tensogram 0.15.0

Fast binary N-tensor message format for scientific data — encode, decode, file I/O, streaming
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
// (C) Copyright 2026- ECMWF and individual contributors.
//
// This software is licensed under the terms of the Apache Licence Version 2.0
// which can be obtained at http://www.apache.org/licenses/LICENSE-2.0.
// In applying this licence, ECMWF does not waive the privileges and immunities
// granted to it by virtue of its status as an intergovernmental organisation nor
// does it submit to any jurisdiction.

//! Level 4: Fidelity validation — full decode, size check, NaN/Inf detection.

use crate::dtype::Dtype;
use crate::encode::build_pipeline_config;
use tensogram_encodings::ByteOrder;

use super::types::*;

/// Run fidelity checks on all objects.
///
/// For each object:
/// 1. Ensure decoded bytes exist (reuse Level 3 cache or decode raw payload).
/// 2. Verify decoded size matches shape * dtype byte width.
/// 3. Scan float types for NaN/Inf values (errors, not warnings).
pub(crate) fn validate_fidelity(
    objects: &mut [ObjectContext<'_>],
    issues: &mut Vec<ValidationIssue>,
) {
    for (i, obj) in objects.iter_mut().enumerate() {
        let desc = match obj.descriptor.as_ref() {
            Some(d) => d,
            None => continue, // Already reported at Level 2
        };

        // Get decoded bytes: reuse Level 3 cache, scan raw payload in-place, or decode now
        let decoded: &[u8] = match &obj.decode_state {
            DecodeState::Decoded(bytes) => bytes,
            DecodeState::DecodeFailed => continue, // Already reported at Level 3
            DecodeState::NotDecoded => {
                if desc.encoding == "none" && desc.filter == "none" && desc.compression == "none" {
                    // Raw object: payload is the decoded bytes
                    obj.payload
                } else {
                    // Non-raw object — Level 3 didn't run, decode now.
                    // Currently unreachable via validate_message (Fidelity implies
                    // Integrity which decodes all non-raw objects), but kept as a
                    // safety net for direct API callers.
                    let shape_product = desc
                        .shape
                        .iter()
                        .try_fold(1u64, |acc, &x| acc.checked_mul(x));
                    let num_elements = match shape_product.and_then(|p| usize::try_from(p).ok()) {
                        Some(n) => n,
                        None => {
                            issues.push(err(
                                IssueCode::DecodeObjectFailed,
                                ValidationLevel::Fidelity,
                                Some(i),
                                Some(obj.frame_offset),
                                "cannot compute element count from shape".to_string(),
                            ));
                            continue;
                        }
                    };
                    match build_pipeline_config(desc, num_elements, desc.dtype) {
                        Ok(config) => {
                            match tensogram_encodings::pipeline::decode_pipeline(
                                obj.payload,
                                &config,
                                false,
                            ) {
                                Ok(decoded_bytes) => {
                                    obj.decode_state = DecodeState::Decoded(decoded_bytes);
                                    // Safe: just set to Decoded above
                                    let DecodeState::Decoded(b) = &obj.decode_state else {
                                        continue;
                                    };
                                    b
                                }
                                Err(e) => {
                                    obj.decode_state = DecodeState::DecodeFailed;
                                    issues.push(err(
                                        IssueCode::DecodeObjectFailed,
                                        ValidationLevel::Fidelity,
                                        Some(i),
                                        Some(obj.frame_offset),
                                        format!("full decode failed: {e}"),
                                    ));
                                    continue;
                                }
                            }
                        }
                        Err(e) => {
                            obj.decode_state = DecodeState::DecodeFailed;
                            issues.push(err(
                                IssueCode::DecodeObjectFailed,
                                ValidationLevel::Fidelity,
                                Some(i),
                                Some(obj.frame_offset),
                                format!("cannot build pipeline config: {e}"),
                            ));
                            continue;
                        }
                    }
                }
            }
        };

