dive-store 0.1.0

State fingerprints and disk-backed state storage for TLA+ model checking
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
//! Ordered frontier records with bounded resident encoded storage.
//!
//! This is a native transient format, not a TLC checkpoint format. It bounds
//! retained encoded frontier bytes; a reader and writer each also hold at most
//! one record. Decoded values, seen sets, candidates, traces and graphs are
//! outside this budget. Checksums detect accidental corruption, not tampering.

use std::fs::{self, File, OpenOptions};
use std::io::{Read, Seek, SeekFrom, Write};
use std::path::{Path, PathBuf};
use std::sync::atomic::{AtomicU64, Ordering};

const DATA_MAGIC: &[u8; 8] = b"DIVEFQ02";
const INDEX_MAGIC: &[u8; 8] = b"DIVEFI02";
const HEADER: usize = 32;
/// Maximum encoded record read into one reader or writer workspace: 64 MiB.
/// Larger states refuse spill mode, rather than allocate an unbounded buffer.
pub const MAX_RECORD_BYTES: usize = 64 * 1024 * 1024;
static NEXT: AtomicU64 = AtomicU64::new(0);

/// Explicit storage failure; never a successful incomplete frontier.
#[derive(Clone, Debug, PartialEq, Eq)]
pub struct SpillError {
    /// Stable storage error code.
    pub code: &'static str,
    /// Operation and underlying failure.
    pub message: String,
}

impl std::fmt::Display for SpillError {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        write!(f, "{}: {}", self.code, self.message)
    }
}
impl std::error::Error for SpillError {}

fn error(operation: &str, cause: impl std::fmt::Display) -> SpillError {
    SpillError {
        code: "DIVE-S0200",
        message: format!("frontier {operation}: {cause}"),
    }
}

/// Resident encoded bytes and cumulative bytes written to frontier files.
#[derive(Clone, Copy, Debug, Default, PartialEq, Eq)]
pub struct SpillStats {
    /// Resident data/index capacity. Excludes one-record codec workspaces.
    pub retained_bytes: u64,
    /// Confirmed complete writes including headers and index. A failed write
    /// can leave additional partial bytes; callers must flag incomplete accounting.
    pub spilled_bytes: u64,
}

#[derive(Debug)]
struct Disk {
    directory: PathBuf,
    data: File,
    index: File,
    bytes: u64,
}

/// A frontier built and read in insertion order, carrying concrete state IDs.
/// Files are isolated below the supplied parent. Call [`Self::close`] before
/// reporting completion: `Drop` cleanup is only a best-effort fallback.
#[derive(Debug)]
pub struct OrderedFrontier {
    limit: usize,
    parent: PathBuf,
    bytes: Vec<u8>,
    offsets: Vec<u64>,
    disk: Option<Disk>,
    len: usize,
    written: u64,
}

impl OrderedFrontier {
    /// Creates an empty queue without touching the filesystem.
    #[must_use]
    pub fn new(memory_bytes: usize, parent: &Path) -> Self {
        Self {
            limit: memory_bytes,
            parent: parent.into(),
            bytes: Vec::new(),
            offsets: Vec::new(),
            disk: None,
            len: 0,
            written: 0,
        }
    }

    /// Number of committed records.
    #[must_use]
    pub const fn len(&self) -> usize {
        self.len
    }

    /// Whether no record has been committed.
    #[must_use]
    pub const fn is_empty(&self) -> bool {
        self.len == 0
    }

    /// The current resident capacities and bytes written.
    #[must_use]
    pub fn stats(&self) -> SpillStats {
        SpillStats {
            retained_bytes: (self.bytes.capacity() + self.offsets.capacity() * 8) as u64,
            spilled_bytes: self.written,
        }
    }

