symforge 7.6.1

SymForge — in-memory code intelligence MCP server for Claude Code, Codex, and Gemini CLI
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
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
use std::path::{Path, PathBuf};

use anyhow::Result;

use crate::domain::{FileClassification, LanguageId};

/// A file found during directory traversal that has a recognized language extension.
#[derive(Debug, Clone)]
pub struct DiscoveredFile {
    /// Relative path from the root, using forward slashes (e.g., "src/lib.rs").
    pub relative_path: String,
    /// Absolute path on disk.
    pub absolute_path: PathBuf,
    /// Language inferred from the file extension.
    pub language: LanguageId,
    /// Deterministic semantic-lane classification captured at discovery time.
    pub classification: FileClassification,
}

/// A file found during a full-filesystem walk (all files, not just known-language ones).
///
/// Used by the admission gate to classify every file — including those with unknown or
/// denylisted extensions — before deciding whether to parse them.
#[derive(Debug, Clone)]
pub struct DiscoveredEntry {
    /// Relative path from the root, using forward slashes.
    pub relative_path: String,
    /// Absolute path on disk.
    pub absolute_path: PathBuf,
    /// File size in bytes from the walk metadata (no extra stat syscall).
    pub file_size: u64,
    /// Language inferred from the extension, if recognized.
    pub language: Option<LanguageId>,
    /// Semantic-lane classification (test/vendor/generated/config flags).
    pub classification: FileClassification,
}

/// Discover all source files under `root` that have a recognized language extension.
///
/// - Respects `.gitignore` files via the `ignore` crate.
/// - Normalizes path separators to `/` in `relative_path`.
/// - Returns files sorted case-insensitively by `relative_path`.
pub fn discover_files(root: &Path) -> Result<Vec<DiscoveredFile>> {
    use ignore::WalkBuilder;

    // Canonicalize root so that strip_prefix succeeds even when the walker
    // resolves symlinks to their canonical targets.
    let root = std::fs::canonicalize(root).unwrap_or_else(|_| root.to_path_buf());

    let mut files: Vec<DiscoveredFile> = WalkBuilder::new(&root)
        .build()
        .filter_map(|entry_result| {
            let entry = entry_result.ok()?;
            let path =
                std::fs::canonicalize(entry.path()).unwrap_or_else(|_| entry.path().to_path_buf());

            // Use the already-known file_type from the walker instead of
            // path.is_file() which would issue a redundant stat() syscall.
            if !entry.file_type().map(|ft| ft.is_file()).unwrap_or(false) {
                return None;
            }

            let ext = path.extension()?.to_str()?;
            let language = LanguageId::from_extension(ext)?;

            // Compute relative path from root
            let relative = path.strip_prefix(&root).ok()?;
            // Normalize backslashes to forward slashes
            let relative_path = relative.to_string_lossy().replace('\\', "/");

            Some(DiscoveredFile {
                classification: FileClassification::for_code_path(&relative_path),
                relative_path,
                absolute_path: path,
                language,
            })
        })
        .collect();

    // Cache sort keys once instead of lowercasing paths on every comparator call.
    files.sort_by_cached_key(|file| file.relative_path.to_lowercase());

    Ok(files)
}

/// Discover ALL files under `root` regardless of extension, for admission-gate classification.
///
/// Unlike `discover_files`, this function:
/// - Yields every file (not just known-language ones), so denylisted/binary files are visible.
/// - Captures file size from walk metadata (avoids a separate stat() call).
/// - Sets `language = None` for files with unrecognized extensions.
/// - Returns files sorted case-insensitively by `relative_path`.
pub fn discover_all_files(root: &Path) -> Result<Vec<DiscoveredEntry>> {
    use ignore::WalkBuilder;

