Keep table row identity attributes through modelled edits

README.md

@@ -39,7 +39,7 @@ rows are the enforced release-mode bounds plus one dated observation.
 
 | Measurement | Value | Version | Platform | Build mode | Input | Command | Statistic | Measured on |
 |---|---|---|---|---|---|---|---|---|
-| Crates.io archive: rdocx | 1,092,256 compressed bytes, 6,498,484 member bytes, 36 members | 0.14.0 | macOS 26.6.2, Apple M5 Max, arm64 | `cargo package --locked --no-verify` | Tracked `rdocx` package inventory | `python3 scripts/readme_doctests.py --record-measurements` | gzip archive bytes, tar member bytes, tar member count | 2026-09-26 |
+| Crates.io archive: rdocx | 1,101,657 compressed bytes, 6,540,109 member bytes, 36 members | 0.14.0 | macOS 26.6.2, Apple M5 Max, arm64 | `cargo package --locked --no-verify` | Tracked `rdocx` package inventory | `python3 scripts/readme_doctests.py --record-measurements` | gzip archive bytes, tar member bytes, tar member count | 2026-09-27 |
 | Large-document layout throughput | minimum 250 pages/s, observed 31,019.1 pages/s | rdocx 0.14.0 | macOS 26.6.2, Apple M5 Max, arm64 | release, one test thread | 1,000 one-page paragraphs with deterministic fonts | `cargo test -p rdocx --test regression_test --release a_thousand_page_document_paginates_and_renders_within_the_declared_limits -- --ignored --exact --nocapture --test-threads=1` | pages per wall-clock second | 2026-09-19 |
 | Large-document layout peak allocation | maximum 64 MiB, observed 29.03 MiB | rdocx 0.14.0 | macOS 26.6.2, Apple M5 Max, arm64 | release, one test thread | 1,000 one-page paragraphs with deterministic fonts | `cargo test -p rdocx --test regression_test --release a_thousand_page_document_paginates_and_renders_within_the_declared_limits -- --ignored --exact --nocapture --test-threads=1` | peak live allocation | 2026-09-19 |
 | Large-document PDF throughput | minimum 1,000 pages/s, observed 60,058.0 pages/s | rdocx 0.14.0 | macOS 26.6.2, Apple M5 Max, arm64 | release, one test thread | 1,000 deterministic layout pages | `cargo test -p rdocx --test regression_test --release a_thousand_page_document_paginates_and_renders_within_the_declared_limits -- --ignored --exact --nocapture --test-threads=1` | pages per wall-clock second | 2026-09-19 |

crates/rdocx-layout/README.md

@@ -18,7 +18,7 @@ sections, and retains source provenance.
 
 | Measurement | Value | Version | Platform | Build mode | Input | Command | Statistic | Measured on |
 |---|---|---|---|---|---|---|---|---|
-| Crates.io archive: rdocx-layout | 255,752 compressed bytes, 1,385,701 member bytes, 15 members | 0.14.0 | macOS 26.6.2, Apple M5 Max, arm64 | `cargo package --locked --no-verify` | Tracked `rdocx-layout` package inventory | `python3 scripts/readme_doctests.py --record-measurements` | gzip archive bytes, tar member bytes, tar member count | 2026-09-26 |
+| Crates.io archive: rdocx-layout | 256,067 compressed bytes, 1,386,937 member bytes, 15 members | 0.14.0 | macOS 26.6.2, Apple M5 Max, arm64 | `cargo package --locked --no-verify` | Tracked `rdocx-layout` package inventory | `python3 scripts/readme_doctests.py --record-measurements` | gzip archive bytes, tar member bytes, tar member count | 2026-09-27 |
 | Large-document layout throughput | minimum 250 pages/s, observed 31,019.1 pages/s | rdocx 0.14.0 | macOS 26.6.2, Apple M5 Max, arm64 | release, one test thread | 1,000 one-page paragraphs with deterministic fonts | `cargo test -p rdocx --test regression_test --release a_thousand_page_document_paginates_and_renders_within_the_declared_limits -- --ignored --exact --nocapture --test-threads=1` | pages per wall-clock second | 2026-09-19 |
 | Large-document layout peak allocation | maximum 64 MiB, observed 29.03 MiB | rdocx 0.14.0 | macOS 26.6.2, Apple M5 Max, arm64 | release, one test thread | 1,000 one-page paragraphs with deterministic fonts | `cargo test -p rdocx --test regression_test --release a_thousand_page_document_paginates_and_renders_within_the_declared_limits -- --ignored --exact --nocapture --test-threads=1` | peak live allocation | 2026-09-19 |
 | Large-document PDF throughput | minimum 1,000 pages/s, observed 60,058.0 pages/s | rdocx 0.14.0 | macOS 26.6.2, Apple M5 Max, arm64 | release, one test thread | 1,000 deterministic layout pages | `cargo test -p rdocx --test regression_test --release a_thousand_page_document_paginates_and_renders_within_the_declared_limits -- --ignored --exact --nocapture --test-threads=1` | pages per wall-clock second | 2026-09-19 |

