Skip to content

Latest commit

 

History

History
485 lines (318 loc) · 27.9 KB

File metadata and controls

485 lines (318 loc) · 27.9 KB

changes.md — compaction

Compaction re-diverges from upstream dcd4619 (2026-08-25)

What changed

  • packages/coding-agent/src/core/compaction/compaction.ts keeps the fork compaction pipeline: image/text content handling in summaries, the retry surface (policies plus callbacks), and the fork's transport-aware message conversion, re-asserted after this sync's resolution regressed it.

Why

These are fork-owned product surfaces (senpi branding, provider wire behavior, fork runtime features) that upstream does not carry; the sync must re-assert them on top of upstream's tree.

Why this lives in the fork

The divergence lives in core wiring, package identity, or build plumbing that executes before any extension loads, so no extension hook can express it.

Expected merge conflict zones

  • The summarization request assembly and retry wiring inside packages/coding-agent/src/core/compaction/compaction.ts.

2026-08-20 - Ignore implausible billed usage for compaction threshold

What changed

  • packages/coding-agent/src/core/compaction/compaction.ts: adds resolveThresholdContextTokens. If the local estimate is at least 50k and billed usage is more than 8× that estimate, compact against the estimate; otherwise use max(usage, estimate).

Why

  • Cursor dashboard-cumulative cacheRead can be millions while the live window is ~150k. Folding that into the threshold forced a useless compact and a 0-token resource_exhausted.

Why an extension could not handle it

  • Compaction threshold math runs in core before any extension compaction hook is consulted.

Expected merge conflict zones

  • packages/coding-agent/src/core/compaction/compaction.ts resolveThresholdContextTokens

Summarization request identity, watchdog, and summary-safe filtering after the 59a71b23 pin (2026-08-19)

What changed

  • packages/coding-agent/src/core/compaction/compaction.ts: completeSummarization() keeps the fork's affinity/request-identity split instead of upstream's single routing id. Upstream (pin 59a71b235dadb4ad0d67557a8abb0aaa093e68b4) sets sessionId: options.sessionId ?? uuidv7(); the fork sets affinitySessionId: options.affinitySessionId ?? options.sessionId and always mints a fresh sessionId per request, so provider affinity follows the caller's session while each summary request stays its own identity. The same function keeps the fork's request-local AbortController plus consumeStreamWithIdleTimeout() (DEFAULT_SUMMARIZATION_IDLE_TIMEOUT_MS / DEFAULT_SUMMARIZATION_MAX_DURATION_MS) over streamSimple, so a stalled summarization is torn down without aborting the caller's signal.
  • compaction.ts also keeps: extraBody/sessionId/transformContext parameters through generateSummary(), generateSummaryWithUsage(), compact(), and generateTurnPrefixSummary(); contentTextForSummary() in place of contentText(); context-excluded custom-message filtering (contextMessagesForCompactionEntry(), filterContextExcludedMessages()) across cut-point scanning and token estimation; base64-run weighting in estimateTokens(); prepareCompaction(forceProgress, allowSummaryOnly); and the fork compaction settings (speculative, restoration, idle) on CompactionSettings/DEFAULT_COMPACTION_SETTINGS.
  • packages/coding-agent/src/core/compaction/branch-summarization.ts: keeps the session_before_compact emission for branch summaries (reason: "branch", fresh requestId, synthesized CompactionPreparation via createBranchCompactionPreparation(), cancel and precomputed-summary handling), extraBody on GenerateBranchSummaryOptions with the env-carrying BranchSummaryStreamOptions, context-excluded custom entries skipped in getMessageFromEntry(), and contentTextForSummary() for the produced summary text.

Why

  • Upstream's centralized summary-request path reuses the caller's session id as the routing id; the fork's providers key prompt-cache affinity and per-request identity separately, so collapsing them would either move a summary onto the main turn's cache identity or lose affinity entirely. The watchdog, summary-safe content extraction, and context-exclusion filters exist because fork sessions carry provider-native replay blocks and fork-only custom messages that must never enter or stall a summarization request.

Why an extension could not handle it

  • These are the core fallback summarization paths: they run exactly when no extension returned a compaction result, and the branch-summary hook is emitted from inside generateBranchSummary() itself.

