docling_core/tree.rs
1//! docling's **item tree**, for a backend that knows the exact shape upstream
2//! gives a document and wants the JSON export to reproduce it.
3//!
4//! [`DoclingDocument::nodes`](crate::DoclingDocument::nodes) is a flat,
5//! reading-order stream tuned for Markdown / DocLang / LaTeX; the JSON export
6//! rebuilds docling's parent/child structure from it with generic rules (runs
7//! of list items become list groups, a heading is a flat sibling of the text
8//! that follows it). Upstream's backends do not all agree on that structure:
9//! the HTML backend nests everything after a heading *under* the heading,
10//! splits a paragraph of mixed formatting into an `inline` group of one text
11//! item per formatting run, parents a rich table cell's content to a group
12//! under the table, keeps site chrome on the `furniture` layer… and numbers
13//! every item in the order it *creates* them. A backend that ports those
14//! rules call-for-call (HTML's `html_tree.rs`, DOCX's `docx_tree.rs`) records
15//! the result here — an arena of items in
16//! creation order, each with its parent and children — and the JSON export
17//! ([`DoclingDocument::export_to_json`](crate::DoclingDocument::export_to_json))
18//! serializes this tree instead of deriving one from the nodes. Every other
19//! serializer keeps reading the flat nodes, so their output is unaffected.
20
21use crate::{ContentLayer, FieldItem, PictureImage, Script, Table};
22
23/// docling-core's `Formatting`: the inline styles an item carries in JSON.
24#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
25pub struct Formatting {
26 pub bold: bool,
27 pub italic: bool,
28 pub underline: bool,
29 pub strikethrough: bool,
30 pub script: Script,
31}
32
33/// A `list_item`'s docling fields.
34#[derive(Debug, Clone, PartialEq, Eq, Default)]
35pub struct ListMeta {
36 pub enumerated: bool,
37 /// docling's `marker` — the HTML backend writes `""` unless an ordered
38 /// list carries an explicit `start`, then `"{n}."`.
39 pub marker: String,
40}
41
42/// What an item in the tree is. Mirrors the docling-core item classes the
43/// JSON `texts` / `groups` / `tables` / `pictures` / `field_regions` buckets
44/// hold.
45#[derive(Debug, Clone, PartialEq)]
46pub enum TreeKind {
47 /// A `TextItem` / `TitleItem` / `SectionHeaderItem` / `ListItem`, told
48 /// apart by `label` (`text`, `title`, `section_header`, `list_item`,
49 /// `caption`, `checkbox_selected`, `checkbox_unselected`, …).
50 Text {
51 label: String,
52 text: String,
53 /// docling's `orig` when it differs from `text` (the heading text
54 /// before unicode cleanup, say); `None` = same as `text`.
55 orig: Option<String>,
56 formatting: Option<Formatting>,
57 hyperlink: Option<String>,
58 /// `section_header` only: docling's heading level.
59 level: Option<u8>,
60 /// `list_item` only.
61 list: Option<ListMeta>,
62 },
63 /// A `CodeItem`.
64 Code {
65 text: String,
66 orig: Option<String>,
67 /// The language hint (a highlighter class token such as `python`),
68 /// mapped onto docling's `CodeLanguageLabel` at export; `None` →
69 /// `unknown`.
70 language: Option<String>,
71 formatting: Option<Formatting>,
72 hyperlink: Option<String>,
73 },
74 /// A `GroupItem`: `label` is docling's `GroupLabel` value (`inline`,
75 /// `list`, `section`, `unspecified`, …), `name` its name (`group`, `list`,
76 /// `ordered list`, `header-2`, `rich_cell_group_1_0_3`, …).
77 Group { label: String, name: String },
78 /// A `TableItem`. `rich_cells` marks the cells docling serialized as a
79 /// `RichTableCell`: `(row, col)` grid anchor → the group item (a child of
80 /// the table) that holds the cell's content. `captions` are caption text
81 /// items in the tree.
82 Table {
83 table: Table,
84 rich_cells: Vec<(usize, usize, usize)>,
85 captions: Vec<usize>,
86 },
87 /// A `PictureItem`, its caption text items and optional payload.
