mbx-cache-cc 0.11.2

mbx internals: conservative C and C++ action analysis and key construction. No API stability -- use the mbx CLI or mbx-cache-protocol.
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
//! Dependency-list parsing and input discovery for C and C++ compiles.

use crate::{
    CcActionContext, CcActionInput, CcBypassReason, CcCompilerFamily, MAX_INPUT_BYTES,
    MAX_MANIFEST_ENTRIES, MAX_PREDICTED_INPUTS, normalize_components,
};
use mbx_cache_core::{
    CacheDigest, FileDigestCache, FileDigestScope, FileIdentity, RecordedFileDigest,
};
use std::collections::{BTreeMap, BTreeSet};
use std::io::Read;
use std::path::{Path, PathBuf};
use std::time::SystemTime;

/// Marker distinguishing an include-directory name manifest from a file input.
pub const INCLUDE_MANIFEST_PREFIX: &str = "@include-manifest:";

/// Macros whose expansion is not a function of the compilation's inputs.
const TIMESTAMP_MACROS: &[&[u8]] = &[b"__DATE__", b"__TIME__", b"__TIMESTAMP__"];

/// Directives whose operands are consumed by the assembler, after the
/// compiler has finished producing its dependency list.
const ASSEMBLER_INPUT_DIRECTIVES: &[&[u8]] = &[b".include", b".incbin", b".sinclude"];

const SCAN_CHUNK_BYTES: usize = 64 * 1024;

/// A parsed GNU-style dependency list.
#[derive(Debug, Clone, PartialEq, Eq, Default)]
pub struct CcDepfile {
    /// Prerequisite files named by the first rule.
    pub files: Vec<PathBuf>,
}

impl CcDepfile {
    /// Read and parse the dependency list the compiler wrote.
    pub fn read(path: &Path) -> Result<Self, CcBypassReason> {
        let contents =
            std::fs::read_to_string(path).map_err(|error| CcBypassReason::DepfileRead {
                path: path.to_path_buf(),
                message: error.to_string(),
            })?;
        Self::parse(&contents)
    }

    /// Read the dependency output emitted by the selected compiler family.
    pub fn read_for(path: &Path, family: CcCompilerFamily) -> Result<Self, CcBypassReason> {
        if family.is_msvc() {
            Self::read_msvc(path)
        } else {
            Self::read(path)
        }
    }

    /// Read MSVC's `/sourceDependencies` JSON output.
    pub fn read_msvc(path: &Path) -> Result<Self, CcBypassReason> {
        let contents = std::fs::read(path).map_err(|error| CcBypassReason::DepfileRead {
            path: path.to_path_buf(),
            message: error.to_string(),
        })?;
        let value: serde_json::Value = serde_json::from_slice(&contents)
            .map_err(|error| CcBypassReason::MalformedDepfile(error.to_string()))?;
        let data = value
            .get("Data")
            .and_then(serde_json::Value::as_object)
            .ok_or_else(|| CcBypassReason::MalformedDepfile("missing Data object".into()))?;
        if data
            .get("ImportedModules")
            .and_then(serde_json::Value::as_array)
            .is_some_and(|modules| !modules.is_empty())
            || data.get("ProvidedModule").is_some_and(|module| {
                !module.is_null() && module.as_str().is_none_or(|s| !s.is_empty())
            })
        {
            return Err(CcBypassReason::MalformedDepfile(
                "C++ module dependencies are not modeled".into(),
            ));
        }
        let includes = data
            .get("Includes")
            .and_then(serde_json::Value::as_array)
            .ok_or_else(|| CcBypassReason::MalformedDepfile("missing Includes array".into()))?;
        let files = includes
            .iter()
            .map(|entry| {
                entry.as_str().map(PathBuf::from).ok_or_else(|| {
                    CcBypassReason::MalformedDepfile("non-string include path".into())
                })
            })
            .collect::<Result<Vec<_>, _>>()?;
        Ok(Self { files })
    }

