain_math import INLINE_MATH_STYLE
+from docreport.word import utf16_length, find_text, MarkerFinder
 
 
 def mm(value):
@@ -179,23 +180,21 @@ class Composer:
         if self._finished:
             raise RuntimeError('Composer.finish() may only be called once.')
         doc = self.doc
+        fields = MarkerFinder(doc)
         for token, code, bookmark in self._pending_fields:
-            rng = find_text(doc, token)
+            rng = fields.find(token)
             field = doc.Fields.Add(rng, -1, code, False)
             field.Code.Font.Bold, field.Code.Font.Italic = 0, 0
             if bookmark:
                 doc.Bookmarks.Add(bookmark, field.Result)
+        numbers = MarkerFinder(doc)
         for begin, end, bookmark in self._pending_numbers:
-            rng = doc.Content.Duplicate
-            if not rng.Find.Execute(FindText=end, Wrap=0):
-                raise RuntimeError(f'Number boundary missing: {end}')
-            end_position = rng.Start
-            rng.Delete()
-            rng = doc.Content.Duplicate
-            if not rng.Find.Execute(FindText=begin, Wrap=0):
-                raise RuntimeError(f'Number boundary missing: {begin}')
-            start_position = rng.Start
-            rng.Delete()
+            opening = numbers.find(begin)
+            closing = numbers.find(end)
+            end_position = closing.Start
+            closing.Delete()
+            start_position = opening.Start
+            opening.Delete()
             doc.Bookmarks.Add(bookmark, doc.Range(
                 start_position, end_position - utf16_length(begin)))
         for table in doc.Tables:
@@ -227,6 +226,7 @@ class ContentComposer(Composer):
         self.heading_lists = {}
         self.appendix_numbering = None
         self.appendix_count = 0
+        self._inline_math_style = None
 
     def point(self):
         rng = self.doc.Content.Duplicate if self.container is None else self.container.Range.Duplicate
@@ -286,7 +286,13 @@ class ContentComposer(Composer):
             if kind == 'text':
                 value = node['text']
             elif kind == 'math':
-                value = self.math_token(node, True)
+                if 'plainText' not in node:
+                    value = self.math_token(node, True)
+                else:
+                    for run in node['plainText']:
+                        spans.append((utf16_length(text), utf16_length(run['text']), ['math'] + run.get('marks', [])))
+                        text += run['text']
+                    continue
             elif kind == 'ref':
                 value = self.ref_token(node)
             elif kind == 'break':
@@ -309,14 +315,39 @@ class ContentComposer(Composer):
     def rich_paragraph(self, nodes, style=None, prefix='', suffix=''):
         text, spans = self.inline(nodes)
         rng = self.paragraph(prefix + text + suffix, style)
+        self.apply_marks(rng, utf16_length(prefix), spans)
+        return rng
+
+    def apply_marks(self, rng, offset, spans):
         for start, length, marks in spans:
-            run = self.part(rng, utf16_length(prefix) + start, length)
+            if not marks:
+                continue
+            run = self.part(rng, offset + start, length)
             for mark in marks:
                 if mark == 'code':
                     run.Font.Name = 'Consolas'
+                elif mark == 'math':
+                    run.Style = self.inline_math_style()
                 else:
                     setattr(run.Font, {'strong': 'Bold', 'em': 'Italic', 'sup': 'Superscript', 'sub': 'Subscript'}[mark], True)
-        return rng
+
+    def inline_math_style(self):
+        """Character style of inline formulas written as text; created in the document, the DOTX stays unchanged."""
+        if self._inline_math_style is None:
+            style = self.doc.Styles.Add(INLINE_MATH_STYLE, 2)
+            style.Font.Name = 'Times New Roman'
+            self._inline_math_style = style
+        return self._inline_math_style
+
+    def single_paragraph_cell(self, cell, node):
+        """Write a one-paragraph cell in one assignment; the general path inserts then trims a paragraph."""
+        text, spans = self.inline(node['content'])
+        rng = cell.Range
+        rng.Text = text
+        rng = self.part(cell.Range, 0, utf16_length(text))
+        rng.Style = self.doc.Styles.Item(self.layout.get(node['id'], {}).get('style') or self.style)
+        self.apply_marks(rng, 0, spans)
+        self.mark(node['id'], rng, '_b')
 