88 /// `classification` is a `PictureClassificationLabel` value written as
89 /// the picture's `meta.classification` (an HTML `<stamp>` / `<signature>`).
90 Picture {
91 captions: Vec<usize>,
92 image: Option<PictureImage>,
93 classification: Option<String>,
94 /// The picture-OCR enrichment's text (#645), serialized like the flat
95 /// node's: `meta.description` + the `description` annotation.
96 description: Option<crate::PictureDescription>,
97 /// The prediction's `confidence`, when the backend writes one: the
98 /// DocLang deserializer stamps `1.0`; the office and HTML backends
99 /// leave it out (`None`).
100 confidence: Option<f64>,
101 /// A native chart's data grid (docling's `meta.tabular_chart.chart_data`,
102 /// the series reconstructed as a `TableData`), for a DOCX chart drawing.
103 chart: Option<Table>,
104 /// The `ImageRef.dpi` docling writes for `image` when the backend
105 /// read one from the file (python-pptx's `Image.dpi`: PIL's `dpi`
106 /// info, rounded, 72 when absent or out of 1–2048); `None` → 72,
107 /// which is what upstream's other office backends pass.
108 dpi: Option<u32>,
109 },
110 /// A form key-value region (`field_regions` / `field_items`).
111 FieldRegion { items: Vec<FieldItem> },
112 /// A `KeyValueItem` (`key_value_items`): docling's `GraphData` of key and
113 /// value cells and their links, written verbatim.
114 KeyValueGraph {
115 cells: Vec<crate::GraphCell>,
116 links: Vec<crate::GraphLink>,
117 },
118}
119
120/// docling's `ProvenanceItem` for a tree item, written verbatim: the
121/// backend's own geometry in the page's units — a PPTX shape's EMU box,
122/// whose `pages` entry is the slide size in EMU — rather than the 0–511
123/// DocLang grid the flat [`Node::Located`](crate::Node::Located) carries
124/// (which cannot round-trip those integers).
125#[derive(Debug, Clone, PartialEq)]
126pub struct TreeProv {
127 /// 1-based page (slide) number.
128 pub page_no: usize,
129 /// `[l, t, r, b]`, exactly as docling computed them.
130 pub bbox: [f64; 4],
131 /// docling's `coord_origin` tag. The office backends tag their
132 /// top-left-based boxes `TOPLEFT` (the PPTX backend since docling#4294 —
133 /// it used to tag them `BOTTOMLEFT`, which read the tuple as `(l, b, r,
134 /// t)` and swapped the vertical edges); a backend that really works in a
135 /// bottom-left space sets this.
136 pub bottom_left: bool,
137 /// `[0, len(text)]` in characters for a text item, `[0, 0]` for a table
138 /// or picture.
139 pub charspan: [usize; 2],
140}
141
142/// docling's `TrackSource` — where in a time-based track (a WebVTT cue) a
143/// text item came from. Written as the item's `source: [{"kind": "track", …}]`.
144#[derive(Debug, Clone, PartialEq)]
145pub struct TreeTrack {
146 /// The cue's start offset in seconds (docling's `WebVTTTimestamp.seconds`:
147 /// `h*3600 + m*60 + s + millis/1000.0`, so the float is bit-identical).
148 pub start_time: f64,
149 /// The cue's end offset in seconds.
150 pub end_time: f64,
151 /// The cue identifier line, when the cue has one.
152 pub identifier: Option<String>,
153 /// The `<v …>` voice annotation the text sits in, when any.
154 pub voice: Option<String>,
155}
156
157/// One item of an [`ItemTree`].
158#[derive(Debug, Clone, PartialEq)]
159pub struct TreeItem {
160 /// The parent item's index; `None` = the document body.
161 pub parent: Option<usize>,
162 /// Child item indices, in docling's `children` order.
163 pub children: Vec<usize>,
164 /// The content layer; `None` = `body`.
165 pub layer: Option<ContentLayer>,
166 pub kind: TreeKind,
167 /// The item's `prov` entry, when the backend has page geometry for it
168 /// (`None` → `prov: []`, what the HTML and DOCX backends write).
