fallow-extract 3.30.0

//! Parsing and extraction engine for fallow codebase intelligence.
//!
//! This crate handles all file parsing: JS/TS via Oxc, Vue/Svelte SFC extraction,
//! Astro frontmatter, MDX import/export extraction, CSS Module class name extraction,
//! HTML asset reference extraction, and incremental caching of parse results.

#![warn(missing_docs)]
#![cfg_attr(not(test), deny(clippy::disallowed_methods))]
#![cfg_attr(
    test,
    allow(
        clippy::unwrap_used,
        clippy::expect_used,
        reason = "tests use unwrap and expect to keep fixture setup concise"
    )
)]

mod asset_url;
pub mod astro;
pub mod cache;
pub(crate) mod complexity;
pub mod css;
pub mod css_classes;
pub mod css_in_js;
pub mod css_metrics;
pub mod federation_runtime;
pub mod flags;
mod function_body;
pub mod glimmer;
pub(crate) mod graphql;
pub(crate) mod html;
pub(crate) mod iconify;
pub mod inventory;
mod jsdoc_attach;
mod jsdoc_deprecated;
pub mod mdx;
mod module_info;
pub mod og_image;
mod parse;
pub mod sfc;
pub mod sfc_css;
mod sfc_props;
mod sfc_template;
pub mod similar_code;
mod source_map;
pub mod suppress;
/// Tailwind CSS arbitrary-value detection.
pub mod tailwind;
pub(crate) mod template_complexity;
mod template_expression_scan;
mod template_usage;
/// Visitor utilities for AST extraction.
pub mod visitor;

use std::path::Path;
use std::sync::atomic::{AtomicBool, Ordering};

use rayon::prelude::*;

use cache::CacheStore;
use fallow_types::discover::{DiscoveredFile, FileId};

pub use fallow_types::extract::{
    AngularComponentFieldArrayTypeFact, AngularTemplateMemberAccessFact, AngularThisSpreadFact,
    ClassHeritageInfo, ClassThisMemberAccessFact, ClassThisWholeObjectUseFact,
    ComputedEnumKeyUseFact, DefaultImportWholeObjectUseFact, DynamicCustomElementRenderFact,
    DynamicImportInfo, DynamicImportPattern, ExportInfo, ExportName,
    ExportedObjectInstancePropertyFact, FactoryCallMemberAccessFact, FactoryFnMemberAccessFact,
    FactoryFnWholeObjectFact, FactoryReturnExport, FactoryReturnObjectPropertyAccessFact,
    FactoryReturnObjectShapeExport, FlagPatterns, FluentChainMemberAccessFact,
    FluentChainNewMemberAccessFact, ImportInfo, ImportedName, InstanceExportBindingFact,
    LocalTypeDeclaration, MemberAccess, MemberInfo, MemberKind, ModuleInfo, ModuleLoadMechanism,
    ParseResult, PlaywrightFixtureAliasFact, PlaywrightFixtureDefinitionFact,
    PlaywrightFixtureTypeFact, PlaywrightFixtureUseFact, PublicSignatureTypeReference,
    QualifiedClassMemberAccessFact, ReExportInfo, RequireCallInfo, RequiredTypeMemberFact,
    SemanticFact, SourceParseDegradation, SourceReadFailure, StringEnumMemberValueFact,
    TypeAliasSurfaceTargetFact, TypeMemberTypeEntry, TypedPropertyMemberAccessFact, VisibilityTag,
    VitestModuleMockAction, VitestModuleMockOperationFact, compute_line_offsets,
};

