Files
codex/codex-rs/core/src/state/session.rs
T
Michael Bolin 32b65bbf7a core: reset context for token budget compaction (#29743)
## Why

When `Feature::TokenBudget` is enabled, compaction should behave like
`new_context`: start a fresh context window with the standard injected
context, without asking the server to summarize old history and without
carrying prior user or assistant messages into the next model request.

This is still a compaction operation from the client lifecycle
perspective. Manual `/compact` and auto-compaction should keep the same
observable side effects that clients and hooks expect, including compact
hooks and `TurnItem::ContextCompaction`.

## What changed

- Added `compact_token_budget` to run token-budget manual and inline
auto-compaction through a shared compaction lifecycle.
- Split pending `new_context` requests from forced context-window
startup: `take_new_context_window_request()` consumes pending requests,
and `start_new_context_window()` installs a fresh context window.
- Routed token-budget manual `/compact` and inline auto-compaction to
install a fresh context window locally instead of calling server/local
summarization.
- Preserved compact lifecycle side effects for token-budget compaction
by running pre/post compact hooks and emitting `ContextCompaction` item
start/completion events.
- Updated token-budget tests to assert fresh window IDs, absence of
server-side compaction calls, dropped prior transcript messages/tool
output after reset, and compact hook/item lifecycle behavior.

## Testing

- `just test -p codex-core
token_budget_context_uses_new_window_after_compaction`
- `just test -p codex-core token_budget_compaction_runs_compact_hooks`
- `just test -p codex-core
token_budget_mid_turn_auto_compaction_resets_before_active_follow_up`