169 pub prov: Option<TreeProv>,
170 /// docling's `DocItem.comments`: the `comment_section` groups (or note
171 /// text items) annotating this item, as item indices — written after
172 /// `prov` when non-empty.
173 pub comments: Vec<usize>,
174 /// docling's `DocItem.source`: the track segment a text item was taken
175 /// from (WebVTT cues) — written after `prov` when set.
176 pub source: Option<TreeTrack>,
177 /// Removed by [`ItemTree::delete`] (docling's `delete_items`): the slot
178 /// stays so every other index keeps its meaning, but the item is not
179 /// numbered or written.
180 pub deleted: bool,
181 /// Footnotes / endnotes referenced from inside this text item (#538): the
182 /// note call's position (chars into the item's text) and the note's text.
183 /// docling keeps notes as unlinked furniture `footnote` items, so the
184 /// JSON never shows this; the Pandoc AST writes each as a `Note` there.
185 pub notes: Vec<TreeNote>,
186 /// This furniture `footnote` item is the body of a note some text item
187 /// calls (it travels in that item's [`Self::notes`]); the Pandoc AST then
188 /// leaves it out as a standalone block.
189 pub note_body: bool,
190}
191
192/// A note call inside a text item ([`TreeItem::notes`]).
193#[derive(Debug, Clone, PartialEq, Eq)]
194pub struct TreeNote {
195 /// The call's position, in chars into the item's `text`.
196 pub offset: usize,
197 /// The note's plain text.
198 pub text: String,
199}
200
201impl ItemTree {
202 /// Where each note call of one paragraph lands (#538): `full_text` is the
203 /// paragraph's text, `offsets` the calls' positions in it (chars), and
204 /// the paragraph's items are those created from `first_new` on. Each text
205 /// item is a trimmed slice of the paragraph (a formatting run, a link, or
206 /// the whole heading / list item), matched in order; a call inside an
207 /// item lands there, one between items ends the earlier (or starts the
208 /// first). When no item matches, the call ends the paragraph's last text
209 /// item; with no text item at all it is `None`.
210 pub fn place_note_calls(
211 &self,
212 first_new: usize,
213 full_text: &str,
214 offsets: &[usize],
215 ) -> Vec<Option<(usize, usize)>> {
216 let texts: Vec<(usize, &str)> = (first_new..self.items.len())
217 .filter(|&i| !self.items[i].deleted)
218 .filter_map(|i| match &self.items[i].kind {
219 TreeKind::Text { text, .. } | TreeKind::Code { text, .. } => {
220 Some((i, text.as_str()))
221 }
222 _ => None,
223 })
224 .collect();
225 // Each found item's (id, start, end) in chars of `full_text`.
226 let mut spans: Vec<(usize, usize, usize)> = Vec::new();
227 let mut cursor = 0usize; // bytes
228 for &(item, text) in &texts {
229 if text.is_empty() {
230 continue;
231 }
232 if let Some(pos) = full_text[cursor..].find(text) {
233 let start_b = cursor + pos;
234 let start = full_text[..start_b].chars().count();
235 spans.push((item, start, start + text.chars().count()));
236 cursor = start_b + text.len();
237 }
238 }
239 offsets
240 .iter()
241 .map(|&offset| {
242 spans
243 .iter()
244 .find(|&&(_, start, end)| offset >= start && offset <= end)
245 .map(|&(item, start, _)| (item, offset - start))
246 .or_else(
247 || match spans.iter().rev().find(|&&(_, _, end)| end <= offset) {
248 Some(&(item, start, end)) => Some((item, end - start)),
249 None => spans.first().map(|&(item, _, _)| (item, 0)),
250 },
251 )
252 .or_else(|| {
253 texts
254 .last()
255 .map(|&(item, text)| (item, text.chars().count()))
256 })
257 })
258 .collect()
259 }
260}
261
262/// docling's item tree in creation order (see the [module docs](self)).
263#[derive(Debug, Clone, PartialEq, Default)]
264pub struct ItemTree {
265 /// Every item, indexed by creation order — which is how docling numbers
266 /// `#/texts/N`, `#/groups/N`, … within each bucket.
