docling 1.48.0

DocumentConverter and format backends for docling.rs (a Rust port of docling).
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
//! docling JSON backend — reads docling's native `DoclingDocument` JSON
//! serialization and re-exports it. docling just pydantic-loads the model; here
//! we walk the `body` tree (children resolved through `$ref` into the
//! `texts`/`groups`/`tables`/`pictures` arrays, skipping `furniture`) and map
//! each item onto the crate's [`Node`] model so the shared Markdown serializer
//! reproduces docling-core's output.

use serde_json::Value;

use crate::backend::markdown::escape_text;
use crate::backend::DeclarativeBackend;
use crate::error::ConversionError;
use crate::source::SourceDocument;
use docling_core::{DoclingDocument, Node, Table};

pub struct DoclingJsonBackend;

impl DeclarativeBackend for DoclingJsonBackend {
    fn convert(&self, source: &SourceDocument) -> Result<DoclingDocument, ConversionError> {
        let root: Value = serde_json::from_str(source.text()?)
            .map_err(|e| ConversionError::with_source("docling-json", e))?;
        let name = root["name"].as_str().unwrap_or(&source.name).to_string();
        let mut doc = DoclingDocument::new(name);
        if let Some(children) = root["body"]["children"].as_array() {
            for c in children {
                walk(c, &root, 0, &mut doc);
            }
        }
        Ok(doc)
    }
}

/// Resolve a `{"$ref": "#/texts/3"}` reference into its array element.
fn resolve<'a>(reference: &Value, root: &'a Value) -> Option<&'a Value> {
    let path = reference["$ref"].as_str()?.strip_prefix("#/")?;
    let (kind, idx) = path.rsplit_once('/')?;
    root.get(kind)?.get(idx.parse::<usize>().ok()?)
}

fn ref_kind(reference: &Value) -> &str {
    reference["$ref"].as_str().unwrap_or("")
}

fn text_of(reference: &Value, root: &Value) -> String {
    resolve(reference, root)
        .map(formatted_text)
        .unwrap_or_default()
}

/// Escaped item text with docling-core's inline markers applied in order:
/// bold → italic → strikethrough → hyperlink (underline/script are no-ops in
/// Markdown).
fn formatted_text(item: &Value) -> String {
    let mut res = escape_text(item["text"].as_str().unwrap_or(""));
    let fmt = &item["formatting"];
    if fmt["bold"].as_bool() == Some(true) {
        res = format!("**{res}**");
    }
    if fmt["italic"].as_bool() == Some(true) {
        res = format!("*{res}*");
    }
    if fmt["strikethrough"].as_bool() == Some(true) {
        res = format!("~~{res}~~");
    }
    if let Some(url) = item["hyperlink"].as_str() {
        res = format!("[{res}]({url})");
    }
    res
}

/// Dispatch one body/child reference by the array it points into.
fn walk(reference: &Value, root: &Value, level: u8, doc: &mut DoclingDocument) {
    let Some(item) = resolve(reference, root) else {
        return;
    };
    if item["content_layer"].as_str() == Some("furniture") {
        return;
    }
    let kind = ref_kind(reference);
    if kind.starts_with("#/texts/") {
        text_item(item, root, level, doc);
        // docling nests a section's content under its heading (and a list
        // item's sub-list under the item), so a text item's `children` are
        // body content too — without this walk a document collapses to its
        // first heading. Tables and pictures are not recursed: their children
        // are rich-cell / caption items already rendered with the parent.
        if let Some(children) = item["children"].as_array() {
            for c in children {
                walk(c, root, level, doc);
            }
        }
    } else if kind.starts_with("#/groups/") {
        group_item(item, root, level, doc);
    } else if kind.starts_with("#/tables/") {
        table_item(item, root, doc);
    } else if kind.starts_with("#/pictures/") {
        picture_item(item, root, doc);
    }
}