     def chapter_sequence(self, kind, bookmark):
         """「章号-序号」 for an equation, figure or table in the current top-level chapter."""
@@ -489,11 +520,14 @@ class ContentComposer(Composer):
         saved = self.container, self.cell, self.style, self.width
         for r, c, rs, cs, cell in placements:
             self.container = cells[r, c].Cells.Item(1)
-            # Merge can concatenate the original empty cell paragraphs.
-            self.container.Range.Text = ''
             self.cell = True
             self.style = 'Table Header' if r < node['headerRows'] else 'Table Text'
             self.width = sum(widths[c:c + cs]) - 8
+            if len(cell['blocks']) == 1 and cell['blocks'][0]['type'] == 'paragraph':
+                self.single_paragraph_cell(self.container, cell['blocks'][0])
+                continue
+            # Merge can concatenate the original empty cell paragraphs.
+            self.container.Range.Text = ''
             self.blocks(cell['blocks'], depth)
             self.trim_container()
         self.container, self.cell, self.style, self.width = saved

package/engine/src/docreport/mathtype.py

@@ -10,7 +10,7 @@ import re
 import struct
 from hashlib import sha256
 
-from .word import find_text
+from .word import MarkerFinder
 
 DEFAULT_DLL = Path(r"C:\Program Files (x86)\MathType\MathPage\64\MathPage.wll")
 
@@ -138,12 +138,13 @@ def fill_and_reopen(draft, destination, pending, record, dll=DEFAULT_DLL):
         word.Visible, word.DisplayAlerts, word.AutomationSecurity = False, 0, 3
         missing = comtypes.automation.VARIANT.missing
         doc = word.Documents.Open(str(draft), False, False, False)
+        markers = MarkerFinder(doc)
         for item in pending:
             equation = item["equation"]
             tex = mathtype_tex(equation["latex"])
             if tex != equation["latex"]:
                 record("tex-adaptation", id=equation["id"], sourceTex=equation["latex"], mathTypeTex=tex)
-            target = find_text(doc, item["token"])
+            target = markers.find(item["token"])
             target.Text = ""
             shape = doc.InlineShapes.AddOLEObject("Equation.DSMT4", missing, False, False,
                                                  missing, missing, missing, target)

package/engine/src/docreport/plain_math.py

@@ -0,0 +1,53 @@
+"""Inline formulas written as Word text: their character style and the saved-package fix-up.
+
+The report marks each such formula with <WordPlainMath> and supplies the runs; the engine
+does not decide which TeX can become text.
+"""
+
+
+INLINE_MATH_STYLE = 'DR Inline Math'
+
+
+def clear_east_asian_hints(path):
+    """Drop w:hint="eastAsia" from runs in the inline-math character style.
+
+    Word adds the hint when it inserts φ, ≈ or ° into a Chinese paragraph, which renders
+    them in the East Asian font; COM exposes no setter for it, so the saved package is edited.
+    """
+    import os
+    import re
+    import tempfile
+    import zipfile
+
+    with zipfile.ZipFile(path) as source:
+        styles = source.read('word/styles.xml').decode('utf8')
+        match = re.search(r'<w:style\b[^>]*w:styleId="([^"]+)"[^>]*>(?:(?!</w:style>).)*?<w:name w:val="'
+                          + re.escape(INLINE_MATH_STYLE) + '"', styles, re.S)
+        if not match:
+            return 0
+        marker = f'<w:rStyle w:val="{match[1]}"/>'
+        cleared = 0
+
+        def clear(run):
+            nonlocal cleared
+            text = run[0]
+            if marker not in text or 'w:hint="eastAsia"' not in text:
+                return text
+            cleared += 1
+            text = text.replace(' w:hint="eastAsia"', '')
+            return text.replace('<w:rFonts/>', '')
+
+        fd, temporary = tempfile.mkstemp(suffix='.docx', dir=os.path.dirname(path))
+        os.close(fd)
+        try:
+            with zipfile.ZipFile(temporary, 'w') as target:
+                for item in source.infolist():
+                    data = source.read(item)
+                    if item.filename.startswith('word/') and item.filename.endswith('.xml') and marker.encode() in data:
+                        data = re.sub(r'<w:r(?:\s[^>]*)?>.*?</w:r>', clear, data.decode('utf8'), flags=re.S).encode('utf8')
+                    target.writestr(item, data, compress_type=item.compress_type)
+        except BaseException:
+            os.unlink(temporary)
+            raise
+    os.replace(temporary, path)
+    return cleared