    // Canonicalize root so that strip_prefix succeeds even when the walker
    // resolves symlinks to their canonical targets.
    let root = std::fs::canonicalize(root).unwrap_or_else(|_| root.to_path_buf());

    let mut entries: Vec<DiscoveredEntry> = WalkBuilder::new(&root)
        .build()
        .filter_map(|entry_result| {
            let entry = entry_result.ok()?;
            let path =
                std::fs::canonicalize(entry.path()).unwrap_or_else(|_| entry.path().to_path_buf());

            if !entry.file_type().map(|ft| ft.is_file()).unwrap_or(false) {
                return None;
            }

            // Get file size from the walk metadata (DirEntry has it on most platforms).
            // Fall back to a stat call only when metadata is unavailable.
            let file_size = entry
                .metadata()
                .ok()
                .map(|m| m.len())
                .unwrap_or_else(|| std::fs::metadata(&path).map(|m| m.len()).unwrap_or(0));

            // Compute relative path from root
            let relative = path.strip_prefix(&root).ok()?;
            let relative_path = relative.to_string_lossy().replace('\\', "/");

            // Attempt language detection; None for unknown/denylisted extensions.
            let language = path
                .extension()
                .and_then(|e| e.to_str())
                .and_then(LanguageId::from_extension);

            let classification = FileClassification::for_code_path(&relative_path);

            Some(DiscoveredEntry {
                relative_path,
                absolute_path: path,
                file_size,
                language,
                classification,
            })
        })
        .collect();

    // Cache sort keys once instead of lowercasing paths on every comparator call.
    entries.sort_by_cached_key(|entry| entry.relative_path.to_lowercase());

    Ok(entries)
}

/// Load all `.gitignore` patterns from a repository root and nested directories.
///
/// Uses `ignore::gitignore::GitignoreBuilder` to build a composite gitignore matcher.
/// Walks nested `.gitignore` files up to `max_depth` levels (default 6).
/// Returns `None` if no `.gitignore` files are found or if loading fails.
pub fn load_gitignore(root: &Path) -> Option<ignore::gitignore::Gitignore> {
    use ignore::gitignore::GitignoreBuilder;
    use std::collections::VecDeque;

    let root_gitignore = root.join(".gitignore");
    if !root_gitignore.exists() {
        return None;
    }

    let mut builder = GitignoreBuilder::new(root);

    // BFS to find nested .gitignore files (max depth 6)
    let max_depth: usize = 6;
    let mut queue: VecDeque<(PathBuf, usize)> = VecDeque::new();
    queue.push_back((root.to_path_buf(), 0));

    while let Some((dir, depth)) = queue.pop_front() {
        let gitignore_path = dir.join(".gitignore");
        if gitignore_path.is_file()
            && let Some(err) = builder.add(&gitignore_path)
        {
            tracing::debug!("failed to load {:?}: {}", gitignore_path, err);
        }

        if depth < max_depth
            && let Ok(entries) = std::fs::read_dir(&dir)
        {
            for entry in entries.flatten() {
                let path = entry.path();
                if path.is_dir() {
                    // Skip common directories that won't have relevant .gitignore files
                    let name = entry.file_name();
                    let name_str = name.to_string_lossy();
                    if name_str.starts_with('.') && name_str != ".github" {
                        continue;
                    }
                    queue.push_back((path, depth + 1));
                }
            }
        }
    }

    match builder.build() {
        Ok(gi) => {
            // Only return Some if there are actual patterns
            if gi.is_empty() { None } else { Some(gi) }
        }
        Err(e) => {
            tracing::debug!("failed to build gitignore matcher: {}", e);
            None
        }
    }
}

/// Walk upward from the current working directory, looking for a `.git` directory.
/// Returns `None` if no git root is found and the cwd is a forbidden directory.
pub fn find_project_root() -> Option<PathBuf> {
    let cwd = std::env::current_dir().unwrap_or_else(|_| PathBuf::from("."));

    // Try to find a git root first (always safe — scoped by repo boundary).
    let mut current = cwd.clone();
    loop {
        if current.join(".git").exists() {
            return Some(current);
        }
        match current.parent() {
            Some(parent) => current = parent.to_path_buf(),
            None => break,
        }
    }