crates/rdocx-layout/src/engine.rs

@@ -3617,7 +3617,9 @@ fn table_is_cache_safe(table: &CT_Tbl, styles: &CT_Styles) -> bool {
             properties.change.is_none() && properties.revision_xml.is_empty()
         })
         && table.rows.iter().all(|row| {
-            row.extra_xml.is_empty()
+            row.extra_xml
+                .iter()
+                .all(|(position, raw)| CT_Row::raw_is_root_attributes(*position, raw))
                 && row.content_controls.is_empty()
                 && row.properties.as_ref().is_none_or(|properties| {
                     properties.revision_markers.is_empty() && properties.revision_xml.is_empty()
@@ -14416,6 +14418,24 @@ mod tests {
             .extra_xml
             .push((0, br#"<w:unknown/>"#.to_vec()));
         assert!(!table_is_cache_safe(&preserved_cell, &input.styles));
+
+        // Word writes revision-save and paragraph identities on every row.
+        // They carry no content, so the row stays cache safe, unlike a raw
+        // child of the row.
+        let document = rdocx_oxml::CT_Document::from_xml(
+            br#"<w:document xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main" xmlns:w14="http://schemas.microsoft.com/office/word/2010/wordml"><w:body><w:tbl><w:tr w:rsidR="00A1B2C3" w:rsidTr="00A1B2C4" w14:paraId="1A2B3C4D"><w:tc><w:p><w:r><w:t>identified row</w:t></w:r></w:p></w:tc></w:tr></w:tbl></w:body></w:document>"#,
+        )
+        .expect("identified row document parses");
+        let Some(BodyContent::Table(mut identified_row)) = document.body.content.into_iter().next()
+        else {
+            panic!("identified row table parses");
+        };
+        assert!(!identified_row.rows[0].extra_xml.is_empty());
+        assert!(table_is_cache_safe(&identified_row, &input.styles));
+        identified_row.rows[0]
+            .extra_xml
+            .push((0, br#"<w:unknown/>"#.to_vec()));
+        assert!(!table_is_cache_safe(&identified_row, &input.styles));
     }
 
     fn restart_input() -> LayoutInput {

crates/rdocx-oxml/README.md

@@ -17,7 +17,7 @@ schema order, and retains unmodelled XML alongside typed edits.
 
 | Measurement | Value | Version | Platform | Build mode | Input | Command | Statistic | Measured on |
 |---|---|---|---|---|---|---|---|---|
-| Crates.io archive: rdocx-oxml | 367,500 compressed bytes, 2,380,047 member bytes, 32 members | 0.14.0 | macOS 26.6.2, Apple M5 Max, arm64 | `cargo package --locked --no-verify` | Tracked `rdocx-oxml` package inventory | `python3 scripts/readme_doctests.py --record-measurements` | gzip archive bytes, tar member bytes, tar member count | 2026-09-19 |
+| Crates.io archive: rdocx-oxml | 372,614 compressed bytes, 2,400,815 member bytes, 32 members | 0.14.0 | macOS 26.6.2, Apple M5 Max, arm64 | `cargo package --locked --no-verify` | Tracked `rdocx-oxml` package inventory | `python3 scripts/readme_doctests.py --record-measurements` | gzip archive bytes, tar member bytes, tar member count | 2026-09-27 |
 