    /// Appends an encoded state without dropping its concrete ID.
    ///
    /// # Errors
    /// Allocation, filesystem or record-size failure. `len` increases only
    /// after both the record and its index entry have been written.
    pub fn push(&mut self, id: u64, state: &[u8]) -> Result<(), SpillError> {
        if state.len() > MAX_RECORD_BYTES {
            return Err(error(
                "record size",
                format!("{} exceeds {MAX_RECORD_BYTES} bytes", state.len()),
            ));
        }
        let record = header(self.len as u64, id, state);
        if self.disk.is_none() {
            let needed = self
                .bytes
                .len()
                .checked_add(HEADER)
                .and_then(|n| n.checked_add(state.len()))
                .ok_or_else(|| error("length", "overflow"))?;
            let total = needed
                .checked_add((self.offsets.len() + 1).saturating_mul(8))
                .ok_or_else(|| error("length", "overflow"))?;
            if total <= self.limit {
                self.bytes
                    .try_reserve_exact(needed - self.bytes.len())
                    .map_err(|e| error("allocation", e))?;
                self.offsets
                    .try_reserve_exact(1)
                    .map_err(|e| error("allocation", e))?;
            }
            if total > self.limit || self.stats().retained_bytes > self.limit as u64 {
                self.start_disk()?;
            }
        }
        if let Some(disk) = &mut self.disk {
            let offset = disk.bytes;
            disk.data
                .write_all(&record)
                .and_then(|()| disk.data.write_all(state))
                .map_err(|e| error("write record", e))?;
            disk.index
                .write_all(&offset.to_le_bytes())
                .map_err(|e| error("write index", e))?;
            disk.bytes += (HEADER + state.len()) as u64;
            self.written += (HEADER + state.len() + 8) as u64;
        } else {
            self.offsets.push(self.bytes.len() as u64);
            self.bytes.extend_from_slice(&record);
            self.bytes.extend_from_slice(state);
        }
        self.len += 1;
        Ok(())
    }

    fn start_disk(&mut self) -> Result<(), SpillError> {
        let serial = NEXT.fetch_add(1, Ordering::Relaxed);
        let directory = self
            .parent
            .join(format!("dive-frontier-{}-{serial}", std::process::id()));
        fs::create_dir(&directory).map_err(|e| error("create directory", e))?;
        let created = (|| {
            let mut data = OpenOptions::new()
                .read(true)
                .write(true)
                .create_new(true)
                .open(directory.join("states"))?;
            let mut index = OpenOptions::new()
                .read(true)
                .write(true)
                .create_new(true)
                .open(directory.join("index"))?;
            data.write_all(DATA_MAGIC)?;
            index.write_all(INDEX_MAGIC)?;
            data.write_all(&self.bytes)?;
            for &offset in &self.offsets {
                index.write_all(&(offset + 8).to_le_bytes())?;
            }
            Ok::<_, std::io::Error>(Disk {
                directory: directory.clone(),
                data,
                index,
                bytes: 8 + self.bytes.len() as u64,
            })
        })();
        match created {
            Ok(disk) => {
                self.written += 16 + self.bytes.len() as u64 + self.offsets.len() as u64 * 8;
                self.disk = Some(disk);
                self.bytes = Vec::new();
                self.offsets = Vec::new();
                Ok(())
            }
            Err(e) => {
                let cleanup = fs::remove_dir_all(&directory);
                Err(error(
                    "initialize files",
                    match cleanup {
                        Ok(()) => e.to_string(),
                        Err(c) => format!("{e}; cleanup also failed: {c}"),
                    },
                ))
            }
        }
    }