pub use astro::{
    extract_astro_frontmatter, extract_astro_style_regions, extract_astro_template_regions,
};
pub use css::{
    StylesheetTokens, ThemeScan, ThemeTokenDef, extract_apply_tokens, extract_apply_tokens_located,
    extract_css_module_exports, extract_css_var_reads_located, scan_stylesheet_tokens,
    scan_theme_blocks,
};
pub use css_classes::{
    MarkupClassScan, MarkupClassToken, is_edit_distance_one, is_typo_edit, scan_markup_class_tokens,
};
pub use css_in_js::{
    ConsumerQuery, CssInJsObjectSheets, CssInJsToken, CssInJsTokenDef, CssInJsTokenOrigin,
    TokenConsumerHit, css_in_js_consumer_scan, css_in_js_object_sheets, css_in_js_theme_token_defs,
    css_in_js_token_defs, css_in_js_virtual_stylesheet,
};
pub use css_metrics::{compute_css_analytics, parse_css_color_rgb};
pub use glimmer::{is_glimmer_file, strip_glimmer_templates};
pub use mdx::{extract_mdx_statements, extract_mdx_statements_mapped};
pub use sfc::{
    SourceRegion, extract_sfc_scripts, extract_sfc_styles, extract_sfc_template_regions,
    is_sfc_file,
};
pub use sfc_css::{
    scoped_unused_classes, sfc_preprocessor_virtual_stylesheet, sfc_virtual_stylesheet,
};
pub use similar_code::extract_similar_code_functions;
pub use source_map::ExtractionResult;
pub use tailwind::{TailwindArbitraryUse, scan_tailwind_arbitrary_values};

#[expect(
    clippy::expect_used,
    reason = "static regex patterns are hard-coded analyzer invariants covered by extraction tests"
)]
fn static_regex(pattern: &str) -> regex::Regex {
    regex::Regex::new(pattern).expect("static regex pattern should compile")
}

pub use parse::{parse_source_to_module, parse_source_to_module_with_flags};

/// Leading UTF-8 byte order mark codepoint.
///
/// Windows editors (Notepad, older VS settings, some IDE plugins) emit a UTF-8
/// BOM at the start of source files. fallow's contract is "UTF-8 with or
/// without BOM; line offsets are computed against the post-BOM view; the BOM,
/// if present on input, is preserved on output by `fallow fix`."
const BOM_CHAR: char = '\u{FEFF}';
// Small, cache-hot inputs are faster on one thread than through Rayon setup.
// Larger file sets still use parallel parsing where parse work dominates.
const PARALLEL_PARSE_FILE_THRESHOLD: usize = 32;

/// Strip the leading UTF-8 BOM if present.
///
/// Called at every file-read entry point in this crate so the rest of the
/// pipeline (content hash, `compute_line_offsets`, oxc parser, downstream
/// analyses) sees a consistent post-BOM view. Mirrors the
/// `fallow_config` layer (`config_writer.rs::BOM`) so config-shaped sources
/// and source-code-shaped sources are processed symmetrically. See issue #475.
#[must_use]
fn strip_bom(source: &str) -> &str {
    source.strip_prefix(BOM_CHAR).unwrap_or(source)
}

/// Parse all files, extracting imports and exports.
///
/// Small file sets use a sequential fast path to avoid parallel scheduling
/// overhead; larger file sets use parallel extraction.
/// Uses the cache to skip reparsing files whose content hasn't changed.
///
/// When `need_complexity` is true, per-function cyclomatic/cognitive complexity
/// metrics are computed during parsing (needed by the `health` command).
/// Pass `false` for dead-code analysis where complexity data is unused.
///
/// Flag detection uses the built-in patterns only. A caller with a resolved
/// config uses [`parse_all_files_cancellable`] with the config's patterns,
/// because the cache keys on them.
pub fn parse_all_files(
    files: &[DiscoveredFile],
    cache: Option<&CacheStore>,
    need_complexity: bool,
) -> ParseResult {
    parse_all_files_cancellable(
        files,
        cache,
        need_complexity,
        None,
        &FlagPatterns::default(),
    )
}

/// Parse all files, abandoning the remaining ones once `cancellation` is set.
///
/// Rayon's `map`/`collect` cannot short-circuit, so cancellation makes the
/// per-file body a no-op instead of stopping the iteration: the scheduled
/// items still drain, but at one atomic load each. The returned
/// [`ParseResult`] is therefore truncated whenever the token flipped, and
/// callers must treat a set token as a failed run rather than as a project
/// with fewer modules.
///
/// `flag_patterns` are the user flag patterns that detection applies on top
/// of the built-in ones. They must match the patterns the cache was keyed on.
pub fn parse_all_files_cancellable(
    files: &[DiscoveredFile],
    cache: Option<&CacheStore>,
    need_complexity: bool,
    cancellation: Option<&AtomicBool>,
    flag_patterns: &FlagPatterns,
) -> ParseResult {
    let parse_one = |file: &DiscoveredFile| {
        if cancellation.is_some_and(|cancelled| cancelled.load(Ordering::SeqCst)) {
            return ParseFileResult::default();
        }
        parse_single_file_cached(file, cache, need_complexity, flag_patterns)
    };
    let results: Vec<ParseFileResult> = if files.len() <= PARALLEL_PARSE_FILE_THRESHOLD {
        files.iter().map(parse_one).collect()
    } else {
        files.par_iter().map(parse_one).collect()
    };

