rs-chunks 0.1.0

Fast, high-fidelity document chunking for RAG — a pure-Rust engine covering 36 file formats (Office, OpenDocument, PDF, email, ebooks, notebooks, and more).
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

//! `.ipynb` Jupyter notebook chunking (markdown pipeline, with images).

pub mod extract;

use crate::chunk::Chunk;
use crate::error::{ChunkError, Result};
use crate::formats::pipeline::{self, Loaded};
use crate::options::ChunkOptions;
use extract::{extract, to_markdown as nb_to_markdown};

fn ensure_ipynb(file_path: &str) -> Result<()> {
    if file_path.to_ascii_lowercase().ends_with(".ipynb") {
        Ok(())
    } else {
        Err(ChunkError::InvalidArg(format!(
            "Expected .ipynb file path, got: {file_path}"
        )))
    }
}

fn load(file_path: &str) -> Result<Loaded> {
    ensure_ipynb(file_path)?;
    let bytes = std::fs::read(file_path).map_err(ChunkError::Io)?;
    load_bytes(&bytes)
}

/// No-filesystem entry (wasm/browser).
pub fn chunk_from_bytes(data: &[u8], mode: &str, window_size: usize, overlap: usize, sentences_per_chunk: usize, paragraphs_per_page: usize) -> Result<Vec<Chunk>> {
    swallow_empty(pipeline::chunk(&load_bytes(data)?, mode, window_size, overlap, sentences_per_chunk, paragraphs_per_page))
}

pub fn to_markdown_from_bytes(data: &[u8]) -> Result<String> {
    Ok(load_bytes(data)?.markdown)
}

fn load_bytes(bytes: &[u8]) -> Result<Loaded> {
    let doc = extract(bytes).map_err(ChunkError::Parse)?;
    let markdown = nb_to_markdown(&doc);
    let metadata = serde_json::json!({
        "source_type": "ipynb",
        "language": doc.language,
        "kernel": doc.kernel,
        "nbformat": doc.nbformat,
        "cell_count": doc.cell_count,
        "code_cell_count": doc.code_cell_count,
        "markdown_cell_count": doc.markdown_cell_count,
    });
    Ok(Loaded { markdown, images: doc.images, metadata })
}

pub fn chunk(
    file_path: &str,
    mode: &str,
    window_size: usize,
    overlap: usize,
    sentences_per_chunk: usize,
    paragraphs_per_page: usize,
) -> Result<Vec<Chunk>> {
    swallow_empty(pipeline::chunk(&load(file_path)?, mode, window_size, overlap, sentences_per_chunk, paragraphs_per_page))
}

/// A notebook whose only content is images (or is otherwise text-empty) yields
/// no markdown chunks. Match the reference engine: return zero chunks rather
/// than surfacing the "empty"/"No chunks" parse error.
fn swallow_empty(res: Result<Vec<Chunk>>) -> Result<Vec<Chunk>> {
    match res {
        Err(ChunkError::Parse(e)) if e.contains("empty") || e.contains("No chunks") => Ok(Vec::new()),
        other => other,
    }
}

pub fn chunk_with_options(file_path: &str, opts: &ChunkOptions) -> Result<Vec<Chunk>> {
    swallow_empty(pipeline::chunk_opts(&load(file_path)?, opts))
}

pub fn chunk_with_images(
    file_path: &str,
    mode: &str,
    window_size: usize,
    overlap: usize,
    sentences_per_chunk: usize,
    paragraphs_per_page: usize,
) -> Result<(Vec<Chunk>, Vec<(String, Vec<u8>)>)> {
    let loaded = load(file_path)?;
    let mut chunks: Vec<Chunk> = loaded
        .images
        .iter()
        .map(|(name, _)| Chunk::new(name.clone(), "image", serde_json::json!({ "image_name": name })))
        .collect();
    let text = swallow_empty(pipeline::chunk(
        &loaded, mode, window_size, overlap, sentences_per_chunk, paragraphs_per_page,
    ))?;
    chunks.extend(text);
    Ok((chunks, loaded.images.clone()))
}

pub fn to_markdown(file_path: &str) -> Result<String> {
    Ok(load(file_path)?.markdown)
}

pub fn to_markdown_with_images(file_path: &str) -> Result<(String, Vec<(String, Vec<u8>)>)> {
    let l = load(file_path)?;
    Ok((l.markdown, crate::formats::pipeline::dedup_images(l.images)))
}

/// No-filesystem `chunk_with_images` (wasm/browser).
pub fn chunk_with_images_from_bytes(data: &[u8], mode: &str, window_size: usize, overlap: usize, sentences_per_chunk: usize, paragraphs_per_page: usize) -> Result<(Vec<Chunk>, Vec<(String, Vec<u8>)>)> {
    let loaded = load_bytes(data)?;
    let mut chunks: Vec<Chunk> = loaded.images.iter().map(|(name, _)| Chunk::new(name.clone(), "image", serde_json::json!({ "image_name": name }))).collect();
    let text = swallow_empty(pipeline::chunk(&loaded, mode, window_size, overlap, sentences_per_chunk, paragraphs_per_page))?;
    chunks.extend(text);
    Ok((chunks, loaded.images.clone()))
}

pub fn to_markdown_with_images_from_bytes(data: &[u8]) -> Result<(String, Vec<(String, Vec<u8>)>)> {
    let l = load_bytes(data)?;
    Ok((l.markdown, crate::formats::pipeline::dedup_images(l.images)))
}


pub fn stream(
    file_path: &str,
    mode: &str,
    window_size: usize,
    overlap: usize,
    sentences_per_chunk: usize,
    paragraphs_per_page: usize,
) -> Result<impl Iterator<Item = Result<Chunk>>> {
    Ok(chunk(file_path, mode, window_size, overlap, sentences_per_chunk, paragraphs_per_page)?
        .into_iter()
        .map(Ok))
}