        // Decoded-size check (unconditional, including size 0)
        let shape_product = desc
            .shape
            .iter()
            .try_fold(1u64, |acc, &x| acc.checked_mul(x));
        let expected_size = match shape_product {
            Some(product) => {
                if desc.dtype == Dtype::Bitmask {
                    usize::try_from(product.div_ceil(8)).ok()
                } else {
                    usize::try_from(product)
                        .ok()
                        .and_then(|p| p.checked_mul(desc.dtype.byte_width()))
                }
            }
            None => {
                issues.push(err(
                    IssueCode::DecodedSizeMismatch,
                    ValidationLevel::Fidelity,
                    Some(i),
                    Some(obj.frame_offset),
                    "shape product overflows, cannot verify decoded size".to_string(),
                ));
                continue;
            }
        };
        let expected_size = match expected_size {
            Some(s) => s,
            None => {
                issues.push(err(
                    IssueCode::DecodedSizeMismatch,
                    ValidationLevel::Fidelity,
                    Some(i),
                    Some(obj.frame_offset),
                    "expected decoded size overflows usize".to_string(),
                ));
                continue;
            }
        };

        if decoded.len() != expected_size {
            issues.push(err(
                IssueCode::DecodedSizeMismatch,
                ValidationLevel::Fidelity,
                Some(i),
                Some(obj.frame_offset),
                format!(
                    "decoded size {} != expected {} (shape {:?}, dtype {})",
                    decoded.len(),
                    expected_size,
                    desc.shape,
                    desc.dtype
                ),
            ));
            continue;
        }

        // NaN/Inf scan for float types
        let byte_order = desc.byte_order;
        match desc.dtype {
            Dtype::Float32 => scan_f32(decoded, byte_order, i, obj.frame_offset, issues),
            Dtype::Float64 => scan_f64(decoded, byte_order, i, obj.frame_offset, issues),
            Dtype::Float16 => scan_f16(decoded, byte_order, i, obj.frame_offset, issues),
            Dtype::Bfloat16 => scan_bf16(decoded, byte_order, i, obj.frame_offset, issues),
            Dtype::Complex64 => scan_complex64(decoded, byte_order, i, obj.frame_offset, issues),
            Dtype::Complex128 => scan_complex128(decoded, byte_order, i, obj.frame_offset, issues),
            _ => {} // Integer/bitmask types: no NaN/Inf possible
        }
    }
}

// ── Float scanners ──────────────────────────────────────────────────────────

fn scan_f32(
    data: &[u8],
    byte_order: ByteOrder,
    obj_idx: usize,
    frame_offset: usize,
    issues: &mut Vec<ValidationIssue>,
) {
    let chunks = data.chunks_exact(4);
    if !chunks.remainder().is_empty() {
        issues.push(err(
            IssueCode::DecodedSizeMismatch,
            ValidationLevel::Fidelity,
            Some(obj_idx),
            Some(frame_offset),
            format!(
                "decoded size {} is not a multiple of 4 (float32)",
                data.len()
            ),
        ));
        return;
    }
    for (idx, chunk) in chunks.enumerate() {
        // chunks_exact(4) guarantees each chunk has length 4. The slice
        // pattern formalises the invariant and keeps this function panic-free;
        // the `else` branch is unreachable under a correct std implementation
        // but lets us skip gracefully instead of panicking if that ever
        // changes.
        let &[b0, b1, b2, b3] = chunk else {
            continue;
        };
        let bytes = [b0, b1, b2, b3];
        let val = match byte_order {
            ByteOrder::Big => f32::from_be_bytes(bytes),
            ByteOrder::Little => f32::from_le_bytes(bytes),
        };
        if val.is_nan() {
            issues.push(err(
                IssueCode::NanDetected,
                ValidationLevel::Fidelity,
                Some(obj_idx),
                Some(frame_offset),
                format!("NaN at element {idx}"),
            ));
            return; // Report first occurrence only
        }
        if val.is_infinite() {
            issues.push(err(
                IssueCode::InfDetected,
                ValidationLevel::Fidelity,
                Some(obj_idx),
                Some(frame_offset),
                format!("Inf at element {idx}"),
            ));
            return;
        }
    }
}

