deepstrike-core 0.2.33

use super::config::ContextConfig;
use super::partitions::ContextPartitions;
use super::token_engine::ContextTokenEngine;

/// Action recommended by the pressure monitor — the *pressure-level* vocabulary.
///
/// This is distinct from [`crate::mm::EvictionOp`] (the *planner-op* vocabulary) and the two must
/// **not** be collapsed: `PressureAction` is what [`PressureMonitor::recommend`] /
/// `ContextManager::should_compress` return, the `Ord`-keyed cascade selector inside the compression
/// pipeline, and the canonical compaction wire label. `EvictionOp` is what `plan_eviction` emits, and
/// carries per-op data this enum has no place for. The single bridge between the two layers is
/// `LoopStateMachine::execute_eviction_op`. See the layer-boundary note on [`crate::mm::EvictionOp`].
#[derive(Debug, Clone, Copy, PartialEq, Eq, PartialOrd, Ord)]
pub enum PressureAction {
    None,
    SnipCompact,
    MicroCompact,
    ContextCollapse,
    AutoCompact,
}

impl PressureAction {
    /// Canonical snake_case label — the single source for compaction-event action strings and the
    /// `[Compressed: {label}]` summary marker (used by both the summarizer and the compaction log).
    pub fn label(self) -> &'static str {
        match self {
            PressureAction::None => "none",
            PressureAction::SnipCompact => "snip_compact",
            PressureAction::MicroCompact => "micro_compact",
            PressureAction::ContextCollapse => "context_collapse",
            PressureAction::AutoCompact => "auto_compact",
        }
    }
}

/// Monitors rho = used_tokens / max_tokens and recommends compression actions.
/// All thresholds come from `ContextConfig` — no hardcoded constants.
pub struct PressureMonitor {
    max_tokens: u32,
    config: ContextConfig,
}

impl PressureMonitor {
    pub fn new(max_tokens: u32, config: ContextConfig) -> Self {
        Self { max_tokens, config }
    }

    pub fn max_tokens(&self) -> u32 {
        self.max_tokens
    }

    /// Current pressure rho ∈ [0, +∞).
    /// Uses provider-reported prompt tokens when available; otherwise estimates from partitions.
    ///
    /// This is the **raw** rho (full partition weight). Making rho paging-aware — i.e. subtracting
    /// non-resident (`Collapsed`/`SpooledOut`/`PagedOut`) handle tokens so paging immediately relieves
    /// pressure — is **not** a drop-in here: [`crate::context::manager::ContextManager::recompute_handle_residency`]
    /// decides the Resident↔Collapsed projection from this very rho, so subtracting collapsed tokens
    /// would drop rho below `collapse_threshold` and immediately un-collapse (oscillation). That needs
    /// a deliberate split into *raw* rho (drives the collapse decision) vs *effective* rho (drives
    /// further compaction/renewal), tracked as remaining W1-1 design work — see `ContextManager::rho`.
    pub fn pressure(
        &self,
        partitions: &ContextPartitions,
        engine: &ContextTokenEngine,
        observed_prompt_tokens: Option<u32>,
    ) -> f64 {
        if self.max_tokens == 0 {
            return 0.0;
        }
        match observed_prompt_tokens {
            Some(tokens) => tokens as f64 / self.max_tokens as f64,
            None => partitions.total_tokens(engine) as f64 / self.max_tokens as f64,
        }
    }

    pub fn recommend(&self, rho: f64) -> PressureAction {
        if rho > self.config.auto_threshold {
            PressureAction::AutoCompact
        } else if rho > self.config.collapse_threshold {
            PressureAction::ContextCollapse
        } else if rho > self.config.micro_threshold {
            PressureAction::MicroCompact
        } else if rho > self.config.snip_threshold {
            PressureAction::SnipCompact
        } else {
            PressureAction::None
        }
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::context::config::ContextConfig;
    use crate::context::partitions::ContextPartitions;
    use crate::context::token_engine::ContextTokenEngine;
    use crate::types::message::Message;

    fn engine() -> ContextTokenEngine {
        ContextTokenEngine::char_approx()
    }
    fn config() -> ContextConfig {
        ContextConfig::default()
    }

    #[test]
    fn thresholds_follow_config() {
        let cfg = config();
        let monitor = PressureMonitor::new(100, cfg.clone());
        assert_eq!(monitor.recommend(0.50), PressureAction::None);
        assert_eq!(
            monitor.recommend(cfg.snip_threshold + 0.01),
            PressureAction::SnipCompact
        );
        assert_eq!(
            monitor.recommend(cfg.micro_threshold + 0.01),
            PressureAction::MicroCompact
        );
        assert_eq!(
            monitor.recommend(cfg.collapse_threshold + 0.01),
            PressureAction::ContextCollapse
        );
        assert_eq!(
            monitor.recommend(cfg.auto_threshold + 0.01),
            PressureAction::AutoCompact
        );
    }

    #[test]
    fn custom_thresholds_respected() {
        let cfg = ContextConfig {
            snip_threshold: 0.50,
            ..Default::default()
        };
        let monitor = PressureMonitor::new(100, cfg);
        assert_eq!(monitor.recommend(0.51), PressureAction::SnipCompact);
        assert_eq!(monitor.recommend(0.49), PressureAction::None);
    }

    #[test]
    fn pressure_calculation_uses_engine() {
        let cfg = config();
        let monitor = PressureMonitor::new(1_000, cfg.clone());
        let mut ctx = ContextPartitions::new(&cfg);
        let baseline = ctx.total_tokens(&engine()) as f64;
        ctx.history.push(Message::user("test"), 500);
        let rho = monitor.pressure(&ctx, &engine(), None);
        let expected = (baseline + 500.0) / 1_000.0;
        assert!((rho - expected).abs() < f64::EPSILON);
    }
}