fn text_item(item: &Value, root: &Value, level: u8, doc: &mut DoclingDocument) {
    let label = item["label"].as_str().unwrap_or("text");
    // docling does not serialize empty text items (an undecoded formula is the
    // one exception — it becomes a placeholder comment).
    if item["text"].as_str().unwrap_or("").is_empty() {
        if label == "formula" {
            doc.push(Node::Paragraph {
                text: "<!-- formula-not-decoded -->".into(),
            });
        }
        return;
    }
    let text = formatted_text(item);
    // Code and formulas are the two items docling serializes *unescaped*
    // (`escape_html = False`, `escape_underscores = False`): a SQL body keeps
    // its `VERIFY_GROUP_FOR_USER`, not `VERIFY\_GROUP\_FOR\_USER`.
    let raw = || item["text"].as_str().unwrap_or("").to_string();
    match label {
        "title" => doc.push(Node::Heading { level: 1, text }),
        "section_header" => {
            let lvl = item["level"].as_u64().unwrap_or(1) as u8;
            doc.push(Node::Heading {
                level: lvl + 1,
                text,
            });
        }
        "code" => {
            doc.push(Node::Code {
                language: item["code_language"]
                    .as_str()
                    .filter(|s| !s.is_empty() && *s != "unknown")
                    .map(String::from),
                text: raw(),
                orig: None,
                pretty: None,
            });
            // A `CodeItem` is docling's only *floating* text item, and the
            // text serializer appends a floating item's captions **after** its
            // own text — the opposite of a picture or a table, which lead with
            // theirs (`Listing 1: …` under the fence, not above it).
            if let Some(cap) = caption_of(item, root) {
                doc.push(Node::Paragraph { text: cap });
            }
        }
        "list_item" => doc.push(Node::ListItem {
            ordered: item["enumerated"].as_bool().unwrap_or(false),
            number: 1,
            first_in_list: true,
            text,
            level,
            marker: None,
            location: None,
            dclx: None,
            href: None,
            layer: None,
        }),
        // docling prefixes the task-list marker to the text of a checkbox item.
        "checkbox_selected" => doc.push(Node::CheckboxItem {
            checked: true,
            text,
        }),
        "checkbox_unselected" => doc.push(Node::CheckboxItem {
            checked: false,
            text,
        }),
        // A decoded formula renders as `$$…$$`, also unescaped.
        "formula" => doc.push(Node::Formula {
            latex: raw(),
            orig: item["orig"].as_str().unwrap_or("").to_string(),
            location: None,
        }),
        // A caption some table/picture/code claims renders with that element;
        // one nobody claims is an ordinary body item and renders where it sits
        // (docling's serializer only skips the refs a floating item consumed).
        "caption" => {
            if !caption_is_claimed(item, root) {
                doc.push(Node::Caption { text, href: None });
            }
        }
        _ => doc.push(Node::Paragraph { text }), // text, paragraph, footnote, …
    }
}

fn group_item(item: &Value, root: &Value, level: u8, doc: &mut DoclingDocument) {
    let label = item["label"].as_str().unwrap_or("unspecified");
    let empty = Vec::new();
    let children = item["children"].as_array().unwrap_or(&empty);
    match label {
        "list" | "ordered_list" => list_group(children, root, level, doc),
        "inline" => {
            // An inline group is one line: serialize each child and join with " ".
            let joined = children
                .iter()
                .map(|c| text_of(c, root))
                .collect::<Vec<_>>()
                .join(" ");
            if !joined.is_empty() {
                doc.push(Node::Paragraph { text: joined });
            }
        }
        // section / chapter / unspecified / sheet / comment_section → transparent
        _ => {
            for c in children {
                walk(c, root, level, doc);
            }
        }
    }
}

/// How docling-core's Markdown serializer renders a list item's marker.
///
/// * A marker it already considers valid Markdown — a bullet, or `12.` — is
///   printed **verbatim** and nothing is computed. This is what carries a
///   split reference list's real numbering (`18.` on the group that continues
///   over a page break) instead of restarting at 1.
/// * Any *other* non-empty marker (`a.`, `[7]`, `(3)`) forces a **bullet**:
///   the computed marker is a number only when the item carries no marker at
///   all (`… and (mode != AUTO or not item.marker)`). The original marker is
///   then kept after it when it holds a letter or digit, so docling renders
///   `- (1) Human Annotation`. In this node model it rides in the item's text,
///   which is where it lands on the rendered line either way; a marker with
///   no alphanumerics (a stray bullet glyph) is dropped, as upstream drops it.
/// * No marker at all leaves the group to decide: `{position}.` when its first
///   child is an enumerated item, `-` otherwise.
enum Marker {
    /// Print verbatim: `Some(n)` numbers the item, `None` bullets it.
    Verbatim(Option<u64>),
    /// Bullet, with this text (if any) kept in front of the item's own.
    Bullet(Option<String>),
    /// Nothing of its own — the group's kind and the item's position decide.
    FromGroup,
}