    // No git root found — use cwd if it's not a forbidden directory.
    if is_forbidden_root(&cwd) {
        tracing::warn!(
            path = %cwd.display(),
            "refusing to auto-index: directory is too broad (home dir, drive root, or system path)"
        );
        None
    } else {
        Some(cwd)
    }
}

/// Returns `true` if `path` is a directory that should never be auto-indexed
/// because it would be too large or contain unrelated files.
fn is_forbidden_root(path: &Path) -> bool {
    // Canonicalize for reliable comparison (resolves symlinks, normalizes separators).
    let path = path.canonicalize().unwrap_or_else(|_| path.to_path_buf());

    // 1. Drive roots: C:\, D:\, /, etc.
    if path.parent().is_none() {
        return true;
    }

    // 2. Windows drive roots that have a parent but are still just "C:\"
    #[cfg(target_os = "windows")]
    {
        let path_str = path.to_string_lossy();
        // Matches patterns like "C:\", "\\?\C:\"
        if path_str.len() <= 7 && path_str.ends_with('\\') {
            return true;
        }
    }

    // 3. User home directories.
    if let Some(home) = home_dir() {
        let home = home.canonicalize().unwrap_or(home);
        if path == home {
            return true;
        }
    }

    // 4. Known system/broad directory names.
    if let Some(name) = path.file_name().and_then(|n| n.to_str()) {
        let lower = name.to_lowercase();
        let forbidden_names = [
            "windows",
            "system32",
            "program files",
            "program files (x86)",
            "programdata",
            "appdata",
            "node_modules",
            ".npm",
            ".cargo",
            "users",
            "home",
            "tmp",
            "temp",
            "var",
        ];
        if forbidden_names.contains(&lower.as_str()) {
            return true;
        }
    }

    // 5. Parent-of-home: e.g. C:\Users or /home
    if let Some(home) = home_dir() {
        let home = home.canonicalize().unwrap_or(home);
        if let Some(parent) = home.parent() {
            let parent = parent
                .canonicalize()
                .unwrap_or_else(|_| parent.to_path_buf());
            if path == parent {
                return true;
            }
        }
    }

    false
}

/// Cross-platform home directory lookup.
fn home_dir() -> Option<PathBuf> {
    // std::env::home_dir is deprecated but dirs::home_dir may not be available.
    // Use environment variables directly for reliability.
    #[cfg(target_os = "windows")]
    {
        std::env::var("USERPROFILE").ok().map(PathBuf::from)
    }
    #[cfg(not(target_os = "windows"))]
    {
        std::env::var("HOME").ok().map(PathBuf::from)
    }
}

/// Check if content appears to be binary.
/// Examines up to BINARY_SNIFF_BYTES of the content using three heuristics:
/// 1. NUL byte present -> binary
/// 2. UTF-8 decode failure -> binary
/// 3. >30% suspicious control bytes (excluding \t, \n, \r) -> binary
pub fn is_binary_content(content: &[u8]) -> bool {
    if content.is_empty() {
        return false;
    }
    let check_len = content.len().min(crate::domain::index::BINARY_SNIFF_BYTES);
    let window = &content[..check_len];

    // Heuristic 1: NUL byte
    if window.contains(&0) {
        return true;
    }

    // Heuristic 2: Invalid UTF-8
    if std::str::from_utf8(window).is_err() {
        return true;
    }

    // Heuristic 3: High control byte ratio
    // Control bytes: 0x01-0x08, 0x0E-0x1F, 0x7F
    // Excludes common text controls: \t (0x09), \n (0x0A), \r (0x0D)
    let suspicious_controls = window
        .iter()
        .filter(|&&b| matches!(b, 0x01..=0x08 | 0x0E..=0x1F | 0x7F))
        .count();
    let ratio = suspicious_controls as f64 / window.len() as f64;
    if ratio > 0.30 {
        return true;
    }

    false
}

use crate::domain::index::{
    AdmissionDecision, AdmissionTier, HARD_SKIP_BYTES, METADATA_ONLY_BYTES, SkipReason,
};