fn scan_f64(
    data: &[u8],
    byte_order: ByteOrder,
    obj_idx: usize,
    frame_offset: usize,
    issues: &mut Vec<ValidationIssue>,
) {
    let chunks = data.chunks_exact(8);
    if !chunks.remainder().is_empty() {
        issues.push(err(
            IssueCode::DecodedSizeMismatch,
            ValidationLevel::Fidelity,
            Some(obj_idx),
            Some(frame_offset),
            format!(
                "decoded size {} is not a multiple of 8 (float64)",
                data.len()
            ),
        ));
        return;
    }
    for (idx, chunk) in chunks.enumerate() {
        // See scan_f32 for the rationale of this slice-pattern idiom.
        let &[b0, b1, b2, b3, b4, b5, b6, b7] = chunk else {
            continue;
        };
        let bytes = [b0, b1, b2, b3, b4, b5, b6, b7];
        let val = match byte_order {
            ByteOrder::Big => f64::from_be_bytes(bytes),
            ByteOrder::Little => f64::from_le_bytes(bytes),
        };
        if val.is_nan() {
            issues.push(err(
                IssueCode::NanDetected,
                ValidationLevel::Fidelity,
                Some(obj_idx),
                Some(frame_offset),
                format!("NaN at element {idx}"),
            ));
            return;
        }
        if val.is_infinite() {
            issues.push(err(
                IssueCode::InfDetected,
                ValidationLevel::Fidelity,
                Some(obj_idx),
                Some(frame_offset),
                format!("Inf at element {idx}"),
            ));
            return;
        }
    }
}

fn scan_f16(
    data: &[u8],
    byte_order: ByteOrder,
    obj_idx: usize,
    frame_offset: usize,
    issues: &mut Vec<ValidationIssue>,
) {
    let chunks = data.chunks_exact(2);
    if !chunks.remainder().is_empty() {
        issues.push(err(
            IssueCode::DecodedSizeMismatch,
            ValidationLevel::Fidelity,
            Some(obj_idx),
            Some(frame_offset),
            format!(
                "decoded size {} is not a multiple of 2 (float16)",
                data.len()
            ),
        ));
        return;
    }
    for (idx, chunk) in chunks.enumerate() {
        // See scan_f32 for the rationale of this slice-pattern idiom.
        let &[b0, b1] = chunk else {
            continue;
        };
        let bytes = [b0, b1];
        let bits = match byte_order {
            ByteOrder::Big => u16::from_be_bytes(bytes),
            ByteOrder::Little => u16::from_le_bytes(bytes),
        };
        // IEEE 754 half: sign(1) + exponent(5) + mantissa(10)
        let exp = (bits >> 10) & 0x1F;
        let mantissa = bits & 0x03FF;
        if exp == 0x1F {
            if mantissa != 0 {
                issues.push(err(
                    IssueCode::NanDetected,
                    ValidationLevel::Fidelity,
                    Some(obj_idx),
                    Some(frame_offset),
                    format!("NaN at element {idx}"),
                ));
                return;
            } else {
                issues.push(err(
                    IssueCode::InfDetected,
                    ValidationLevel::Fidelity,
                    Some(obj_idx),
                    Some(frame_offset),
                    format!("Inf at element {idx}"),
                ));
                return;
            }
        }
    }
}

fn scan_bf16(
    data: &[u8],
    byte_order: ByteOrder,
    obj_idx: usize,
    frame_offset: usize,
    issues: &mut Vec<ValidationIssue>,
) {
    let chunks = data.chunks_exact(2);
    if !chunks.remainder().is_empty() {
        issues.push(err(
            IssueCode::DecodedSizeMismatch,
            ValidationLevel::Fidelity,
            Some(obj_idx),
            Some(frame_offset),
            format!(
                "decoded size {} is not a multiple of 2 (bfloat16)",
                data.len()
            ),
        ));
        return;
    }
    for (idx, chunk) in chunks.enumerate() {
        // See scan_f32 for the rationale of this slice-pattern idiom.
        let &[b0, b1] = chunk else {
            continue;
        };
        let bytes = [b0, b1];
        let bits = match byte_order {
            ByteOrder::Big => u16::from_be_bytes(bytes),
            ByteOrder::Little => u16::from_le_bytes(bytes),
        };
        // BFloat16: sign(1) + exponent(8) + mantissa(7)
        let exp = (bits >> 7) & 0xFF;
        let mantissa = bits & 0x7F;
        if exp == 0xFF {
            if mantissa != 0 {
                issues.push(err(
                    IssueCode::NanDetected,
                    ValidationLevel::Fidelity,
                    Some(obj_idx),
                    Some(frame_offset),
                    format!("NaN at element {idx}"),
                ));
                return;
            } else {
                issues.push(err(
                    IssueCode::InfDetected,
                    ValidationLevel::Fidelity,
                    Some(obj_idx),
                    Some(frame_offset),
                    format!("Inf at element {idx}"),
                ));
                return;
            }
        }
    }
}