---------

Co-authored-by: pakrym-oai <pakrym@openai.com>
2026-06-23 16:59:04 -07:00

339 lines
12 KiB
Rust

//! Session-wide mutable state.
use codex_protocol::models::AdditionalPermissionProfile;
use codex_protocol::models::ResponseItem;
use codex_sandboxing::policy_transforms::merge_permission_profiles;
use std::collections::HashMap;
use std::collections::HashSet;
use std::collections::VecDeque;
use super::AdditionalContextStore;
use super::auto_compact_window::AutoCompactWindow;
use super::auto_compact_window::AutoCompactWindowIds;
use super::auto_compact_window::AutoCompactWindowSnapshot;
use crate::context_manager::ContextManager;
use crate::session::PreviousTurnSettings;
use crate::session::session::SessionConfiguration;
use crate::session::time_reminder::CurrentTimeReminderState;
use crate::session_startup_prewarm::SessionStartupPrewarmHandle;
use codex_protocol::protocol::RateLimitSnapshot;
use codex_protocol::protocol::TokenUsage;
use codex_protocol::protocol::TokenUsageInfo;
use codex_protocol::protocol::TurnContextItem;
use codex_utils_output_truncation::TruncationPolicy;
/// Persistent, session-scoped state previously stored directly on `Session`.
pub(crate) struct SessionState {
pub(crate) session_configuration: SessionConfiguration,
pub(crate) history: ContextManager,
pub(crate) latest_rate_limits: Option<RateLimitSnapshot>,
pub(crate) server_reasoning_included: bool,
pub(crate) mcp_dependency_prompted: HashSet<String>,
pub(crate) additional_context: AdditionalContextStore,
/// Settings used by the latest regular user turn, used for turn-to-turn
/// model/realtime handling on subsequent regular turns (including full-context
/// reinjection after resume or `/compact`).
previous_turn_settings: Option<PreviousTurnSettings>,
/// Runtime accounting state for the active auto-compaction window.
auto_compact_window: AutoCompactWindow,
/// Startup prewarmed session prepared during session initialization.
pub(crate) startup_prewarm: Option<SessionStartupPrewarmHandle>,
pub(crate) current_time_reminder: CurrentTimeReminderState,
pub(crate) active_connector_selection: HashSet<String>,
pub(crate) pending_session_start_sources: VecDeque<codex_hooks::SessionStartSource>,
granted_permissions_by_environment_id: HashMap<String, AdditionalPermissionProfile>,
next_turn_is_first: bool,
}
impl SessionState {
/// Create a new session state mirroring previous `State::default()` semantics.
#[cfg(test)]
pub(crate) fn new(session_configuration: SessionConfiguration) -> Self {
Self::new_with_auto_compact_window_ids(
session_configuration,
AutoCompactWindowIds::new_initial(),
)
}
pub(crate) fn new_with_auto_compact_window_ids(
session_configuration: SessionConfiguration,
auto_compact_window_ids: AutoCompactWindowIds,
) -> Self {
let history = ContextManager::new();
Self {
session_configuration,
history,
latest_rate_limits: None,
server_reasoning_included: false,
mcp_dependency_prompted: HashSet::new(),
additional_context: AdditionalContextStore::default(),
previous_turn_settings: None,
auto_compact_window: AutoCompactWindow::new_with_ids(auto_compact_window_ids),
startup_prewarm: None,
current_time_reminder: CurrentTimeReminderState::default(),
active_connector_selection: HashSet::new(),
pending_session_start_sources: VecDeque::new(),
granted_permissions_by_environment_id: HashMap::new(),
next_turn_is_first: true,
}
}
// History helpers
pub(crate) fn record_items<I>(&mut self, items: I, policy: TruncationPolicy)
where
I: IntoIterator,
I::Item: std::ops::Deref<Target = ResponseItem>,
{
self.history.record_items(items, policy);
}
pub(crate) fn previous_turn_settings(&self) -> Option<PreviousTurnSettings> {
self.previous_turn_settings.clone()
}
pub(crate) fn set_previous_turn_settings(
&mut self,
previous_turn_settings: Option<PreviousTurnSettings>,
) {
self.previous_turn_settings = previous_turn_settings;
}
pub(crate) fn set_next_turn_is_first(&mut self, value: bool) {
self.next_turn_is_first = value;
}
pub(crate) fn take_next_turn_is_first(&mut self) -> bool {
let is_first_turn = self.next_turn_is_first;
self.next_turn_is_first = false;
is_first_turn
}
pub(crate) fn clone_history(&self) -> ContextManager {
self.history.clone()
}
pub(crate) fn replace_history(
&mut self,
items: Vec<ResponseItem>,
reference_context_item: Option<TurnContextItem>,
) {
self.history.replace(items);
self.history
.set_reference_context_item(reference_context_item);
self.auto_compact_window.clear_prefill();
}
pub(crate) fn set_token_info(&mut self, info: Option<TokenUsageInfo>) {
self.history.set_token_info(info);
}
pub(crate) fn set_reference_context_item(&mut self, item: Option<TurnContextItem>) {
self.history.set_reference_context_item(item);
}
pub(crate) fn reference_context_item(&self) -> Option<TurnContextItem> {
self.history.reference_context_item()
}
// Token/rate limit helpers
pub(crate) fn update_token_info_from_usage(
&mut self,
usage: &TokenUsage,
model_context_window: Option<i64>,
) {
self.history.update_token_info(usage, model_context_window);
}
pub(crate) fn ensure_auto_compact_window_server_prefill_from_usage(
&mut self,
usage: &TokenUsage,
) {
self.auto_compact_window