    /// Parse a GNU-style dependency list.
    ///
    /// Only the first rule is read. The adapter never passes `-MP`, so a
    /// well-formed file the adapter asked for has exactly one rule, and
    /// anything further is ignored rather than guessed at.
    pub fn parse(contents: &str) -> Result<Self, CcBypassReason> {
        let joined = join_continuations(contents)?;
        let (_, prerequisites) = joined
            .lines()
            .find_map(|line| line.split_once(RULE_SEPARATOR))
            .ok_or_else(|| CcBypassReason::MalformedDepfile("no dependency rule".into()))?;
        let files = split_prerequisites(prerequisites)?;
        Ok(Self { files })
    }
}

const RULE_SEPARATOR: &str = ": ";

impl CcDepfile {
    /// Render the dependency list a caller asked for, the way the driver
    /// writes one: the rule's targets, then every prerequisite on its own
    /// continued line, then, for `-MP`, an empty rule per header so make does
    /// not fail when one is deleted. `source` is the file that was compiled
    /// and gets no phony rule, exactly as the driver leaves it out.
    ///
    /// Escapes are the ones `parse` reads back: a space, a `#`, and a `$` in
    /// a path. A quoted target gets the same treatment; a literal one is
    /// written as given, which is what `-MT` promises.
    pub fn render(
        targets: &[crate::DepfileTarget],
        files: &[PathBuf],
        source: &Path,
        phony_targets: bool,
    ) -> String {
        let mut rendered = String::new();
        for (index, target) in targets.iter().enumerate() {
            if index > 0 {
                rendered.push(' ');
            }
            if target.quoted {
                rendered.push_str(&escape_make_word(&target.name));
            } else {
                rendered.push_str(&target.name);
            }
        }
        rendered.push(':');
        for file in files {
            rendered.push_str(" \\\n ");
            rendered.push_str(&escape_make_word(&file.to_string_lossy()));
        }
        rendered.push('\n');
        if phony_targets {
            for file in files.iter().filter(|file| file.as_path() != source) {
                rendered.push_str(&escape_make_word(&file.to_string_lossy()));
                rendered.push_str(":\n");
            }
        }
        rendered
    }
}

/// Quote a word for make the way the driver does in a dependency list.
fn escape_make_word(word: &str) -> String {
    let mut escaped = String::with_capacity(word.len());
    for character in word.chars() {
        match character {
            ' ' => escaped.push_str("\\ "),
            '#' => escaped.push_str("\\#"),
            '$' => escaped.push_str("$$"),
            other => escaped.push(other),
        }
    }
    escaped
}

/// Join physical lines the compiler wrapped with a trailing backslash.
fn join_continuations(contents: &str) -> Result<String, CcBypassReason> {
    let mut joined = String::with_capacity(contents.len());
    let mut continued = false;
    for line in contents.lines() {
        let trimmed = line.strip_suffix('\r').unwrap_or(line);
        let (text, continues) = match trimmed.strip_suffix('\\') {
            Some(text) => (text, true),
            None => (trimmed, false),
        };
        if continued {
            joined.push(' ');
        }
        joined.push_str(text.trim_end_matches(['\t']));
        if !continues {
            joined.push('\n');
        }
        continued = continues;
    }
    if continued {
        return Err(CcBypassReason::MalformedDepfile(
            "unterminated line continuation".into(),
        ));
    }
    Ok(joined)
}

/// Split a prerequisite list, honoring exactly the escapes make defines.
///
/// Anything else escaped is a spelling this parser does not model, and a
/// mis-parsed prerequisite would silently drop an input from the key.
fn split_prerequisites(value: &str) -> Result<Vec<PathBuf>, CcBypassReason> {
    let mut files = Vec::new();
    let mut current = String::new();
    let mut characters = value.chars().peekable();
    while let Some(character) = characters.next() {
        match character {
            ' ' | '\t' => {
                if !current.is_empty() {
                    files.push(PathBuf::from(std::mem::take(&mut current)));
                }
            }
            '\\' => match characters.next() {
                Some(' ') => current.push(' '),
                Some('#') => current.push('#'),
                Some(other) => {
                    return Err(CcBypassReason::MalformedDepfile(format!(
                        "unmodeled escape \\{other}"
                    )));
                }
                None => {
                    return Err(CcBypassReason::MalformedDepfile(
                        "trailing escape character".into(),
                    ));
                }
            },
            '$' => match characters.next() {
                Some('$') => current.push('$'),
                Some(other) => {
                    return Err(CcBypassReason::MalformedDepfile(format!(
                        "unmodeled variable reference ${other}"
                    )));
                }
                None => {
                    return Err(CcBypassReason::MalformedDepfile(
                        "trailing variable reference".into(),
                    ));
                }
            },
            other => current.push(other),
        }
    }
    if !current.is_empty() {
        files.push(PathBuf::from(current));
    }
    Ok(files)
}