Expected merge conflict zones

  • MEDIUM: completeSummarization() request-option construction and the produce/watchdog body — upstream edits this function whenever it changes caching or routing; keep affinitySessionId plus the fresh sessionId.
  • LOW-MEDIUM: the trailing parameter lists on the four public summary functions; the session_before_compact block in branch-summarization.ts.

Repository audit baseline for the compaction tracker (2026-08-17)

What changed

  • This entry is the canonical inventory for the repository-wide changes.md audit (scripts/audit-changes-md.mjs, pin 914cf1472e715297caa30db4b9535d534a9eb718). The audited production paths whose exact nearest tracker is this file: packages/coding-agent/src/core/compaction/compaction.ts, packages/coding-agent/src/core/compaction/branch-summarization.ts, and packages/coding-agent/src/core/compaction/utils.ts.

Why

  • The audit requires every upstream-owned production divergence to be covered by one entry with all four canonical sections in its exact nearest tracker; utils.ts divergence predated the gate and had no entry naming it.

Why an extension could not handle it

  • Tracker coverage is repository policy enforced by repository scripts before any extension loader exists.

Expected merge conflict zones

  • NONE: this tracker merges to ours; the inventory names pin-relative paths so it survives edits below.

Summary content normalization and apply_patch file operations (2026-08-17)

What changed

  • utils.ts: new contentTextForSummary() extracts text from content that may retain provider-native replay blocks by filtering providerNative blocks out of a copy before contentText(); serializeConversation() uses it for user, assistant, and tool-result arms. Provider-native blocks stay on the persisted message for same-provider replay and are never mutated.
  • utils.ts: extractFileOpsFromMessage() recognizes apply_patch calls — patched paths extracted from the patch text via extractPatchedPaths() are recorded as edited — and each tool arm validates its own argument shape instead of sharing one pre-checked path variable.

Why

  • Provider-native blocks are outside pi-ai's portable contentText contract, so summarization either cast them away or crashed; apply_patch mutations were invisible to the file-operation lists that summaries use to describe what the compacted history changed on disk.

Why an extension could not handle it

  • These helpers run inside the core compaction fallback (compact() with no extension result) and inside branch summarization before any extension hook can substitute content.

Expected merge conflict zones

  • LOW: utils.ts extractFileOpsFromMessage() switch and the contentTextForSummary() helper; the gpt-apply-patch import is a fork-only dependency direction.

Summarization request options: extra body, session affinity, env (2026-08-17)

What changed

  • compaction.ts: generateSummary(), generateSummaryWithUsage(), compact(), and generateTurnPrefixSummary() accept extraBody (merged into the outgoing provider payload) and sessionId; createSummarizationOptions() passes extraBody and sets affinitySessionId from the session id.
  • compaction.ts: completeSummarization() keeps summary requests standalone — cacheRetention: "none", a fresh sessionId per request, affinity inherited from the caller's session — and its option type is SimpleStreamOptions & { env } so provider-scoped environment values ride the request.
  • branch-summarization.ts: GenerateBranchSummaryOptions gains extraBody alongside env, and the branch-summary stream options carry the same environment typing.

Why

  • Summaries are one-shot requests: they must not write cache entries nothing will reuse, they must be attributable to the session that produced them, and fork providers need per-request body fields (routing, affinity) applied to the summarization request exactly as to main turns.

Why an extension could not handle it

  • The core fallback summarization request is dispatched inside compact() where extensions that returned no result never see the request; option plumbing at this seam is the only path those requests have.

Expected merge conflict zones

  • LOW-MEDIUM: the option-parameter lists on the four public functions (upstream adds parameters here periodically); re-apply the trailing extraBody/sessionId parameters on sync.

Cut-point fallback, forced progress, and summary-only regeneration (2026-08-17)

What changed

  • compaction.ts: findCutPoint() falls back to the last valid cut point when the token budget is exceeded before any cut point exists at or after the scan index (previously it returned the boundary start and prepared nothing); the backward cut-extension scan also stops at context-excluded custom entries.
  • compaction.ts: prepareCompaction() accepts forceProgress — when the natural cut point equals the boundary start, it advances to the next valid cut point and recomputes the split-turn window — and allowSummaryOnly, which permits regeneration of an existing summary even when no new messages would otherwise be summarized.

Why

  • Overflow recovery and retry-fallback model switches need a compaction that provably shrinks the next prompt: without the fallback, a session whose kept-window landed before the first cut point could not compact at all, and a model switch to a smaller context window could not regenerate an oversized summary.