fn marker_of(item: &Value) -> Marker {
    let raw = item["marker"].as_str().unwrap_or("");
    if raw.is_empty() {
        return Marker::FromGroup;
    }
    if matches!(raw, "-" | "*" | "+") {
        return Marker::Verbatim(None);
    }
    if let Some(digits) = raw.strip_suffix('.') {
        if !digits.is_empty() && digits.bytes().all(|b| b.is_ascii_digit()) {
            return Marker::Verbatim(digits.parse().ok());
        }
    }
    Marker::Bullet(
        raw.chars()
            .any(|c| c.is_ascii_alphanumeric())
            .then(|| raw.to_string()),
    )
}

/// Emit a list group's items, recursing into nested lists at the next level.
fn list_group(children: &[Value], root: &Value, level: u8, doc: &mut DoclingDocument) {
    // docling-core's `first_item_is_enumerated`: the group renders computed
    // markers as numbers only when its *first child* is an enumerated item.
    let enumerated_group = children
        .first()
        .and_then(|c| resolve(c, root))
        .is_some_and(|it| {
            it["label"].as_str() == Some("list_item") && it["enumerated"].as_bool() == Some(true)
        });
    let mut first = true;
    for (pos, c) in children.iter().enumerate() {
        let kind = ref_kind(c);
        if kind.starts_with("#/groups/") {
            // A bare nested list (no enclosing item). It still occupies a
            // position in this group, which is why a later item can be
            // numbered past it.
            walk(c, root, level + 1, doc);
            continue;
        }
        let Some(item) = resolve(c, root) else {
            continue;
        };
        if item["label"].as_str() != Some("list_item") {
            continue;
        }
        let position = pos as u64 + 1;
        let body = formatted_text(item);
        let (ordered, number, text) = match marker_of(item) {
            Marker::Verbatim(Some(n)) => (true, n, body),
            Marker::Verbatim(None) => (false, position, body),
            Marker::Bullet(kept) => {
                let text = match kept {
                    Some(m) if !body.is_empty() => format!("{m} {body}"),
                    Some(m) => m,
                    None => body,
                };
                (false, position, text)
            }
            Marker::FromGroup => (enumerated_group, position, body),
        };
        doc.push(Node::ListItem {
            ordered,
            number,
            first_in_list: first,
            text,
            level,
            marker: None,
            location: None,
            dclx: None,
            href: None,
            layer: None,
        });
        first = false;
        if let Some(sub) = item["children"].as_array() {
            for s in sub {
                walk(s, root, level + 1, doc);
            }
        }
    }
}

fn table_item(item: &Value, root: &Value, doc: &mut DoclingDocument) {
    let mut rows = Vec::new();
    // Per-cell flags carried through so DocTags export and the chunker's
    // triplet serialization see docling's header/span structure, not just the
    // flattened text grid.
    let mut structure = docling_core::TableStructure::default();
    if let Some(grid) = item["data"]["grid"].as_array() {
        for (r, row) in grid.iter().enumerate() {
            let Some(cells) = row.as_array() else {
                continue;
            };
            rows.push(
                cells
                    .iter()
                    .map(|cell| cell["text"].as_str().unwrap_or("").to_string())
                    .collect::<Vec<_>>(),
            );
            // A spanning cell is repeated at each grid position it covers; its
            // start offsets mark the anchor, everything past it a continuation.
            let past = |cell: &Value, key: &str, idx: usize| {
                cell[key].as_u64().is_some_and(|v| (v as usize) < idx)
            };
            structure
                .col_header
                .push(flags(cells, |c| c["column_header"].as_bool() == Some(true)));
            structure
                .row_header
                .push(flags(cells, |c| c["row_header"].as_bool() == Some(true)));
            structure.col_continuation.push(flags(cells.as_slice(), {
                let mut col = 0;
                move |c| {
                    let cont = past(c, "start_col_offset_idx", col);
                    col += 1;
                    cont
                }
            }));
            structure
                .row_continuation
                .push(flags(cells, |c| past(c, "start_row_offset_idx", r)));
        }
    }
    if !rows.is_empty() {
        let has_structure = structure.col_header.iter().flatten().any(|&b| b)
            || structure.row_header.iter().flatten().any(|&b| b)
            || structure.col_continuation.iter().flatten().any(|&b| b)
            || structure.row_continuation.iter().flatten().any(|&b| b);
        doc.push(Node::Table(Table {
            rows,
            location: None,
            structure: has_structure.then_some(structure),
            cell_blocks: None,
            cells: None,
            caption: caption_of(item, root),
        }));
    }
}