package/engine/src/docreport/verify.py

@@ -59,6 +59,39 @@ def abstract_state(doc, content):
     return results
 
 
+WORD_NS = '{http://schemas.openxmlformats.org/wordprocessingml/2006/main}'
+# Word writes these into a field's result when a REF/SEQ target or switch is broken; readers then see them as body text.
+FIELD_ERROR_PREFIXES = ('错误!不能识别的开关参数', '错误!未找到引用源', '错误!找不到引用源', '错误!未定义书签',
+                        'error!unknownswitchargument', 'error!referencesourcenotfound', 'error!bookmarknotdefined')
+
+
+def field_results(root):
+    """Yield (field code, displayed result) for every simple and complex field in one XML part."""
+    open_fields = []  # [code, result, past separator] per nested complex field
+    for node in root.iter():
+        if node.tag == WORD_NS + 'fldSimple':
+            yield node.get(WORD_NS + 'instr', ''), ''.join(t.text or '' for t in node.iter(WORD_NS + 't'))
+        elif node.tag == WORD_NS + 'fldChar':
+            kind = node.get(WORD_NS + 'fldCharType')
+            if kind == 'begin':
+                open_fields.append(['', '', False])
+            elif kind == 'separate' and open_fields:
+                open_fields[-1][2] = True
+            elif kind == 'end' and open_fields:
+                code, result, _ = open_fields.pop()
+                yield code, result
+        elif open_fields and node.tag == WORD_NS + 'instrText' and not open_fields[-1][2]:
+            open_fields[-1][0] += node.text or ''
+        elif open_fields and node.tag == WORD_NS + 't' and open_fields[-1][2]:
+            open_fields[-1][1] += node.text or ''
+
+
+def reject_field_error_results(root):
+    for code, result in field_results(root):
+        if re.sub(r'\s+', '', result).lower().replace('！', '!').startswith(FIELD_ERROR_PREFIXES):
+            raise RuntimeError(f'Word field result error: {code.strip()}: {result}')
+
+
 def package_state(path, expected_equations):
     with ZipFile(path) as archive:
         objects, omml = [], []
@@ -66,6 +99,7 @@ def package_state(path, expected_equations):
             if not part.startswith('word/') or not part.endswith('.xml'):
                 continue
             root = ET.fromstring(archive.read(part))
+            reject_field_error_results(root)
             found = [o for o in root.findall(".//{urn:schemas-microsoft-com:office:office}OLEObject")
                      if o.get("ProgID") == "Equation.DSMT4"]
             objects.extend(found)
@@ -103,6 +137,11 @@ def content_state(path, content, directory):
                 if compact(node['text']) not in available:
                     raise RuntimeError(f"Source text missing after save: {node['text'][:100]}")
                 checked += 1
+            elif node.get('type') == 'math' and 'plainText' in node:
+                written = ''.join(run['text'] for run in node['plainText'])
+                if compact(written) not in available:
+                    raise RuntimeError(f"Inline formula text missing after save: {written[:100]}")
+                checked += 1
             elif node.get('type') == 'abstract':
                 keywords = '关键词：' + '；'.join(node['keywords'])
                 if compact(node.get('title', '摘要')) not in available or compact(keywords) not in available:

package/engine/src/docreport/word.py

@@ -72,6 +72,29 @@ def find_text(doc, text):
     raise ValueError(f"Generation marker missing: {text}")
 
 
+class MarkerFinder:
+    """Find generation markers written in body order without rescanning from the top each time.
+
+    The body story is searched from the previous body hit; a marker not found there
+    (notes, headers, or out of order) falls back to the full search of every story.
+    """
+
+    MAIN_STORY = 1
+
+    def __init__(self, doc):
+        self.doc = doc
+        self.position = 0
+
+    def find(self, text):
+        target = self.doc.Range(self.position, self.doc.Content.End)
+        target.Find.ClearFormatting()
+        if not target.Find.Execute(text, True, False, False, False, False, True, 0):
+            target = find_text(self.doc, text)
+        if target.StoryType == self.MAIN_STORY:
+            self.position = target.Start
+        return target
+
+
 def refresh(doc):
     """Resolve sequence fields before cross-references, then refresh TOC pages."""
     for _ in range(2):

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@godcreator/word-report",
-  "version": "0.4.1",
+  "version": "0.5.1",
   "author": {
     "name": "godcreator"
   },