Why an extension could not handle it

  • Cut-point selection and preparation are pure core functions feeding both the extension hook (session_before_compact) and the core fallback; the admission gates compare against these results.

Expected merge conflict zones

  • MEDIUM: findCutPoint() scan loop and prepareCompaction() cut-point block; keep the fallback and the forceProgress/allowSummaryOnly parameters together with their admission callers in agent-session.ts.

Extension context transform for summarization sources (2026-08-17)

What changed

  • compaction.ts: generateSummaryWithUsage() and compact() accept transformContext; transformSummarySource() runs it over the previous summary (injected as a sentinel message with a unique negative timestamp) plus the current messages, then splits the transformed previous summary back out and re-serializes it for the <previous-summary>; untransformed paths are unchanged.

Why

  • Providers and extensions that rewrite context (sanitization, format conversion, provider-native replay) must apply the same transform to what compaction summarizes, or the summary and the kept window diverge from what the provider actually saw.

Why an extension could not handle it

  • The transform must wrap the exact message array handed to the summarization request inside core; the session_before_compact hook replaces summary content but cannot transform the source window itself.

Expected merge conflict zones

  • LOW: transformSummarySource() and the two call sites; the sentinel-timestamp scheme is fork-owned.

Wall-clock budget includes provider stream acquisition (2026-07-28)

What changed

  • stream-watchdog.ts: consumeStreamWithIdleTimeout() now accepts a promised stream and starts its absolute duration budget before waiting for that promise to resolve.
  • compaction.ts: completeSummarization() passes the provider stream promise directly into the watchdog instead of awaiting connection setup outside the protected interval.

Why

  • A provider adapter that never returned its event stream left compaction permanently stuck before either the idle or wall-clock watchdog existed. The request-local abort controller and normal compaction failure cleanup now run after the same 120s bound whether the provider stalls before or after stream creation.
  • Session 019fa809-5ef4-7db3-bdc3-048da7e0fd9d exposed the user-visible failure mode: the TUI stayed in compaction long enough to appear permanently frozen while provider-side summarization work held the session lifecycle open.

Expected merge conflict zones

  • LOW: stream-watchdog.ts around promised-stream acquisition.
  • LOW: compaction.ts around the completeSummarization() stream setup.

Wall-clock budget for summarization streams (2026-07-28)

What changed

  • stream-watchdog.ts: consumeStreamWithIdleTimeout() accepts an optional maxDurationMs and throws the new StreamDurationBudgetError when one stream outlives it. The budget is a single absolute deadline for the whole stream, not a per-read timer, and it is cleared alongside the idle timer. Caller aborts still win over the budget.
  • DEFAULT_SUMMARIZATION_MAX_DURATION_MS = 120s, applied by compaction.ts completeSummarization() and the extension's speculative.ts request path. retryAssistantCall applies it per attempt.

Why

  • The idle watchdog only catches a silent connection. A summarization stream that keeps trickling events stays under the 300s idle budget indefinitely, and that work is serialized on AgentSession's agent-event queue, which beforeToolCall waits on before every tool prepare. A live-but-slow summarization therefore froze a whole session: tool results withheld at the parallel-batch barrier, typed input queued, TUI stuck on "Working", recoverable only by ESC (which releases compaction before the run signal in _abortActiveAgentAndRetry).
  • Observed in a real session: two freezes of 241s and 208s, both under the idle cap, on a session whose earlier auto-compaction had already blocked the same queue for 44s.

Expected merge conflict zones

  • LOW: stream-watchdog.ts around the contender race in consumeStreamWithIdleTimeout().
  • LOW: compaction.ts around the consumeStreamWithIdleTimeout call in completeSummarization().