267 pub items: Vec<TreeItem>,
268 /// The body's `children`, as item indices.
269 pub body: Vec<usize>,
270}
271
272impl ItemTree {
273 /// Append an item under `parent` (`None` = body) on `layer`, registering
274 /// it as its parent's last child — docling's `add_*` calls do exactly that.
275 pub fn add(
276 &mut self,
277 parent: Option<usize>,
278 layer: Option<ContentLayer>,
279 kind: TreeKind,
280 ) -> usize {
281 let id = self.items.len();
282 self.items.push(TreeItem {
283 parent,
284 children: Vec::new(),
285 layer,
286 kind,
287 prov: None,
288 comments: Vec::new(),
289 source: None,
290 deleted: false,
291 notes: Vec::new(),
292 note_body: false,
293 });
294 match parent {
295 Some(p) => self.items[p].children.push(id),
296 None => self.body.push(id),
297 }
298 id
299 }
300
301 /// [`add`](Self::add) with the item's provenance — docling's
302 /// `add_text(…, prov=prov)`.
303 pub fn add_with_prov(
304 &mut self,
305 parent: Option<usize>,
306 layer: Option<ContentLayer>,
307 kind: TreeKind,
308 prov: TreeProv,
309 ) -> usize {
310 let id = self.add(parent, layer, kind);
311 self.items[id].prov = Some(prov);
312 id
313 }
314
315 /// Append every item of `other` after this tree's, renumbering its
316 /// indices (parents, children, comments, table/picture caption and
317 /// rich-cell refs) and adding its body children to this body — so a
318 /// backend can build independent fragments in parallel (one per PPTX
319 /// slide) and still hand the export one tree in creation order, exactly
320 /// as if it had been built sequentially.
321 pub fn append(&mut self, other: ItemTree) {
322 let off = self.items.len();
323 let shift = |i: usize| i + off;
324 for mut item in other.items {
325 item.parent = item.parent.map(shift);
326 for c in item.children.iter_mut().chain(item.comments.iter_mut()) {
327 *c = shift(*c);
328 }
329 match &mut item.kind {
330 TreeKind::Table {
331 rich_cells,
332 captions,
333 ..
334 } => {
335 for (_, _, g) in rich_cells.iter_mut() {
336 *g = shift(*g);
337 }
338 for c in captions.iter_mut() {
339 *c = shift(*c);
340 }
341 }
342 TreeKind::Picture { captions, .. } => {
343 for c in captions.iter_mut() {
344 *c = shift(*c);
345 }
346 }
347 _ => {}
348 }
349 self.items.push(item);
350 }
351 self.body.extend(other.body.into_iter().map(shift));
352 }
353
354 /// Move `id` under `new_parent`, dropping it from its current parent's
355 /// children and appending it to the new one's — docling's
356 /// `group_cell_elements` re-parenting of a rich cell's items.
357 pub fn reparent(&mut self, id: usize, new_parent: Option<usize>) {
358 let old = self.items[id].parent;
359 let siblings = match old {
360 Some(p) => &mut self.items[p].children,
361 None => &mut self.body,
362 };
363 siblings.retain(|&c| c != id);
364 self.items[id].parent = new_parent;
365 match new_parent {
366 Some(p) => self.items[p].children.push(id),
367 None => self.body.push(id),
368 }
369 }
370
371 /// Re-number the items in traversal order — a pre-order walk of the body
372 /// through every layer, which is how docling's
373 /// `DoclingDocument.concatenate` (and `_normalize_references`) re-creates
374 /// a document's items: a group created after the content it was later
375 /// wrapped around comes before that content afterwards. Items the walk
376 /// does not reach (deleted ones) are dropped. Returns each old id's new
377 /// id.
