python-hwpx 3.5.0__py3-none-any.whl → 3.6.0__py3-none-any.whl

hwpx/form_fit/engine.py

@@ -55,6 +55,26 @@ class FitEngine:
         policy: FitPolicy | None = None,
         *,
         field_id: str | None = None,
+    ) -> FitResult:
+        result = self._fit_decide(value, slot, policy, field_id=field_id)
+        inline_count = int(getattr(slot, "inline_object_count", 0) or 0)
+        if inline_count and (value or "") != "":
+            inline_width = float(getattr(slot, "inline_object_width", 0.0) or 0.0)
+            result.warnings.append(
+                f"cell line shared with {inline_count} inline control(s) "
+                f"({inline_width:.0f} HWPUNIT deducted from the width budget); "
+                "control-sharing cells are unusual fill targets — verify with a "
+                "render gate or pick the intended value cell"
+            )
+        return result
+
+    def _fit_decide(
+        self,
+        value: str,
+        slot: SlotMetrics,
+        policy: FitPolicy | None = None,
+        *,
+        field_id: str | None = None,
     ) -> FitResult:
         policy = policy or FitPolicy()
         value = "" if value is None else str(value)
@@ -77,6 +97,29 @@ class FitEngine:
         if policy.mode == "expand_row":
             return self._expand_row(value, slot, policy, field_id)
 
+        # Inline objects can consume the whole line; with less than a glyph of
+        # width left there is nothing to reshape into — refuse with a message
+        # that names the cause instead of an absurd wrapped-line estimate.
+        if (
+            int(getattr(slot, "inline_object_count", 0) or 0)
+            and slot.available_width < slot.font_pt * 100.0 * 0.6
+        ):
+            return FitResult(
+                ok=False,
+                value=value,
+                applied_value=value,
+                lines=0,
+                font_pt=slot.font_pt,
+                overflow_detected=True,
+                confidence="high",
+                errors=[
+                    "FIELD_OVERFLOW: inline control(s) leave no usable width in "
+                    "this cell; choose the intended value cell or allow row "
+                    "expansion"
+                ],
+                field_id=field_id,
+            )
+
         # Reshaping ladder: wrap, shrink, or wrap_then_shrink, then fail_on_overflow.
         wrap_lines = policy.effective_max_lines if policy.may_wrap else 1
         wrap_slot = replace(slot, max_lines=wrap_lines)

hwpx/form_fit/measure.py

@@ -221,6 +221,13 @@ class SlotMetrics:
     # Per-line advance as a multiple of the em, from the cell's declared paragraph
     # line spacing (PERCENT). ``None`` falls back to ``DEFAULT_LINE_SPACING_RATIO``.
     line_spacing_ratio: float | None = None
+    # Width already consumed on the line by inline treat-as-char objects
+    # (checkboxes, form controls, pictures) that share the target paragraph.
+    # Their declared ``hp:sz/@width`` is subtracted from the usable width —
+    # ignoring them made the engine call "fits" on a fill that real Hancom
+    # wrapped, growing the row and repaginating a 10-page form (S-087 P0).
+    inline_object_width: float = 0.0
+    inline_object_count: int = 0
     # A cell height existed but was unusable (merged row-span fragment, or an
     # auto-grow floor shorter than one line). Records "height budget unavailable"
     # so the fit reports width-only honestly rather than guessing a vertical fit.
@@ -473,6 +480,11 @@ def resolve_slot_metrics(
     element = getattr(cell, "element", None)
     left, right = _cell_margin(element) if element is not None else (0, 0)
     inner = max(raw_width - left - right, 0.0) * safety
+    inline_width, inline_count = (
+        _inline_object_width(element) if element is not None else (0.0, 0)
+    )
+    if inline_width:
+        inner = max(inner - inline_width, 0.0)
     resolved_pt = font_pt if font_pt is not None else _first_run_font_pt(cell, document)
     line_ratio = _first_para_line_spacing_ratio(cell, document)
 