Lifecycle ownership and required-admission safety (2026-07-23)

What changed

  • lifecycle.ts now owns the active compaction controller together with reducer transitions, so feedback from an older generation cannot progress or terminate a newer one. Feedback-only cancellation emits one terminal compaction_end, and accepted compactions emit their terminal event before session_compact handlers can start another generation.
  • Extension contexts retain the signal returned by beginCompaction() and supply it to legacy updateCompaction() / endCompaction() calls that omit one. Core accepts feedback mutations only from the current signal.
  • Provider admissions now share one required-compaction gate for prompt preflight, extension-triggered turns, and next turns. Silent provider overflow and threshold-required compaction synchronously stop agent-core's post-agent_end queue drain so only an accepted AgentSession recovery may resume queued work, and overflow can force a split-turn preparation when keeping the only oversized prompt would otherwise leave no compactable source.
  • Compaction rejects stale source snapshots with stale-revision before the durable entry append.
  • Retry fallback model changes invalidate prior-model compaction and re-check the selected model's context window. Summary-only re-compaction is allowed only for this retry boundary.
  • Assistant history is classified around the latest compaction by persisted branch order; an older payload timestamp cannot hide a message whose entry was appended after the compaction boundary.
  • Execution routes pass their own controller into core compaction; an auto request supersedes unrelated feedback instead of inheriting/promoting its controller and leaving outer compaction state stuck.
  • The one-turn post-compaction and post-retry stale-usage exemptions are shared across synchronous queue ownership, asynchronous checking, and admission resampling, while explicit provider overflow is never exempt.

Why

A late extension completion could overwrite fresh feedback, and some continuation routes skipped required compaction. Compacting a source that changed during summary generation could also append a stale checkpoint over intervening work.

Expected merge conflict zones

  • LOW: lifecycle.ts and the compaction admission calls in agent-session.ts.

Operation lifecycle reducer (2026-07-23)

What changed

  • lifecycle.ts adds the pure idle / running / completed / failed / aborted transition model used by AgentSession, including monotonic generations, feedback-to-execution promotion, and stale terminal-event rejection.

Why

  • Compaction completion must remain observable after controllers are released, while delayed work from an older generation must not overwrite the active operation.

Expected merge conflict zones

  • NONE: lifecycle.ts is a new fork-owned module.

Summarization stream idle watchdog (2026-07-21)

What changed

  • stream-watchdog.ts (new, fork-owned): consumeStreamWithIdleTimeout() drains an event stream and throws StreamIdleTimeoutError when no provider event arrives within the idle budget (default 300s, DEFAULT_SUMMARIZATION_IDLE_TIMEOUT_MS, matching the agent stream idle-timeout default). On trip it aborts a request-local controller and returns the iterator; caller aborts end the wait quietly so ESC still reads as the stream's own aborted result.
  • compaction.ts completeSummarization(): both the streamSimple and custom-streamFn routes now consume the summarization stream through the watchdog under a request-local AbortController linked to the caller's signal, instead of awaiting completeSimple() / stream.result() with no bound.

Why

Local compaction summarization had no timeout at any layer: a stalled provider/gateway connection hung the session on "Compacting…" forever (observed: 11+ minutes, recovered only by ESC abort). The agent loop has had this protection for main turns (StreamIdleTimeoutError in packages/agent); this ports the same guarantee to compaction requests.

Why extension system couldn't handle this

  • The core compact() fallback route (session_before_compact handlers returning no result) dispatches its own summarization request inside core; extensions cannot bound a request they never see.

Expected merge conflict zones

  • MEDIUM: compaction.ts around completeSummarization() and the pi-ai/compat import (completeSimplestreamSimple).
  • NONE: stream-watchdog.ts is a new file.

Base64-aware token estimation (2026-07-18)

What changed

  • compaction.ts: estimateTokens() now weights long unbroken base64-ish runs (512+ chars of [A-Za-z0-9+/=_-]) at ~1 token per character instead of the chars/4 prose heuristic. Applied to string/text-block content, tool-call arguments, and bash output via a shared weightedChars() helper.

Why

  • Providers tokenize base64 near 1 token/char. A tool result carrying a ~1 MB inline screenshot data URL estimated at ~256K tokens while Anthropic counted ~1M, so pre-flight compaction never triggered and the provider rejected the request (prompt is too long: 1029893 tokens > 1000000 maximum). Real reproducer: session 019f711b-587a-75ba-9eda-48fd5b2c2c01 (compaction recorded tokensBefore: 319506 for a context the provider counted at 1.03M).

Why extension system couldn't handle this

  • estimateTokens() is core and feeds estimateContextTokens(), which agent-session.ts uses for the pre-prompt compaction gate before any extension sees the turn.

Expected merge conflict zones

  • LOW: compaction.ts around estimateTextAndImageContentChars() and the estimateTokens() switch arms. Keep the weighting applied to every text surface the estimator counts.

