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

fallow_extract/
lib.rs
