fallow_extract/
lib.rs

1//! Parsing and extraction engine for fallow codebase intelligence.
2//!
3//! This crate handles all file parsing: JS/TS via Oxc, Vue/Svelte SFC extraction,
4//! Astro frontmatter, MDX import/export extraction, CSS Module class name extraction,
5//! HTML asset reference extraction, and incremental caching of parse results.
6
7#![warn(missing_docs)]
8#![cfg_attr(not(test), deny(clippy::disallowed_methods))]
9#![cfg_attr(
10    test,
11    allow(
12        clippy::unwrap_used,
13        clippy::expect_used,
14        reason = "tests use unwrap and expect to keep fixture setup concise"
15    )
16)]
17
18pub mod alchemy;
19mod asset_url;
20pub mod astro;
21mod binding_references;
22pub mod cache;
23pub(crate) mod complexity;
24mod complexity_names;
25mod component_contracts;
26pub(crate) mod cost;
27pub mod css;
28pub mod css_classes;
29pub mod css_in_js;
30mod css_mask;
31mod css_masked_bytes;
32pub mod css_metrics;
33mod css_module_classes;
34pub mod effect_facts;
35pub mod federation_runtime;
36pub mod flags;
37mod function_body;
38pub(crate) mod function_hash;
39pub mod glimmer;
40pub(crate) mod graphql;
41pub(crate) mod html;
42pub(crate) mod iconify;
43pub mod inventory;
44pub mod js_text;
45mod jsdoc_attach;
46mod jsdoc_deprecated;
47mod literal_membership;
48pub mod mdx;
49mod module_info;
50mod module_parse;
51mod nesting;
52pub mod og_image;
53mod parse;
54mod parse_degradation;
55pub mod sfc;
56pub mod sfc_css;
57mod sfc_props;
58mod sfc_template;
59pub mod similar_code;
60mod source_map;
61mod stack;
62pub mod suppress;
63/// Tailwind CSS arbitrary-value detection.
64pub mod tailwind;
65pub(crate) mod template_complexity;
66mod template_expression_scan;
67mod template_usage;
68/// Visitor utilities for AST extraction.
69pub mod visitor;
70
71use std::path::Path;
72use std::sync::atomic::{AtomicBool, Ordering};
73
74use rayon::prelude::*;
75
76use cache::CacheStore;
77use fallow_types::discover::{DiscoveredFile, FileId};
78
79pub use fallow_types::extract::{
80    AngularComponentFieldArrayTypeFact, AngularTemplateMemberAccessFact, AngularThisSpreadFact,
81    ClassHeritageInfo, ClassThisMemberAccessFact, ClassThisWholeObjectUseFact,
82    ComputedEnumKeyUseFact, DefaultImportWholeObjectUseFact, DynamicCustomElementRenderFact,
83    DynamicImportInfo, DynamicImportPattern, ExportInfo, ExportName,
84    ExportedObjectInstancePropertyFact, FactoryCallMemberAccessFact, FactoryFnMemberAccessFact,
85    FactoryFnWholeObjectFact, FactoryPropertySource, FactoryReturnExport,
86    FactoryReturnObjectPropertyAccessFact, FactoryReturnObjectPropertySourceFact,
87    FactoryReturnObjectShapeExport, FlagPatterns, FluentChainMemberAccessFact,
88    FluentChainNewMemberAccessFact, ImportInfo, ImportedName, InstanceExportBindingFact,
89    LocalTypeDeclaration, MemberAccess, MemberInfo, MemberKind, ModuleInfo, ModuleLoadMechanism,
90    ParseResult, PlaywrightConditionalFixtureDefinitionFact, PlaywrightFixtureAliasFact,
91    PlaywrightFixtureDefinitionFact, PlaywrightFixtureTypeFact, PlaywrightFixtureUseFact,
92    PlaywrightTestBindingFact, PublicSignatureTypeReference, QualifiedClassMemberAccessFact,
93    ReExportInfo, RequireCallInfo, RequiredTypeMemberFact, SemanticFact, SourceParseDegradation,
94    SourceReadFailure, StringEnumMemberValueFact, TypeAliasSurfaceTargetFact, TypeMemberTypeEntry,
