fix(layout): keep rows with vertical text as declared and their tables whole

CHANGELOG.md

@@ -6,6 +6,12 @@
 
 ### 고침
 
+- 쪽 수 추정(실험, `estimate_pages`)이 세로쓰기 칸(`hp:subList@textDirection="VERTICAL"`)의 글을 가로 글처럼
+  칸 폭에서 나누어 행을 키우던 것을 고친다.
+  - 한/글은 세로쓰기 칸의 줄을 칸 높이를 따라 세로로 놓고 옆으로 늘어놓는다. 넘치는 글은 잘리고, 행은 선언
+    높이 그대로다.
+  - 세로쓰기 칸이 있는 흐르는 표는 쪽 끝에서 나뉘지 않고, 나누지 않게 둔 표처럼 통째로 다음 쪽으로 간다.
+    기준 문단 줄과 그 뒤 글은 앞 쪽에 남는다.
 - 쪽 수 추정(실험, `estimate_pages`)이 줄 캐시 없는 표 칸에서 글자처럼 둔 표 옆에 글 앞·뒤로 둔 개체가 있는
   문단을 지원 밖(`a nested table`)으로 두던 것을 고친다.
   - 그런 개체는 자리를 차지하지 않으므로, 그 문단은 표 혼자인 문단처럼 표 높이의 한 줄이다.

src/hwpx/layout/pages.py

@@ -918,7 +918,30 @@ def _down(pos: Any) -> int:
     return 0 if offset < 0 or offset >= 1 << 31 else offset
 
 
+def _vertical(cell: Any) -> bool:
+    """Whether *cell* holds vertical text (``hp:subList@textDirection``): Hancom lays its lines down the
+    cell's height, side by side across it, cuts what does not fit and never makes its row taller."""
+
+    sub_list = cell.find(f"{HP}subList")
+    return sub_list is not None and sub_list.get("textDirection", "HORIZONTAL") != "HORIZONTAL"
+
+
+def _page_break(table: Any) -> str:
+    """How a flowing table goes over a page end (``hp:tbl@pageBreak``): one holding a cell of vertical text
+    does not split, as if set not to (NONE): it moves to the next page whole."""
+
+    if any(_vertical(cell) for row in table.findall(f"{HP}tr") for cell in row.findall(f"{HP}tc")):
+        return "NONE"
+    return table.get("pageBreak", "CELL")
+
+
 def _cell_row(measure: _Measure, table: Any, cell: Any) -> _Row:
+    if _vertical(cell):  # its lines go across it, the row as declared: one block, not split
+        margins = cell_margins_of(cell, table)
+        vertical = margins.top + margins.bottom
+        height = max(int(cell.find(f"{HP}cellSz").get("height", 0)), vertical)
+        return _Row(height, 1, height - vertical, height - vertical, vertical, cell.get("header") == "1",
+                    first=height - vertical)
     paragraphs = cell.findall(f"{HP}subList/{HP}p")
     nested = _holds_table(cell)
     drawn: tuple[tuple[int, int], ...] = ()
@@ -1660,7 +1683,7 @@ def _anchor(measure: _Measure, paragraph: Any, runs: list[Any], text: str, obj:
     top, bottom = (0, 0) if margin is None else (_margin(margin, "top"), _margin(margin, "bottom"))
     if _local(obj) == "tbl":
         rows, cells = _table_rows(measure, obj)
-        table = _FlowTable(rows, obj.get("pageBreak", "CELL"), obj.get("repeatHeader") == "1", (top, bottom),
+        table = _FlowTable(rows, _page_break(obj), obj.get("repeatHeader") == "1", (top, bottom),
                            tuple(cells), caption=_caption(measure, obj), cut=_spare_cut(obj))
         return _Anchor(line, table, 0)
     return _Anchor(line, None, _drawn_height(obj, measure) + top + bottom)
@@ -1688,7 +1711,7 @@ def _object_line(
         if name == "tbl":
             rows, cells = _table_rows(measure, obj)
             offset = int(pos.get("vertOffset", 0))  # one up (a negative offset, kept unsigned) starts at the line
-            table = _FlowTable(rows, obj.get("pageBreak", "CELL"), obj.get("repeatHeader") == "1", (top, bottom),
+            table = _FlowTable(rows, _page_break(obj), obj.get("repeatHeader") == "1", (top, bottom),
                                tuple(cells), 0 if offset < 0 or offset >= 1 << 31 else offset,
                                _caption(measure, obj), _spare_cut(obj))
             return count, size, pitch, table
@@ -1928,7 +1951,7 @@ def _wrapped_paragraph(measure: _Measure, page: _Page, paragraph: Any,
             rows, cells = _table_rows(measure, pusher)
             margin = pusher.find(f"{HP}outMargin")
             ends = (0, 0) if margin is None else (_margin(margin, "top"), _margin(margin, "bottom"))
-            table = _FlowTable(rows, pusher.get("pageBreak", "CELL"), pusher.get("repeatHeader") == "1", ends,
+            table = _FlowTable(rows, _page_break(pusher), pusher.get("repeatHeader") == "1", ends,
                                tuple(cells), caption=_caption(measure, pusher), cut=_spare_cut(pusher))
             return replace(para, band=_Band(para.wrap_anchor, offset, table)), None
         top = para.span(0, para.wrap_anchor) + offset
@@ -2029,7 +2052,7 @@ def _stack(measure: _Measure, objects: list[Any], column: int, shape: _Shape) ->
             rows, cells = _table_rows(measure, obj)
             margin = obj.find(f"{HP}outMargin")
             ends = (0, 0) if margin is None else (_margin(margin, "top"), _margin(margin, "bottom"))
-            table = _FlowTable(rows, obj.get("pageBreak", "CELL"), obj.get("repeatHeader") == "1", ends,
+            table = _FlowTable(rows, _page_break(obj), obj.get("repeatHeader") == "1", ends,
                                tuple(cells), caption=_caption(measure, obj), cut=_spare_cut(obj))
         stacked.append(_Stacked(int(pos.get("vertOffset", 0)), left, left + width, height, table))
     return tuple(stacked)