@@ -503,9 +515,49 @@ def resolve_slot_metrics(
         available_height=available_height,
         line_spacing_ratio=line_ratio,
         height_unavailable=height_unavailable,
+        inline_object_width=inline_width,
+        inline_object_count=inline_count,
     )
 
 
+def _inline_object_width(cell_element: object) -> tuple[float, int]:
+    """Total declared width of inline treat-as-char objects in the cell.
+
+    Checkboxes and similar form controls flow on the text line
+    (``hp:pos/@treatAsChar='1'``) and consume their ``hp:sz/@width``; a fill
+    value shares whatever width remains. Objects without a declared size are
+    counted but contribute 0 width (the count still signals reduced trust).
+    """
+
+    total = 0.0
+    count = 0
+    iter_fn = getattr(cell_element, "iter", None)
+    if iter_fn is None:
+        return 0.0, 0
+    for node in iter_fn():
+        tag = getattr(node, "tag", "")
+        if not isinstance(tag, str):
+            continue
+        local = tag.rsplit("}", 1)[-1].lower()
+        if local != "pos":
+            continue
+        if (node.get("treatAsChar") or "").strip() not in {"1", "true", "TRUE"}:
+            continue
+        holder = node.getparent() if hasattr(node, "getparent") else None
+        if holder is None:
+            continue
+        count += 1
+        for sibling in holder:
+            sib_local = str(getattr(sibling, "tag", "")).rsplit("}", 1)[-1].lower()
+            if sib_local == "sz":
+                try:
+                    total += float(sibling.get("width") or 0)
+                except (TypeError, ValueError):
+                    pass
+                break
+    return total, count
+
+
 __all__ = [
     "SlotMetrics",
     "Measurement",

hwpx/opc/package.py

@@ -159,10 +159,69 @@ def _local_name(element: etree._Element) -> str:
         return tag
 
 
+def _paragraph_plain_text_length(paragraph: etree._Element) -> int | None:
+    """Plain text length of a simple paragraph, or ``None`` when unjudgeable.
+
+    Mirrors the oxml-side stale detector: only paragraphs made purely of runs
+    with text/tab-like children can be judged at the byte boundary; anything
+    else (controls, tables, fields) returns ``None`` so its cache is kept.
+    """
+    total = 0
+    for child in paragraph:
+        child_name = _local_name(child).lower()
+        if child_name in _LAYOUT_CACHE_ELEMENT_NAMES:
+            continue
+        if child_name != "run":
+            return None
+        for run_child in child:
+            run_child_name = _local_name(run_child).lower()
+            if run_child_name == "t":
+                total += len("".join(run_child.itertext()))
+            elif run_child_name in {"tab", "linebreak", "hyphen", "nbspace"}:
+                total += 1
+            else:
+                return None
+    return total
+
+
+def _paragraph_layout_cache_is_stale(paragraph: etree._Element) -> bool:
+    text_length = _paragraph_plain_text_length(paragraph)
+    if text_length is None:
+        return False
+    for child in paragraph:
+        if _local_name(child).lower() not in _LAYOUT_CACHE_ELEMENT_NAMES:
+            continue
+        for line_seg in child:
+            if _local_name(line_seg).lower() != "lineseg":
+                continue
+            textpos = line_seg.get("textpos")
+            if textpos is None:
+                continue
+            try:
+                if int(textpos) > text_length:
+                    return True
+            except ValueError:
+                return True
+    return False
+
+
 def _strip_section_layout_caches(payload: bytes) -> bytes:
+    """Drop only *stale* paragraph layout caches from a section payload.
+
+    Valid caches are the authored layout of untouched paragraphs; removing
+    them forces the editor to re-lay-out entire multi-page documents, which
+    shifted page counts and stacked glyphs in the wild form-fill differential
+    (specs/031 P0 receipt). The mutating APIs upstream clear the caches of
+    exactly the paragraphs they touch, so this byte-boundary sweep only
+    guards against writers that left a provably stale cache behind.
+    """
     root = parse_xml(payload)
     removed = False
     for parent in root.iter():
+        if _local_name(parent).lower() != "p":
+            continue
+        if not _paragraph_layout_cache_is_stale(parent):
+            continue
         for child in list(parent):
             if _local_name(child).lower() in _LAYOUT_CACHE_ELEMENT_NAMES:
                 parent.remove(child)