    /// Reads exactly one record with bounded buffering. File handles are
    /// opened separately for each call: seeks are never shared across workers.
    ///
    /// # Errors
    /// Missing files, invalid version/index/length/checksum, or read failure.
    pub fn read(&self, position: usize) -> Result<(u64, Vec<u8>), SpillError> {
        if position >= self.len {
            return Err(error("index", "position outside frontier"));
        }
        if let Some(disk) = &self.disk {
            let mut index =
                File::open(disk.directory.join("index")).map_err(|e| error("open index", e))?;
            let mut data =
                File::open(disk.directory.join("states")).map_err(|e| error("open record", e))?;
            check_magic(&mut index, *INDEX_MAGIC)?;
            check_magic(&mut data, *DATA_MAGIC)?;
            index
                .seek(SeekFrom::Start(8 + position as u64 * 8))
                .map_err(|e| error("seek index", e))?;
            let mut offset = [0; 8];
            index
                .read_exact(&mut offset)
                .map_err(|e| error("read index", e))?;
            let offset = u64::from_le_bytes(offset);
            if offset < 8 {
                return Err(error("index", "record offset overlaps header"));
            }
            data.seek(SeekFrom::Start(offset))
                .map_err(|e| error("seek record", e))?;
            let mut record = [0; HEADER];
            data.read_exact(&mut record)
                .map_err(|e| error("read header", e))?;
            let length = number(&record[16..24]);
            let end = offset
                .checked_add(HEADER as u64)
                .and_then(|n| n.checked_add(length));
            let file_len = data.metadata().map_err(|e| error("metadata", e))?.len();
            if length > MAX_RECORD_BYTES as u64 || end.is_none_or(|n| n > file_len) {
                return Err(error(
                    "length",
                    "truncated, overflowing or oversized record",
                ));
            }
            let length = usize::try_from(length).map_err(|e| error("length", e))?;
            let mut bytes = Vec::new();
            bytes
                .try_reserve_exact(length)
                .map_err(|e| error("read allocation", e))?;
            bytes.resize(length, 0);
            data.read_exact(&mut bytes)
                .map_err(|e| error("read payload", e))?;
            validate(position, &record, &bytes)?;
            Ok((number(&record[8..16]), bytes))
        } else {
            let offset = usize::try_from(self.offsets[position]).map_err(|e| error("index", e))?;
            let record = &self.bytes[offset..offset + HEADER];
            let length =
                usize::try_from(number(&record[16..24])).map_err(|e| error("length", e))?;
            let bytes = self
                .bytes
                .get(offset + HEADER..offset + HEADER + length)
                .ok_or_else(|| error("length", "invalid resident record"))?;
            validate(position, record, bytes)?;
            Ok((number(&record[8..16]), bytes.to_vec()))
        }
    }

    /// Transient storage directory, for diagnostics and storage controls.
    /// This is not a portable checkpoint or recovery interface.
    #[must_use]
    pub fn disk_directory(&self) -> Option<&Path> {
        self.disk.as_ref().map(|d| d.directory.as_path())
    }

    /// Removes owned files explicitly so cleanup failures reach the caller.
    ///
    /// # Errors
    /// Filesystem cleanup failure. The queue remains available for retry.
    pub fn close(&mut self) -> Result<(), SpillError> {
        if let Some(disk) = &self.disk {
            fs::remove_dir_all(&disk.directory).map_err(|e| error("cleanup", e))?;
            self.disk = None;
        }
        self.bytes = Vec::new();
        self.offsets = Vec::new();
        self.len = 0;
        Ok(())
    }
}

impl Drop for OrderedFrontier {
    fn drop(&mut self) {
        if let Some(disk) = &self.disk {
            let _ = fs::remove_dir_all(&disk.directory);
        }
    }
}

fn check_magic(file: &mut File, magic: [u8; 8]) -> Result<(), SpillError> {
    let mut found = [0; 8];
    file.read_exact(&mut found)
        .map_err(|e| error("read version", e))?;
    if found != magic {
        return Err(error("version", "unsupported or corrupt frontier format"));
    }
    Ok(())
}

fn number(bytes: &[u8]) -> u64 {
    let mut n = [0; 8];
    n.copy_from_slice(bytes);
    u64::from_le_bytes(n)
}
fn checksum(position: u64, id: u64, bytes: &[u8]) -> u64 {
    position
        .to_le_bytes()
        .into_iter()
        .chain(id.to_le_bytes())
        .chain(bytes.iter().copied())
        .fold(0xcbf2_9ce4_8422_2325, |h, b| {
            (h ^ u64::from(b)).wrapping_mul(0x100_0000_01b3)
        })
}
fn header(position: u64, id: u64, bytes: &[u8]) -> [u8; HEADER] {
    let mut out = [0; HEADER];
    for (i, n) in [
        position,
        id,
        bytes.len() as u64,
        checksum(position, id, bytes),
    ]
    .into_iter()
    .enumerate()
    {
        out[i * 8..i * 8 + 8].copy_from_slice(&n.to_le_bytes());
    }
    out
}
fn validate(position: usize, record: &[u8], bytes: &[u8]) -> Result<(), SpillError> {
    if number(&record[..8]) != position as u64
        || number(&record[24..32]) != checksum(position as u64, number(&record[8..16]), bytes)
    {
        return Err(error("checksum", "record identity or payload mismatch"));
    }
    Ok(())
}