/// A complete, content-addressed compiler input manifest.
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct CcDiscoveredInputs {
    working_dir: PathBuf,
    /// Content-addressed inputs, including include-directory manifests.
    pub inputs: Vec<CcActionInput>,
    /// What each file input looked like on disk when its digest was
    /// established, index-aligned with `inputs`; `None` for manifests and
    /// where the filesystem gave nothing to compare against later.
    identities: Vec<Option<FileIdentity>>,
}

impl CcDiscoveredInputs {
    /// Digest every file the compilation read, and summarize the directories it
    /// searched.
    ///
    /// Digesting the files answers "did any input change". The directory
    /// manifests answer the question a dependency list cannot: whether a header
    /// that was *not* read now exists somewhere that would shadow one that was.
    pub fn collect(
        working_dir: &Path,
        files: BTreeSet<PathBuf>,
        directories: BTreeSet<PathBuf>,
        digests: &dyn FileDigestCache,
    ) -> Result<Self, CcBypassReason> {
        if !working_dir.is_absolute() {
            return Err(CcBypassReason::RelativeWorkingDirectory(
                working_dir.to_path_buf(),
            ));
        }
        let directories = minimal_manifest_directories(directories);
        if files.len() + directories.len() > MAX_PREDICTED_INPUTS {
            return Err(CcBypassReason::TooManyInputs);
        }
        let working_dir = normalize_components(working_dir);
        let mut inputs = Vec::with_capacity(files.len() + directories.len());
        let mut total_bytes = 0_u64;
        // Stat everything first so one batched ledger lookup can stand in for
        // rereading headers the session already scanned and hashed. A ledger
        // entry in the cc scope was recorded after the timestamp-macro scan
        // passed, so a hit skips the scan for the same reason it skips the
        // hash: the identity says the contents have not changed since both
        // were established.
        let mut identified = Vec::with_capacity(files.len());
        for path in files {
            let metadata = std::fs::metadata(&path).map_err(|error| CcBypassReason::InputRead {
                path: path.clone(),
                message: error.to_string(),
            })?;
            if !metadata.is_file() {
                return Err(CcBypassReason::InputRead {
                    path,
                    message: "input is not a regular file".into(),
                });
            }
            total_bytes = total_bytes.saturating_add(metadata.len());
            if total_bytes > MAX_INPUT_BYTES {
                return Err(CcBypassReason::TooManyInputs);
            }
            let identity = FileIdentity::describe(&path, &metadata);
            identified.push((path, identity));
        }
        let queries = identified
            .iter()
            .filter_map(|(_, identity)| identity.clone())
            .collect::<Vec<_>>();
        let mut recorded = digests.find(FileDigestScope::CcInput, &queries).into_iter();
        let mut identities = Vec::with_capacity(inputs.capacity());
        let mut fresh = Vec::new();
        for (path, identity) in identified {
            identities.push(identity.clone());
            let remembered = identity
                .as_ref()
                .and_then(|_| recorded.next().flatten())
                .filter(|digest| {
                    identity
                        .as_ref()
                        .is_some_and(|identity| identity.len == digest.size)
                });
            let digest = match remembered {
                Some(digest) => digest,
                None => {
                    if contains_timestamp_macro(&path)? {
                        return Err(CcBypassReason::EmbeddedTimestampMacro(path));
                    }
                    let digest = CacheDigest::blake3_file(&path).map_err(|error| {
                        CcBypassReason::InputRead {
                            path: path.clone(),
                            message: error.to_string(),
                        }
                    })?;
                    if let Some(identity) = identity
                        && identity.len == digest.size
                    {
                        fresh.push(RecordedFileDigest {
                            file: identity,
                            digest: digest.clone(),
                        });
                    }
                    digest
                }
            };
            inputs.push(CcActionInput { path, digest });
        }
        if !fresh.is_empty() {
            digests.record(FileDigestScope::CcInput, fresh);
        }
        let mut manifest_entries = 0_usize;
        for directory in directories {
            let digest = include_manifest(&directory, &mut manifest_entries)?;
            inputs.push(CcActionInput {
                path: PathBuf::from(format!("{INCLUDE_MANIFEST_PREFIX}{}", directory.display())),
                digest,
            });
            identities.push(None);
        }
        Ok(Self {
            working_dir,
            inputs,
            identities,
        })
    }

