v0.5.2

Что то сделал

md2gost/page_geometry.py

@@ -2,13 +2,22 @@
 
 from __future__ import annotations
 
+import os
+import tempfile
+import zipfile
+from pathlib import Path
+
 from docx.enum.section import WD_ORIENT
 from docx.enum.text import WD_PARAGRAPH_ALIGNMENT
 from docx.oxml.ns import qn
 from docx.shared import Mm, Pt
+from lxml import etree
 
 from .util import create_element
 
+_W_NS = "http://schemas.openxmlformats.org/wordprocessingml/2006/main"
+_W = f"{{{_W_NS}}}"
+
 # GOST-like A4 margins (same as styles._apply_common_page_and_body)
 MARGIN_LEFT = Mm(30)
 MARGIN_RIGHT = Mm(10)
@@ -91,6 +100,10 @@ def apply_section_geometry(section, *, landscape: bool) -> None:
     if landscape:
         sect_pr.append(create_element("w:vAlign", {"w:val": "center"}))
 
+    # add_section copies template sectPr including w:pgNumType/@w:start — strip so
+    # landscape / return-to-portrait sections continue numbering instead of restarting.
+    clear_section_page_start(section)
+
 
 def content_size(*, landscape: bool) -> tuple:
     """Return (max_height, max_width) usable content area for LayoutTracker."""
@@ -131,31 +144,31 @@ def clear_section_footer(section) -> None:
         footer.add_paragraph()
 
 
+def clear_section_page_start(section) -> None:
+    """Remove w:pgNumType/@w:start so PAGE continues from the previous section."""
+    sect_pr = section._sectPr
+    for child in list(sect_pr):
+        if child.tag != qn("w:pgNumType"):
+            continue
+        if child.get(qn("w:start")) is not None:
+            del child.attrib[qn("w:start")]
+        # Empty pgNumType (no start/fmt) is useless — drop the element.
+        if len(child.attrib) == 0:
+            sect_pr.remove(child)
+
+
 def ensure_continuous_page_numbers(document) -> None:
     """
     After docxcompose of title/assignment: no PAGE on early sections,
     continuous numbering (do not restart at 1 on body section).
     """
-    from docx.oxml.ns import qn
-
     sections = list(document.sections)
     if not sections:
         return
 
-    # Heuristic: sections before the first that already has a PAGE field /
-    # or first N-1 if title was prepended — clear footer on all but keep
-    # continuous pgNumType. Body renderer already puts PAGE on body section;
-    # after compose, early sections may inherit footers — clear empty ones
-    # that belong to front matter (no Heading-like body content is hard to
-    # detect), so: clear footer on every section that has no PAGE field, and
-    # strip w:pgNumType start=1 everywhere.
+    # Strip w:pgNumType start everywhere so landscape / body sections stay continuous.
     for section in sections:
-        sect_pr = section._sectPr
-        for child in list(sect_pr):
-            if child.tag == qn("w:pgNumType"):
-                # Keep continuous: remove start attribute if present
-                if child.get(qn("w:start")) is not None:
-                    del child.attrib[qn("w:start")]
+        clear_section_page_start(section)
 
 
 def set_section_page_start(section, start: int) -> None:
@@ -191,3 +204,100 @@ def apply_page_number_start(
     if idx >= len(sections):
         return
     set_section_page_start(sections[idx], int(start))
+
+
+def _iter_sect_pr(document_root) -> list:
+    """Return sectPr elements in document order (inline + final body sectPr)."""
+    body = document_root.find(f"{_W}body")
+    if body is None:
+        return []
+    sects: list = []
+    for child in body:
+        tag = child.tag
+        if tag == f"{_W}sectPr":
+            sects.append(child)
+            continue
+        if tag != f"{_W}p":
+            continue
+        p_pr = child.find(f"{_W}pPr")
+        if p_pr is None:
+            continue
+        sect = p_pr.find(f"{_W}sectPr")
+        if sect is not None:
+            sects.append(sect)
+    return sects
+
+
+def _strip_pg_num_start_xml(sect_pr) -> None:
+    for child in list(sect_pr):
+        if child.tag != f"{_W}pgNumType":
+            continue
+        if child.get(f"{_W}start") is not None:
+            del child.attrib[f"{_W}start"]
+        if len(child.attrib) == 0:
+            sect_pr.remove(child)
+
+
+def _set_pg_num_start_xml(sect_pr, start: int) -> None:
+    pg = sect_pr.find(f"{_W}pgNumType")
+    if pg is None:
+        pg = etree.SubElement(sect_pr, f"{_W}pgNumType")
+    pg.set(f"{_W}start", str(int(start)))
+
+
+def patch_docx_page_starts(
+    path: str | Path,
+    start: int | None = None,
+    *,
+    front_sections: int = 0,
+) -> None:
+    """
+    Strip (and optionally re-apply) w:pgNumType/@w:start inside word/document.xml.
+
+    Used after Word COM Save, which often rewrites start=\"1\" onto every section.
+    Zip/lxml only — does not round-trip through python-docx (preserves
+    continuation captions and other Word-written parts).
+    """
+    abs_path = os.path.abspath(str(path))
+    if not os.path.isfile(abs_path):
+        return
+
+    with zipfile.ZipFile(abs_path, "r") as zin:
+        names = zin.namelist()
+        if "word/document.xml" not in names:
+            return
+        doc_bytes = zin.read("word/document.xml")
+        other = {n: zin.read(n) for n in names if n != "word/document.xml"}
+
+    root = etree.fromstring(doc_bytes)
+    sects = _iter_sect_pr(root)
+    for sect in sects:
+        _strip_pg_num_start_xml(sect)
+
+    if start is not None and int(start) >= 1:
+        idx = max(0, int(front_sections))
+        if idx < len(sects):
+            _set_pg_num_start_xml(sects[idx], int(start))
+
+    new_doc = etree.tostring(
+        root,
+        xml_declaration=True,
+        encoding="UTF-8",
+        standalone=True,
+    )
+
+    parent = os.path.dirname(abs_path) or "."
+    fd, tmp_name = tempfile.mkstemp(suffix=".docx", dir=parent)
+    os.close(fd)
+    try:
+        with zipfile.ZipFile(tmp_name, "w", compression=zipfile.ZIP_DEFLATED) as zout:
+            for name, data in other.items():
+                zout.writestr(name, data)
+            zout.writestr("word/document.xml", new_doc)
+        os.replace(tmp_name, abs_path)
+    except Exception:
+        try:
+            os.remove(tmp_name)
+        except OSError:
+            pass
+        raise