Gap 19: ODT page margins default as LibreOffice's; loose drawings join a paragraph

More fixes from LibreOffice's bundled templates:
- A page margin the file leaves unsaid was 1 in, and a page layout with
  no properties at all was ignored. Both now take LibreOffice's
  defaults: 2 cm margins (observed black-box). LibreOffice's CV template
  now lays out as LibreOffice draws it.
- Drawings anchored outside any paragraph (LibreOffice's
  page-anchored frames before the first paragraph) made a paragraph of
  their own. That paragraph started the document on the default page
  style rather than the one the first paragraph asks for, which put the
  running header on a letter's first page and pushed the letter onto
  two pages. They now join the next paragraph; before a table or at the
  end, they still get a paragraph of their own.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GBVQLyDnrG8Tn1yDQcUabC

crates/formats/src/odt/body.rs

@@ -7,6 +7,9 @@ use super::{
     model,
 };
 
+/// Most drawings waiting outside a paragraph.
+const MAX_LOOSE: usize = 10_000;
+
 pub(in crate::odt) mod drawing;
 mod indexes;
 mod notes;
@@ -57,6 +60,10 @@ pub(super) struct Body<'a> {
     forms: super::forms::FormsReader,
     pub(super) manifest: Option<&'a HashMap<String, String>>,
     containers: Vec<Vec<FBlock>>,
+    /// Drawings anchored in the text but outside any paragraph (LibreOffice's page-anchored
+    /// frames before the first one): the next paragraph's, so they don't make a paragraph of
+    /// their own (which would start the document on another page style).
+    loose: Vec<Inline>,
     fmt: Vec<Fmt>,
     para: Option<Para>,
     /// The text read last ended in a space collapsed from XML whitespace (not a `text:s`).
@@ -170,6 +177,7 @@ impl<'a> Body<'a> {
             para_seq: 0,
             manifest: None,
             containers: vec![Vec::new()],
+            loose: Vec::new(),
             fmt: Vec::new(),
             para: None,
             xml_space_end: false,
@@ -341,6 +349,11 @@ impl<'a> Body<'a> {
         std::mem::take(&mut self.forms.hidden)
     }
 
+    /// Whether the reader is in the text itself: not in a table, note, comment or frame.
+    pub(in crate::odt) fn at_body_level(&self) -> bool {
+        self.tables.is_empty() && self.cells.is_empty() && self.note.is_none() && self.frame.is_none() && self.containers.len() == 1
+    }
+
     pub(super) fn take_blocks(&mut self) -> Vec<FBlock> {
         // What truncated XML left open: a field, a note or comment (its paragraph goes on).
         self.end_field();
@@ -348,6 +361,11 @@ impl<'a> Body<'a> {
             self.end_note();
         }
         self.end_para();
+        // Drawings after the last paragraph: a paragraph of their own.
+        if !self.loose.is_empty() {
+            self.start_para(None, None);
+            self.end_para();
+        }
         while self.containers.len() > 1 {
             let c = self.containers.pop().unwrap_or_default();
             self.container().extend(c);
@@ -513,6 +531,12 @@ impl Body<'_> {
             self.para_cont = false;
         }
         self.fmt.push(base);
+        // Drawings waiting outside a paragraph come first in it.
+        if !self.loose.is_empty() && self.note.is_none() {
+            let mut loose = std::mem::take(&mut self.loose);
+            loose.append(&mut p.inlines);
+            p.inlines = loose;
+        }
         self.para = Some(p);
     }
 
@@ -788,6 +812,10 @@ impl Body<'_> {
                 self.start_para(get(a, "text:style-name"), h);
             }
             "table" if q.starts_with("table:") => {
+                // Drawings waiting for a paragraph get their own before a table.
+                if !self.loose.is_empty() && self.tables.is_empty() {
+                    self.start_para(None, None);
+                }
                 self.end_para();
                 // A body table whose style names a master page starts that page (ODF 1.3 §19.509).
                 if let Some(next) = get(a, "table:style-name").and_then(|s| self.page_section(s)) {

crates/formats/src/odt/body/notes.rs

@@ -190,6 +190,12 @@ impl Body<'_> {
 
     /// `inline` at the end of the paragraph being read (a new one if none is).
     pub(in crate::odt) fn push_inline(&mut self, inline: Inline) {
+        // A drawing in the text outside any paragraph waits for the next one.
+        let drawing = matches!(inline, Inline::Image(_) | Inline::Shape { .. } | Inline::Chart(_) | Inline::Group { .. });
+        if self.para.is_none() && drawing && self.at_body_level() && self.loose.len() < super::MAX_LOOSE {
+            self.loose.push(inline);
+            return;
+        }
         if self.para.is_none() {
             self.start_para(None, None);
         }

crates/formats/src/odt/import.rs

@@ -435,6 +435,9 @@ fn hyphenation(blocks: &mut [FBlock]) -> bool {
 /// The content of a package that has none: an empty text.
 const EMPTY_CONTENT: &str = r#"<office:document-content xmlns:office="urn:oasis:names:tc:opendocument:xmlns:office:1.0" xmlns:text="urn:oasis:names:tc:opendocument:xmlns:text:1.0"><office:body><office:text><text:p/></office:text></office:body></office:document-content>"#;
 
