libmagic_rs/lib.rs
1// Copyright (c) 2025-2026 the libmagic-rs contributors
2// SPDX-License-Identifier: Apache-2.0
3
4//! Rust Libmagic - A pure-Rust implementation of libmagic
5//!
6//! This library provides safe, efficient file type identification through magic rule evaluation.
7//! It parses magic files into an Abstract Syntax Tree (AST) and evaluates them against file
8//! buffers using memory-mapped I/O for optimal performance.
9//!
10//! # Security Features
11//!
12//! This implementation prioritizes security through:
13//! - **Memory Safety**: Pure Rust with no unsafe code (except in vetted dependencies)
14//! - **Bounds Checking**: Comprehensive validation of all buffer accesses
15//! - **Resource Limits**: Configurable limits to prevent resource exhaustion attacks
16//! - **Input Validation**: Strict validation of magic files and configuration
17//! - **Error Handling**: Secure error messages that don't leak sensitive information
18//! - **Timeout Protection**: Configurable timeouts to prevent denial of service
19//!
20//! # Examples
21//!
22//! ## Complete Workflow: Load → Evaluate → Output
23//!
24//! ```rust,no_run
25//! use libmagic_rs::MagicDatabase;
26//!
27//! // Load magic rules from a text file
28//! let db = MagicDatabase::load_from_file("/usr/share/misc/magic")?;
29//!
30//! // Evaluate a file to determine its type
31//! let result = db.evaluate_file("sample.bin")?;
32//! println!("File type: {}", result.description);
33//!
34//! // Access metadata about loaded rules
35//! if let Some(path) = db.source_path() {
36//! println!("Rules loaded from: {}", path.display());
37//! }
38//! # Ok::<(), Box<dyn std::error::Error>>(())
39//! ```
40//!
41//! ## Loading from a Directory
42//!
43//! ```rust,no_run
44//! use libmagic_rs::MagicDatabase;
45//!
46//! // Load all magic files from a directory (Magdir pattern)
47//! let db = MagicDatabase::load_from_file("/usr/share/misc/magic.d")?;
48//!
49//! // Evaluate multiple files
50//! for file in &["file1.bin", "file2.bin", "file3.bin"] {
51//! let result = db.evaluate_file(file)?;
52//! println!("{}: {}", file, result.description);
53//! }
54//! # Ok::<(), Box<dyn std::error::Error>>(())
55//! ```
56//!
57//! ## Error Handling for Binary Files
58//!
59//! ```rust,no_run
60//! use libmagic_rs::MagicDatabase;
61//!
62//! // Attempt to load a binary .mgc file
63//! match MagicDatabase::load_from_file("/usr/share/misc/magic.mgc") {
64//! Ok(db) => {
65//! let result = db.evaluate_file("sample.bin")?;
66//! println!("File type: {}", result.description);
67//! }
68//! Err(e) => {
69//! eprintln!("Error loading magic file: {}", e);
70//! eprintln!("Hint: Binary .mgc files are not supported.");
71//! eprintln!("Use --use-builtin option to use built-in rules,");
72//! eprintln!("or provide a text-based magic file or directory.");
73//! }
74//! }
75//! # Ok::<(), Box<dyn std::error::Error>>(())
76//! ```
77//!
78//! ## Debugging with Source Path Metadata
79//!
80//! ```rust,no_run
81//! use libmagic_rs::MagicDatabase;
82//!
83//! let db = MagicDatabase::load_from_file("/usr/share/misc/magic")?;
84//!
85//! // Use source_path() for debugging and logging
86//! if let Some(source) = db.source_path() {
87//! println!("Loaded {} from {}",
88//! "magic rules",
89//! source.display());
90//! }
91//!
92//! // Evaluate files with source tracking
93//! let result = db.evaluate_file("sample.bin")?;
94//! println!("Detection result: {}", result.description);
95//! # Ok::<(), Box<dyn std::error::Error>>(())
96//! ```
97
98#![deny(missing_docs)]
99#![deny(unsafe_code)]
100#![deny(clippy::all)]
101#![warn(clippy::pedantic)]
102
103use std::path::{Path, PathBuf};
104
105use serde::{Deserialize, Serialize};
106
107// Re-export modules
108pub mod builtin_rules;
109mod config;
110pub mod error;
111pub mod evaluator;
112pub mod io;
113pub mod mime;
114pub mod output;
115pub mod parser;
116pub mod tags;
117
118pub use config::EvaluationConfig;
119
120/// Build-time helpers for compiling magic rules.
