Struct CompactionContext 

pub struct CompactionContext {
    pub tokens_before: u64,
    pub reason: CompactReason,
    pub context_window: u64,
    pub turn: usize,
    pub counter: Arc<dyn TokenCounter>,
}

Metadata passed to ContextCompactor::compact describing the compaction trigger and current state.

Compactors can use this information to decide how aggressively to compact — e.g. an emergency compaction may use more aggressive summarization than a routine threshold check.

Fields

tokens_before: u64

Estimated token count before compaction.

The compactor’s input size, computed by the configured TokenCounter. Compaction aims to bring the post-compaction size below this so the conversation fits with headroom for the next turn.

reason: CompactReason

Why compaction was triggered.

Compactors may use the trigger to pick a strategy — an Emergency trigger warrants more aggressive summarization than a routine ThresholdExceeded.

context_window: u64

The model’s context window size.

The hard upper bound on tokens the model accepts in one request. This is the denominator every compaction threshold and target is expressed against, so the compactor can decide how much to keep.

turn: usize

The current turn number in the session.

Zero-indexed within the run. Useful for compaction strategies that weight recent turns more heavily, or for correlating a compaction pass back to the turn that triggered it in logs.

counter: Arc<dyn TokenCounter>

The token counter for estimating message sizes.

The same counter the driver uses for its compaction trigger — so the compactor can self-report tokens_after consistently. Compact implementations should use context.counter.count(&messages) instead of the static CompactionOutcome::estimate_tokens to match the driver’s configured counter.

Trait Implementations

impl Clone for CompactionContext

fn clone(&self) -> CompactionContext

Returns a duplicate of the value.

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

Performs copy-assignment from source.