Struct ToolResultCache 

pub struct ToolResultCache { /* private fields */ }

Tool result cache with LRU eviction

Implementations

impl ToolResultCache

pub fn new(capacity: usize) -> Self

Create a new cache with specified capacity.

pub fn insert(&mut self, key: ToolCacheKey, output: String)

Insert a result into the cache

pub fn insert_arc(&mut self, key: ToolCacheKey, output: Arc<String>)

Insert an Arc<String>-wrapped result into the cache to avoid cloning when the caller already has an Arc<String> available.

pub fn get(&self, key: &ToolCacheKey) -> Option<Arc<String>>

Retrieve a result if cached and fresh - now returns zero-copy Arc by default

pub fn get_owned(&self, key: &ToolCacheKey) -> Option<String>

Get owned value (explicitly clones when needed)

pub fn invalidate_for_path(&mut self, path: &str)

Clear cache entries for a specific file (selective eviction, not full clear)

This now uses granular eviction to remove only entries related to the changed file, avoiding the cache thrashing that occurred when the entire cache was cleared.

Impact

• Before: Full cache clear on any file change → 90% hit rate drop
• After: Selective removal → 10-15% hit rate impact only

pub fn invalidate_key(&mut self, key: &ToolCacheKey)

Invalidate a single cache entry by exact key.

pub fn invalidate_for_paths<I, S>(&mut self, paths: I)

where I: IntoIterator<Item = S>, S: AsRef<str>,

Invalidate cache entries for multiple paths.

This keeps invalidation policy centralized in one place and avoids duplicating path-iteration logic at call sites.

pub fn clear(&mut self)

Clear entire cache

pub fn check_pressure_and_evict(&mut self)

Check memory pressure and evict entries if necessary

This is a lightweight version of eviction that triggers when the cache size exceeds a heuristic threshold (e.g., 50MB for results).

pub fn stats(&self) -> CacheStats

Get cache statistics