Struct DiagnosticsStore 

pub struct DiagnosticsStore { /* private fields */ }

Stores diagnostics from all LSP servers, keyed per (ServerKey, file).

Key design points (driven by the v0.16 LSP audit):

1. Per-server state. A single file can be served by multiple LSP servers (e.g., pyright + ty, or tsserver + ESLint). The cache key is (ServerKey, PathBuf) so each server’s view is tracked independently.

2. Empty publishes are kept. Earlier the store deleted entries on empty publishes, making “checked clean” indistinguishable from “never checked”. Now we preserve the entry with epoch = ... so callers can answer the question honestly.

3. LRU cap. capacity (default 5000, configurable via Config::diagnostic_cache_size) bounds memory. Set to 0 to disable. On insert when at capacity, the least-recently-touched entry is evicted. Eviction is tracked so directory-mode callers can list those files as unchecked rather than silently lose them.

Implementations

impl DiagnosticsStore

pub fn new() -> Self

pub fn with_capacity(capacity: usize) -> Self

pub fn set_capacity(&mut self, capacity: usize)

Set or change the LRU cap. If the new cap is smaller than the current entry count, the oldest entries are evicted immediately to fit.

pub fn len(&self) -> usize

Number of currently-tracked entries.

pub fn is_empty(&self) -> bool

pub fn publish( &mut self, server: ServerKey, file: PathBuf, diagnostics: Vec<StoredDiagnostic>, )

Replace diagnostics for a (server_kind, file) pair using the server’s lifecycle root from the active manager. Empty diagnostics are preserved as “checked clean” (NOT deleted as before).

Note: the (server, file) key uses ServerKey { kind, root } so concurrent multi-workspace usage doesn’t collapse different roots. Callers without the root (legacy push handler) should call publish_with_kind which derives the key.

pub fn publish_with_result_id( &mut self, server: ServerKey, file: PathBuf, diagnostics: Vec<StoredDiagnostic>, result_id: Option<String>, )

Replace diagnostics and record a pull resultId for the next request. Empty diagnostics are preserved as “checked clean”.

pub fn publish_full( &mut self, server: ServerKey, file: PathBuf, diagnostics: Vec<StoredDiagnostic>, result_id: Option<String>, version: Option<i32>, )

Replace diagnostics with full provenance (resultId + document version). version should be the LSP version field from publishDiagnostics when the server provided one, or None otherwise.

pub fn publish_with_kind( &mut self, kind: ServerKind, file: PathBuf, diagnostics: Vec<StoredDiagnostic>, )

Compatibility wrapper for the legacy push path that knows only the ServerKind. Builds a ServerKey with an empty root, which is adequate for the single-root-per-kind case the manager currently uses for push diagnostics. Multi-root callers should use publish directly with a real ServerKey.

pub fn for_file(&self, file: &Path) -> Vec<&StoredDiagnostic>

Get all diagnostics for a specific file (across all servers). Updates LRU position for each touched entry.

pub fn entries_for_file( &self, file: &Path, ) -> Vec<(&ServerKey, &DiagnosticEntry)>

Get the full per-server entry for a file. Useful when callers need to know epoch/resultId, not just the diagnostics array.

pub fn has_any_report_for_file(&self, file: &Path) -> bool

True if any server has reported (even an empty result) for this file.

pub fn has_report_for_server_file( &self, server: &ServerKey, file: &Path, ) -> bool

True if this exact server instance has reported (even an empty result) for this exact file.

pub fn has_publish_for_file_after( &self, server: &ServerKey, file: &Path, since: Instant, ) -> bool

True if this exact server instance published/replaced diagnostics for this exact file after since. This is intentionally per (kind, root, file); a publish for another file must not prove freshness here.

pub fn for_directory(&self, dir: &Path) -> Vec<&StoredDiagnostic>

Get all diagnostics for files under a directory.

pub fn all(&self) -> Vec<&StoredDiagnostic>

All stored diagnostics, flattened.

pub fn error_warning_counts(&self) -> (usize, usize)

Count of errors and warnings across the entire warm set (every file any server has published for). Allocation-free — used to refresh the agent status bar on each request drain without materializing a Vec.

pub fn clear_server(&mut self, server: ServerKind)

Drop all entries for a server kind (e.g., on server crash/restart). Prefer clear_for_server for real manager cleanup so peer roots of the same kind are not wiped.

pub fn clear_for_server_file(&mut self, key: &ServerKey, file: &Path)

Drop one cached report for a specific server/file pair.

pub fn clear_for_file(&mut self, file: &Path) -> bool

Drop every cached report for a file across all servers. Used when a file is deleted/renamed away — its diagnostics would otherwise linger in the warm set forever (no server republishes for a path that no longer exists), inflating the error/warning counts surfaced in the status bar and aft_inspect. Returns true if any entry was removed.

pub fn clear_for_server(&mut self, key: &ServerKey)

Drop all entries for a specific server instance.

pub fn clear_server_instance(&mut self, key: &ServerKey)

Backward-compatible alias for tests/callers that already used the instance-scoped name.

Trait Implementations

impl Default for DiagnosticsStore

fn default() -> Self

Returns the “default value” for a type.