95    TypedPropertyMemberAccessFact, VisibilityTag, VitestModuleMockAction,
96    VitestModuleMockOperationFact, compute_line_offsets,
97};
98
99pub use astro::{
100    extract_astro_frontmatter, extract_astro_style_regions, extract_astro_template_regions,
101};
102pub use css::{
103    StylesheetTokens, ThemeScan, ThemeTokenDef, extract_apply_tokens, extract_apply_tokens_located,
104    extract_css_module_exports, extract_css_var_reads_located, scan_stylesheet_tokens,
105    scan_theme_blocks,
106};
107pub use css_classes::{
108    MarkupClassScan, MarkupClassToken, is_edit_distance_one, is_typo_edit, scan_markup_class_tokens,
109};
110pub use css_in_js::{
111    ConsumerQuery, CssInJsObjectSheets, CssInJsToken, CssInJsTokenDef, CssInJsTokenOrigin,
112    TokenConsumerHit, css_in_js_consumer_scan, css_in_js_object_sheets, css_in_js_theme_token_defs,
113    css_in_js_token_defs, css_in_js_virtual_stylesheet,
114};
115pub use css_metrics::{compute_css_analytics, parse_css_color_rgb};
116pub use glimmer::{is_glimmer_file, strip_glimmer_templates};
117pub use js_text::JsText;
118pub use mdx::{extract_mdx_statements, extract_mdx_statements_mapped};
119pub use nesting::{
120    MAX_NESTING_DEPTH, exceeds_nesting_limit, parse_bounded, parse_bounded_or_default,
121    parse_style_bounded,
122};
123pub use sfc::{
124    SourceRegion, extract_sfc_scripts, extract_sfc_styles, extract_sfc_template_regions,
125    is_sfc_file, sfc_merged_script_source,
126};
127pub use sfc_css::{
128    scoped_unused_classes, sfc_preprocessor_virtual_stylesheet, sfc_virtual_stylesheet,
129};
130pub use similar_code::extract_similar_code_functions;
131pub use source_map::{ExtractionResult, ScriptSection, embedded_script_sections};
132pub use stack::register_thread_stack;
133pub use tailwind::{TailwindArbitraryUse, scan_tailwind_arbitrary_values};
134
135#[expect(
136    clippy::expect_used,
137    reason = "static regex patterns are hard-coded analyzer invariants covered by extraction tests"
138)]
139fn static_regex(pattern: &str) -> regex::Regex {
140    regex::Regex::new(pattern).expect("static regex pattern should compile")
141}
142
143pub use module_parse::{parse_source_to_module, parse_source_to_module_with_flags};
144
145/// Leading UTF-8 byte order mark codepoint.
146///
147/// Windows editors (Notepad, older VS settings, some IDE plugins) emit a UTF-8
148/// BOM at the start of source files. fallow's contract is "UTF-8 with or
149/// without BOM; line offsets are computed against the post-BOM view; the BOM,
150/// if present on input, is preserved on output by `fallow fix`."
151const BOM_CHAR: char = '\u{FEFF}';
152// Small, cache-hot inputs are faster on one thread than through Rayon setup.
153// Larger file sets still use parallel parsing where parse work dominates.
154const PARALLEL_PARSE_FILE_THRESHOLD: usize = 32;
155
156/// Strip the leading UTF-8 BOM if present.
157///
158/// Called at every file-read entry point in this crate so the rest of the
159/// pipeline (content hash, `compute_line_offsets`, oxc parser, downstream
160/// analyses) sees a consistent post-BOM view. Mirrors the
161/// `fallow_config` layer (`config_writer.rs::BOM`) so config-shaped sources
162/// and source-code-shaped sources are processed symmetrically. See issue #475.