/// Classify a file's admission tier. Returns AdmissionDecision with both tier and reason.
///
/// Precedence (first match wins):
/// 1. Hard-skip size ceiling (>100MB) → Tier 3
/// 2. Extension denylist → Tier 2
/// 3. Metadata-only size threshold (>1MB) → Tier 2
/// 4. Binary sniff (null bytes in first 8KB) → Tier 2
/// 5. All else → Tier 1
pub fn classify_admission(
    path: &std::path::Path,
    file_size: u64,
    content_sample: Option<&[u8]>,
) -> AdmissionDecision {
    use crate::domain::index::is_denylisted_extension;

    if file_size > HARD_SKIP_BYTES {
        return AdmissionDecision::skip(AdmissionTier::HardSkip, SkipReason::SizeCeiling);
    }
    if let Some(ext) = path.extension().and_then(|e| e.to_str())
        && is_denylisted_extension(ext)
    {
        return AdmissionDecision::skip(
            AdmissionTier::MetadataOnly,
            SkipReason::DenylistedExtension,
        );
    }
    if file_size > METADATA_ONLY_BYTES {
        return AdmissionDecision::skip(AdmissionTier::MetadataOnly, SkipReason::SizeThreshold);
    }
    if let Some(content) = content_sample
        && is_binary_content(content)
    {
        return AdmissionDecision::skip(AdmissionTier::MetadataOnly, SkipReason::BinaryContent);
    }
    AdmissionDecision::normal()
}

#[cfg(test)]
mod tests {
    use super::*;
    use std::fs;
    use tempfile::TempDir;

    fn create_file(dir: &Path, name: &str, content: &str) {
        let path = dir.join(name);
        if let Some(parent) = path.parent() {
            fs::create_dir_all(parent).unwrap();
        }
        fs::write(path, content).unwrap();
    }

    #[test]
    fn test_discover_files_finds_rs_py_js() {
        let tmp = TempDir::new().unwrap();
        create_file(tmp.path(), "main.rs", "fn main() {}");
        create_file(tmp.path(), "script.py", "def foo(): pass");
        create_file(tmp.path(), "app.js", "function bar() {}");

        let files = discover_files(tmp.path()).unwrap();
        let extensions: Vec<&str> = files
            .iter()
            .map(|f| f.relative_path.rsplit('.').next().unwrap())
            .collect();

        assert!(extensions.contains(&"rs"), "should find .rs");
        assert!(extensions.contains(&"py"), "should find .py");
        assert!(extensions.contains(&"js"), "should find .js");
        assert_eq!(files.len(), 3);
    }

    #[test]
    fn test_discover_files_includes_config_files() {
        let tmp = TempDir::new().unwrap();
        create_file(tmp.path(), "config.json", "{}");
        create_file(tmp.path(), "README.md", "# readme");
        create_file(tmp.path(), "Cargo.toml", "[package]");
        create_file(tmp.path(), "main.rs", "fn main() {}");

        let files = discover_files(tmp.path()).unwrap();
        assert_eq!(files.len(), 4, "should discover .rs + .json + .md + .toml");
        let paths: Vec<&str> = files.iter().map(|f| f.relative_path.as_str()).collect();
        assert!(paths.contains(&"config.json"), "should find .json");
        assert!(paths.contains(&"README.md"), "should find .md");
        assert!(paths.contains(&"Cargo.toml"), "should find .toml");
        assert!(paths.contains(&"main.rs"), "should find .rs");
    }

    #[test]
    fn test_discover_files_respects_gitignore() {
        let tmp = TempDir::new().unwrap();
        // Must create .git dir for gitignore to be respected
        fs::create_dir(tmp.path().join(".git")).unwrap();
        fs::write(tmp.path().join(".gitignore"), "ignored.rs\n").unwrap();

        create_file(tmp.path(), "main.rs", "fn main() {}");
        create_file(tmp.path(), "ignored.rs", "fn ignored() {}");

        let files = discover_files(tmp.path()).unwrap();
        assert_eq!(files.len(), 1, "ignored.rs should be excluded");
        assert_eq!(files[0].relative_path, "main.rs");
    }