121///
122/// This module contains functionality used by the build script to parse magic files
123/// and generate Rust code for built-in rules. It is only available during tests and
124/// documentation builds to enable comprehensive testing of the build process.
125#[cfg(any(test, doc))]
126pub mod build_helpers;
127
128// Re-export core AST types for convenience
129pub use parser::ast::{
130 Endianness, MagicRule, OffsetSpec, Operator, PStringLengthWidth, StrengthModifier, TypeKind,
131 Value,
132};
133
134// Re-export evaluator types for convenience
135pub use evaluator::{EvaluationContext, RuleMatch};
136
137// Re-export error types for convenience
138pub use error::{EvaluationError, LibmagicError, ParseError};
139
140/// Result type for library operations
141pub type Result<T> = std::result::Result<T, LibmagicError>;
142
143impl From<crate::io::IoError> for LibmagicError {
144 fn from(err: crate::io::IoError) -> Self {
145 // Preserve the structured error message (includes path and operation context)
146 LibmagicError::FileError(err.to_string())
147 }
148}
149
150/// Main interface for magic rule database
151#[derive(Debug)]
152pub struct MagicDatabase {
153 /// Named subroutine definitions extracted from magic file `name` rules,
154 /// keyed by identifier. The evaluator consults this table when a rule of
155 /// type `TypeKind::Meta(MetaType::Use(name))` is reached.
156 name_table: std::sync::Arc<crate::parser::name_table::NameTable>,
157 /// Top-level rules as a shared immutable slice. This is the primary rule
158 /// storage for the database. Passed through the evaluation context as part
159 /// of the rule environment so whole-database operations (e.g. `indirect`)
160 /// can re-enter at the root without re-sorting or cloning the rule tree.
161 root_rules: std::sync::Arc<[MagicRule]>,
162 config: EvaluationConfig,
163 /// Optional path to the source magic file or directory from which rules were loaded.
164 /// This is used for debugging and logging purposes.
165 source_path: Option<PathBuf>,
166 /// Cached MIME type mapper to avoid rebuilding the lookup table on every evaluation
167 mime_mapper: mime::MimeMapper,
168}
169
170impl MagicDatabase {
171 /// Create a database using built-in magic rules.
172 ///
173 /// Loads magic rules that are compiled into the library binary at build time
174 /// from `src/builtin_rules.magic`. These rules provide high-confidence detection
175 /// for common file types including executables (ELF, PE/DOS), archives (ZIP, TAR,
176 /// GZIP), images (JPEG, PNG, GIF, BMP), and documents (PDF).
177 ///
178 /// # Security
179 ///
180 /// This constructor uses [`EvaluationConfig::default()`], which leaves
181 /// `timeout_ms` unset (unbounded). When processing untrusted input
182 /// (adversarial file buffers, large uploads, etc.), prefer
183 /// [`MagicDatabase::with_builtin_rules_and_config`] with
184 /// [`EvaluationConfig::performance()`] (which sets a 1-second timeout)
185 /// or construct a config explicitly with a non-`None` timeout sized
186 /// for your workload. The `Default` impl intentionally targets CLI
187 /// one-shot usage rather than long-running services.
188 ///
189 /// # Thread safety
190 ///
191 /// `MagicDatabase` is `Send + Sync` and holds no interior mutability,
192 /// so an `Arc<MagicDatabase>` can be shared across threads for
193 /// parallel file scanning. A fresh evaluation context is constructed
194 /// per `evaluate_buffer` / `evaluate_file` call, so concurrent calls
195 /// do not interfere.
196 ///
197 /// # Errors
198 ///
199 /// Currently always returns `Ok`. In future implementations, this may return
200 /// an error if the built-in rules fail to load or validate.
