Merge pull request #6047 from vangel76/fix/5922-doc-header-empty-para

.doc: first body line no longer starts too low under a header

crates/docbin/src/lib.rs

@@ -348,7 +348,7 @@ fn read_limited(bytes: &[u8], password: Option<&str>, max_paragraphs: usize) ->
                 0 | 1 | 4 => PartKind::Header,
                 _ => PartKind::Footer,
             };
-            let blocks = story_blocks(&ctx, a, b, &mut pics, max_paragraphs)?;
+            let blocks = header_story_blocks(&ctx, a, b, &mut pics, max_paragraphs)?;
             if !blocks.is_empty() {
                 let id = doc.add_part(kind, blocks);
                 hf_ids.push((si, k, id));
@@ -436,6 +436,36 @@ fn story_blocks(ctx: &WalkCtx<'_>, a: u32, b: u32, pics: &mut media::Pictures, m
     Ok(table::assemble(paras))
 }
 
+/// A header or footer story's blocks without its guard paragraph mark. [MS-DOC] 2.3.3: a
+/// non-empty header story ends with a paragraph mark that only separates it from the next story
+/// and is not part of its contents, so a one-line header is stored as `text\r\r`. Laid out, the
+/// guard would add an empty paragraph to every header and footer and push the body down when
+/// the header reaches past the top margin (#5922).
+///
+/// Only an empty last paragraph after an ordinary one is dropped: a story that is nothing but a
+/// mark stays a part (an empty header, not one linked to the previous section), and a mark right
+/// after a table row stays the paragraph that closes the table.
+fn header_story_blocks(
+    ctx: &WalkCtx<'_>,
+    a: u32,
+    b: u32,
+    pics: &mut media::Pictures,
+    max_paragraphs: usize,
+) -> Result<wordcraft_doc::Blocks, DocbinError> {
+    let mut paras = walk(ctx, a, b, pics);
+    if ctx.paragraphs.exceeded.get() {
+        return Err(paragraph_limit(max_paragraphs));
+    }
+    let outside_table = |p: &table::ParaOut| p.row.as_ref().is_none_or(|r| !r.in_table && !r.ttp && !r.cell_mark);
+    if let [.., prev, last] = paras.as_slice() {
+        let guard = outside_table(last) && last.para.text.is_empty() && last.para.objects.is_empty() && last.para.section.is_none();
+        if guard && outside_table(prev) {
+            paras.pop();
+        }
+    }
+    Ok(table::assemble(paras))
+}
+
 /// A story's CP range `[a, b)`, relative to a subdocument starting at `base` with `len`
 /// CPs, as absolute CPs clamped to that subdocument; `None` when empty.
 fn story(base: u32, len: u32, (a, b): (u32, u32)) -> Option<(u32, u32)> {

crates/docbin/src/tests.rs

@@ -1157,6 +1157,58 @@ fn header_story_becomes_part() {
     assert!(part.blocks.iter().any(|b| b.as_para().is_some_and(|p| p.text.contains("Header text"))));
 }
 
+/// The odd header's blocks for a one-section document whose header story is `story`, with
+/// `papx_entries` over the whole stream (the header story starts at CP 5, after `body\r`):
+/// paragraph texts, a table as `<table>`; `None` when the slot has no part.
+fn odd_header_blocks(story: &'static str, papx_entries: Vec<(u32, u16, Vec<u8>)>) -> Option<Vec<String>> {
+    let spec = StructSpec {
+        text: "body\r",
+        header_stories: vec![story],
+        sections: vec![(5, Vec::new())],
+        chpx_runs: vec![(0, Vec::new())],
+        papx_entries,
+        plf_lst: Vec::new(),
+        plf_lfo: Vec::new(),
+        note_stories: vec![],
+        endnote_stories: vec![],
+        extra_pairs: Vec::new(),
+        data_stream: Vec::new(),
+    };
+    let doc = read(&struct_doc(&spec)).expect("opens");
+    let id = doc.last_section.headers.default?;
+    let part = doc.parts.get(&id).expect("part");
+    Some(part.blocks.iter().map(|b| b.as_para().map_or_else(|| "<table>".to_string(), |p| p.text.clone())).collect())
+}
+
+/// #5922: [MS-DOC] 2.3.3 ends every non-empty header story with a guard paragraph mark that
+/// is not part of its contents. Read as a paragraph, it put an empty line under every header
+/// and pushed the first body line down when the header reached past the top margin.
+#[test]
+fn header_story_guard_mark_is_not_a_paragraph() {
+    let plain = || vec![(0, 0, Vec::new())];
+    // One line typed into the header: content "Header\r" + guard.
+    assert_eq!(odd_header_blocks("Header\r\r", plain()), Some(vec!["Header".to_string()]));
+    // Enter pressed after it: the empty paragraph is content and stays, only the guard goes.
+    assert_eq!(odd_header_blocks("Header\r\r\r", plain()), Some(vec!["Header".to_string(), String::new()]));
+    // An empty header (one empty content paragraph + guard) is still a header of its own.
+    assert_eq!(odd_header_blocks("\r\r", plain()), Some(vec![String::new()]));
+    // A story that is only a mark (no guard) keeps its paragraph and its part.
+    assert_eq!(odd_header_blocks("\r", plain()), Some(vec![String::new()]));
+    // A story without a guard keeps its last paragraph.
+    assert_eq!(odd_header_blocks("Header\r", plain()), Some(vec!["Header".to_string()]));
+}
+
+/// #5922: a mark right after a table row in a header story is kept: it is the paragraph that
+/// closes the table, not a guard to drop.
+#[test]
+fn header_story_mark_after_a_table_row_is_kept() {
+    let in_tbl = grpprl(&[(0x2416, &[0x01])]);
+    let ttp: Vec<u8> = [grpprl(&[(0x2416, &[0x01]), (0x2417, &[0x01])]), t_def_table(&[0, 2880])].concat();
+    // Stream: "body\r" (CPs 0-4), then the header story "a\x07" (5-6), row end "\x07" (7), "\r" (8).
+    let papx = vec![(0, 0, Vec::new()), (5, 0, in_tbl), (7, 0, ttp), (8, 0, Vec::new())];
+    assert_eq!(odd_header_blocks("a\u{7}\u{7}\r", papx), Some(vec!["<table>".to_string(), String::new()]));
+}
+
 /// A `sprmTDefTable` Prl for `edges` twip boundaries and TC80s of 20 bytes per column,
 /// built per the spec (cb counts the bytes after it, plus one).
 fn t_def_table(edges: &[i16]) -> Vec<u8> {