 ## Use it when
 

crates/rdocx-oxml/src/comments.rs

@@ -10,7 +10,7 @@ use crate::namespace::{W_NS, matches_local_name};
 use crate::numbering::word_prefixes_at;
 use crate::properties::is_word_element;
 use crate::raw_xml::{capture_element, capture_empty_element};
-use crate::text::CT_P;
+use crate::text::{CT_P, declare_w14_on_part_root};
 
 const W14_NS: &str = "http://schemas.microsoft.com/office/word/2010/wordml";
 
@@ -137,7 +137,9 @@ impl CT_Comments {
             write_raw_at(&mut writer, &self.extra_xml, index + 1)?;
         }
         writer.write_event(Event::End(BytesEnd::new("w:comments")))?;
-        Ok(writer.into_inner())
+        let mut xml = writer.into_inner();
+        declare_w14_on_part_root(&mut xml)?;
+        Ok(xml)
     }
 }
 
@@ -447,6 +449,19 @@ mod tests {
         assert!(output.contains(r#"ext:item="kept""#));
     }
 
+    #[test]
+    fn a_retained_text_identity_stays_bound_without_a_paragraph_identity() {
+        // The root declares `w14` for an authored paragraph identity only, so
+        // a paragraph that carries `w14:textId` alone relies on the check of
+        // the written part.
+        let xml = br#"<w:comments xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main" xmlns:w14="http://schemas.microsoft.com/office/word/2010/wordml"><w:comment w:id="1"><w:p w14:textId="5E6F7A8B"><w:r><w:t>thread</w:t></w:r></w:p></w:comment></w:comments>"#;
+        let comments = CT_Comments::from_xml(xml).unwrap();
+        let output = comments.to_xml().unwrap();
+        let reparsed = CT_Comments::from_xml(&output)
+            .unwrap_or_else(|error| panic!("{error}: {}", String::from_utf8_lossy(&output)));
+        assert_eq!(reparsed.to_xml().unwrap(), output);
+    }
+
     #[test]
     fn malformed_comment_id_is_rejected() {
         let xml = br#"<w:comments xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main"><w:comment w:id="not-a-number"><w:p/></w:comment></w:comments>"#;

crates/rdocx-oxml/src/content_control.rs

@@ -1242,6 +1242,7 @@ fn parse_content(
                             reader,
                             &prefixes,
                             &owner_bindings,
+                            Some(&child),
                         )?,
                     ));
                 } else if is_word_element(child.name().as_ref(), b"tc", &prefixes) {
@@ -1298,7 +1299,7 @@ fn parse_content(
                 } else if is_word_element(child.name().as_ref(), b"tbl", &prefixes) {
                     content.push(SdtContent::Table(CT_Tbl::new()));
                 } else if is_word_element(child.name().as_ref(), b"tr", &prefixes) {
-                    content.push(SdtContent::Row(CT_Row::new()));
+                    content.push(SdtContent::Row(CT_Row::from_empty_root(&child, &prefixes)?));
                 } else if is_word_element(child.name().as_ref(), b"tc", &prefixes) {
                     content.push(SdtContent::Cell(CT_Tc {
                         properties: None,

crates/rdocx-oxml/src/document.rs

@@ -17,7 +17,8 @@ use crate::revision::CT_Revision;
 use crate::shared::{ST_PageOrientation, ST_SectionType};
 use crate::table::{CT_Tbl, ST_VerticalJc};
 use crate::text::{
-    CT_P, capture_root_attribute_record, is_root_attribute_record, push_root_attribute_record,
+    CT_P, capture_root_attribute_record, declare_w14_on_part_root, is_root_attribute_record,
+    push_root_attribute_record,
 };
 use crate::units::Twips;
 
@@ -2998,7 +2999,9 @@ impl CT_Document {
 
         writer.write_event(Event::End(BytesEnd::new("w:document")))?;
 
-        Ok(writer.into_inner())
+        let mut xml = writer.into_inner();
+        declare_w14_on_part_root(&mut xml)?;
+        Ok(xml)
     }
 }
 

crates/rdocx-oxml/src/drawing.rs

@@ -1,7 +1,7 @@
 //! Drawing elements for inline and anchor images: `CT_Drawing`, `CT_Inline`, `CT_Anchor`.
 