378 pub fn renumber_in_traversal_order(&mut self) -> Vec<Option<usize>> {
379 let mut order = Vec::with_capacity(self.items.len());
380 let mut stack: Vec<usize> = self.body.iter().rev().copied().collect();
381 while let Some(id) = stack.pop() {
382 if self.items[id].deleted {
383 continue;
384 }
385 order.push(id);
386 stack.extend(self.items[id].children.iter().rev());
387 }
388 let mut new_of: Vec<Option<usize>> = vec![None; self.items.len()];
389 for (new, &old) in order.iter().enumerate() {
390 new_of[old] = Some(new);
391 }
392 let remap = |ids: &mut Vec<usize>| {
393 *ids = ids.iter().filter_map(|&i| new_of[i]).collect();
394 };
395 let mut old_items: Vec<Option<TreeItem>> = std::mem::take(&mut self.items)
396 .into_iter()
397 .map(Some)
398 .collect();
399 for &old in &order {
400 let mut item = old_items[old].take().expect("each item visited once");
401 item.parent = item.parent.and_then(|p| new_of[p]);
402 remap(&mut item.children);
403 remap(&mut item.comments);
404 match &mut item.kind {
405 TreeKind::Table {
406 rich_cells,
407 captions,
408 ..
409 } => {
410 rich_cells.retain_mut(|(_, _, g)| match new_of[*g] {
411 Some(n) => {
412 *g = n;
413 true
414 }
415 None => false,
416 });
417 remap(captions);
418 }
419 TreeKind::Picture { captions, .. } => remap(captions),
420 _ => {}
421 }
422 self.items.push(item);
423 }
424 remap(&mut self.body);
425 new_of
426 }
427
428 /// Remove `id` from the tree — docling's `delete_items`, which the DOCX
429 /// backend uses to drop the empty text item a blank spacer paragraph left
430 /// between two items of a resumed list. The item leaves its parent's
431 /// children and is neither numbered nor written; its slot stays so the
432 /// indices held elsewhere stay valid.
433 pub fn delete(&mut self, id: usize) {
434 match self.items[id].parent {
435 Some(p) => self.items[p].children.retain(|&c| c != id),
436 None => self.body.retain(|&c| c != id),
437 }
438 self.items[id].deleted = true;
439 }
440
441 /// docling-core's `DoclingDocument.validate_misplaced_list_items` (a
442 /// model validator, so it runs whenever docling-core serializes or loads
443 /// a document): every `list_item` whose parent is not a `list` group is
444 /// re-homed into a new one. A pre-order walk of the body (groups
445 /// included) collects them; consecutive misplaced items directly on the
446 /// body share one group, any other misplaced item gets its own. Working
447 /// from the last run back, each run gets a `ListGroup` (name `group`)
448 /// inserted where its first item stood, the items are deleted and
449 /// re-added under the group as fresh items, so they move to the end of
450 /// the text numbering (#527: the DOCX backend leaves such items in rich
451 /// table cells). A no-op for a well-formed tree.
452 ///
453 /// One deliberate divergence (#586): docling-core deletes each item *with
454 /// its children* and re-adds it from its text alone, so a mixed-format
455 /// item — an empty `list_item` over an `inline` group of text runs —
456 /// comes back empty and its text is gone from the JSON (and from the
457 /// Markdown/HTML docling serializes after that validator ran; the JSON
458 /// docling saves *before* exporting still has it). Here the children
459 /// follow the item into the group: the structure docling-core requires,
460 /// the content the document had.
461 pub fn wrap_misplaced_list_items(&mut self) {
462 let is_list_item = |t: &Self, id: usize| matches!(&t.items[id].kind, TreeKind::Text { label, .. } if label == "list_item");
463 let in_list_group = |t: &Self, id: usize| {
464 t.items[id].parent.is_some_and(
465 |p| matches!(&t.items[p].kind, TreeKind::Group { label, .. } if label == "list"),
466 )
467 };
468 let mut runs: Vec<Vec<usize>> = Vec::new();
469 // `None` = the body itself, which the walk yields first.