.ensure_server_observed_prefill_from_usage(usage);
}
pub(crate) fn set_auto_compact_window_estimated_prefill(&mut self, tokens: i64) {
self.auto_compact_window.set_estimated_prefill(tokens);
}
pub(crate) fn auto_compact_window_snapshot(&self) -> AutoCompactWindowSnapshot {
self.auto_compact_window.snapshot()
}
pub(crate) fn claim_token_budget_reminder(&mut self) -> bool {
self.auto_compact_window.claim_token_budget_reminder()
}
pub(crate) fn auto_compact_window_number(&self) -> u64 {
self.auto_compact_window.window_number()
}
pub(crate) fn auto_compact_window_ids(&self) -> AutoCompactWindowIds {
self.auto_compact_window.ids()
}
pub(crate) fn restore_auto_compact_window(
&mut self,
window_number: u64,
ids: AutoCompactWindowIds,
) {
self.auto_compact_window.restore(window_number, ids);
}
pub(crate) fn advance_auto_compact_window(&mut self) -> (u64, AutoCompactWindowIds) {
self.auto_compact_window.advance()
}
pub(crate) fn request_new_context_window(&mut self) {
self.auto_compact_window.request_new_context_window();
}
pub(crate) fn take_new_context_window_request(&mut self) -> bool {
self.auto_compact_window.take_new_context_window_request()
}
pub(crate) fn start_new_context_window(&mut self) -> (u64, AutoCompactWindowIds) {
let window = self.auto_compact_window.advance();
self.auto_compact_window.clear_prefill();
window
}
pub(crate) fn token_info(&self) -> Option<TokenUsageInfo> {
self.history.token_info()
}
pub(crate) fn set_rate_limits(&mut self, snapshot: RateLimitSnapshot) {
self.latest_rate_limits = Some(merge_rate_limit_fields(
self.latest_rate_limits.as_ref(),
snapshot,
));
}
pub(crate) fn token_info_and_rate_limits(
&self,
) -> (Option<TokenUsageInfo>, Option<RateLimitSnapshot>) {
(self.token_info(), self.latest_rate_limits.clone())
}
pub(crate) fn set_token_usage_full(&mut self, context_window: i64) {
self.history.set_token_usage_full(context_window);
}
pub(crate) fn get_total_token_usage(&self, server_reasoning_included: bool) -> i64 {
self.history
.get_total_token_usage(server_reasoning_included)
}
pub(crate) fn set_server_reasoning_included(&mut self, included: bool) {
self.server_reasoning_included = included;
}
pub(crate) fn server_reasoning_included(&self) -> bool {
self.server_reasoning_included
}
pub(crate) fn record_mcp_dependency_prompted<I>(&mut self, names: I)
where
I: IntoIterator<Item = String>,
{
self.mcp_dependency_prompted.extend(names);
}
pub(crate) fn mcp_dependency_prompted(&self) -> HashSet<String> {
self.mcp_dependency_prompted.clone()
}
pub(crate) fn set_session_startup_prewarm(
&mut self,
startup_prewarm: SessionStartupPrewarmHandle,
) {
self.startup_prewarm = Some(startup_prewarm);
}
pub(crate) fn take_session_startup_prewarm(&mut self) -> Option<SessionStartupPrewarmHandle> {
self.startup_prewarm.take()
}
// Adds connector IDs to the active set and returns the merged selection.
pub(crate) fn merge_connector_selection<I>(&mut self, connector_ids: I) -> HashSet<String>
where
I: IntoIterator<Item = String>,
{
self.active_connector_selection.extend(connector_ids);
self.active_connector_selection.clone()
}
// Returns the current connector selection tracked on session state.
pub(crate) fn get_connector_selection(&self) -> HashSet<String> {
self.active_connector_selection.clone()
}
// Removes all currently tracked connector selections.
pub(crate) fn clear_connector_selection(&mut self) {
self.active_connector_selection.clear();
}
pub(crate) fn queue_pending_session_start_source(
&mut self,
value: codex_hooks::SessionStartSource,
) {
self.pending_session_start_sources.push_back(value);
}
pub(crate) fn take_pending_session_start_source(
&mut self,
) -> Option<codex_hooks::SessionStartSource> {
self.pending_session_start_sources.pop_front()
}
pub(crate) fn record_granted_permissions(
&mut self,
environment_id: &str,
permissions: AdditionalPermissionProfile,
) {
let granted_permissions = merge_permission_profiles(
self.granted_permissions_by_environment_id
.get(environment_id),
Some(&permissions),
);
if let Some(granted_permissions) = granted_permissions {
self.granted_permissions_by_environment_id
.insert(environment_id.to_string(), granted_permissions);
}
}
pub(crate) fn granted_permissions(
&self,
environment_id: &str,
) -> Option<AdditionalPermissionProfile> {
self.granted_permissions_by_environment_id
.get(environment_id)
.cloned()
}
}
// Sometimes new snapshots don't include credits or plan information.
// Preserve those from the previous snapshot when missing. For `limit_id`, treat
// missing values as the default `"codex"` bucket.
fn merge_rate_limit_fields(
previous: Option<&RateLimitSnapshot>,
mut snapshot: RateLimitSnapshot,
) -> RateLimitSnapshot {
if snapshot.limit_id.is_none() {
snapshot.limit_id = Some("codex".to_string());
}
if snapshot.credits.is_none() {
snapshot.credits = previous.and_then(|prior| prior.credits.clone());
}
if snapshot.individual_limit.is_none() {
snapshot.individual_limit = previous.and_then(|prior| prior.individual_limit.clone());
}
if snapshot.plan_type.is_none() {
snapshot.plan_type = previous.and_then(|prior| prior.plan_type);
}
snapshot
}
#[cfg(test)]
#[path = "session_tests.rs"]
mod tests;