xberg 1.1.1

High-performance document intelligence library for Rust. Extract text, metadata, and structured data from PDFs, Office documents, images, and 107 formats and 371 programming languages via tree-sitter code intelligence with async/sync APIs.
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

//! LLM-driven translation backend.
//!
//! Builds a Minijinja prompt per text segment (whole content, formatted
//! content, chunks) and calls
//! [`crate::llm::text_completion::complete_text`] for each. Empty segments are
//! skipped so we do not waste tokens on whitespace.

use crate::core::config::TranslationConfig;
use crate::types::translation::Translation;
use crate::types::{ExtractedDocument, LlmUsage};

/// Default Jinja2 template for LLM translation. Receives `target_lang`,
/// `source_lang` (may be `"auto"`), `preserve_markup`, and `text` variables.
pub const DEFAULT_TRANSLATION_TEMPLATE: &str = "\
You are a precise translation engine. Translate the text below {% if source_lang and source_lang != 'auto' %}from {{ source_lang }} {% endif %}\
into {{ target_lang }}.

Rules:
- Preserve the original meaning exactly.
- Do not add commentary, explanations, or surrounding quotes.
{% if preserve_markup %}- Preserve Markdown formatting (headings, lists, emphasis, links, code blocks) and HTML tags exactly as they appear.\
{% else %}- Return plain text only.{% endif %}
- If the text is already in {{ target_lang }}, return it unchanged.
- If the text is empty, return an empty string.

Text:
{{ text }}";

/// Render the prompt for a single text segment.
fn render_prompt(config: &TranslationConfig, text: &str, preserve_markup: bool) -> crate::Result<String> {
    let ctx = minijinja::context! {
        target_lang => &config.target_lang,
        source_lang => config.source_lang.as_deref().unwrap_or("auto"),
        preserve_markup => preserve_markup,
        text => text,
    };
    crate::llm::prompts::render_template(DEFAULT_TRANSLATION_TEMPLATE, &ctx)
}

/// Translate a single segment, collecting any usage entry produced.
///
/// `pub(super)` so the sibling [`super::fields`] module — which translates
/// every secondary text field (tables, pages, metadata, elements, document
/// structure) — can reuse it for fields that are naturally singular (page
/// content, table markdown) rather than duplicating this logic.
pub(super) async fn translate_segment(
    config: &TranslationConfig,
    text: &str,
    preserve_markup: bool,
    source_label: &str,
    usages: &mut Vec<LlmUsage>,
) -> crate::Result<String> {
    if text.trim().is_empty() {
        return Ok(text.to_string());
    }
    let prompt = render_prompt(config, text, preserve_markup)?;
    let (translated, usage) = crate::llm::text_completion::complete_text(&config.llm, &prompt, source_label).await?;
    if let Some(u) = usage {
        usages.push(u);
    }
    Ok(translated)
}

/// Translate the extraction result in place.
///
/// Populates `result.translation` with the translated `content`, optionally the
/// translated `formatted_content` (when `preserve_markup = true`), and rewrites
/// every chunk's `content` field. It also rewrites every other text-bearing
/// field — tables, pages, metadata, semantic elements, and the structured
/// document tree — via [`super::fields::translate_secondary_fields`]
/// (xberg-io/xberg#254). Every LLM call's usage is appended to
/// `result.llm_usage`.
pub async fn translate_result(result: &mut ExtractedDocument, config: &TranslationConfig) -> crate::Result<()> {
    if config.target_lang.trim().is_empty() {
        return Err(crate::XbergError::validation(
            "TranslationConfig.target_lang must not be empty",
        ));
    }

    let mut usages: Vec<LlmUsage> = Vec::new();

    let translated_content =
        translate_segment(config, &result.content, false, "translation_content", &mut usages).await?;

    let translated_formatted = if config.preserve_markup
        && let Some(formatted) = result.formatted_content.as_deref()
        && !formatted.trim().is_empty()
    {
        Some(translate_segment(config, formatted, true, "translation_formatted", &mut usages).await?)
    } else {
        None
    };

    if let Some(chunks) = result.chunks.as_mut() {
        for chunk in chunks.iter_mut() {
            let translated = translate_segment(config, &chunk.content, false, "translation_chunk", &mut usages).await?;
            chunk.content = translated;
        }
    }

    super::fields::translate_secondary_fields(result, config, &mut usages).await?;

    result.translation = Some(Translation {
        target_lang: config.target_lang.clone(),
        source_lang: config.source_lang.clone(),
        content: translated_content,
        formatted_content: translated_formatted,
    });

    if !usages.is_empty() {
        result.llm_usage.get_or_insert_with(Vec::new).extend(usages);
    }