    #[test]
    fn test_discover_files_normalizes_backslashes() {
        let tmp = TempDir::new().unwrap();
        // Create a file in a subdirectory — the path separator will be OS-native
        create_file(tmp.path(), "src/lib.rs", "pub fn lib() {}");

        let files = discover_files(tmp.path()).unwrap();
        assert_eq!(files.len(), 1);
        // Must use forward slashes regardless of OS
        assert!(
            !files[0].relative_path.contains('\\'),
            "should have no backslashes: {:?}",
            files[0].relative_path
        );
        assert!(files[0].relative_path.contains('/') || files[0].relative_path == "src/lib.rs");
    }

    #[test]
    fn test_discover_files_deterministic_sorted_order() {
        let tmp = TempDir::new().unwrap();
        create_file(tmp.path(), "Zoo.rs", "fn zoo() {}");
        create_file(tmp.path(), "apple.rs", "fn apple() {}");
        create_file(tmp.path(), "Mango.rs", "fn mango() {}");

        let files = discover_files(tmp.path()).unwrap();
        assert_eq!(files.len(), 3);
        // Case-insensitive alphabetical order
        let names: Vec<&str> = files.iter().map(|f| f.relative_path.as_str()).collect();
        // "apple" < "Mango" < "Zoo" case-insensitively
        let lower: Vec<String> = names.iter().map(|n| n.to_lowercase()).collect();
        let mut sorted = lower.clone();
        sorted.sort();
        assert_eq!(
            lower, sorted,
            "files should be in case-insensitive sorted order"
        );
    }

    #[test]
    fn test_discover_files_assigns_classification_tags_from_path() {
        let tmp = TempDir::new().unwrap();
        create_file(tmp.path(), "tests/unit_spec.rs", "fn spec_case() {}");
        create_file(tmp.path(), "vendor/pkg/lib.rs", "fn vendored() {}");
        create_file(
            tmp.path(),
            "src/generated/client.generated.rs",
            "fn generated() {}",
        );

        let files = discover_files(tmp.path()).unwrap();
        let by_path: std::collections::HashMap<&str, &DiscoveredFile> = files
            .iter()
            .map(|file| (file.relative_path.as_str(), file))
            .collect();

        assert!(
            by_path["tests/unit_spec.rs"].classification.is_test,
            "tests path should set is_test"
        );
        assert!(
            by_path["vendor/pkg/lib.rs"].classification.is_vendor,
            "vendor path should set is_vendor"
        );
        assert!(
            by_path["src/generated/client.generated.rs"]
                .classification
                .is_generated,
            "generated path should set is_generated"
        );
    }

    #[test]
    fn test_is_forbidden_root_blocks_home_dir() {
        let home = home_dir();
        if let Some(h) = home {
            assert!(is_forbidden_root(&h), "home directory should be forbidden");
        }
    }

    #[test]
    fn test_is_forbidden_root_blocks_drive_root() {
        #[cfg(target_os = "windows")]
        assert!(is_forbidden_root(Path::new("C:\\")));
        #[cfg(not(target_os = "windows"))]
        assert!(is_forbidden_root(Path::new("/")));
    }

    #[test]
    fn test_is_forbidden_root_blocks_system_dirs() {
        assert!(is_forbidden_root(Path::new("/tmp")));
        assert!(is_forbidden_root(Path::new("/home")));
    }

    #[test]
    fn test_is_forbidden_root_allows_project_dirs() {
        let tmp = TempDir::new().unwrap();
        assert!(
            !is_forbidden_root(tmp.path()),
            "temp project dir should be allowed"
        );
    }

    #[test]
    fn test_binary_sniff_detects_null_bytes() {
        let content = b"hello\x00world";
        assert!(is_binary_content(content));
    }