163#[must_use]
164fn strip_bom(source: &str) -> &str {
165    source.strip_prefix(BOM_CHAR).unwrap_or(source)
166}
167
168/// Parse all files, extracting imports and exports.
169///
170/// Small file sets use a sequential fast path to avoid parallel scheduling
171/// overhead; larger file sets use parallel extraction.
172/// Uses the cache to skip reparsing files whose content hasn't changed.
173///
174/// When `need_complexity` is true, per-function cyclomatic/cognitive complexity
175/// metrics are computed during parsing (needed by the `health` command).
176/// Pass `false` for dead-code analysis where complexity data is unused.
177///
178/// Flag detection uses the built-in patterns only. A caller with a resolved
179/// config uses [`parse_all_files_cancellable`] with the config's patterns,
180/// because the cache keys on them.
181pub fn parse_all_files(
182    files: &[DiscoveredFile],
183    cache: Option<&CacheStore>,
184    need_complexity: bool,
185) -> ParseResult {
186    parse_all_files_cancellable(
187        files,
188        cache,
189        need_complexity,
190        None,
191        &FlagPatterns::default(),
192    )
193}
194
195/// Parse all files, abandoning the remaining ones once `cancellation` is set.
196///
197/// Rayon's `map`/`collect` cannot short-circuit, so cancellation makes the
198/// per-file body a no-op instead of stopping the iteration: the scheduled
199/// items still drain, but at one atomic load each. The returned
200/// [`ParseResult`] is therefore truncated whenever the token flipped, and
201/// callers must treat a set token as a failed run rather than as a project
202/// with fewer modules.
203///
204/// `flag_patterns` are the user flag patterns that detection applies on top
205/// of the built-in ones. They must match the patterns the cache was keyed on.
206pub fn parse_all_files_cancellable(
207    files: &[DiscoveredFile],
208    cache: Option<&CacheStore>,
209    need_complexity: bool,
210    cancellation: Option<&AtomicBool>,
211    flag_patterns: &FlagPatterns,
212) -> ParseResult {
213    let parse_one = |file: &DiscoveredFile| {
214        if cancellation.is_some_and(|cancelled| cancelled.load(Ordering::SeqCst)) {
215            return ParseFileResult::default();
216        }
217        parse_single_file_cached(file, cache, need_complexity, flag_patterns)
218    };
219    let results: Vec<ParseFileResult> = if files.len() <= PARALLEL_PARSE_FILE_THRESHOLD {
220        files.iter().map(parse_one).collect()
221    } else {
222        files.par_iter().map(parse_one).collect()
223    };
224
225    let mut modules = Vec::with_capacity(results.len());
226    let mut read_failures = Vec::new();
227    let mut parse_degradations = Vec::new();
228    let mut hits = 0usize;
229    let mut misses = 0usize;
230    let mut parse_cpu_nanos = 0u64;
231    let mut files_read = 0u64;
232    let mut source_bytes_read = 0u64;
233    let mut css_masked_bytes = 0u64;
234
235    // `results` is a positional map over `files`, so zipping recovers the path
236    // for a module without carrying one on `ModuleInfo`.
