Struct PluginsGenerator 

pub struct PluginsGenerator {
    pub meta: Arc<PluginMetaRegistry>,
}

gen:plugins — one candidate per loaded plugin. Drives :describe-plugin <Tab> and the plugin manager’s :plugin-unload / :plugin-reload / :plugin-update.

Both call sites carried this as a deferred comment — “gen:plugins completion is a follow-up … empty until Phase-8” in ex_commands.rs, and the same note on lattice-plugin-loader’s string_arg. Phase 8 landed and the comments did not get grepped, which is why the rule is now a test (ex_string_args_have_completion.rs) rather than a convention.

Holds a strong Arc<PluginMetaRegistry>, the same instance the host’s register_plugin writes through and :list-plugins reads. It is registered once at boot and never swapped, so there is no stale-Arc hazard; the interior RwLock is what a mid-session load mutates, and reading it per generate is what makes a plugin loaded after boot appear under <Tab>.

Fields

meta: Arc<PluginMetaRegistry>

Trait Implementations

impl CandidateGenerator for PluginsGenerator

fn generate(&self, _ctx: &GenerateContext<'_>) -> Vec<RawCandidate>

fn cache_key(&self, _ctx: &GenerateContext<'_>) -> Option<CacheKey>

Optional cache key. When two contexts produce the same Some(key), the pipeline serves the second from cache. The query (prefix) is implicitly part of the key only if the generator includes it – many generators key on a registry version (cache covers all queries until the registry changes) and let the matcher do the per-query filtering.

fn cache_ttl(&self) -> Duration

Soft TTL for cache entries. After elapsing, the entry is regenerated on next access. Defaults to no expiry.