/// Map each grid cell to a flag.
fn flags(cells: &[Value], f: impl FnMut(&Value) -> bool) -> Vec<bool> {
    cells.iter().map(f).collect()
}

fn picture_item(item: &Value, root: &Value, doc: &mut DoclingDocument) {
    doc.push(Node::Picture {
        caption: caption_of(item, root),
        caption_href: None,
        image: None,
        classification: None,
    });
}

/// Whether a floating item lists this caption in its own `captions` — tables,
/// pictures and code items are the three that can. Scanning per caption keeps
/// the walk's signature unchanged; a document has few of either.
fn caption_is_claimed(item: &Value, root: &Value) -> bool {
    let Some(me) = item["self_ref"].as_str() else {
        return false;
    };
    ["tables", "pictures", "texts"].iter().any(|bucket| {
        root[bucket].as_array().is_some_and(|items| {
            items.iter().any(|it| {
                it["captions"]
                    .as_array()
                    .is_some_and(|caps| caps.iter().any(|c| c["$ref"].as_str() == Some(me)))
            })
        })
    })
}

/// An item's `captions` (refs into `texts`), joined as docling-core joins them
/// (`caption_delim`, a space). It belongs *on* the table or picture, not after
/// it: every serializer renders a caption before its element, and emitting the
/// caption items as trailing paragraphs put them on the wrong side (#384).
fn caption_of(item: &Value, root: &Value) -> Option<String> {
    let caps = item["captions"].as_array()?;
    let joined = caps
        .iter()
        .map(|c| text_of(c, root))
        .collect::<Vec<_>>()
        .join(" ");
    (!joined.trim().is_empty()).then_some(joined)
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::format::InputFormat;

    fn md(json: &str) -> String {
        DoclingJsonBackend
            .convert(&SourceDocument::from_bytes(
                "t.json",
                InputFormat::JsonDocling,
                json.as_bytes().to_vec(),
            ))
            .unwrap()
            .export_to_markdown()
    }

    /// #384: a caption leads its picture and its table, and trails its *code*
    /// block — a `CodeItem` is docling's only floating text item, and the text
    /// serializer appends a floating item's captions after its own text.
    #[test]
    fn captions_lead_a_picture_and_a_table_but_trail_code() {
        let json = r##"{
          "name":"n","body":{"children":[{"$ref":"#/pictures/0"},{"$ref":"#/tables/0"},{"$ref":"#/texts/2"}]},
          "texts":[
            {"self_ref":"#/texts/0","label":"caption","text":"Figure 1: a duck","children":[]},
            {"self_ref":"#/texts/1","label":"caption","text":"Table 1: the counts","children":[]},
            {"self_ref":"#/texts/2","label":"code","text":"let x = 1;","children":[],
             "captions":[{"$ref":"#/texts/3"}]},
            {"self_ref":"#/texts/3","label":"caption","text":"Listing 1: a binding","children":[]}
          ],
          "groups":[],
          "tables":[{"self_ref":"#/tables/0","label":"table","captions":[{"$ref":"#/texts/1"}],
                     "data":{"grid":[[{"text":"a"}]]},"children":[]}],
          "pictures":[{"self_ref":"#/pictures/0","label":"picture","captions":[{"$ref":"#/texts/0"}],"children":[]}]
        }"##;
        assert_eq!(
            md(json),
            "Figure 1: a duck\n\n<!-- image -->\n\nTable 1: the counts\n\n| a   |\n|-----|\n\n```\nlet x = 1;\n```\n\nListing 1: a binding\n"
        );
    }

    /// A caption no floating item claims is an ordinary body item and renders
    /// where it sits; one that is claimed renders only with its element.
    #[test]
    fn an_unclaimed_caption_still_renders() {
        let json = r##"{
          "name":"n","body":{"children":[{"$ref":"#/texts/0"},{"$ref":"#/pictures/0"},{"$ref":"#/texts/1"}]},
          "texts":[
            {"self_ref":"#/texts/0","label":"caption","text":"Figure 6: nobody claims me","children":[]},
            {"self_ref":"#/texts/1","label":"caption","text":"Figure 7: claimed","children":[]}
          ],
          "groups":[],"tables":[],
          "pictures":[{"self_ref":"#/pictures/0","label":"picture","captions":[{"$ref":"#/texts/1"}],"children":[]}]
        }"##;
        assert_eq!(
            md(json),
            "Figure 6: nobody claims me\n\nFigure 7: claimed\n\n<!-- image -->\n"
        );
    }