    let mut modules = Vec::with_capacity(results.len());
    let mut read_failures = Vec::new();
    let mut parse_degradations = Vec::new();
    let mut hits = 0usize;
    let mut misses = 0usize;
    let mut parse_cpu_nanos = 0u64;
    let mut files_read = 0u64;
    let mut source_bytes_read = 0u64;
    let mut css_masked_bytes = 0u64;

    // `results` is a positional map over `files`, so zipping recovers the path
    // for a module without carrying one on `ModuleInfo`.
    for (file, result) in files.iter().zip(results) {
        hits += result.cache_hits;
        misses += result.cache_misses;
        parse_cpu_nanos = parse_cpu_nanos.saturating_add(result.parse_cpu_nanos);
        if let Some(bytes) = result.source_bytes_read {
            files_read += 1;
            source_bytes_read += bytes;
        }
        css_masked_bytes += result.css_masked_bytes;
        if let Some(module) = result.module {
            if module.parse_error_count > 0 {
                parse_degradations.push(SourceParseDegradation {
                    file_id: module.file_id,
                    path: file.path.clone(),
                    error_count: module.parse_error_count,
                    panicked: module.parse_panicked,
                });
            }
            modules.push(module);
        }
        if let Some(failure) = result.read_failure {
            read_failures.push(failure);
        }
    }

    if hits > 0 || misses > 0 {
        tracing::info!(
            cache_hits = hits,
            cache_misses = misses,
            "incremental cache stats"
        );
    }

    ParseResult {
        modules,
        read_failures,
        parse_degradations,
        cache_hits: hits,
        cache_misses: misses,
        parse_cpu_ms: parse_cpu_nanos as f64 / 1_000_000.0,
        files_read,
        source_bytes_read,
        css_masked_bytes,
    }
}

#[derive(Default)]
struct ParseFileResult {
    module: Option<ModuleInfo>,
    read_failure: Option<SourceReadFailure>,
    cache_hits: usize,
    cache_misses: usize,
    parse_cpu_nanos: u64,
    /// Source bytes read from disk for this file, or `None` when the file was
    /// served from cache metadata without a read.
    source_bytes_read: Option<u64>,
    /// Source bytes that the CSS comment mask read during the parse.
    css_masked_bytes: u64,
}

impl ParseFileResult {
    fn cache_hit(module: ModuleInfo) -> Self {
        Self {
            module: Some(module),
            read_failure: None,
            cache_hits: 1,
            cache_misses: 0,
            parse_cpu_nanos: 0,
            source_bytes_read: None,
            css_masked_bytes: 0,
        }
    }

    fn cache_miss(module: ModuleInfo, parse_cpu_nanos: u64) -> Self {
        Self {
            module: Some(module),
            read_failure: None,
            cache_hits: 0,
            cache_misses: 1,
            parse_cpu_nanos,
            source_bytes_read: None,
            css_masked_bytes: 0,
        }
    }

    const fn with_source_bytes_read(mut self, bytes: usize) -> Self {
        self.source_bytes_read = Some(bytes as u64);
        self
    }

    fn read_failure(file: &DiscoveredFile, error: &std::io::Error) -> Self {
        Self {
            module: None,
            read_failure: Some(SourceReadFailure {
                file_id: file.id,
                path: file.path.clone(),
                error: error.to_string(),
            }),
            cache_hits: 0,
            cache_misses: 0,
            parse_cpu_nanos: 0,
            source_bytes_read: None,
            css_masked_bytes: 0,
        }
    }
}