hwpx/oxml/document_parts.py

@@ -1086,10 +1086,14 @@ class HwpxOxmlDocument:
         if self._manifest_dirty:
             updates[self._manifest_path] = _serialize_xml(self._manifest)
         for section in self._sections:
-            if section.dirty:
-                section.remove_layout_caches()
-            else:
-                section.remove_stale_layout_caches()
+            # Edit-scoped invalidation: the mutating APIs (cell/paragraph/run
+            # text and style setters) clear the caches of exactly the
+            # paragraphs they touch, so even a dirty section only needs the
+            # stale sweep as a safety net. Nuking every cache here forced
+            # Hancom to re-lay-out untouched pages of multi-page forms, which
+            # is what stacked glyphs and shifted page counts in the wild
+            # form-fill differential (specs/031 P0 receipt).
+            section.remove_stale_layout_caches()
         for section in self._sections:
             if section.dirty:
                 updates[section.part_name] = section.to_bytes()

hwpx/oxml/paragraph.py

@@ -914,6 +914,9 @@ class HwpxOxmlParagraph:
                     changed = True
 
         if changed:
+            # A style swap changes glyph metrics, so the cached line layout of
+            # this paragraph no longer holds.
+            _clear_paragraph_layout_cache(self.element)
             self.section.mark_dirty()
 
 __all__ = ["HwpxOxmlParagraph"]

hwpx/oxml/run.py

@@ -320,12 +320,16 @@ class HwpxOxmlRun:
         if value is None:
             if "charPrIDRef" in self.element.attrib:
                 del self.element.attrib["charPrIDRef"]
+                _clear_paragraph_layout_cache(self.paragraph.element)
                 self.paragraph.section.mark_dirty()
             return
 
         new_value = str(value)
         if self.element.get("charPrIDRef") != new_value:
             self.element.set("charPrIDRef", new_value)
+            # A style swap changes glyph metrics, so the cached line layout of
+            # the containing paragraph no longer holds.
+            _clear_paragraph_layout_cache(self.paragraph.element)
             self.paragraph.section.mark_dirty()
 
     def _plain_text_nodes(self) -> list[ET.Element]:

hwpx/oxml/table.py

@@ -209,6 +209,7 @@ class HwpxOxmlTableCell:
                 sublist.set("lineWrap", "BREAK")
         if split_paragraphs:
             self._set_split_paragraph_text(sanitized_value)
+            self._clear_own_layout_caches()
             self.element.set("dirty", "1")
             self.table.mark_dirty()
             return
@@ -226,9 +227,19 @@ class HwpxOxmlTableCell:
                 current = current.getparent() if hasattr(current, "getparent") else None
             if current is not None:
                 current.set("charPrIDRef", "0")
+        self._clear_own_layout_caches()
         self.element.set("dirty", "1")
         self.table.mark_dirty()
 
+    def _clear_own_layout_caches(self) -> None:
+        # Edit-scoped invalidation: only this cell's paragraphs lose their
+        # layout caches, so Hancom re-lays-out just the filled cell instead of
+        # the whole section (specs/031 P0 — the whole-section nuke shifted
+        # pages and stacked glyphs on multi-page forms).
+        for node in self.element.iter():
+            if _element_local_name(node) == "p":
+                _clear_paragraph_layout_cache(node)
+
     def remove(self) -> None:
         self._row_element.remove(self.element)
         self.table.mark_dirty()

{python_hwpx-3.5.0.dist-info → python_hwpx-3.6.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-hwpx
-Version: 3.5.0
+Version: 3.6.0
 Summary: 한글 없이 HWPX 문서를 열고, 편집하고, 생성하고, 검증하는 Python 자동화 라이브러리
 Author: python-hwpx Maintainers
 License-Expression: Apache-2.0
@@ -70,7 +70,7 @@ Dynamic: license-file
 `hwpx-mcp-server`와 `hwpx-plugin`은 같은 프로젝트가 직접 유지보수하는 first-party 연동 구성요소입니다.
 “first-party”는 프로젝트 유지보수 관계를 뜻하며, 한컴 또는 제3자의 공식 인증을 뜻하지 않습니다.
 