    /// docling-core prints a marker it already considers valid Markdown
    /// verbatim, so a reference list that continues over a page break keeps its
    /// real numbering instead of restarting at 1.
    #[test]
    fn a_numeric_marker_is_printed_verbatim() {
        let json = r##"{
          "name":"n","body":{"children":[{"$ref":"#/groups/0"}]},
          "texts":[
            {"self_ref":"#/texts/0","label":"list_item","text":"Xue, W.","marker":"18.","enumerated":true,"children":[]},
            {"self_ref":"#/texts/1","label":"list_item","text":"Ye, J.","marker":"19.","enumerated":true,"children":[]}
          ],
          "groups":[{"self_ref":"#/groups/0","label":"list","name":"list",
                     "children":[{"$ref":"#/texts/0"},{"$ref":"#/texts/1"}]}],
          "tables":[],"pictures":[]
        }"##;
        assert_eq!(md(json), "18. Xue, W.\n19. Ye, J.\n");
    }

    /// Any *other* non-empty marker forces a bullet and is kept in front of the
    /// text when it holds a letter or digit (docling: `- (1) Human Annotation`),
    /// because a number is computed only for an item with no marker at all.
    #[test]
    fn a_non_markdown_marker_forces_a_bullet_and_is_kept() {
        let json = r##"{
          "name":"n","body":{"children":[{"$ref":"#/groups/0"}]},
          "texts":[
            {"self_ref":"#/texts/0","label":"list_item","text":"Human Annotation","marker":"(1)","enumerated":true,"children":[]},
            {"self_ref":"#/texts/1","label":"list_item","text":"Red - PDF cells","marker":"a.","enumerated":true,"children":[]},
            {"self_ref":"#/texts/2","label":"list_item","text":"a stray glyph","marker":"\u0084","enumerated":false,"children":[]}
          ],
          "groups":[{"self_ref":"#/groups/0","label":"list","name":"list",
                     "children":[{"$ref":"#/texts/0"},{"$ref":"#/texts/1"},{"$ref":"#/texts/2"}]}],
          "tables":[],"pictures":[]
        }"##;
        assert_eq!(
            md(json),
            "- (1) Human Annotation\n- a. Red - PDF cells\n- a stray glyph\n"
        );
    }

    /// With no marker at all the group decides: its first child being an
    /// enumerated item numbers every item by its *position among the children*,
    /// nested groups included — so an item after a sublist is numbered past it.
    #[test]
    fn an_unmarked_item_is_numbered_by_its_position_in_the_group() {
        let json = r##"{
          "name":"n","body":{"children":[{"$ref":"#/groups/0"}]},
          "texts":[
            {"self_ref":"#/texts/0","label":"list_item","text":"one","marker":"","enumerated":true,"children":[]},
            {"self_ref":"#/texts/1","label":"list_item","text":"nested","marker":"","enumerated":true,"children":[]},
            {"self_ref":"#/texts/2","label":"list_item","text":"after","marker":"","enumerated":true,"children":[]}
          ],
          "groups":[
            {"self_ref":"#/groups/0","label":"list","name":"list",
             "children":[{"$ref":"#/texts/0"},{"$ref":"#/groups/1"},{"$ref":"#/texts/2"}]},
            {"self_ref":"#/groups/1","label":"list","name":"list","children":[{"$ref":"#/texts/1"}]}
          ],
          "tables":[],"pictures":[]
        }"##;
        // The nested group takes position 2, so "after" is the third child.
        assert_eq!(md(json), "1. one\n    1. nested\n3. after\n");
    }