201 ///
202 /// # Examples
203 ///
204 /// ```rust,no_run
205 /// use libmagic_rs::MagicDatabase;
206 ///
207 /// let db = MagicDatabase::with_builtin_rules()?;
208 /// let result = db.evaluate_buffer(b"\x7fELF")?;
209 /// // Returns actual file type detection (e.g., "ELF")
210 /// # Ok::<(), Box<dyn std::error::Error>>(())
211 /// ```
212 pub fn with_builtin_rules() -> Result<Self> {
213 Self::with_builtin_rules_and_config(EvaluationConfig::default())
214 }
215
216 /// Create database with built-in rules and custom configuration.
217 ///
218 /// Loads built-in magic rules compiled at build time and applies the specified
219 /// evaluation configuration (e.g., custom timeout settings).
220 ///
221 /// # Security
222 ///
223 /// For untrusted input (adversarial file buffers, web uploads, mail
224 /// scanning), pass a config with an explicit timeout such as
225 /// [`EvaluationConfig::performance()`]. The default config has
226 /// `timeout_ms = None` which leaves evaluation unbounded; see the
227 /// rationale on [`EvaluationConfig::default`].
228 ///
229 /// # Arguments
230 ///
231 /// * `config` - Custom evaluation configuration to use with the built-in rules
232 ///
233 /// # Errors
234 ///
235 /// Returns `LibmagicError` if the configuration is invalid (e.g., timeout is zero).
236 ///
237 /// # Examples
238 ///
239 /// ```rust,no_run
240 /// use libmagic_rs::{MagicDatabase, EvaluationConfig};
241 ///
242 /// // Prefer the performance() preset over default() when processing
243 /// // untrusted input. default() has no timeout by design.
244 /// let config = EvaluationConfig::performance();
245 /// let db = MagicDatabase::with_builtin_rules_and_config(config)?;
246 /// # Ok::<(), Box<dyn std::error::Error>>(())
247 /// ```
248 pub fn with_builtin_rules_and_config(config: EvaluationConfig) -> Result<Self> {
249 config.validate()?;
250 let mut rules = crate::builtin_rules::get_builtin_rules();
251 crate::evaluator::strength::sort_rules_by_strength_recursive(&mut rules);
252 let root_rules: std::sync::Arc<[MagicRule]> =
253 std::sync::Arc::from(rules.into_boxed_slice());
254 Ok(Self {
255 name_table: std::sync::Arc::new(crate::parser::name_table::NameTable::empty()),
256 root_rules,
257 config,
258 source_path: None,
259 mime_mapper: mime::MimeMapper::new(),
260 })
261 }
262
263 /// Load magic rules from a file
264 ///
265 /// # Security
266 ///
267 /// This constructor uses [`EvaluationConfig::default()`], which
268 /// leaves `timeout_ms` unset. See the security note on
269 /// [`Self::with_builtin_rules`] for the implications and prefer
270 /// [`Self::load_from_file_with_config`] with an explicit timeout
271 /// when processing untrusted input.
272 ///
273 /// # Arguments
274 ///
275 /// * `path` - Path to the magic file to load
276 ///
277 /// # Errors
278 ///
279 /// Returns `LibmagicError::IoError` if the file cannot be read.
280 /// Returns `LibmagicError::ParseError` if the magic file format is invalid.
281 ///
282 /// # Examples
283 ///
284 /// ```rust,no_run
285 /// use libmagic_rs::MagicDatabase;
286 ///
287 /// let db = MagicDatabase::load_from_file("magic.db")?;
288 /// # Ok::<(), Box<dyn std::error::Error>>(())
289 /// ```
290 pub fn load_from_file<P: AsRef<Path>>(path: P) -> Result<Self> {
291 Self::load_from_file_with_config(path, EvaluationConfig::default())
292 }
293
294 /// Load from file with custom config (e.g., timeout).
295 ///
296 /// # Security
297 ///
298 /// For untrusted input, pass [`EvaluationConfig::performance()`] or
299 /// a config with an explicit non-`None` `timeout_ms`. See
300 /// [`Self::with_builtin_rules`] for the full rationale.