-현재 PyPI 공개 릴리스는 `python-hwpx 3.5.0`입니다. 일반
+현재 PyPI 공개 릴리스는 `python-hwpx 3.6.0`입니다. 일반
 `pip install python-hwpx`로 이 릴리스를 설치할 수 있습니다.
 현재 패키지 분류는 `Development Status :: 3 - Alpha`입니다. 이 분류는 API와 제품의
 성숙도를 나타내며, 공개 버전이나 플러그인의 최소 호환 버전을 대신하지 않습니다.

{python_hwpx-3.5.0.dist-info → python_hwpx-3.6.0.dist-info}/RECORD

@@ -94,8 +94,8 @@ hwpx/exam/parser.py,sha256=hA_oq_8i1XpmzJ8xw1L7R6LenxREKRvlOwMMXbs9Mpc,5026
 hwpx/exam/profile.py,sha256=pEYpgwoNUuMToHMJjiknHykjXs0UobGuYW67N_3VnEQ,4071
 hwpx/form_fit/__init__.py,sha256=l1HcFge4u_wL2vogB-VxSMrb-8h7zUT_IgTWMtnH9CM,1779
 hwpx/form_fit/apply.py,sha256=X2q7K-mLku4AXxm4-xP80QGzhBlpzpqA_Ade8aPXzNY,3450
-hwpx/form_fit/engine.py,sha256=N6twcIS-xjVdkCtVy03cHG8dGtRDtYV7pTRMhsPQ8rc,19799
-hwpx/form_fit/measure.py,sha256=HZfcsCrvv4GOPIp8Tiz7BHpw_Hv0LGWrpkpV19PKFPg,20221
+hwpx/form_fit/engine.py,sha256=DiAGTRbpaH3WJNAiimRiv9KWHA-ijtUd5W2za-QhaWQ,21576
+hwpx/form_fit/measure.py,sha256=gnSfqccm7_3oTKv3HH6Zs77r2bToxzmfvUWN2zshDE8,22323
 hwpx/form_fit/policy.py,sha256=iXAatjS7ZycOdKgMCL4Q_kYg0qnogFPdRNlr_ecOvjc,3106
 hwpx/form_fit/report.py,sha256=h1NzQfbj1JYvZtKUi0DckJNWGnUHMZtQ3iCzbUxUjHA,3305
 hwpx/form_fit/seal.py,sha256=QsPzFagaEi7UKAXS-iFwiZs8rLG-_Fo9RixLc2OaqbU,17640
@@ -106,7 +106,7 @@ hwpx/ingest/hwpx_converter.py,sha256=HOUCFYO-HK4WRpWiN5sA_kZIvdv439mkZ6PMCFU3hLg
 hwpx/layout/__init__.py,sha256=qotAVhol_yrPlOHvQfs68nx7ADYxPcwITYNA9gMutUM,1158
 hwpx/layout/lint.py,sha256=byhIkP5av_il1cDdm71Oj_Em7-HUuU0w1nwhzu8tpto,15410
 hwpx/layout/report.py,sha256=lwTIEWrZRHXc02k8_vpcaXNEcU1dDQ3dZ48y7Fxkxkc,3957
-hwpx/opc/package.py,sha256=4Ydaw-1uyTZ7DwNZ89ICJ7CHYwUGzeGR-FbZiyJ5XWk,35484
+hwpx/opc/package.py,sha256=7i9h5HU2Vwfl5-SWPy8OHIfmLxrxTbkGnSqIc-Z0IUk,37824
 hwpx/opc/relationships.py,sha256=tPWLHRMlw0Spvtwou2jCDRfHdcm9FEKKLd95YVHLwYI,6971
 hwpx/opc/security.py,sha256=hsA73sZxUoGIgy2zue9EnV7ChHMfasnplmqzxFs3Mp4,4419