tests/test_layout_page_estimate.py

@@ -305,6 +305,14 @@
     "pages_table_nested_in_table_as_character": 1,  # a table in a cell of a table set as a character
     "pages_table_nested_row_split": 2,  # a row holding a table splits after its first line of text
     "pages_table_nested_row_moved": 2,  # none of a row holding a table fits: it goes on whole
+    # A flowing table whose first column holds vertical text (10 pt, spaced 160%): Hancom lays its lines down
+    # the cell's height, side by side across it, and cuts what does not fit; the row stays as declared:
+    "pages_table_vertical_cell_short_text": 1,  # three characters in a cell 3000 x 4000: one line
+    "pages_table_vertical_cell_text_past_its_height": 1,  # thirty: ten lines, the row still 4000 tall
+    "pages_table_vertical_cell_text_past_its_width": 1,  # sixty in a cell 1500 wide: twenty lines
+    "pages_table_vertical_cell_merged_over_rows": 1,  # ten in a cell merged over four rows 2000 tall
+    "pages_table_merged_cells_over_the_page_end": 2,  # twenty rows, cells of four merged, horizontal text:
+                                                     # the table splits at the page end
     "pages_table_nested_top_and_bottom_alone_then_a_line": 1,  # a table placed top and bottom alone in
                                                               # a cell paragraph, then a line: right
                                                               # below the table
@@ -888,6 +896,27 @@ def test_a_cell_holding_a_table_among_text_or_top_and_bottom_is_as_tall_as_hanco
         assert estimate_pages(_without_caches(data)).unsupported == ("section 0: a nested table",)
 
 
+@pytest.mark.parametrize("fixture", ["pages_table_vertical_cells_over_the_page_end",
+                                     "pages_table_vertical_cell_of_all_rows_over_the_page_end",
+                                     "pages_table_vertical_cell_in_every_row_over_the_page_end",
+                                     "pages_table_vertical_cells_taller_than_a_page"])
+def test_a_table_holding_vertical_text_moves_to_the_next_page_whole(fixture: str) -> None:
+    # pages_table_merged_cells_over_the_page_end with vertical text in its first column (cells of four rows
+    # merged, one of all twenty, or one in every row), and one of forty rows taller than a page: Hancom does
+    # not split it but moves it whole to the next page's top (running past the page's foot when taller), the
+    # anchor's line and the paragraphs after it staying on the first page.
+    data = (FIXTURES / f"{fixture}.hwpx").read_bytes()
+    hancom = _hancom_lines(data)
+
+    for document in (data, _without_caches(data)):
+        estimate = estimate_pages(document)
+        estimated = [[line.vertpos for line in lines] for lines in estimate.lines]
+        assert estimate.unsupported == ()
+        assert estimate.pages == 2
+        assert [mine for mine, theirs in zip(estimated, hancom) if theirs] == [theirs for theirs in hancom if theirs]
+        assert {line.page for lines in estimate.lines for line in lines} == {0}
+
+
 def test_a_negative_outer_margin_counts_as_none() -> None:
     # A table over two pages whose outer margins were -500 (kept as unsigned 32-bit numbers): Hancom laid it
     # out as one without them and saved them as 0. With the margins back at -500 the estimate is the same.