301 ///
302 /// # Errors
303 ///
304 /// Returns error if file cannot be read, parsed, or config is invalid
305 pub fn load_from_file_with_config<P: AsRef<Path>>(
306 path: P,
307 config: EvaluationConfig,
308 ) -> Result<Self> {
309 config.validate()?;
310 let parsed = parser::load_magic_file(path.as_ref()).map_err(|e| match e {
311 ParseError::IoError(io_err) => LibmagicError::IoError(io_err),
312 other => LibmagicError::ParseError(other),
313 })?;
314 let parser::ParsedMagic {
315 mut rules,
316 mut name_table,
317 } = parsed;
318 crate::evaluator::strength::sort_rules_by_strength_recursive(&mut rules);
319 // Each named subroutine body must be sorted by the same strength
320 // ordering so evaluation of a `use` site is deterministic and
321 // matches the ordering applied to top-level rules.
322 name_table.sort_subroutines(|rules| {
323 crate::evaluator::strength::sort_rules_by_strength_recursive(rules);
324 });
325
326 let root_rules: std::sync::Arc<[MagicRule]> =
327 std::sync::Arc::from(rules.into_boxed_slice());
328 Ok(Self {
329 name_table: std::sync::Arc::new(name_table),
330 root_rules,
331 config,
332 source_path: Some(path.as_ref().to_path_buf()),
333 mime_mapper: mime::MimeMapper::new(),
334 })
335 }
336
337 /// Evaluate magic rules against a file
338 ///
339 /// # Arguments
340 ///
341 /// * `path` - Path to the file to evaluate
342 ///
343 /// # Errors
344 ///
345 /// Returns `LibmagicError::IoError` if the file cannot be accessed.
346 /// Returns `LibmagicError::EvaluationError` if rule evaluation fails.
347 ///
348 /// # Security
349 ///
350 /// This method has a time-of-check/time-of-use (TOCTOU) window between
351 /// resolving the path and memory-mapping the file
352 /// ([CWE-367](https://cwe.mitre.org/data/definitions/367.html)).
353 /// Internally, `evaluate_file` first calls `std::fs::metadata(path)` to
354 /// detect the empty-file case, then opens and memory-maps the file via
355 /// [`io::FileBuffer::new`], which itself re-validates file metadata
356 /// (regular file, size bounds) before calling `create_memory_mapping`.
357 /// Between these validation steps and the final `mmap` call, the path
358 /// may be swapped (for example, via a symlink replacement or rename)
359 /// by another process. The content that gets mapped may therefore
360 /// differ from the file that passed validation.
361 ///
362 /// The I/O layer mitigates the common shapes of this attack by
363 /// canonicalizing the path and rejecting special file types, and the
364 /// mapping itself is read-only, so a successful exploit cannot corrupt
365 /// the victim file. The residual risk is that `evaluate_file` may
366 /// classify a different file than the caller intended.
367 ///
368 /// **For adversarial or untrusted environments, prefer
369 /// [`MagicDatabase::evaluate_buffer`]**: load the bytes yourself using
370 /// whatever resource-bounded, TOCTOU-aware I/O strategy your
371 /// application requires (e.g., `openat` with `O_NOFOLLOW`, holding an
372 /// open file descriptor across validation and read), then pass the
373 /// in-memory slice directly to `evaluate_buffer`. See
374 /// [the security assurance case](https://evilbit-labs.github.io/libmagic-rs/security-assurance.html)
375 /// for the residual-risk discussion.
376 ///
377 /// # Examples
378 ///
379 /// ```rust,no_run
380 /// use libmagic_rs::MagicDatabase;
381 ///
382 /// let db = MagicDatabase::load_from_file("magic.db")?;
383 /// let result = db.evaluate_file("sample.bin")?;
384 /// println!("File type: {}", result.description);
385 /// # Ok::<(), Box<dyn std::error::Error>>(())
386 /// ```
387 pub fn evaluate_file<P: AsRef<Path>>(&self, path: P) -> Result<EvaluationResult> {
388 use crate::io::FileBuffer;
389 use std::fs;
390 use std::time::Instant;
391
392 let start_time = Instant::now();
393 let path = path.as_ref();
394
395 // Check if file is empty - if so, evaluate as empty buffer
396 // This allows empty files to be processed like any other file
397 let file_metadata = fs::metadata(path)?;
398 let file_size = file_metadata.len();
399
400 if file_size == 0 {
401 // Empty file - evaluate as empty buffer but preserve file metadata
402 let mut result = self.evaluate_buffer_internal(b"", start_time)?;
403 result.metadata.file_size = 0;
404 result.metadata.magic_file.clone_from(&self.source_path);
405 return Ok(result);
406 }
407
408 // Load the file into memory. Reuse the metadata we just read instead
409 // of having FileBuffer::new call canonicalize+metadata again.