 hwpx/opc/xml_utils.py,sha256=L_fHY1-D5I_TfdRkDQV-bn55EnXc6AqEDWItfMpawVs,3840
@@ -117,22 +117,22 @@ hwpx/oxml/body.py,sha256=j-MJFHgInfOuFYYf-tsjFXvzPvKLt7HqUvc5Lhmp0is,30586
 hwpx/oxml/canonical_defaults.py,sha256=WHAK7u_W-PDv3P3N1-1-aWAoiUgBh4x-UybFL4NqW-w,3791
 hwpx/oxml/common.py,sha256=TJkafzg7x4T3J29tZchRZk57ZTsrM9PEiqGT3rX3w5o,1044
 hwpx/oxml/document.py,sha256=0Ml-t2fwa4f2xMfu5jbkOoh_KDKCCf3gNpaYICRV-KY,1172
-hwpx/oxml/document_parts.py,sha256=yd12u15Yzb10gtpdJfSLzOZkKdj1xY2xHYs5CbbB_TA,41281
+hwpx/oxml/document_parts.py,sha256=fbj_MA8imnKhz6XN2ufNAMPYNI5xi0dCNjEv0nGESwo,41692
 hwpx/oxml/header.py,sha256=4LtTDPe7YW0o_4jPv3Rd-vyv5suRfebecxtTWbF65Uw,50011
 hwpx/oxml/header_part.py,sha256=BIwdzDlwJSxcVIErZ8jQuT2vQ4Tndr4Xa87beYbpSFU,50640
 hwpx/oxml/memo.py,sha256=QlmuELEzywnmywgjm94S3vzJB8wPpoFOmI8KVuzoNPw,9484
 hwpx/oxml/namespaces.py,sha256=c7JfdOdJbzrhyHvjbxoeeeloRE3xPuoB7v3YI8SmKDk,6524
 hwpx/oxml/numbering.py,sha256=9a0ARGW1DkKsTF7mEEXM4FA5loQmzIBuqE4APA8UFIQ,669
 hwpx/oxml/objects.py,sha256=44778UkUuoNZoAxIbaA6VjAbzDm0auV7TPtllhQyln0,15767
-hwpx/oxml/paragraph.py,sha256=YQXMkcjUVwPe0FanPY1VdJXzIT9ZDNOWTgyD3XwSVLk,32712
+hwpx/oxml/paragraph.py,sha256=-E8A-DjKagjm47xNPnPkbIUYWNdEdAQowzl4X-EP6uI,32893
 hwpx/oxml/parser.py,sha256=pIfyNdW3WdFkCcE8JFY7hn1fobRbMngzfhVZICvWeVs,2937
-hwpx/oxml/run.py,sha256=L-ZhZ145eFG8OhrkSHHTWJ2Obp5O4SgXkA1X8fvyuHo,15232
+hwpx/oxml/run.py,sha256=F939J1W6zQjr-JQ1Z8-nKh2s7NQ_0Okov4QJAEyw_5M,15503
 hwpx/oxml/schema.py,sha256=ElR3_IIhhPPZEqJtNKMNCHA-VFVLdhL454W_4spuvlc,1284
 hwpx/oxml/section.py,sha256=i3t1SKIMneA1qWAajDlmZ-yrWJF9VvqT-aoq3mRXtxE,11638
 hwpx/oxml/section_format.py,sha256=EKqod989mzcaPz-lDKM1Y7atdpFX9x_SdQ8zfCL3BpU,18320
 hwpx/oxml/section_story.py,sha256=zvUV7O4dtogWcM9-zudRAszCHRwfzTBMW2GGI1Q24kg,22410
 hwpx/oxml/simple_parts.py,sha256=dOVKWlUCQV8GePH0WRWPUmRylvH5O8bu0B6jj_4lejk,1864
-hwpx/oxml/table.py,sha256=w34AEoBC1Jkhks7OUwQM8MnQpEQMZ7ptdMlnS_mEY7k,39145