fn scan_complex64(
    data: &[u8],
    byte_order: ByteOrder,
    obj_idx: usize,
    frame_offset: usize,
    issues: &mut Vec<ValidationIssue>,
) {
    // Complex64 = two f32 (real, imaginary) = 8 bytes per element
    let chunks = data.chunks_exact(8);
    if !chunks.remainder().is_empty() {
        issues.push(err(
            IssueCode::DecodedSizeMismatch,
            ValidationLevel::Fidelity,
            Some(obj_idx),
            Some(frame_offset),
            format!(
                "decoded size {} is not a multiple of 8 (complex64)",
                data.len()
            ),
        ));
        return;
    }
    for (idx, chunk) in chunks.enumerate() {
        // See scan_f32 for the rationale of this slice-pattern idiom.
        let &[r0, r1, r2, r3, i0, i1, i2, i3] = chunk else {
            continue;
        };
        let real_bytes = [r0, r1, r2, r3];
        let imag_bytes = [i0, i1, i2, i3];
        let (real, imag) = match byte_order {
            ByteOrder::Big => (
                f32::from_be_bytes(real_bytes),
                f32::from_be_bytes(imag_bytes),
            ),
            ByteOrder::Little => (
                f32::from_le_bytes(real_bytes),
                f32::from_le_bytes(imag_bytes),
            ),
        };
        for (val, component) in [(real, "real"), (imag, "imaginary")] {
            if val.is_nan() {
                issues.push(err(
                    IssueCode::NanDetected,
                    ValidationLevel::Fidelity,
                    Some(obj_idx),
                    Some(frame_offset),
                    format!("NaN at element {idx} ({component} component)"),
                ));
                return;
            }
            if val.is_infinite() {
                issues.push(err(
                    IssueCode::InfDetected,
                    ValidationLevel::Fidelity,
                    Some(obj_idx),
                    Some(frame_offset),
                    format!("Inf at element {idx} ({component} component)"),
                ));
                return;
            }
        }
    }
}

fn scan_complex128(
    data: &[u8],
    byte_order: ByteOrder,
    obj_idx: usize,
    frame_offset: usize,
    issues: &mut Vec<ValidationIssue>,
) {
    // Complex128 = two f64 (real, imaginary) = 16 bytes per element
    let chunks = data.chunks_exact(16);
    if !chunks.remainder().is_empty() {
        issues.push(err(
            IssueCode::DecodedSizeMismatch,
            ValidationLevel::Fidelity,
            Some(obj_idx),
            Some(frame_offset),
            format!(
                "decoded size {} is not a multiple of 16 (complex128)",
                data.len()
            ),
        ));
        return;
    }
    for (idx, chunk) in chunks.enumerate() {
        // See scan_f32 for the rationale of this slice-pattern idiom.
        let &[
            r0,
            r1,
            r2,
            r3,
            r4,
            r5,
            r6,
            r7,
            i0,
            i1,
            i2,
            i3,
            i4,
            i5,
            i6,
            i7,
        ] = chunk
        else {
            continue;
        };
        let real_bytes = [r0, r1, r2, r3, r4, r5, r6, r7];
        let imag_bytes = [i0, i1, i2, i3, i4, i5, i6, i7];
        let (real, imag) = match byte_order {
            ByteOrder::Big => (
                f64::from_be_bytes(real_bytes),
                f64::from_be_bytes(imag_bytes),
            ),
            ByteOrder::Little => (
                f64::from_le_bytes(real_bytes),
                f64::from_le_bytes(imag_bytes),
            ),
        };
        for (val, component) in [(real, "real"), (imag, "imaginary")] {
            if val.is_nan() {
                issues.push(err(
                    IssueCode::NanDetected,
                    ValidationLevel::Fidelity,
                    Some(obj_idx),
                    Some(frame_offset),
                    format!("NaN at element {idx} ({component} component)"),
                ));
                return;
            }
            if val.is_infinite() {
                issues.push(err(
                    IssueCode::InfDetected,
                    ValidationLevel::Fidelity,
                    Some(obj_idx),
                    Some(frame_offset),
                    format!("Inf at element {idx} ({component} component)"),
                ));
                return;
            }
        }
    }
}