    #[test]
    fn test_binary_sniff_allows_pure_utf8() {
        let content = b"fn main() { println!(\"hello\"); }";
        assert!(!is_binary_content(content));
    }

    #[test]
    fn test_binary_sniff_empty_file() {
        assert!(!is_binary_content(b""));
    }

    #[test]
    fn test_binary_sniff_detects_invalid_utf8() {
        let content: &[u8] = &[0x80, 0x81, 0x82, 0x83, 0x84];
        assert!(is_binary_content(content));
    }

    #[test]
    fn test_binary_sniff_detects_high_control_ratio() {
        let mut content = Vec::new();
        content.extend(std::iter::repeat_n(0x01, 80)); // SOH — control char
        content.extend(std::iter::repeat_n(b'A', 20)); // printable
        // 80% control bytes > 30% threshold -> binary
        assert!(is_binary_content(&content));
    }

    #[test]
    fn test_binary_sniff_allows_low_control_ratio() {
        let content = b"line1\tvalue1\nline2\tvalue2\nline3\tvalue3\n";
        assert!(!is_binary_content(content));
    }

    #[test]
    fn test_binary_sniff_allows_common_whitespace_controls() {
        let content = b"col1\tcol2\tcol3\r\nval1\tval2\tval3\r\n";
        assert!(!is_binary_content(content));
    }

    // ── classify_admission tests ──

    use crate::domain::index::{AdmissionDecision, AdmissionTier, SkipReason};

    #[test]
    fn test_huge_text_file_is_hard_skip() {
        let decision =
            classify_admission(std::path::Path::new("huge.txt"), 150 * 1024 * 1024, None);
        assert_eq!(decision.tier, AdmissionTier::HardSkip);
        assert_eq!(decision.reason, Some(SkipReason::SizeCeiling));
    }

    #[test]
    fn test_small_ckpt_is_metadata_only() {
        let decision = classify_admission(std::path::Path::new("model.ckpt"), 50 * 1024, None);
        assert_eq!(decision.tier, AdmissionTier::MetadataOnly);
        assert_eq!(decision.reason, Some(SkipReason::DenylistedExtension));
    }

    #[test]
    fn test_huge_ckpt_is_hard_skip() {
        let decision = classify_admission(std::path::Path::new("big.ckpt"), 4_200_000_000, None);
        assert_eq!(decision.tier, AdmissionTier::HardSkip);
        assert_eq!(decision.reason, Some(SkipReason::SizeCeiling));
    }

    #[test]
    fn test_large_json_is_metadata_only() {
        let decision = classify_admission(std::path::Path::new("big.json"), 2 * 1024 * 1024, None);
        assert_eq!(decision.tier, AdmissionTier::MetadataOnly);
        assert_eq!(decision.reason, Some(SkipReason::SizeThreshold));
    }

    #[test]
    fn test_small_txt_is_normal() {
        let decision = classify_admission(std::path::Path::new("readme.txt"), 50 * 1024, None);
        assert_eq!(decision, AdmissionDecision::normal());
    }

    #[test]
    fn test_medium_rust_source_is_normal() {
        let decision = classify_admission(std::path::Path::new("big_module.rs"), 500 * 1024, None);
        assert_eq!(decision, AdmissionDecision::normal());
    }

    #[test]
    fn test_binary_content_is_metadata_only() {
        let content = b"ELF\x00\x00\x00binary";
        let decision =
            classify_admission(std::path::Path::new("unknown_file"), 1024, Some(content));
        assert_eq!(decision.tier, AdmissionTier::MetadataOnly);
        assert_eq!(decision.reason, Some(SkipReason::BinaryContent));
    }

    #[test]
    fn test_svg_not_denylisted() {
        let decision = classify_admission(std::path::Path::new("icon.svg"), 50 * 1024, None);
        assert_eq!(decision, AdmissionDecision::normal());
    }