    /// File inputs, excluding include-directory manifests.
    pub fn files(&self) -> impl Iterator<Item = &CcActionInput> {
        self.inputs
            .iter()
            .filter(|input| !is_manifest_input(&input.path))
    }

    /// Reject inputs whose modification time overlaps the compiler invocation.
    ///
    /// Contents are hashed after the compiler reports the paths it read. This
    /// timestamp barrier prevents a write that landed during the compile from
    /// being mistaken for the contents that produced the object; `verify`
    /// closes the remaining race after hashing.
    pub fn verify_not_modified_since(&self, started_at: SystemTime) -> Result<(), CcBypassReason> {
        for input in self.files() {
            let modified = std::fs::metadata(&input.path)
                .and_then(|metadata| metadata.modified())
                .map_err(|error| CcBypassReason::InputRead {
                    path: input.path.clone(),
                    message: error.to_string(),
                })?;
            if modified >= started_at {
                return Err(CcBypassReason::InputModifiedDuringCompilation(
                    input.path.clone(),
                ));
            }
        }
        Ok(())
    }

    /// Confirm every discovered file before publication, degrading a changed
    /// input to a miss rather than storing an object under a stale key.
    ///
    /// A file still wearing the identity `collect` recorded is confirmed by
    /// that stat alone where the identity carries a change time, which cannot
    /// be set from user space and so shows a rewrite that restored the old
    /// modification time. One whose identity moved, that had none, or that a
    /// platform without change times described, is read and hashed again.
    pub fn verify(&self) -> Result<(), CcBypassReason> {
        for (index, input) in self.inputs.iter().enumerate() {
            if is_manifest_input(&input.path) {
                continue;
            }
            let read_error = |error: std::io::Error| CcBypassReason::InputRead {
                path: input.path.clone(),
                message: error.to_string(),
            };
            if let Some(Some(identity)) = self.identities.get(index)
                && identity.changed.is_some()
                && identity.still_describes().map_err(read_error)?
            {
                continue;
            }
            let matches = input.digest.matches_file(&input.path).map_err(|error| {
                CcBypassReason::InputRead {
                    path: input.path.clone(),
                    message: error.to_string(),
                }
            })?;
            if !matches {
                return Err(CcBypassReason::InputChanged(input.path.clone()));
            }
        }
        Ok(())
    }

    /// Merge the manifest into an action context after confirming both use the
    /// same compiler working directory.
    pub fn apply_to(self, context: &mut CcActionContext) -> Result<(), CcBypassReason> {
        if normalize_components(&context.working_dir) != self.working_dir {
            return Err(CcBypassReason::DiscoveryWorkingDirectory);
        }
        context.inputs.extend(self.inputs);
        Ok(())
    }
}

/// Drop include directories already covered by an ancestor's recursive manifest.
///
/// Discovered headers often contribute hundreds of nested parent directories,
/// especially for amalgamated C sources. Keeping both an ancestor and its
/// descendants walks and hashes the same subtree repeatedly, and can exhaust
/// the manifest-entry budget even though the ancestor already names every
/// includable file below it.
fn minimal_manifest_directories(directories: BTreeSet<PathBuf>) -> Vec<PathBuf> {
    let mut directories = directories
        .into_iter()
        .map(|directory| {
            let normalized = normalize_components(&directory);
            (directory, normalized)
        })
        .collect::<Vec<_>>();
    directories.sort_by(|(left, left_normalized), (right, right_normalized)| {
        left_normalized
            .components()
            .count()
            .cmp(&right_normalized.components().count())
            .then_with(|| left_normalized.cmp(right_normalized))
            .then_with(|| left.cmp(right))
    });

    let mut minimal = Vec::<(PathBuf, PathBuf)>::new();
    for (directory, normalized) in directories {
        if !minimal
            .iter()
            .any(|(_, ancestor)| manifest_covers(ancestor, &normalized))
        {
            minimal.push((directory, normalized));
        }
    }
    minimal
        .into_iter()
        .map(|(directory, _)| directory)
        .collect()
}