470 let mut prev: Option<usize> = None;
471 let mut stack: Vec<usize> = self.body.iter().rev().copied().collect();
472 while let Some(id) = stack.pop() {
473 if self.items[id].deleted {
474 continue;
475 }
476 if is_list_item(self, id) && !in_list_group(self, id) {
477 let continues =
478 prev.is_some_and(|p| is_list_item(self, p) && self.items[p].parent.is_none());
479 match runs.last_mut() {
480 Some(run) if continues => run.push(id),
481 _ => runs.push(vec![id]),
482 }
483 }
484 prev = Some(id);
485 stack.extend(self.items[id].children.iter().rev().copied());
486 }
487 for run in runs.into_iter().rev() {
488 let parent = self.items[run[0]].parent;
489 let group = self.add(
490 parent,
491 None,
492 TreeKind::Group {
493 label: "list".into(),
494 name: "group".into(),
495 },
496 );
497 let siblings = match parent {
498 Some(p) => &mut self.items[p].children,
499 None => &mut self.body,
500 };
501 siblings.pop();
502 let at = siblings
503 .iter()
504 .position(|&c| c == run[0])
505 .unwrap_or(siblings.len());
506 siblings.insert(at, group);
507 for &li in &run {
508 // Not `delete_subtree`: the children move to the copy (#586).
509 self.delete(li);
510 }
511 // `add_list_item` keeps the text, marker, formatting, hyperlink
512 // and first provenance — not comments or a source. The children
513 // (the inline group of a mixed-format item) are carried over.
514 for &li in &run {
515 let children = std::mem::take(&mut self.items[li].children);
516 let copy = TreeItem {
517 parent: Some(group),
518 children: children.clone(),
519 comments: Vec::new(),
520 source: None,
521 deleted: false,
522 ..self.items[li].clone()
523 };
524 let id = self.items.len();
525 self.items.push(copy);
526 for c in children {
527 self.items[c].parent = Some(id);
528 }
529 self.items[group].children.push(id);
530 }
531 }
532 }
533
534 /// The last live text-bucket item (docling's `doc.texts[-1]`).
535 pub fn last_text(&self) -> Option<usize> {
536 self.items.iter().rposition(|it| {
537 !it.deleted && matches!(it.kind, TreeKind::Text { .. } | TreeKind::Code { .. })
538 })
539 }
540
541 /// How many items of a bucket precede `id` — its `#/{bucket}/N` index.
542 pub fn bucket_index(&self, id: usize) -> usize {
543 let same = |k: &TreeKind| {
544 std::mem::discriminant(k) == std::mem::discriminant(&self.items[id].kind)
545 || matches!(
546 (k, &self.items[id].kind),
547 (TreeKind::Text { .. }, TreeKind::Code { .. })
548 | (TreeKind::Code { .. }, TreeKind::Text { .. })
549 )
550 };
551 self.items[..id]
552 .iter()
553 .filter(|it| !it.deleted && same(&it.kind))
554 .count()
555 }
556
557 /// The number of tables created so far (docling's `len(doc.tables)`).
558 pub fn table_count(&self) -> usize {
559 self.items
560 .iter()
561 .filter(|it| !it.deleted && matches!(it.kind, TreeKind::Table { .. }))
562 .count()
563 }
564}
565
566#[cfg(test)]
567mod tests {
568 use super::*;
569
570 fn text(t: &str) -> TreeKind {
571 TreeKind::Text {
572 label: "text".into(),
573 text: t.into(),
574 orig: None,
575 formatting: None,
576 hyperlink: None,
577 level: None,
578 list: None,
579 }
580 }
581
582 /// `add` registers the item as its parent's (or the body's) last child;
583 /// `reparent` moves it — a rich cell's items leave the heading they were
584 /// created under for the table's group.
585 #[test]
586 fn add_and_reparent_keep_docling_children_order() {
587 let mut t = ItemTree::default();
588 let title = t.add(None, None, text("Title"));
589 let a = t.add(Some(title), None, text("a"));
590 let b = t.add(Some(title), None, text("b"));
591 let table = t.add(
592 Some(title),
593 None,
594 TreeKind::Table {
595 table: Table::default(),
596 rich_cells: Vec::new(),
597 captions: Vec::new(),
598 },
599 );
600 let group = t.add(
601 Some(table),
602 None,
603 TreeKind::Group {
604 label: "unspecified".into(),
605 name: "rich_cell_group_1_0_0".into(),
606 },
607 );
608 assert_eq!(t.body, vec![title]);
609 assert_eq!(t.items[title].children, vec![a, b, table]);
610 t.reparent(a, Some(group));
611 assert_eq!(t.items[title].children, vec![b, table]);
612 assert_eq!(t.items[group].children, vec![a]);
613 assert_eq!(t.items[a].parent, Some(group));
614 assert_eq!(t.table_count(), 1);
615 // Text and code share the `texts` bucket.