    #[test]
    fn test_large_svg_is_metadata_only_by_size() {
        let decision = classify_admission(std::path::Path::new("huge.svg"), 2 * 1024 * 1024, None);
        assert_eq!(decision.tier, AdmissionTier::MetadataOnly);
        assert_eq!(decision.reason, Some(SkipReason::SizeThreshold));
    }

    // ── discover_all_files + admission gate integration tests ──

    #[test]
    fn test_discovery_skips_denylisted_extension() {
        let tmp = TempDir::new().unwrap();
        create_file(tmp.path(), "main.rs", "fn main() {}");
        // Write a fake .safetensors file (extension is on the denylist)
        fs::write(tmp.path().join("model.safetensors"), b"fake model bytes").unwrap();

        let entries = discover_all_files(tmp.path()).unwrap();

        // Classify each entry and collect skipped ones
        let mut rs_found = false;
        let mut safetensors_skipped = false;
        let mut safetensors_reason = None;

        for entry in &entries {
            let size = entry.file_size;
            let decision = classify_admission(&entry.absolute_path, size, None);
            if entry.relative_path == "main.rs" {
                assert_eq!(
                    decision.tier,
                    AdmissionTier::Normal,
                    ".rs file should be Normal"
                );
                rs_found = true;
            }
            if entry.relative_path == "model.safetensors" {
                assert_eq!(
                    decision.tier,
                    AdmissionTier::MetadataOnly,
                    ".safetensors should be MetadataOnly"
                );
                safetensors_skipped = true;
                safetensors_reason = decision.reason;
            }
        }

        assert!(rs_found, ".rs file must appear in discovered entries");
        assert!(
            safetensors_skipped,
            ".safetensors must appear in discovered entries and be skipped"
        );
        assert_eq!(
            safetensors_reason,
            Some(SkipReason::DenylistedExtension),
            ".safetensors skip reason must be DenylistedExtension"
        );
    }

    #[test]
    fn test_discovery_deferred_binary_sniff_reclassifies() {
        let tmp = TempDir::new().unwrap();
        create_file(tmp.path(), "lib.rs", "pub fn hello() {}");

        // Write a .dat file with NUL-heavy content: not on denylist, under 1MB,
        // but binary sniff (NUL bytes) should reclassify to MetadataOnly.
        let mut binary_content = vec![0u8; 512]; // NUL bytes — triggers binary sniff
        binary_content.extend_from_slice(b"some trailing text");
        fs::write(tmp.path().join("custom.dat"), &binary_content).unwrap();

        let entries = discover_all_files(tmp.path()).unwrap();

        let mut rs_normal = false;
        let mut dat_skipped = false;
        let mut dat_reason = None;

        for entry in &entries {
            let size = entry.file_size;

            // Phase 1: pre-content check
            let pre = classify_admission(&entry.absolute_path, size, None);

            if entry.relative_path == "lib.rs" {
                assert_eq!(pre.tier, AdmissionTier::Normal);
                rs_normal = true;
            }

            if entry.relative_path == "custom.dat" {
                // Pre-content: should be Normal (not denylisted, under 1MB)
                assert_eq!(
                    pre.tier,
                    AdmissionTier::Normal,
                    "custom.dat should be Normal before binary sniff"
                );

                // Phase 2: with content — binary sniff should reclassify
                let content = fs::read(&entry.absolute_path).unwrap();
                let post = classify_admission(&entry.absolute_path, size, Some(&content));
                assert_eq!(
                    post.tier,
                    AdmissionTier::MetadataOnly,
                    "custom.dat should be MetadataOnly after binary sniff"
                );
                dat_skipped = true;
                dat_reason = post.reason;
            }
        }

        assert!(rs_normal, "lib.rs must be Normal");
        assert!(
            dat_skipped,
            "custom.dat must be discovered and reclassified"
        );
        assert_eq!(
            dat_reason,
            Some(SkipReason::BinaryContent),
            "custom.dat skip reason must be BinaryContent"
        );
    }
}