 use quick_xml::events::{BytesEnd, BytesStart, BytesText, Event};
-use quick_xml::name::{Namespace, ResolveResult};
+use quick_xml::name::{Namespace, PrefixDeclaration, ResolveResult};
 use quick_xml::reader::NsReader;
 use quick_xml::{Reader, Writer, XmlVersion};
 
@@ -698,29 +698,7 @@ impl CT_Anchor {
                                 Ok(Event::Start(ref ie))
                                     if matches_local_name(ie.name().as_ref(), b"p") =>
                                 {
-                                    let raw = capture_ns_element(reader, ie)?;
-                                    let mut paragraph_reader = Reader::from_reader(raw.as_slice());
-                                    let mut paragraph_buffer = Vec::new();
-                                    loop {
-                                        match paragraph_reader
-                                            .read_event_into(&mut paragraph_buffer)?
-                                        {
-                                            Event::Start(ref paragraph_start)
-                                                if matches_local_name(
-                                                    paragraph_start.name().as_ref(),
-                                                    b"p",
-                                                ) =>
-                                            {
-                                                paragraphs.push(crate::text::CT_P::from_xml(
-                                                    &mut paragraph_reader,
-                                                )?);
-                                                break;
-                                            }
-                                            Event::Eof => break,
-                                            _ => {}
-                                        }
-                                        paragraph_buffer.clear();
-                                    }
+                                    paragraphs.extend(text_box_paragraph(reader, ie)?);
                                 }
                                 Ok(Event::End(ref ie))
                                     if matches_local_name(ie.name().as_ref(), b"txbxContent") =>
@@ -1384,6 +1362,54 @@ fn canonical_wp_element(
     namespace_matches(&namespace, drawing_ns::WP, b"wp") && local.as_ref() == expected_local
 }
 
+/// Parse a text-box paragraph out of its anchor.
+///
+/// The paragraph is parsed on its own, so its start tag gets the bindings in
+/// scope here, on top of the scope `CT_P::from_xml` assumes. A run attribute
+/// under any prefix the part binds then resolves as it does in the body.
+fn text_box_paragraph(
+    reader: &mut NsReader<&[u8]>,
+    start: &BytesStart<'_>,
+) -> Result<Option<crate::text::CT_P>> {
+    let mut bindings = Vec::new();
+    for (prefix, namespace) in reader.resolver().bindings() {
+        let prefix = match prefix {
+            PrefixDeclaration::Default => "",
+            PrefixDeclaration::Named(prefix) => std::str::from_utf8(prefix)?,
+        };
+        let namespace = quick_xml::escape::unescape(std::str::from_utf8(namespace.as_ref())?)
+            .map_err(quick_xml::Error::from)?;
+        bindings.push((prefix.to_owned(), namespace.into_owned()));
+    }
+    let raw =
+        crate::text::raw_with_external_bindings(&capture_ns_element(reader, start)?, &bindings)?;
+    let mut paragraph_reader = Reader::from_reader(raw.as_slice());
+    let mut buffer = Vec::new();
+    loop {
+        match paragraph_reader.read_event_into(&mut buffer)? {
+            Event::Start(ref paragraph_start)
+                if matches_local_name(paragraph_start.name().as_ref(), b"p") =>
+            {
+                let prefixes = crate::numbering::word_prefixes_at(
+                    paragraph_start,
+                    &[
+                        "w".to_owned(),
+                        format!("\0r\0{}", crate::namespace::R_NS),
+                        format!("\0mc\0{}", crate::namespace::MC_NS),
+                    ],
+                )?;
+                return Ok(Some(crate::text::CT_P::from_xml_with_prefixes(
+                    &mut paragraph_reader,
+                    &prefixes,
+                )?));
+            }
+            Event::Eof => return Ok(None),
+            _ => {}
+        }
+        buffer.clear();
+    }
+}
+
 fn capture_ns_element(reader: &mut NsReader<&[u8]>, start: &BytesStart<'_>) -> Result<Vec<u8>> {
     let mut writer = Writer::new(Vec::new());
     writer.write_event(Event::Start(start.to_owned()))?;

crates/rdocx-oxml/src/footnotes.rs

@@ -7,7 +7,7 @@ use crate::error::Result;
 use crate::namespace::{W_NS, matches_local_name};
 use crate::numbering::word_prefixes_at;
 use crate::properties::is_word_element;
-use crate::text::CT_P;
+use crate::text::{CT_P, declare_w14_on_part_root};
 