/// Whether walking `ancestor` recursively is guaranteed to visit `descendant`.
///
/// Component-aware normalization rejects a lexical prefix that escapes through
/// `..`. Directory symlinks need an explicit check because `read_dir` follows
/// the directory it starts at but the recursive walk deliberately does not
/// follow symlink entries beneath it.
fn manifest_covers(ancestor: &Path, descendant: &Path) -> bool {
    let Ok(relative) = descendant.strip_prefix(ancestor) else {
        return false;
    };
    if relative.as_os_str().is_empty() {
        return false;
    }
    let mut current = ancestor.to_path_buf();
    for component in relative.components() {
        current.push(component);
        let Ok(metadata) = std::fs::symlink_metadata(&current) else {
            return false;
        };
        if !metadata.is_dir() || metadata.file_type().is_symlink() {
            return false;
        }
    }
    true
}

fn is_manifest_input(path: &Path) -> bool {
    path.to_str()
        .is_some_and(|path| path.starts_with(INCLUDE_MANIFEST_PREFIX))
}

/// Digest the includable names in each directory, reading no file contents.
///
/// Taken once before the compiler runs and again before publishing, this is
/// what detects a header that appeared in a search directory *while* the
/// compilation was in flight. The manifest recorded in the key is the one from
/// after the compile, and without this check that later state would be claimed
/// as the state the compiler saw.
pub fn manifest_snapshot(
    directories: &BTreeSet<PathBuf>,
) -> Result<BTreeMap<PathBuf, CacheDigest>, CcBypassReason> {
    let mut budget = 0_usize;
    minimal_manifest_directories(directories.iter().cloned().collect())
        .into_iter()
        .map(|directory| {
            include_manifest(&directory, &mut budget).map(|digest| (directory, digest))
        })
        .collect()
}

/// Extensions a file must carry to be a plausible `#include` target.
///
/// An extensionless name also qualifies: C++ standard headers are spelled that
/// way and projects ship their own.
///
/// `gch` and `pch` are here because a precompiled header answers an `#include`
/// without being named by one. GCC prefers `foo.h.gch` over `foo.h` on its own,
/// with nothing on the command line to say so, which is precisely the
/// substitution these manifests exist to notice -- and the one case the
/// adapter's explicit precompiled-header bypass cannot see.
const INCLUDABLE_EXTENSIONS: &[&str] = &[
    "c", "c++", "cc", "cpp", "cxx", "def", "gch", "h", "h++", "hh", "hpp", "hxx", "inc", "inl",
    "ipp", "pch", "s", "tcc",
];

/// Whether a file name could be what an `#include` directive names.
///
/// The manifest exists to notice a file appearing where it would shadow a
/// header that was read. A build writes its own objects, dependency files, and
/// archives into these directories -- often the very directory a generated
/// header lives in -- and none of those can shadow an include. Counting them
/// would make the key depend on how many sibling compilations had finished,
/// which is not a property of this compilation at all.
fn is_includable(name: &str) -> bool {
    match name.rsplit_once('.') {
        Some((stem, extension)) if !stem.is_empty() => INCLUDABLE_EXTENSIONS
            .binary_search(&extension.to_ascii_lowercase().as_str())
            .is_ok(),
        // No extension, or a leading-dot name like `.keep`.
        _ => !name.starts_with('.'),
    }
}

/// Digest the sorted includable file names beneath a directory.
///
/// Names only: the contents of anything actually read are digested as inputs,
/// so this exists purely to notice a file appearing where it could shadow one
/// of them. A directory that does not exist has an empty manifest, which is
/// what makes "the directory was created" a key change rather than an error.
fn include_manifest(directory: &Path, budget: &mut usize) -> Result<CacheDigest, CcBypassReason> {
    let mut names = Vec::new();
    let mut pending = vec![(directory.to_path_buf(), String::new())];
    while let Some((current, prefix)) = pending.pop() {
        let entries = match std::fs::read_dir(&current) {
            Ok(entries) => entries,
            Err(error) if error.kind() == std::io::ErrorKind::NotFound => continue,
            Err(error) => {
                return Err(CcBypassReason::InputRead {
                    path: current,
                    message: error.to_string(),
                });
            }
        };
        for entry in entries {
            let entry = entry.map_err(|error| CcBypassReason::InputRead {
                path: current.clone(),
                message: error.to_string(),
            })?;
            let name = entry.file_name();
            let Some(name) = name.to_str() else {
                return Err(CcBypassReason::NonUtf8Path(entry.path()));
            };
            let relative = if prefix.is_empty() {
                name.to_string()
            } else {
                format!("{prefix}/{name}")
            };
            let file_type = entry
                .file_type()
                .map_err(|error| CcBypassReason::InputRead {
                    path: entry.path(),
                    message: error.to_string(),
                })?;
            if file_type.is_dir() {
                pending.push((entry.path(), relative));
                continue;
            }
            if !is_includable(name) {
                continue;
            }
            *budget += 1;
            if *budget > MAX_MANIFEST_ENTRIES {
                return Err(CcBypassReason::TooManyInputs);
            }
            names.push(relative);
        }
    }
    names.sort();
    Ok(CacheDigest::blake3(names.join("\n").as_bytes()))
}