    Ok(())
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::core::config::LlmConfig;

    fn cfg() -> TranslationConfig {
        TranslationConfig {
            target_lang: "de".to_string(),
            source_lang: None,
            preserve_markup: false,
            llm: LlmConfig {
                model: "openai/gpt-4o-mini".to_string(),
                ..Default::default()
            },
        }
    }

    #[test]
    fn render_prompt_includes_target_lang() {
        let prompt = render_prompt(&cfg(), "Hello world", false).unwrap();
        assert!(prompt.contains("de"));
        assert!(prompt.contains("Hello world"));
    }

    #[test]
    fn render_prompt_includes_source_lang_when_set() {
        let mut c = cfg();
        c.source_lang = Some("en".to_string());
        let prompt = render_prompt(&c, "Hello", false).unwrap();
        assert!(prompt.contains("from en"));
    }

    #[test]
    fn render_prompt_preserves_markup_clause_when_enabled() {
        let prompt = render_prompt(&cfg(), "**hi**", true).unwrap();
        assert!(prompt.contains("Markdown"));
    }

    /// Regression test for xberg-io/xberg#254: `translate_result` used to
    /// rewrite only `content`, `formatted_content`, and chunk content,
    /// leaving every other text-bearing field (tables, pages, metadata, the
    /// structured document tree) in the source language. This drives
    /// `translate_result` end-to-end against a table cell — a field
    /// `translate_result` never touched before the fix — through a local
    /// loopback HTTP stub (no real network call), and asserts both the exact
    /// translated value and that exactly one LLM call was made for the one
    /// non-empty secondary field present, which is the call-volume contract
    /// documented in `super::fields`.
    #[cfg(feature = "api")]
    #[tokio::test]
    async fn translate_result_translates_table_cells_via_secondary_fields() {
        use crate::types::Table;
        use std::sync::Arc;
        use std::sync::atomic::{AtomicUsize, Ordering};

        let call_count = Arc::new(AtomicUsize::new(0));
        let call_count_handler = call_count.clone();

        let app = axum::Router::new().fallback(axum::routing::post(move || {
            let call_count = call_count_handler.clone();
            async move {
                call_count.fetch_add(1, Ordering::SeqCst);
                axum::response::Json(serde_json::json!({
                    "id": "test",
                    "object": "chat.completion",
                    "created": 0,
                    "model": "test",
                    "choices": [{
                        "index": 0,
                        "message": { "role": "assistant", "content": "[\"CELDA TRADUCIDA\"]" },
                        "finish_reason": "stop"
                    }]
                }))
            }
        }));

        let listener = tokio::net::TcpListener::bind("127.0.0.1:0").await.unwrap();
        let addr = listener.local_addr().unwrap();
        tokio::spawn(async move {
            axum::serve(listener, app).await.unwrap();
        });

        let base_url = format!("http://{addr}/v1/");
        let config = TranslationConfig {
            target_lang: "es".to_string(),
            source_lang: None,
            preserve_markup: false,
            llm: LlmConfig {
                model: "openai/gpt-4o-mini".to_string(),
                api_key: Some("test-key".to_string()),
                base_url: Some(base_url),
                ..Default::default()
            },
        };

        // `content` is left empty so the only field this document has to
        // translate is the single table cell — isolating the assertion to
        // the field `translate_result` used to skip.
        let mut result = ExtractedDocument {
            content: String::new(),
            mime_type: std::borrow::Cow::Borrowed("text/plain"),
            tables: vec![Table {
                cells: vec![vec!["hola".to_string()]],
                markdown: String::new(),
                ..Table::default()
            }],
            ..Default::default()
        };

        translate_result(&mut result, &config).await.unwrap();

        assert_eq!(result.tables[0].cells[0][0], "CELDA TRADUCIDA");
        assert_eq!(
            call_count.load(Ordering::SeqCst),
            1,
            "expected exactly one batched LLM call for the one non-empty table cell, not one call per field"
        );
    }
}