#[cfg(test)]
#[allow(
    clippy::expect_used,
    clippy::unwrap_used,
    reason = "storage failure controls"
)]
mod tests {
    use super::*;
    fn parent() -> PathBuf {
        let p = std::env::temp_dir().join(format!(
            "dive-frontier-test-{}-{}",
            std::process::id(),
            NEXT.fetch_add(1, Ordering::Relaxed)
        ));
        fs::create_dir(&p).expect("test directory");
        p
    }
    fn queue() -> (PathBuf, OrderedFrontier) {
        let p = parent();
        let mut q = OrderedFrontier::new(0, &p);
        q.push(37, b"abc").expect("record");
        (p, q)
    }
    #[test]
    fn memory_and_disk_keep_ids_order_and_capacity_bound() {
        for budget in [0, 1, 64, 1024] {
            let p = parent();
            let mut q = OrderedFrontier::new(budget, &p);
            for i in 0..20 {
                q.push(100 + i, &[u8::try_from(i).expect("small index"); 19])
                    .expect("push");
                assert!(q.stats().retained_bytes <= budget as u64);
            }
            for i in 0..20 {
                assert_eq!(
                    q.read(i).unwrap(),
                    (
                        100 + i as u64,
                        vec![u8::try_from(i).expect("small index"); 19]
                    )
                );
            }
            assert!(q.read(20).is_err());
            q.close().unwrap();
            assert_eq!(fs::read_dir(&p).unwrap().count(), 0);
            fs::remove_dir(p).unwrap();
        }
    }
    #[test]
    fn corrupt_truncated_wrong_index_and_oversized_records_refuse() {
        for case in 0..6 {
            let (p, mut q) = queue();
            let d = q.disk.as_ref().unwrap();
            let file = d.directory.join("states");
            let mut bytes = fs::read(&file).unwrap();
            match case {
                0 => bytes[0] ^= 1,
                1 => bytes.truncate(8 + HEADER - 1),
                2 => *bytes.last_mut().unwrap() ^= 1,
                3 => bytes[8 + 16..8 + 24].copy_from_slice(&u64::MAX.to_le_bytes()),
                4 => bytes[8..16].copy_from_slice(&1u64.to_le_bytes()),
                _ => fs::write(
                    d.directory.join("index"),
                    [INDEX_MAGIC.as_slice(), &0u64.to_le_bytes()].concat(),
                )
                .unwrap(),
            }
            fs::write(file, bytes).unwrap();
            assert!(q.read(0).is_err(), "case{case}");
            q.close().unwrap();
            fs::remove_dir(p).unwrap();
        }
    }
    #[test]
    fn old_value_bearing_frontier_versions_refuse_without_payload_guessing() {
        for (file, magic) in [("states", b"DIVEFQ01"), ("index", b"DIVEFI01")] {
            let (p, mut q) = queue();
            let path = q.disk.as_ref().unwrap().directory.join(file);
            let mut bytes = fs::read(&path).unwrap();
            bytes[..8].copy_from_slice(magic);
            fs::write(path, bytes).unwrap();
            let failure = q.read(0).expect_err("version one refuses");
            assert_eq!(failure.code, "DIVE-S0200");
            assert_eq!(
                failure.message,
                "frontier version: unsupported or corrupt frontier format"
            );
            q.close().unwrap();
            fs::remove_dir(p).unwrap();
        }
    }
    #[test]
    fn missing_read_write_and_cleanup_failures_are_explicit() {
        let (p, mut q) = queue();
        let disk = q.disk.as_mut().unwrap();
        let path = disk.directory.join("states");
        disk.data = File::open(&path).unwrap();
        assert!(q.push(38, b"def").is_err());
        assert_eq!(q.len(), 1);
        fs::remove_file(path).unwrap();
        assert!(q.read(0).is_err());
        let actual = q.disk.as_ref().unwrap().directory.clone();
        q.disk.as_mut().unwrap().directory = actual.join("index");
        assert!(q.close().is_err());
        q.disk.as_mut().unwrap().directory = actual;
        q.close().unwrap();
        fs::remove_dir(p).unwrap();
        let p = parent();
        let mut q = OrderedFrontier::new(0, &p.join("missing"));
        assert!(q.push(1, b"x").is_err());
        assert_eq!(q.len(), 0);
        fs::remove_dir(p).unwrap();
    }
}