    /// A group whose first child is not an enumerated item bullets every item,
    /// whatever each one's own flag says.
    #[test]
    fn a_group_starting_on_a_bullet_stays_bulleted() {
        let json = r##"{
          "name":"n","body":{"children":[{"$ref":"#/groups/0"}]},
          "texts":[
            {"self_ref":"#/texts/0","label":"list_item","text":"bullet first","marker":"","enumerated":false,"children":[]},
            {"self_ref":"#/texts/1","label":"list_item","text":"still a bullet","marker":"","enumerated":true,"children":[]}
          ],
          "groups":[{"self_ref":"#/groups/0","label":"list","name":"list",
                     "children":[{"$ref":"#/texts/0"},{"$ref":"#/texts/1"}]}],
          "tables":[],"pictures":[]
        }"##;
        assert_eq!(md(json), "- bullet first\n- still a bullet\n");
    }

    /// Checkbox items carry docling's task-list marker, and code and formulas
    /// are the two items it serializes unescaped.
    #[test]
    fn checkboxes_render_and_code_is_not_escaped() {
        let json = r##"{
          "name":"n","body":{"children":[{"$ref":"#/texts/0"},{"$ref":"#/texts/1"},{"$ref":"#/texts/2"},{"$ref":"#/texts/3"}]},
          "texts":[
            {"self_ref":"#/texts/0","label":"checkbox_selected","text":"done","children":[]},
            {"self_ref":"#/texts/1","label":"checkbox_unselected","text":"todo","children":[]},
            {"self_ref":"#/texts/2","label":"code","text":"VERIFY_GROUP_FOR_USER ( SESSION_USER )","children":[]},
            {"self_ref":"#/texts/3","label":"formula","text":"a_1 + b_2","orig":"a_1 + b_2","children":[]}
          ],
          "groups":[],"tables":[],"pictures":[]
        }"##;
        assert_eq!(
            md(json),
            "- [x] done\n\n- [ ] todo\n\n```\nVERIFY_GROUP_FOR_USER ( SESSION_USER )\n```\n\n$$a_1 + b_2$$\n"
        );
    }

    /// docling ≥ 2.5x nests body content under its `section_header`; the
    /// nested items must be walked, not dropped with the heading's subtree.
    #[test]
    fn walks_children_nested_under_headings() {
        let json = r##"{
          "name": "n", "body": {"children": [{"$ref":"#/texts/0"}]},
          "texts": [
            {"self_ref":"#/texts/0","label":"section_header","level":1,"text":"Intro",
             "children":[{"$ref":"#/texts/1"},{"$ref":"#/texts/2"}]},
            {"self_ref":"#/texts/1","label":"text","text":"First para","children":[]},
            {"self_ref":"#/texts/2","label":"section_header","level":2,"text":"Sub",
             "children":[{"$ref":"#/texts/3"}]},
            {"self_ref":"#/texts/3","label":"text","text":"Deep para","children":[]}
          ],
          "groups": [], "tables": [], "pictures": []
        }"##;
        let doc = DoclingJsonBackend
            .convert(&SourceDocument::from_bytes(
                "t.json",
                InputFormat::JsonDocling,
                json.as_bytes().to_vec(),
            ))
            .unwrap();
        assert_eq!(
            doc.export_to_markdown(),
            "## Intro\n\nFirst para\n\n### Sub\n\nDeep para\n"
        );
    }

    #[test]
    fn walks_body_tree_with_formatting_and_lists() {
        let json = r##"{
          "schema_name": "DoclingDocument", "name": "t",
          "body": {"children": [{"$ref":"#/texts/0"},{"$ref":"#/texts/1"},{"$ref":"#/groups/0"}]},
          "texts": [
            {"self_ref":"#/texts/0","label":"title","text":"Doc"},
            {"self_ref":"#/texts/1","label":"section_header","level":1,"text":"Sec","hyperlink":"http://x"},
            {"self_ref":"#/texts/2","label":"list_item","text":"one","enumerated":false},
            {"self_ref":"#/texts/3","label":"list_item","text":"two","enumerated":false,
             "formatting":{"bold":true,"italic":false,"strikethrough":false}}
          ],
          "groups": [{"self_ref":"#/groups/0","label":"list",
                      "children":[{"$ref":"#/texts/2"},{"$ref":"#/texts/3"}]}],
          "tables": [], "pictures": []
        }"##;
        let src =
            SourceDocument::from_bytes("t", InputFormat::JsonDocling, json.as_bytes().to_vec());
        let md = DoclingJsonBackend
            .convert(&src)
            .unwrap()
            .export_to_markdown();
        assert!(
            md.starts_with("# Doc\n\n## [Sec](http://x)\n\n- one\n- **two**"),
            "got:\n{md}"
        );
    }
}