616 let code = t.add(
617 None,
618 None,
619 TreeKind::Code {
620 text: "x".into(),
621 orig: None,
622 language: None,
623 formatting: None,
624 hyperlink: None,
625 },
626 );
627 assert_eq!(t.bucket_index(code), 3, "title, a, b precede it in `texts`");
628 assert_eq!(t.bucket_index(group), 0);
629 assert_eq!(t.body, vec![title, code]);
630 }
631
632 /// `append` renumbers a fragment built on its own (a slide converted in
633 /// parallel) so the merged tree reads as if built in one pass: parents,
634 /// children, comment back-refs and caption refs all shift together.
635 #[test]
636 fn append_renumbers_a_fragment_into_creation_order() {
637 let mut whole = ItemTree::default();
638 let slide0 = whole.add(
639 None,
640 None,
641 TreeKind::Group {
642 label: "chapter".into(),
643 name: "slide-0".into(),
644 },
645 );
646 whole.add(Some(slide0), None, text("first"));
647
648 let mut frag = ItemTree::default();
649 let slide1 = frag.add(
650 None,
651 None,
652 TreeKind::Group {
653 label: "chapter".into(),
654 name: "slide-1".into(),
655 },
656 );
657 let cap = frag.add_with_prov(
658 Some(slide1),
659 None,
660 TreeKind::Text {
661 label: "caption".into(),
662 text: "Title".into(),
663 orig: None,
664 formatting: None,
665 hyperlink: None,
666 level: None,
667 list: None,
668 },
669 TreeProv {
670 page_no: 2,
671 bbox: [1.0, 2.0, 3.0, 4.0],
672 bottom_left: true,
673 charspan: [0, 5],
674 },
675 );
676 let pic = frag.add(
677 Some(slide1),
678 None,
679 TreeKind::Picture {
680 captions: vec![cap],
681 image: None,
682 classification: Some("bar_chart".into()),
683 description: None,
684 confidence: None,
685 chart: None,
686 dpi: None,
687 },
688 );
689 let note = frag.add(
690 None,
691 Some(ContentLayer::Notes),
692 TreeKind::Group {
693 label: "comment_section".into(),
694 name: "comment-slide2-1".into(),
695 },
696 );
697 frag.items[pic].comments.push(note);
698
699 whole.append(frag);
700 assert_eq!(whole.body, vec![slide0, 2, 5]);
701 assert_eq!(whole.items[2].children, vec![3, 4]);
702 assert_eq!(whole.items[3].parent, Some(2));
703 assert_eq!(whole.items[3].prov.as_ref().map(|p| p.page_no), Some(2));
704 assert!(
705 matches!(&whole.items[4].kind, TreeKind::Picture { captions, .. } if captions == &[3])
706 );
707 assert_eq!(whole.items[4].comments, vec![5]);
708 assert_eq!(whole.items[5].parent, None);
709 assert_eq!(
710 whole.bucket_index(4),
711 0,
712 "the fragment's picture is #/pictures/0"
713 );
714 assert_eq!(
715 whole.bucket_index(5),
716 2,
717 "slide-0, slide-1 precede it in `groups`"
718 );
719 }
720
721 /// `renumber_in_traversal_order`: docling's `concatenate` re-creates the
722 /// items as a pre-order walk meets them, so a group created after the
723 /// content it was wrapped around (a rich cell's group) comes first, and a
724 /// deleted item disappears; every cross-reference follows.