+/// The margin a page layout leaves unsaid (LibreOffice's page margins: 2 cm).
+const DEFAULT_MARGIN: f32 = 2.0 * 72.0 / 2.54;
+
 /// The page size of a page layout that states none: WordCraft's default page's.
 fn default_paper() -> (f32, f32) {
     let d = SectionProps::default();
@@ -725,7 +728,15 @@ fn read_pages(xml: &[u8], paper: (f32, f32)) -> Pages {
                 }
             }
             "linenumbering-configuration" if q.starts_with("text:") => lines = line_numbering(&a),
-            "page-layout" if q.starts_with("style:") => cur = get(&a, "style:name").map(String::from),
+            "page-layout" if q.starts_with("style:") => {
+                cur = get(&a, "style:name").map(String::from);
+                // A layout stating nothing is still one: the defaults' page.
+                if let Some(n) = &cur
+                    && layouts.len() < MAX_SECTIONS
+                {
+                    layouts.entry(n.clone()).or_default();
+                }
+            }
             "header-style" => in_footer = false,
             "footer-style" => in_footer = true,
             "header-footer-properties" => {
@@ -737,10 +748,8 @@ fn read_pages(xml: &[u8], paper: (f32, f32)) -> Pages {
                 }
             }
             "page-layout-properties" => {
-                if let Some(n) = &cur
-                    && layouts.len() < MAX_SECTIONS
-                {
-                    layouts.insert(n.clone(), Layout { attrs: a, ..Default::default() });
+                if let Some(l) = cur.as_ref().and_then(|n| layouts.get_mut(n)) {
+                    l.attrs = a;
                 }
             }
             "columns" if q.starts_with("style:") => {
@@ -784,7 +793,16 @@ fn read_pages(xml: &[u8], paper: (f32, f32)) -> Pages {
     let mut pages = Pages { doc, next: nexts, ..Default::default() };
     for (name, layout, header, footer) in masters {
         let Some(Layout { attrs: a, columns: cols, header: hb, footer: fb }) = layouts.get(&layout) else { continue };
-        let mut sec = SectionProps { page_w: paper.0, page_h: paper.1, ..SectionProps::default() };
+        // What a layout leaves unsaid: the default paper, and LibreOffice's 2 cm margins.
+        let mut sec = SectionProps {
+            page_w: paper.0,
+            page_h: paper.1,
+            margin_top: DEFAULT_MARGIN,
+            margin_bottom: DEFAULT_MARGIN,
+            margin_left: DEFAULT_MARGIN,
+            margin_right: DEFAULT_MARGIN,
+            ..SectionProps::default()
+        };
         apply_layout(a, &mut sec);
         sec.line_numbers.clone_from(&lines);
         if pages.first.is_none() {

crates/formats/src/odt/tests.rs

@@ -276,7 +276,8 @@ fn page_layout_follows_the_first_master_page() {
     let mut sec = wordcraft_doc::SectionProps::default();
     apply_page_layout(xml, &mut sec);
     assert!(sec.landscape);
-    assert_eq!((sec.page_w, sec.page_h, sec.margin_left, sec.margin_top), (720.0, 360.0, 72.0, 72.0));
+    // A margin left unsaid is LibreOffice's 2 cm.
+    assert_eq!((sec.page_w, sec.page_h, sec.margin_left, sec.margin_top), (720.0, 360.0, 72.0, 2.0 * 72.0 / 2.54));
 }
 
 #[test]
@@ -296,8 +297,9 @@ fn hostile_page_layout_values_are_clamped() {
     apply_page_layout(styles("1e30pt", "1e30in", "1e30pt").as_bytes(), &mut sec);
     assert_eq!((sec.page_w, sec.page_h), (1584.0, 1584.0));
     assert_eq!((sec.margin_top, sec.margin_left), (1584.0, 1584.0));
-    // Negative, zero, tiny and non-numeric values leave the defaults alone.
-    let def = wordcraft_doc::SectionProps::default();
+    // Negative, zero, tiny and non-numeric values leave the defaults alone: the default page, and
+    // LibreOffice's 2 cm margins.
+    let def = wordcraft_doc::SectionProps { margin_top: 2.0 * 72.0 / 2.54, margin_right: 2.0 * 72.0 / 2.54, ..Default::default() };
     for (w, h, m) in [("-10in", "-5in", "-1in"), ("0pt", "0pt", "NaNpt"), ("1pt", "2pt", "infpt"), ("NaN", "inf", "x")] {
         let mut sec = wordcraft_doc::SectionProps::default();
         apply_page_layout(styles(w, h, m).as_bytes(), &mut sec);