237    for (file, result) in files.iter().zip(results) {
238        hits += result.cache_hits;
239        misses += result.cache_misses;
240        parse_cpu_nanos = parse_cpu_nanos.saturating_add(result.parse_cpu_nanos);
241        if let Some(bytes) = result.source_bytes_read {
242            files_read += 1;
243            source_bytes_read += bytes;
244        }
245        css_masked_bytes += result.css_masked_bytes;
246        if let Some(module) = result.module {
247            if module.parse_error_count > 0 {
248                parse_degradations.push(SourceParseDegradation {
249                    file_id: module.file_id,
250                    path: file.path.clone(),
251                    error_count: module.parse_error_count,
252                    panicked: module.parse_panicked,
253                    nesting_limit: module.parse_nesting_too_deep.then_some(MAX_NESTING_DEPTH),
254                });
255            }
256            modules.push(module);
257        }
258        if let Some(failure) = result.read_failure {
259            read_failures.push(failure);
260        }
261    }
262
263    if hits > 0 || misses > 0 {
264        tracing::info!(
265            cache_hits = hits,
266            cache_misses = misses,
267            "incremental cache stats"
268        );
269    }
270
271    ParseResult {
272        modules,
273        read_failures,
274        parse_degradations,
275        cache_hits: hits,
276        cache_misses: misses,
277        parse_cpu_ms: parse_cpu_nanos as f64 / 1_000_000.0,
278        files_read,
279        source_bytes_read,
280        css_masked_bytes,
281    }
282}
283
284#[derive(Default)]
285struct ParseFileResult {
286    module: Option<ModuleInfo>,
287    read_failure: Option<SourceReadFailure>,
288    cache_hits: usize,
289    cache_misses: usize,
290    parse_cpu_nanos: u64,
291    /// Source bytes read from disk for this file, or `None` when the file was
292    /// served from cache metadata without a read.
293    source_bytes_read: Option<u64>,
294    /// Source bytes that the CSS comment mask read during the parse.
295    css_masked_bytes: u64,
296}
297
298impl ParseFileResult {
299    fn cache_hit(module: ModuleInfo) -> Self {
300        Self {
301            module: Some(module),
302            read_failure: None,
303            cache_hits: 1,
304            cache_misses: 0,
305            parse_cpu_nanos: 0,
306            source_bytes_read: None,
307            css_masked_bytes: 0,
308        }
309    }
310
311    fn cache_miss(module: ModuleInfo, parse_cpu_nanos: u64) -> Self {
312        Self {
313            module: Some(module),
314            read_failure: None,
315            cache_hits: 0,
316            cache_misses: 1,
317            parse_cpu_nanos,
318            source_bytes_read: None,
319            css_masked_bytes: 0,
320        }
321    }
322
323    const fn with_source_bytes_read(mut self, bytes: usize) -> Self {
324        self.source_bytes_read = Some(bytes as u64);
325        self
326    }
327
328    fn read_failure(file: &DiscoveredFile, error: &std::io::Error) -> Self {
329        Self {
330            module: None,
331            read_failure: Some(SourceReadFailure {
332                file_id: file.id,
333                path: file.path.clone(),
334                error: error.to_string(),
335            }),
336            cache_hits: 0,
337            cache_misses: 0,
338            parse_cpu_nanos: 0,
339            source_bytes_read: None,
340            css_masked_bytes: 0,
341        }
342    }
343}
344
345/// Parse a single file, consulting the cache first.
346///
347/// Cache validation strategy (fast path -> slow path):
348/// 1. Open the file so unreadable sources cannot use stale cached analysis
349/// 2. Read mtime + ctime + size from the open handle
350/// 3. If all three match the cached entry -> cache hit, return immediately
351/// 4. Otherwise -> read file, compute content hash
352/// 5. If content hash matches cached entry -> cache hit (file was rewritten or
353///    `touch`ed but its content is unchanged)
354/// 6. Otherwise -> cache miss, full parse
355///
356/// Step 3 requires ctime as well as mtime because mtime is writer-controlled:
357/// a same-length rewrite whose mtime is restored (`touch -r`, a codemod, a
358/// `git checkout` of an equal-length revision) leaves `(mtime, size)`
359/// unchanged, and serving the cached module for it means reporting the OLD
360/// file's unused exports with an auto-fixable `remove-export` action. A file
361/// whose ctime moved falls through to step 4 and still hits on the content
362/// hash, so the cost of the stricter gate is one read, not a reparse.