725 #[test]
726 fn renumbering_follows_the_traversal() {
727 let mut t = ItemTree::default();
728 let a = t.add(None, None, text("a"));
729 let table = t.add(
730 None,
731 None,
732 TreeKind::Table {
733 table: Table::default(),
734 rich_cells: Vec::new(),
735 captions: Vec::new(),
736 },
737 );
738 let cell_text = t.add(None, None, text("cell"));
739 let group = t.add(
740 Some(table),
741 None,
742 TreeKind::Group {
743 label: "unspecified".into(),
744 name: "rich_cell_group_1_0_0".into(),
745 },
746 );
747 t.reparent(cell_text, Some(group));
748 if let TreeKind::Table { rich_cells, .. } = &mut t.items[table].kind {
749 rich_cells.push((0, 0, group));
750 }
751 let gone = t.add(None, None, text("gone"));
752 t.delete(gone);
753 let z = t.add(None, None, text("z"));
754
755 let new_of = t.renumber_in_traversal_order();
756 assert_eq!(
757 new_of,
758 vec![Some(0), Some(1), Some(3), Some(2), None, Some(4)]
759 );
760 assert_eq!(t.items.len(), 5);
761 assert_eq!(t.body, vec![0, 1, 4]);
762 assert_eq!(t.items[1].children, vec![2], "the group follows its table");
763 assert_eq!(t.items[2].parent, Some(1));
764 assert_eq!(
765 t.items[3].parent,
766 Some(2),
767 "the cell text follows its group"
768 );
769 assert!(matches!(&t.items[3].kind, TreeKind::Text { text, .. } if text == "cell"));
770 assert!(
771 matches!(&t.items[1].kind, TreeKind::Table { rich_cells, .. } if rich_cells == &[(0, 0, 2)])
772 );
773 let _ = (a, z);
774 }
775
776 /// #586: a misplaced `list_item` keeps its children when it is re-homed.
777 /// docling-core's `validate_misplaced_list_items` re-adds the item from
778 /// its text alone, so a mixed-format item (empty `list_item` over an
779 /// `inline` group of runs) lost every run — a DOCX table cell whose list
780 /// paragraph followed a `numId 0` spacer came out as an empty bullet in
781 /// the JSON and the HTML. The group goes where the item stood, the copy
782 /// is numbered last, the inline group and its texts move under it.
783 #[test]
784 fn wrapping_a_misplaced_list_item_keeps_its_runs() {
785 let mut t = ItemTree::default();
786 let cell = t.add(
787 None,
788 None,
789 TreeKind::Group {
790 label: "unspecified".into(),
791 name: "rich_cell_group_1_0_1".into(),
792 },
793 );
794 let before = t.add(Some(cell), None, text("before"));
795 let item = t.add(
796 Some(cell),
797 None,
798 TreeKind::Text {
799 label: "list_item".into(),
800 text: String::new(),
801 orig: None,
802 formatting: None,
803 hyperlink: None,
804 level: None,
805 list: Some(ListMeta {
806 enumerated: false,
807 marker: String::new(),
808 }),
809 },
810 );
811 let inline = t.add(
812 Some(item),
813 None,
814 TreeKind::Group {
815 label: "inline".into(),
816 name: "group".into(),
817 },
818 );
819 let run_a = t.add(Some(inline), None, text("Second item text"));
820 let run_b = t.add(Some(inline), None, text("[Optional]"));
821 let after = t.add(Some(cell), None, text("after"));
822
823 t.wrap_misplaced_list_items();
824
825 let group = t.items[cell].children[1];
826 assert_eq!(t.items[cell].children, vec![before, group, after]);
827 assert!(
828 matches!(&t.items[group].kind, TreeKind::Group { label, name } if label == "list" && name == "group")
829 );
830 assert!(t.items[item].deleted, "the original item is deleted");
831 let copy = t.items[group].children[0];
832 assert_ne!(copy, item);
833 assert!(
834 copy > after,
835 "the copy is numbered after every existing item"
836 );
837 assert_eq!(t.items[copy].children, vec![inline]);
838 assert_eq!(t.items[inline].parent, Some(copy));
839 assert_eq!(t.items[inline].children, vec![run_a, run_b]);
840 for id in [inline, run_a, run_b] {
841 assert!(!t.items[id].deleted, "item {id} must survive the re-homing");
842 }
843 }
844}