weirwood 0.3.2

Privacy-preserving XGBoost inference via Fully Homomorphic Encryption
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

//! UBJSON parser for XGBoost `.ubj` model files.
//!
//! XGBoost follows the UBJSON specification, which mandates **big-endian** byte
//! order for all multi-byte numeric types. This parser implements that format.
//!
//! The output is a `serde_json::Value`, which is then deserialized through the
//! same `RawModel` path used by the JSON loader.

use serde_json::{Map, Number, Value};

use crate::Error;

/// Parse XGBoost UBJ bytes into a `serde_json::Value`.
pub(crate) fn parse(data: &[u8]) -> Result<Value, Error> {
    let mut pos: usize = 0usize;
    let value: Value = parse_value(data, &mut pos)?;
    Ok(value)
}

// ---------------------------------------------------------------------------
// Primitive readers (big-endian, per the UBJSON specification)
// ---------------------------------------------------------------------------

fn read_byte(data: &[u8], pos: &mut usize) -> Result<u8, Error> {
    if *pos >= data.len() {
        return Err(Error::Format("unexpected end of UBJ data".into()));
    }
    let b: u8 = data[*pos];
    *pos += 1;
    Ok(b)
}

fn read_n<const N: usize>(data: &[u8], pos: &mut usize) -> Result<[u8; N], Error> {
    if *pos + N > data.len() {
        return Err(Error::Format("unexpected end of UBJ data".into()));
    }
    let arr: [u8; N] = data[*pos..*pos + N].try_into().unwrap();
    *pos += N;
    Ok(arr)
}

fn read_i8(data: &[u8], pos: &mut usize) -> Result<i8, Error> {
    Ok(read_byte(data, pos)? as i8)
}

fn read_u8(data: &[u8], pos: &mut usize) -> Result<u8, Error> {
    read_byte(data, pos)
}

fn read_i16(data: &[u8], pos: &mut usize) -> Result<i16, Error> {
    Ok(i16::from_be_bytes(read_n::<2>(data, pos)?))
}

fn read_i32(data: &[u8], pos: &mut usize) -> Result<i32, Error> {
    Ok(i32::from_be_bytes(read_n::<4>(data, pos)?))
}

fn read_i64(data: &[u8], pos: &mut usize) -> Result<i64, Error> {
    Ok(i64::from_be_bytes(read_n::<8>(data, pos)?))
}

fn read_f32(data: &[u8], pos: &mut usize) -> Result<f32, Error> {
    Ok(f32::from_be_bytes(read_n::<4>(data, pos)?))
}

fn read_f64(data: &[u8], pos: &mut usize) -> Result<f64, Error> {
    Ok(f64::from_be_bytes(read_n::<8>(data, pos)?))
}

/// Read a UBJSON string: a typed-integer length followed by UTF-8 bytes.
/// Object keys use this format (without the preceding `S` marker).
fn read_string(data: &[u8], pos: &mut usize) -> Result<String, Error> {
    let len: usize = read_count(data, pos)?;
    if *pos + len > data.len() {
        return Err(Error::Format("UBJ string truncated".into()));
    }
    let s: String = std::str::from_utf8(&data[*pos..*pos + len])
        .map_err(|_| Error::Format("UBJ string is not valid UTF-8".into()))?
        .to_owned();
    *pos += len;
    Ok(s)
}

/// Read any UBJSON integer type and return it as a `usize` (used for counts
/// and string lengths — must be non-negative).
fn read_count(data: &[u8], pos: &mut usize) -> Result<usize, Error> {
    let marker: u8 = read_byte(data, pos)?;
    let n: i64 = match marker {
        b'i' => read_i8(data, pos)? as i64,
        b'U' => read_u8(data, pos)? as i64,
        b'I' => read_i16(data, pos)? as i64,
        b'l' => read_i32(data, pos)? as i64,
        b'L' => read_i64(data, pos)?,
        m => {
            return Err(Error::Format(format!(
                "expected integer type for count, got {m:#04x}"
            )));
        }
    };
    if n < 0 {
        return Err(Error::Format(format!("negative UBJ count: {n}")));
    }
    Ok(n as usize)
}

// ---------------------------------------------------------------------------
// Value constructors
// ---------------------------------------------------------------------------

fn int_value(n: i64) -> Value {
    Value::Number(Number::from(n))
}

fn uint_value(n: u64) -> Value {
    Value::Number(Number::from(n))
}

fn float_value(f: f64) -> Result<Value, Error> {
    Number::from_f64(f)
        .map(Value::Number)
        .ok_or_else(|| Error::Format(format!("non-finite float {f} in UBJ")))
}