+hwpx/oxml/table.py,sha256=SPm4a3s_xSq1DNBh-2VSK87mIGroPnSyLeBSM50-4ok,39708
 hwpx/oxml/utils.py,sha256=2L9au3SD_MFkFt2DXx3udXBT5caYuYO623lhGhgldVo,2972
 hwpx/presets/__init__.py,sha256=I3gSJDT2tuvxXbw3FOmI3CnbxvjDMPZfppyg6OlMHOI,535
 hwpx/presets/proposal.py,sha256=2yzLfDxOAT0P19dPX1RBWh56_AIOMsWqINDadtnKQoc,21176
@@ -199,10 +199,10 @@ hwpx/visual/page_qa.py,sha256=AuJCySfLIdXjPeekhAc3YnrIWOsN0sI35Pnz8BZ6Pro,7278
 hwpx/visual/qa_contracts.py,sha256=1cjoiRTAWgi7NgE8SPc6bKxdoUfKx9nyou1mgC47TxY,10506
 hwpx/visual/qa_metrics.py,sha256=I7RdybN-f1Fvcv2j-ceDPoek51nOkMui1Zrd7TEX_C8,9838
 hwpx/visual/report.py,sha256=2RhXN1KBYOZTim9FNpeUhaaDHR7oFxI6Z2DLUkDIiwE,1717
-python_hwpx-3.5.0.dist-info/licenses/LICENSE,sha256=_ubz4wv-BkkT3l3gu-QuH7JGeVjuRYGZoZK95eNsCHU,9688
-python_hwpx-3.5.0.dist-info/licenses/NOTICE,sha256=k48h6EaGQE8Y1c0dS9sIOOcz4YqkbcImWClF7pBOgsg,2473
-python_hwpx-3.5.0.dist-info/METADATA,sha256=dokrEArymZrHXyslBq1OhwQwYg3NhnNSeFLpHx10ppk,13561
-python_hwpx-3.5.0.dist-info/WHEEL,sha256=K260EYznzXsJYBQGqmI8VTxEdiZYNvDZwW9cBh9-_MA,91
-python_hwpx-3.5.0.dist-info/entry_points.txt,sha256=PVUBTBQtBHiflQ9XBjfd004w-LrDibgZjwzxgu8Tx7w,480
-python_hwpx-3.5.0.dist-info/top_level.txt,sha256=R1iToqDh80Nf2oQhRjTN0rbN2X6kyDUizIocZjkhuxc,5
-python_hwpx-3.5.0.dist-info/RECORD,,
+python_hwpx-3.6.0.dist-info/licenses/LICENSE,sha256=_ubz4wv-BkkT3l3gu-QuH7JGeVjuRYGZoZK95eNsCHU,9688
+python_hwpx-3.6.0.dist-info/licenses/NOTICE,sha256=k48h6EaGQE8Y1c0dS9sIOOcz4YqkbcImWClF7pBOgsg,2473
+python_hwpx-3.6.0.dist-info/METADATA,sha256=L4frTvOU4Mi8VP5EDr3ps2UPQ5c5Qx1Yadn0uu_CXVU,13561
+python_hwpx-3.6.0.dist-info/WHEEL,sha256=K260EYznzXsJYBQGqmI8VTxEdiZYNvDZwW9cBh9-_MA,91
+python_hwpx-3.6.0.dist-info/entry_points.txt,sha256=PVUBTBQtBHiflQ9XBjfd004w-LrDibgZjwzxgu8Tx7w,480
+python_hwpx-3.6.0.dist-info/top_level.txt,sha256=R1iToqDh80Nf2oQhRjTN0rbN2X6kyDUizIocZjkhuxc,5
+python_hwpx-3.6.0.dist-info/RECORD,,