/// Parse a single file, consulting the cache first.
///
/// Cache validation strategy (fast path -> slow path):
/// 1. Open the file so unreadable sources cannot use stale cached analysis
/// 2. Read mtime + ctime + size from the open handle
/// 3. If all three match the cached entry -> cache hit, return immediately
/// 4. Otherwise -> read file, compute content hash
/// 5. If content hash matches cached entry -> cache hit (file was rewritten or
///    `touch`ed but its content is unchanged)
/// 6. Otherwise -> cache miss, full parse
///
/// Step 3 requires ctime as well as mtime because mtime is writer-controlled:
/// a same-length rewrite whose mtime is restored (`touch -r`, a codemod, a
/// `git checkout` of an equal-length revision) leaves `(mtime, size)`
/// unchanged, and serving the cached module for it means reporting the OLD
/// file's unused exports with an auto-fixable `remove-export` action. A file
/// whose ctime moved falls through to step 4 and still hits on the content
/// hash, so the cost of the stricter gate is one read, not a reparse.
fn parse_single_file_cached(
    file: &DiscoveredFile,
    cache: Option<&CacheStore>,
    need_complexity: bool,
    flag_patterns: &FlagPatterns,
) -> ParseFileResult {
    let cached_by_path = cache.and_then(|store| store.get_by_path_only(&file.path));

    if let Some(cached) = cached_by_path
        && cached.file_size == file.size_bytes
    {
        let source_file = match std::fs::File::open(&file.path) {
            Ok(source_file) => source_file,
            Err(error) => return ParseFileResult::read_failure(file, &error),
        };
        if let Ok(metadata) = source_file.metadata()
            && metadata.len() == cached.file_size
        {
            let fingerprint =
                fallow_types::source_fingerprint::SourceFingerprint::from_metadata(&metadata);
            if cached.source_fingerprint() == fingerprint
                && fingerprint.is_trustworthy_without_content()
                && (!need_complexity || cached.complexity_extracted)
            {
                return ParseFileResult::cache_hit(cache::cached_to_module_opts(
                    cached,
                    file.id,
                    need_complexity,
                ));
            }
        }
    }

    let raw = match std::fs::read_to_string(&file.path) {
        Ok(raw) => raw,
        Err(error) => return ParseFileResult::read_failure(file, &error),
    };
    let source = strip_bom(&raw);
    let content_hash = xxhash_rust::xxh3::xxh3_64(source.as_bytes());

    if let Some(cached) = cached_by_path
        && cached.content_hash == content_hash
        && (!need_complexity || cached.complexity_extracted)
    {
        return ParseFileResult::cache_hit(cache::cached_to_module_opts(
            cached,
            file.id,
            need_complexity,
        ))
        .with_source_bytes_read(raw.len());
    }

    let parse_start = std::time::Instant::now();
    // Drop a count that a scan outside a parse left on this thread.
    css::take_comment_masked_bytes();
    let module = parse_source_to_module_with_flags(
        file.id,
        &file.path,
        source,
        content_hash,
        need_complexity,
        flag_patterns,
    );
    let parse_cpu_nanos = u64::try_from(parse_start.elapsed().as_nanos()).unwrap_or(u64::MAX);
    let mut result =
        ParseFileResult::cache_miss(module, parse_cpu_nanos).with_source_bytes_read(raw.len());
    result.css_masked_bytes = css::take_comment_masked_bytes();
    result
}

/// Parse a single file and extract module information (without complexity).
#[must_use]
pub fn parse_single_file(file: &DiscoveredFile) -> Option<ModuleInfo> {
    let raw = std::fs::read_to_string(&file.path).ok()?;
    let source = strip_bom(&raw);
    let content_hash = xxhash_rust::xxh3::xxh3_64(source.as_bytes());
    Some(parse_source_to_module(
        file.id,
        &file.path,
        source,
        content_hash,
        false,
    ))
}

/// Parse from in-memory content (for LSP, includes complexity).
#[must_use]
pub fn parse_from_content(file_id: FileId, path: &Path, content: &str) -> ModuleInfo {
    let content = strip_bom(content);
    let content_hash = xxhash_rust::xxh3::xxh3_64(content.as_bytes());
    parse_source_to_module(file_id, path, content, content_hash, true)
}

#[cfg(all(test, not(miri)))]
mod tests;