[DemchaAV]

Why

A layer stack is atomic in the layout, so a page break moves it whole. A stack of one layer is
written in the flow as plain paragraphs, and nothing told Word to keep them together. Word broke
between them, which left part of a stack the page sets on the next page at the foot of the page
above.

Measured in Word 16 and LibreOffice on Windows: a stack whose layer is a section of three lines,
"Opening", "Body" and "Tail", that the page moves to page 2 (baselines; p = page).

Stack of one layer	Page	Before, Word / LibreOffice	After, Word / LibreOffice
three plain lines	all on p2 (86.2, 99.2, 112.1)	Opening and Body on p1, Tail on p2 / same	all on p2 (85.3, 98.2, 111.2) / same
an 8pt spacer, then the lines	all on p2 (94.2, 107.2, 120.1)	Opening and Body on p1, Tail on p2 / same	all on p2 (93.0, 105.9, 118.9) / same
a 10pt margin on the first line	all on p2 (96.2, 109.2, 122.1)	Opening and Body on p1, Tail on p2 / same	all on p2 (95.2, 108.2, 121.2) / same

Nothing was named. A stack that stays on its page stood where the page sets it before, and still
does.

What changed

• writePlacedNode keeps a stack of one layer in the flow on its page as it keeps a block
  that asks for keepTogether(): through keepOnOnePage, when the layout put it on one page.
  • Each paragraph keeps its lines, and every one but the last keeps with the next. A table in
    it is chained row by row.
  • In a cell a stack's paragraphs are its table's, and are left to the row.
  • A stack of more than one layer is not kept here.
• The spacer's own keep is removed. fix(docx): keep the space above a spacer opening a stack of one layer #930 added followedInItsStack for a spacer opening such
  a stack on a new page. The stack's keep now holds every paragraph but its last to the next, so
  the separate keep is gone. The spacer's line still holds the space above it.
• Docs:
  • the capability matrix's keep row;
  • the recipe's sections row;
  • CHANGELOG v2.5.0.

Verification

• Corpus, 62 DOCX before and after: all 62 are byte-identical, and the report's 1246 notes
  are identical. A corpus document with a stack of one layer in the flow would have gained its
  keeps, so none of the 62 has one.

• Example: word-export-companion.docx was regenerated. It differs from the committed one only
  in docProps/core.xml, the export's time.

• fix(docx): keep the space above a spacer opening a stack of one layer #930's spacer cases, measured again on this code in both editors: the text stands at 107.0
  where the page sets it at 108.2, as before.

• DocxKeepTogetherTest, 5 new tests:

  • a stack is chained into one unit, and its last line is tied to nothing after it;
  • a row in a stack gets the keeps and the one-row table of a section kept together;
  • a stack in a row's cell is left to its row;
  • a band is not chained here;
  • without a layout nothing is kept.

  The fix(docx): keep the space above a spacer opening a stack of one layer #930 spacer tests in DocxStackOfOneLayerOnANewPageTest pass through the stack's keep.

• 4 sabotages, all caught:

  • no keep for a stack;
  • kept in a cell;
  • a band kept too;
  • kept with no layout.

• ./mvnw -B -ntp clean verify: BUILD SUCCESS.

  • Since then only comments, one test and the docs changed.
  • After them: render-docx's 1881 tests and qa's 27 documentation and guard tests pass, and so
    do the knowledge checks.

What this leaves, each recorded as a separate step:

• A stack of one layer inside a painted panel's cell: a panel is a table whose row may break,
  and the stack there is not kept. Whether Word holds a cell's paragraph keeps in a row that may
  break is not measured.
• A band (a stack of more than one layer) opening a page still breaks in Word. That is its
  own step.

Lane: render-docx (semantic backend). DOCX output changes only for a stack of one layer in
the flow, outside any cell.