410 let file_buffer = FileBuffer::from_path_and_metadata(path, &file_metadata)?;
411 let buffer = file_buffer.as_slice();
412
413 // Route the evaluation through `evaluate_buffer_internal` so the
414 // rule environment (name table + root rules) is attached to the
415 // context identically for in-memory and on-disk paths.
416 let mut result = self.evaluate_buffer_internal(buffer, start_time)?;
417 result.metadata.file_size = file_size;
418 Ok(result)
419 }
420
421 /// Evaluate magic rules against an in-memory buffer
422 ///
423 /// This method evaluates a byte buffer directly without reading from disk,
424 /// which is useful for stdin input or pre-loaded data.
425 ///
426 /// # Arguments
427 ///
428 /// * `buffer` - Byte buffer to evaluate
429 ///
430 /// # Errors
431 ///
432 /// Returns `LibmagicError::EvaluationError` if rule evaluation fails.
433 ///
434 /// # Examples
435 ///
436 /// ```rust,no_run
437 /// use libmagic_rs::MagicDatabase;
438 ///
439 /// let db = MagicDatabase::load_from_file("/usr/share/misc/magic")?;
440 /// let buffer = b"test data";
441 /// let result = db.evaluate_buffer(buffer)?;
442 /// println!("Buffer type: {}", result.description);
443 /// # Ok::<(), Box<dyn std::error::Error>>(())
444 /// ```
445 pub fn evaluate_buffer(&self, buffer: &[u8]) -> Result<EvaluationResult> {
446 use std::time::Instant;
447 self.evaluate_buffer_internal(buffer, Instant::now())
448 }
449
450 /// Internal buffer evaluation with externally provided start time
451 fn evaluate_buffer_internal(
452 &self,
453 buffer: &[u8],
454 start_time: std::time::Instant,
455 ) -> Result<EvaluationResult> {
456 use crate::evaluator::{EvaluationContext, RuleEnvironment, evaluate_rules};
457
458 let file_size = buffer.len() as u64;
459
460 // Validate config once at the entry point to match the previous
461 // behavior of `evaluate_rules_with_config`.
462 self.config.validate()?;
463
464 // Reset the thread-local regex compile cache so it is bounded to
465 // the lifetime of a single top-level evaluation call.
466 crate::evaluator::types::regex::reset_regex_cache();
467
468 let env = std::sync::Arc::new(RuleEnvironment {
469 name_table: self.name_table.clone(),
470 root_rules: self.root_rules.clone(),
471 });
472
473 let mut context = EvaluationContext::new(self.config.clone()).with_rule_env(env);
474
475 // `evaluate_rules` returns `Ok(vec![])` for an empty rule list,
476 // so no `is_empty()` guard is needed here.
477 let matches = evaluate_rules(&self.root_rules, buffer, &mut context)?;
478
479 Ok(self.build_result(matches, file_size, start_time))
480 }
481
482 /// Build an `EvaluationResult` from match results, file size, and start time.
483 ///
484 /// This is shared between `evaluate_file` and `evaluate_buffer_internal` to
485 /// avoid duplicating the result-construction logic.