// ---------------------------------------------------------------------------
// Core parser
// ---------------------------------------------------------------------------

fn parse_value(data: &[u8], pos: &mut usize) -> Result<Value, Error> {
    let marker: u8 = read_byte(data, pos)?;
    match marker {
        b'N' => parse_value(data, pos), // no-op: skip and read next
        b'Z' => Ok(Value::Null),
        b'T' => Ok(Value::Bool(true)),
        b'F' => Ok(Value::Bool(false)),
        b'i' => Ok(int_value(read_i8(data, pos)? as i64)),
        b'U' => Ok(uint_value(read_u8(data, pos)? as u64)),
        b'I' => Ok(int_value(read_i16(data, pos)? as i64)),
        b'l' => Ok(int_value(read_i32(data, pos)? as i64)),
        b'L' => Ok(int_value(read_i64(data, pos)?)),
        b'd' => float_value(read_f32(data, pos)? as f64),
        b'D' => float_value(read_f64(data, pos)?),
        b'C' => Ok(Value::String((read_byte(data, pos)? as char).to_string())),
        b'S' => Ok(Value::String(read_string(data, pos)?)),
        b'H' => Ok(Value::String(read_string(data, pos)?)), // high-precision: keep as string
        b'[' => parse_array(data, pos),
        b'{' => parse_object(data, pos),
        m => Err(Error::Format(format!("unknown UBJ type marker {m:#04x}"))),
    }
}

/// Read a single value whose type is already known (used in typed containers).
fn read_typed(data: &[u8], pos: &mut usize, type_marker: u8) -> Result<Value, Error> {
    match type_marker {
        b'Z' => Ok(Value::Null),
        b'T' => Ok(Value::Bool(true)),
        b'F' => Ok(Value::Bool(false)),
        b'i' => Ok(int_value(read_i8(data, pos)? as i64)),
        b'U' => Ok(uint_value(read_u8(data, pos)? as u64)),
        b'I' => Ok(int_value(read_i16(data, pos)? as i64)),
        b'l' => Ok(int_value(read_i32(data, pos)? as i64)),
        b'L' => Ok(int_value(read_i64(data, pos)?)),
        b'd' => float_value(read_f32(data, pos)? as f64),
        b'D' => float_value(read_f64(data, pos)?),
        b'C' => Ok(Value::String((read_byte(data, pos)? as char).to_string())),
        b'S' => Ok(Value::String(read_string(data, pos)?)),
        m => Err(Error::Format(format!(
            "unsupported typed container element: {m:#04x}"
        ))),
    }
}

fn parse_array(data: &[u8], pos: &mut usize) -> Result<Value, Error> {
    if *pos >= data.len() {
        return Err(Error::Format("truncated UBJ array".into()));
    }

    // Optimized typed array: `[` `$` type `#` count  values...
    if data[*pos] == b'$' {
        *pos += 1;
        let type_marker = read_byte(data, pos)?;
        match read_byte(data, pos)? {
            b'#' => {}
            m => {
                return Err(Error::Format(format!(
                    "expected '#' after '$' in array, got {m:#04x}"
                )));
            }
        }
        let count = read_count(data, pos)?;
        let arr = (0..count)
            .map(|_| read_typed(data, pos, type_marker))
            .collect::<Result<Vec<_>, _>>()?;
        return Ok(Value::Array(arr));
    }

    // Counted untyped array: `[` `#` count  values...
    if data[*pos] == b'#' {
        *pos += 1;
        let count: usize = read_count(data, pos)?;
        let arr: Vec<Value> = (0..count)
            .map(|_| parse_value(data, pos))
            .collect::<Result<Vec<_>, _>>()?;
        return Ok(Value::Array(arr));
    }

    // Standard terminated array: `[` values... `]`
    let mut arr: Vec<Value> = Vec::new();
    loop {
        if *pos >= data.len() {
            return Err(Error::Format("unterminated UBJ array".into()));
        }
        if data[*pos] == b']' {
            *pos += 1;
            break;
        }
        arr.push(parse_value(data, pos)?);
    }
    Ok(Value::Array(arr))
}

fn parse_object(data: &[u8], pos: &mut usize) -> Result<Value, Error> {
    if *pos >= data.len() {
        return Err(Error::Format("truncated UBJ object".into()));
    }

