Struct LspConfig 

pub struct LspConfig {

    pub php_version: Option<String>,
    pub exclude_paths: Vec<String>,
    pub include_paths: Vec<String>,
    pub stub_dirs: Vec<String>,
    pub diagnostics: DiagnosticsConfig,
    pub features: FeaturesConfig,
    pub max_indexed_files: usize,
    pub index_vendor: bool,
    pub debug: bool,
    pub debounce_ms: u64,
    pub warm_analysis: bool,
    pub warm_vendor_analysis: bool,
    pub cache_path: Option<PathBuf>,
    pub flush_interval_ms: u64,
    pub external_tools: ExternalToolsConfig,
}

Configuration received from the client via initializationOptions.

Fields

php_version: Option<String>

PHP version string, e.g. "8.1". Set explicitly via initializationOptions or auto-detected from composer.json / the php binary at startup.

exclude_paths: Vec<String>

Glob patterns for paths to exclude from workspace indexing.

include_paths: Vec<String>

Glob patterns for paths that must be indexed even if they match an excludePaths entry. Patterns are matched against path components (same semantics as excludePaths). Example: ["vendor/yiisoft"].

stub_dirs: Vec<String>

Directories of user-supplied PHP stub files to load in addition to the bundled built-ins. Each entry is resolved relative to the workspace root if not already absolute. Every .php file found (recursively) is registered as a read-only, highest-precedence symbol source — useful for extensions or frameworks the bundled stubs don’t cover.

diagnostics: DiagnosticsConfig

Per-category diagnostic toggles.

features: FeaturesConfig

Per-feature capability toggles.

max_indexed_files: usize

Hard cap on the number of PHP files indexed during a workspace scan. Defaults to MAX_INDEXED_FILES. Set lower via initializationOptions to reduce memory on projects with very large vendor trees.

index_vendor: bool

Whether to eagerly index vendor/ during the workspace scan.

Default true: vendor/ is walked and declaration-extracted (name/namespace/signature, no type inference) alongside the rest of the workspace, so bare-name completion, workspace symbols, and find-implementations/type-hierarchy see vendor classes without extra configuration (issues #240, #246). This scan is I/O-bound, not CPU-bound, and cheap relative to the fear it usually gets: on real fixtures the walk+read+parse+extract for ~5k files runs in ~200ms, and unchanged files replay from the on-disk content-hash cache (index/cache.rs) on subsequent starts. What eager vendor indexing does NOT do is pull vendor into the background warm_analysis_sweep (full type analysis for find-references/rename) — that sweep skips vendor files unconditionally (see document_store::is_vendor_uri) since sweeping the whole vendor tree buys nothing lasting (the analysis-result cache is capped, see ANALYSIS_CACHE_CAP) and would multiply background CPU cost by vendor’s file count. A vendor file still gets fully analyzed on demand — when a request touches it directly, or when it’s referenced by a currently-open file (which the sweep prioritizes regardless of this flag).

Set false to skip vendor/ entirely on scan (old default): vendor files then load only on demand via PSR-4 resolution (composer autoload + per-file parse). Useful for very large vendor trees where even the cheap declaration scan isn’t worth paying for.

debug: bool

Emit extra diagnostic log messages on startup: cache hit/miss ratio, workspace root paths, and PSR-4 namespace count. Enable via initializationOptions.debug = true.

debounce_ms: u64

Debounce delay in milliseconds between the last textDocument/didChange and the parse + analysis run. Defaults to 100 ms. Set lower for fast machines / Neovim users; set higher for slow machines or large files to reduce thrashing.

warm_analysis: bool

Background-analyze the workspace after indexing (and re-warm after edits settle) so find-references / rename answer from warm memos instead of paying a cold per-file analysis at request time. Default true; set warmAnalysis: false to trade slower references for a smaller resident footprint.

warm_vendor_analysis: bool

Extend the background warm-analysis sweep to vendor/ files (see the index_vendor docs above for why the ambient sweep skips vendor by default). Opt-in and throttled/idle-priority once implemented — ships off by default since it’s new background CPU cost with no prior production data on real vendor-tree sizes (ROADMAP 0c step 2, ~/.claude/plans/crispy-noodling-key.md). Not implemented yet: setting this to true is currently a no-op.

cache_path: Option<PathBuf>

Override the on-disk cache directory. When set, used verbatim (no schema-version or workspace-hash subdirectories appended). Primarily useful in tests and CI environments with non-standard cache locations. When absent, falls back to the platform default ($XDG_CACHE_HOME / $HOME/.cache on Unix, %LOCALAPPDATA% on Windows).

flush_interval_ms: u64

How often the background loop persists staged analysis-cache postings (reference postings from an in-progress warm sweep or an on-demand query freshness pass) to disk. Bounds data loss on an unclean exit (crash, kill — anything that skips shutdown) to roughly one interval. Defaults to 20 s; lower mainly useful in tests.

external_tools: ExternalToolsConfig

External static-analysis tool integration (PHPStan / PHPCS). Both default off; see ExternalToolsConfig.

Trait Implementations

impl Clone for LspConfig

fn clone(&self) -> LspConfig

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for LspConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for LspConfig

fn default() -> Self

Returns the “default value” for a type.