486 fn build_result(
487 &self,
488 matches: Vec<evaluator::RuleMatch>,
489 file_size: u64,
490 start_time: std::time::Instant,
491 ) -> EvaluationResult {
492 let (description, confidence) = if matches.is_empty() {
493 ("data".to_string(), 0.0)
494 } else {
495 (
496 Self::concatenate_messages(&matches),
497 matches.first().map_or(0.0, |m| m.confidence),
498 )
499 };
500
501 let mime_type = if self.config.enable_mime_types {
502 self.mime_mapper
503 .get_mime_type(&description)
504 .map(String::from)
505 } else {
506 None
507 };
508
509 EvaluationResult {
510 description,
511 mime_type,
512 confidence,
513 matches,
514 metadata: EvaluationMetadata {
515 file_size,
516 evaluation_time_ms: start_time.elapsed().as_secs_f64() * 1000.0,
517 rules_evaluated: self.root_rules.len(),
518 magic_file: self.source_path.clone(),
519 timed_out: false,
520 },
521 }
522 }
523
524 /// Concatenate match messages following libmagic behavior
525 ///
526 /// Each match's `message` is first run through
527 /// [`crate::output::format::format_magic_message`], which substitutes
528 /// printf-style specifiers (`%lld`, `%02x`, `%s`, etc.) with the
529 /// rule's read value. The resulting rendered strings are then joined
530 /// with spaces, except when a rendered string starts with the
531 /// backspace character (`\b`, U+0008) which suppresses both the
532 /// separating space and the backspace itself (GOTCHAS.md S14.1).
533 ///
534 /// The backspace check runs on the *post-substitution* text so rules
535 /// like `\b, version %s` compose correctly once the specifier has been
536 /// rendered.
537 fn concatenate_messages(matches: &[evaluator::RuleMatch]) -> String {
538 use crate::output::format::format_magic_message;
539
540 let capacity: usize = matches.iter().map(|m| m.message.len() + 1).sum();
541 let mut result = String::with_capacity(capacity);
542 for m in matches {
543 let rendered = format_magic_message(&m.message, &m.value, &m.type_kind);
544 if let Some(rest) = rendered.strip_prefix('\u{0008}') {
545 // Backspace suppresses the space and the character itself
546 result.push_str(rest);
547 } else if !result.is_empty() {
548 result.push(' ');
549 result.push_str(&rendered);
550 } else {
551 result.push_str(&rendered);
552 }
553 }
554 result
555 }
556
557 /// Returns the evaluation configuration used by this database.
558 ///
559 /// This provides read-only access to the evaluation configuration for
560 /// callers that need to inspect resource limits or evaluation options.
561 #[must_use]
562 pub fn config(&self) -> &EvaluationConfig {
563 &self.config
564 }
565
566 /// Returns the path from which magic rules were loaded.
567 ///
568 /// This method returns the source path that was used to load the magic rules
569 /// into this database. It is useful for debugging, logging, and tracking the
570 /// origin of magic rules.
571 ///
572 /// # Returns
573 ///
574 /// - `Some(&Path)` - If the database was loaded from a file or directory using
575 /// [`load_from_file()`](Self::load_from_file)
576 /// - `None` - If the database was constructed programmatically or the source
577 /// path was not recorded
578 ///
579 /// # Examples
580 ///
581 /// ```rust,no_run
582 /// use libmagic_rs::MagicDatabase;
583 ///
584 /// let db = MagicDatabase::load_from_file("/usr/share/misc/magic")?;
585 /// if let Some(path) = db.source_path() {
586 /// println!("Rules loaded from: {}", path.display());
587 /// }
588 /// # Ok::<(), Box<dyn std::error::Error>>(())
589 /// ```
590 #[must_use]
591 pub fn source_path(&self) -> Option<&Path> {
592 self.source_path.as_deref()
593 }
594}
595
596/// Metadata about the evaluation process
597///
598/// Contains diagnostic information about how the evaluation was performed,
599/// including performance metrics and statistics about rule processing.