crates/formats/src/odt/tests_more/floats.rs

@@ -215,3 +215,28 @@ fn a_contour_outside_its_view_box_is_worked_out_instead() {
     let a: crate::odt::xml::Attrs = vec![("svg:viewBox".into(), "0 0 100 100".into()), ("draw:points".into(), "0,0 100,100 0,100".into())];
     assert!(crate::odt::frames::contour_polygon(&a).is_some());
 }
+
+#[test]
+fn a_drawing_outside_any_paragraph_belongs_to_the_next_one() {
+    // LibreOffice puts page-anchored frames before the first paragraph. They join it rather
+    // than making a paragraph of their own, which would start the document on the default page
+    // style instead of the one the first paragraph asks for.
+    let ns = crate::odt::tests::NS_T;
+    let styles = format!(
+        r#"<office:document-styles {ns}><office:styles><style:style style:name="Opening" style:family="paragraph" style:master-page-name="First"/></office:styles><office:automatic-styles><style:page-layout style:name="pm1"><style:page-layout-properties fo:page-width="8.5in" fo:page-height="11in"/></style:page-layout></office:automatic-styles><office:master-styles><style:master-page style:name="Standard" style:page-layout-name="pm1"><style:header><text:p>Running head</text:p></style:header></style:master-page><style:master-page style:name="First" style:page-layout-name="pm1" style:next-style-name="Standard"/></office:master-styles></office:document-styles>"#
+    );
+    let content = format!(
+        r#"<office:document-content {ns} xmlns:draw="urn:oasis:names:tc:opendocument:xmlns:drawing:1.0" xmlns:svg="urn:oasis:names:tc:opendocument:xmlns:svg-compatible:1.0"><office:body><office:text><draw:frame text:anchor-type="page" text:anchor-page-number="1" svg:x="1in" svg:y="2in" svg:width="2in" svg:height="1in"><draw:text-box><text:p>Address</text:p></draw:text-box></draw:frame><text:p text:style-name="Opening">Dear reader,</text:p><text:p>Body.</text:p></office:text></office:body></office:document-content>"#
+    );
+    let d = import(&crate::odt::tests::package(&content, Some(&styles))).unwrap();
+    let paras: Vec<&wordcraft_doc::Paragraph> = d.body.iter().filter_map(|b| b.as_para()).collect();
+    assert_eq!(paras.len(), 2, "no paragraph of the frame's own");
+    assert_eq!(paras[0].text.replace(wordcraft_doc::para::OBJ, ""), "Dear reader,");
+    assert_eq!(paras[0].objects.len(), 1);
+    // One section, starting on the first page style: a different first page without the head.
+    assert!(d.last_section.title_page && d.last_section.headers.default.is_some());
+    // Drawings with no paragraph after them still get one.
+    let tail = content.replace(r#"<text:p text:style-name="Opening">Dear reader,</text:p><text:p>Body.</text:p>"#, "");
+    let d = import(&crate::odt::tests::package(&tail, Some(&styles))).unwrap();
+    assert_eq!(d.body.iter().filter_map(|b| b.as_para()).map(|p| p.objects.len()).sum::<usize>(), 1);
+}

crates/formats/src/odt/tests_more/page_setup.rs

@@ -141,3 +141,23 @@ fn a_page_layout_without_a_size_is_the_default_pages_size() {
     let d = crate::import_on("odt", &crate::odt::tests::package(&content, Some(&sized)), &a4).unwrap().unwrap();
     assert_eq!((d.last_section.page_w, d.last_section.page_h), (360.0, 504.0));
 }
+
+#[test]
+fn a_page_layout_stating_nothing_has_libreoffices_margins() {
+    // LibreOffice's CV template: a page layout with no properties at all. LibreOffice draws it
+    // on its default paper with 2 cm margins.
+    let ns = crate::odt::tests::NS_T;
+    let styles = format!(
+        r#"<office:document-styles {ns}><office:automatic-styles><style:page-layout style:name="Mpm1"><style:header-style/><style:footer-style/></style:page-layout></office:automatic-styles><office:master-styles><style:master-page style:name="Standard" style:page-layout-name="Mpm1"/></office:master-styles></office:document-styles>"#
+    );
+    let content =
+        format!("<office:document-content {ns}><office:body><office:text><text:p>x</text:p></office:text></office:body></office:document-content>");
+    let d = crate::import("odt", &crate::odt::tests::package(&content, Some(&styles))).unwrap().unwrap();
+    let s = &d.last_section;
+    let cm2 = 2.0 * 72.0 / 2.54;
+    assert_eq!([s.margin_top, s.margin_bottom, s.margin_left, s.margin_right], [cm2; 4]);
+    // One margin stated: that one is the file's.
+    let one = styles.replace("<style:header-style/>", r#"<style:page-layout-properties fo:margin-left="1in"/><style:header-style/>"#);
+    let d = crate::import("odt", &crate::odt::tests::package(&content, Some(&one))).unwrap().unwrap();
+    assert_eq!((d.last_section.margin_left, d.last_section.margin_top), (72.0, cm2));
+}