    // Optimized typed object: `{` `$` type `#` count  (key value)...
    if data[*pos] == b'$' {
        *pos += 1;
        let type_marker: u8 = read_byte(data, pos)?;
        match read_byte(data, pos)? {
            b'#' => {}
            m => {
                return Err(Error::Format(format!(
                    "expected '#' after '$' in object, got {m:#04x}"
                )));
            }
        }
        let count: usize = read_count(data, pos)?;
        let mut map: Map<String, Value> = Map::with_capacity(count);
        for _ in 0..count {
            let key = read_string(data, pos)?;
            let value = read_typed(data, pos, type_marker)?;
            map.insert(key, value);
        }
        return Ok(Value::Object(map));
    }

    // Counted untyped object: `{` `#` count  (key value)...
    if data[*pos] == b'#' {
        *pos += 1;
        let count: usize = read_count(data, pos)?;
        let mut map: Map<String, Value> = Map::with_capacity(count);
        for _ in 0..count {
            let key = read_string(data, pos)?;
            let value = parse_value(data, pos)?;
            map.insert(key, value);
        }
        return Ok(Value::Object(map));
    }

    // Standard terminated object: `{` (key value)... `}`
    let mut map: Map<String, Value> = Map::new();
    loop {
        if *pos >= data.len() {
            return Err(Error::Format("unterminated UBJ object".into()));
        }
        if data[*pos] == b'}' {
            *pos += 1;
            break;
        }
        // Object keys are strings without the leading `S` marker.
        let key = read_string(data, pos)?;
        let value = parse_value(data, pos)?;
        map.insert(key, value);
    }
    Ok(Value::Object(map))
}

// ---------------------------------------------------------------------------
// Tests
// ---------------------------------------------------------------------------

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

    /// Hand-encode a minimal UBJSON object and verify the parser round-trips it.
    #[test]
    fn parse_simple_object() {
        // Encodes: {"a": 1, "b": true}  (little-endian, terminated style)
        let data: &[u8] = &[
            b'{', // key "a" (length i8=1, then 'a')
            b'i', 1, b'a', // value: int8 42
            b'i', 42, // key "b"
            b'i', 1, b'b', // value: true
            b'T', b'}',
        ];
        let v: Value = parse(data).unwrap();
        assert_eq!(v["a"], Value::Number(42.into()));
        assert_eq!(v["b"], Value::Bool(true));
    }

    #[test]
    fn parse_typed_int32_array() {
        // Encodes: [1, -1, 2]  as an optimized i32 array (big-endian)
        let mut data: Vec<u8> = vec![b'[', b'$', b'l', b'#', b'i', 3];
        data.extend_from_slice(&1_i32.to_be_bytes());
        data.extend_from_slice(&(-1_i32).to_be_bytes());
        data.extend_from_slice(&2_i32.to_be_bytes());

        let v = parse(&data).unwrap();
        let arr = v.as_array().unwrap();
        assert_eq!(arr[0], Value::Number(1.into()));
        assert_eq!(arr[1], Value::Number((-1_i64).into()));
        assert_eq!(arr[2], Value::Number(2.into()));
    }

    #[test]
    fn parse_typed_f32_array() {
        // Encodes: [0.5, -0.5]  as an optimized f32 array (big-endian)
        let mut data: Vec<u8> = vec![b'[', b'$', b'd', b'#', b'i', 2];
        data.extend_from_slice(&0.5_f32.to_be_bytes());
        data.extend_from_slice(&(-0.5_f32).to_be_bytes());

        let v: Value = parse(&data).unwrap();
        let arr: &Vec<Value> = v.as_array().unwrap();
        let a: f32 = arr[0].as_f64().unwrap() as f32;
        let b: f32 = arr[1].as_f64().unwrap() as f32;
        approx::assert_abs_diff_eq!(a, 0.5_f32, epsilon = 1e-7);
        approx::assert_abs_diff_eq!(b, -0.5_f32, epsilon = 1e-7);
    }

    #[test]
    fn parse_nested_string_values() {
        // Encodes: {"name": "hello"}
        let data: &[u8] = &[
            b'{', b'i', 4, b'n', b'a', b'm', b'e', b'S', b'i', 5, b'h', b'e', b'l', b'l', b'o',
            b'}',
        ];
        let v = parse(data).unwrap();
        assert_eq!(v["name"], Value::String("hello".into()));
    }
}