363///
364/// Step 3 also needs a stored fingerprint that settled before the run that
365/// wrote it read the file. A same-length write in the same timestamp tick as
366/// the previous write keeps mtime, ctime and size, so the cache writer stores
367/// a recent fingerprint without its ctime
368/// (`SourceFingerprint::for_content_read_at`). Such an entry never matches at
369/// step 3, and the next run compares the content hash.
370fn parse_single_file_cached(
371    file: &DiscoveredFile,
372    cache: Option<&CacheStore>,
373    need_complexity: bool,
374    flag_patterns: &FlagPatterns,
375) -> ParseFileResult {
376    let cached_by_path = cache.and_then(|store| store.get_by_path_only(&file.path));
377
378    if let Some(cached) = cached_by_path
379        && cached.file_size == file.size_bytes
380    {
381        let source_file = match std::fs::File::open(&file.path) {
382            Ok(source_file) => source_file,
383            Err(error) => return ParseFileResult::read_failure(file, &error),
384        };
385        if let Ok(metadata) = source_file.metadata()
386            && metadata.len() == cached.file_size
387        {
388            let fingerprint =
389                fallow_types::source_fingerprint::SourceFingerprint::from_metadata(&metadata);
390            if cached.source_fingerprint() == fingerprint
391                && fingerprint.is_trustworthy_without_content()
392                && (!need_complexity || cached.complexity_extracted)
393            {
394                return ParseFileResult::cache_hit(cache::cached_to_module_opts(
395                    cached,
396                    file.id,
397                    need_complexity,
398                ));
399            }
400        }
401    }
402
403    let raw = match std::fs::read_to_string(&file.path) {
404        Ok(raw) => raw,
405        Err(error) => return ParseFileResult::read_failure(file, &error),
406    };
407    let source = strip_bom(&raw);
408    let content_hash = xxhash_rust::xxh3::xxh3_64(source.as_bytes());
409
410    if let Some(cached) = cached_by_path
411        && cached.content_hash == content_hash
412        && (!need_complexity || cached.complexity_extracted)
413    {
414        return ParseFileResult::cache_hit(cache::cached_to_module_opts(
415            cached,
416            file.id,
417            need_complexity,
418        ))
419        .with_source_bytes_read(raw.len());
420    }
421
422    let parse_start = std::time::Instant::now();
423    // Drop a count that a scan outside a parse left on this thread.
424    css_masked_bytes::take();
425    let module = parse_source_to_module_with_flags(
426        file.id,
427        &file.path,
428        source,
429        content_hash,
430        need_complexity,
431        flag_patterns,
432    );
433    let parse_cpu_nanos = u64::try_from(parse_start.elapsed().as_nanos()).unwrap_or(u64::MAX);
434    let mut result =
435        ParseFileResult::cache_miss(module, parse_cpu_nanos).with_source_bytes_read(raw.len());
436    result.css_masked_bytes = css_masked_bytes::take();
437    result
438}
439
440/// Parse a single file and extract module information (without complexity).
441#[must_use]
442pub fn parse_single_file(file: &DiscoveredFile) -> Option<ModuleInfo> {
443    let raw = std::fs::read_to_string(&file.path).ok()?;
444    let source = strip_bom(&raw);
445    let content_hash = xxhash_rust::xxh3::xxh3_64(source.as_bytes());
446    Some(parse_source_to_module(
447        file.id,
448        &file.path,
449        source,
450        content_hash,
451        false,
452    ))
453}
454
455/// Parse from in-memory content (for LSP, includes complexity).
456#[must_use]
457pub fn parse_from_content(file_id: FileId, path: &Path, content: &str) -> ModuleInfo {
458    let content = strip_bom(content);
459    let content_hash = xxhash_rust::xxh3::xxh3_64(content.as_bytes());
460    parse_source_to_module(file_id, path, content, content_hash, true)
461}
462
463#[cfg(all(test, not(miri)))]
464mod tests;
fallow_extract/lib.rs

fallow_extract/
lib.rs