 /// `ST_FtnEdn` — what a note in the stream is for.
 ///
@@ -212,7 +212,9 @@ impl CT_Footnotes {
 
         writer.write_event(Event::End(BytesEnd::new(root_tag)))?;
 
-        Ok(writer.into_inner())
+        let mut xml = writer.into_inner();
+        declare_w14_on_part_root(&mut xml)?;
+        Ok(xml)
     }
 }
 
@@ -273,6 +275,18 @@ fn parse_footnote_content(
 mod tests {
     use super::*;
 
+    #[test]
+    fn a_note_paragraph_identity_stays_bound_under_the_written_root() {
+        // The written root declares only `w` and `r`, while Word declares
+        // `w14` on the root of the notes part it writes.
+        let xml = br#"<w:footnotes xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main" xmlns:w14="http://schemas.microsoft.com/office/word/2010/wordml"><w:footnote w:id="1"><w:p w14:paraId="1A2B3C4D" w14:textId="5E6F7A8B"><w:r><w:t>note</w:t></w:r></w:p></w:footnote></w:footnotes>"#;
+        let footnotes = CT_Footnotes::from_xml(xml).unwrap();
+        let output = footnotes.to_xml_footnotes().unwrap();
+        let reparsed = CT_Footnotes::from_xml(&output)
+            .unwrap_or_else(|error| panic!("{error}: {}", String::from_utf8_lossy(&output)));
+        assert_eq!(reparsed.to_xml_footnotes().unwrap(), output);
+    }
+
     #[test]
     fn parse_footnotes_xml() {
         let xml = br#"<?xml version="1.0" encoding="UTF-8" standalone="yes"?>

crates/rdocx-oxml/src/header_footer.rs

@@ -10,7 +10,7 @@ use crate::namespace::{W_NS, matches_local_name};
 use crate::numbering::{namespace_bindings, word_prefixes_at};
 use crate::properties::is_word_element;
 use crate::raw_xml::{capture_element, capture_empty_element};
-use crate::text::CT_P;
+use crate::text::{CT_P, declare_w14_on_part_root};
 
 const VML_NS: &str = "urn:schemas-microsoft-com:vml";
 const OFFICE_NS: &str = "urn:schemas-microsoft-com:office:office";
@@ -352,7 +352,9 @@ impl CT_HdrFtr {
 
         writer.write_event(Event::End(BytesEnd::new(root_tag)))?;
 
-        Ok(writer.into_inner())
+        let mut xml = writer.into_inner();
+        declare_w14_on_part_root(&mut xml)?;
+        Ok(xml)
     }
 }
 
@@ -1186,6 +1188,20 @@ pub struct HdrFtrRef {
 mod tests {
     use super::*;
 
+    #[test]
+    fn a_paragraph_identity_stays_bound_when_only_the_paragraph_declares_w14() {
+        let w14 = "http://schemas.microsoft.com/office/word/2010/wordml";
+        let xml = format!(
+            r#"<w:hdr xmlns:w="{W_NS}"><w:p xmlns:w14="{w14}" w14:paraId="1A2B3C4D"><w:r><w:t>header</w:t></w:r></w:p></w:hdr>"#
+        );
+        let header = CT_HdrFtr::from_xml(xml.as_bytes()).unwrap();
+        let output = String::from_utf8(header.to_xml_header().unwrap()).unwrap();
+        let root = &output[output.find("<w:hdr").unwrap()..];
+        let root = &root[..root.find('>').unwrap()];
+        assert!(root.contains(&format!(r#"xmlns:w14="{w14}""#)), "{output}");
+        assert!(output.contains(r#"w14:paraId="1A2B3C4D""#), "{output}");
+    }
+
     #[test]
     fn round_trip_header() {
         let mut hdr = CT_HdrFtr::new();