/// Whether a file mentions a macro whose expansion is not a function of the
/// compilation's inputs.
///
/// The token is looked for rather than its expansion: a match inside a comment
/// or a string literal bypasses a compilation that would in fact have been
/// cacheable, which is the conservative direction.
fn contains_timestamp_macro(path: &Path) -> Result<bool, CcBypassReason> {
    let file = std::fs::File::open(path).map_err(|error| CcBypassReason::InputRead {
        path: path.to_path_buf(),
        message: error.to_string(),
    })?;
    let longest = TIMESTAMP_MACROS
        .iter()
        .map(|macro_name| macro_name.len())
        .max()
        .unwrap_or_default();
    let mut reader = std::io::BufReader::new(file);
    let mut window = Vec::with_capacity(SCAN_CHUNK_BYTES + longest);
    let mut chunk = vec![0_u8; SCAN_CHUNK_BYTES];
    loop {
        let read = reader
            .read(&mut chunk)
            .map_err(|error| CcBypassReason::InputRead {
                path: path.to_path_buf(),
                message: error.to_string(),
            })?;
        if read == 0 {
            return Ok(false);
        }
        window.extend_from_slice(&chunk[..read]);
        if TIMESTAMP_MACROS
            .iter()
            .any(|macro_name| contains_subslice(&window, macro_name))
        {
            return Ok(true);
        }
        // Keep the tail so a token split across two reads is still found.
        let keep = window.len().saturating_sub(longest.saturating_sub(1));
        window.drain(..keep);
    }
}

/// Whether a preprocessor input can make the assembler read another file.
///
/// Searching for the directive text, including in comments and inactive
/// conditional branches, deliberately accepts false positives. Missing a real
/// directive would publish an object whose complete inputs are absent from the
/// key; bypassing an otherwise cacheable object is the safe outcome instead.
pub(crate) fn contains_assembler_input_directive(path: &Path) -> Result<bool, CcBypassReason> {
    contains_any(path, ASSEMBLER_INPUT_DIRECTIVES)
}

fn contains_any(path: &Path, needles: &[&[u8]]) -> Result<bool, CcBypassReason> {
    let file = std::fs::File::open(path).map_err(|error| CcBypassReason::InputRead {
        path: path.to_path_buf(),
        message: error.to_string(),
    })?;
    let longest = needles
        .iter()
        .map(|needle| needle.len())
        .max()
        .unwrap_or_default();
    let mut reader = std::io::BufReader::new(file);
    let mut window = Vec::with_capacity(SCAN_CHUNK_BYTES + longest);
    let mut chunk = vec![0_u8; SCAN_CHUNK_BYTES];
    loop {
        let read = reader
            .read(&mut chunk)
            .map_err(|error| CcBypassReason::InputRead {
                path: path.to_path_buf(),
                message: error.to_string(),
            })?;
        if read == 0 {
            return Ok(false);
        }
        window.extend_from_slice(&chunk[..read]);
        if needles
            .iter()
            .any(|needle| contains_subslice_ascii_case_insensitive(&window, needle))
        {
            return Ok(true);
        }
        let keep = window.len().saturating_sub(longest.saturating_sub(1));
        window.drain(..keep);
    }
}

fn contains_subslice_ascii_case_insensitive(haystack: &[u8], needle: &[u8]) -> bool {
    if needle.is_empty() || haystack.len() < needle.len() {
        return false;
    }
    haystack.windows(needle.len()).any(|window| {
        window
            .iter()
            .zip(needle)
            .all(|(left, right)| left.eq_ignore_ascii_case(right))
    })
}

fn contains_subslice(haystack: &[u8], needle: &[u8]) -> bool {
    if needle.is_empty() || haystack.len() < needle.len() {
        return false;
    }
    haystack
        .windows(needle.len())
        .any(|window| window == needle)
}

#[cfg(test)]
#[path = "depfile_tests.rs"]
mod tests;