600///
601/// # Examples
602///
603/// ```
604/// use libmagic_rs::EvaluationMetadata;
605/// use std::path::PathBuf;
606///
607/// let metadata = EvaluationMetadata {
608/// file_size: 8192,
609/// evaluation_time_ms: 2.5,
610/// rules_evaluated: 42,
611/// magic_file: Some(PathBuf::from("/usr/share/misc/magic")),
612/// timed_out: false,
613/// };
614///
615/// assert_eq!(metadata.file_size, 8192);
616/// assert!(!metadata.timed_out);
617/// ```
618#[derive(Debug, Clone, Serialize, Deserialize)]
619pub struct EvaluationMetadata {
620 /// Size of the analyzed file or buffer in bytes
621 pub file_size: u64,
622 /// Time taken to evaluate rules in milliseconds
623 pub evaluation_time_ms: f64,
624 /// Number of top-level rules that were evaluated
625 pub rules_evaluated: usize,
626 /// Path to the magic file used, or None for built-in rules
627 #[serde(skip_serializing_if = "Option::is_none", default)]
628 pub magic_file: Option<PathBuf>,
629 /// Whether evaluation was stopped due to timeout
630 pub timed_out: bool,
631}
632
633impl Default for EvaluationMetadata {
634 fn default() -> Self {
635 Self {
636 file_size: 0,
637 evaluation_time_ms: 0.0,
638 rules_evaluated: 0,
639 magic_file: None,
640 timed_out: false,
641 }
642 }
643}
644
645/// Result of magic rule evaluation
646///
647/// Contains the file type description, optional MIME type, confidence score,
648/// individual match details, and evaluation metadata.
649///
650/// # Relationship to [`crate::output::EvaluationResult`]
651///
652/// This is the **library-facing** result type returned by [`MagicDatabase::evaluate_file`]
653/// and [`MagicDatabase::evaluate_buffer`].
654/// It carries a rolled-up description, MIME type, and confidence score along with
655/// raw [`evaluator::RuleMatch`] values. It intentionally does **not** carry the
656/// analyzed filename or a surface-level error string, because those are caller
657/// concerns (a caller may evaluate an in-memory buffer that has no filename).
658///
659/// The parallel type [`crate::output::EvaluationResult`] is the **output-facing**
660/// result used by the CLI and JSON/text formatters. It adds `filename` and
661/// `error`, carries enriched [`crate::output::MatchResult`] values (with
662/// extracted tags), and uses `u32` counters in its metadata to match the JSON
663/// output schema.
664///
665/// The two types are **intentionally distinct** — do not try to unify them.
666/// Convert library → output explicitly via
667/// [`crate::output::EvaluationResult::from_library_result`], which is the single
668/// named conversion point. Any drift between the two hierarchies should be
669/// resolved there, not by back-channel field copying in call sites.
670///
671/// # Examples
672///
673/// ```
674/// use libmagic_rs::{EvaluationResult, EvaluationMetadata};
675///
676/// let result = EvaluationResult {
677/// description: "ELF 64-bit executable".to_string(),
678/// mime_type: Some("application/x-executable".to_string()),
679/// confidence: 0.9,
680/// matches: vec![],
681/// metadata: EvaluationMetadata::default(),
682/// };
683///
684/// assert_eq!(result.description, "ELF 64-bit executable");
685/// assert!(result.confidence > 0.5);
686/// ```
687#[derive(Debug, Clone, Serialize, Deserialize)]
688pub struct EvaluationResult {
689 /// Human-readable file type description
690 ///
691 /// This is the concatenated message from all matching rules,
692 /// following libmagic behavior where hierarchical matches
693 /// are joined with spaces (unless backspace character is used).
694 pub description: String,
695 /// Optional MIME type for the detected file type
696 ///
697 /// Only populated when `enable_mime_types` is set in the configuration.
698 /// Omitted from the serialized form when unset (rather than emitted
699 /// as `"mime_type": null`) so downstream JSON consumers can treat
700 /// presence as the "MIME type is known" indicator.
701 #[serde(skip_serializing_if = "Option::is_none", default)]
702 pub mime_type: Option<String>,
703 /// Confidence score (0.0 to 1.0)
704 ///
705 /// Based on the depth of the match in the rule hierarchy.
706 /// Deeper matches indicate more specific identification.
707 pub confidence: f64,
708 /// Individual match results from rule evaluation
709 ///
710 /// Contains details about each rule that matched, including
711 /// offset, matched value, and per-match confidence.
712 pub matches: Vec<evaluator::RuleMatch>,
713 /// Metadata about the evaluation process
714 pub metadata: EvaluationMetadata,
715}
716
717#[cfg(test)]
718mod tests;