Split-turn compaction serialization sync (2026-07-02)

What changed

  • compaction.ts: accepted upstream serialization of split-turn compaction summaries so single-concurrency providers do not receive overlapping generations.

Why

  • Split-turn compaction can be triggered while the session is still processing summary work. Serializing those summaries avoids provider-side 429/concurrency failures and keeps compaction state deterministic.

Why extension system couldn't handle this

  • The serialization boundary is inside core compaction preparation/execution. Extensions can provide or observe summaries, but they cannot serialize the underlying core summary request queue from outside.

Expected merge conflict zones

  • LOW: compaction.ts around summary generation scheduling and split-turn helper calls.

Plugsuit-style Threshold Foundation (2026-04-28)

What changed

  • compaction.ts: Added speculative compaction settings fields (speculativeEnabled, speculativeFraction, speculativeCooldownMs) to CompactionSettings and defaults.
  • extensions/builtin/compaction/policy.ts: Removed the 0.78 OMO threshold floor. Effective threshold now follows the adaptive plugsuit-style tiers directly (0.45/0.50/0.55/0.60/0.65), with yield adjustment clamped to the existing 0.4-0.7 adaptive range.
  • extensions/builtin/compaction/policy.ts: Added SPECULATIVE_FRACTION, shouldStartSpeculativeCompaction(), computeEffectiveKeepRecentTokens(), and isAtHardLimit() for later speculative/emergency phases.
  • settings-manager.ts: Resolved compaction settings now include speculative and restoration fields.
  • extensions/builtin/compaction/index.ts and speculative.ts: Builtin compaction uses resolved settings from ExtensionContext instead of hardcoded defaults for before-turn threshold checks and snapshot preparation.

Why

  • Plugsuit starts compaction much earlier than the OMO 78% floor. Keeping the floor made senpi's auto-compaction late and mostly reactive.
  • Removing the floor alone is unsafe for small context windows because the default keepRecentTokens (20000) can exceed the useful compactable range. The effective keep-recent cap prevents early thresholds from producing empty preparations.
  • Speculative and emergency phases need stable policy functions and settings keys before they can be wired safely.

Why extension system couldn't handle this

  • The policy constants live in the builtin compaction extension and must be shared by unit tests, speculative snapshots, and future emergency pruning.
  • Resolved settings are owned by core SettingsManager; builtin extensions needed a typed ExtensionContext reader to avoid bypassing user settings.json.

Modified upstream files

  • compaction.ts — additive CompactionSettings fields and defaults.
  • settings-manager.ts — resolved setting defaults for new compaction fields.

Expected merge conflict zones

  • LOW: compaction.ts settings interface/defaults.
  • MEDIUM: settings-manager.ts CompactionSettings and getCompactionSettings() if upstream changes settings shape.

Migration notes

  • Preserve the invariant that adaptive threshold and effective keep-recent cap are updated together. Do not reintroduce a hard floor without also proving small-context compaction can still prepare non-empty summaries.

prepareCompaction Rejects Empty Summarization (2026-04-28)

What changed

  • compaction.ts: prepareCompaction() now returns undefined when both messagesToSummarize and turnPrefixMessages are empty.
  • _executeCompaction() (unchanged) reaches its existing "Nothing to compact (session too small)" error path, which surfaces as a clear failure instead of silently invoking the LLM with an empty <conversation> block.

Why

When keepRecentTokens (default 20000) is larger than the total session token count, findCutPoint defaults to the first valid cut point and then findCutPoint's backward scan extends the cut all the way to entry 0 (model_change / thinking_level_change). The result was a preparation with messagesToSummarize: [], turnPrefixMessages: [], and firstKeptEntryId pointing at the very first non-message entry. The new builtin compaction extension then called the LLM with an empty <conversation></conversation> block and the 9-section prompt's R2 rule ("If a section has no content, write 'None.'") forced the model to emit None. for every section. That all-None. summary was persisted as a real compaction entry, destroying the conversation that should have been summarized.

A real reproducer: ~/.senpi/agent/sessions/--Users-yeongyu-local-workspaces-senpi-mono--/2026-04-28T01-50-51-950Z_*.jsonl contains two consecutive compactions on a tiny Kimi K2.6 hello session, both stored as all-None. summaries with tokensBefore of 11527 and 11690.

Why extension system couldn't handle this

prepareCompaction() is core; it computes the cut point, the messages to summarize, and the previous summary. Extensions can override the summary content via session_before_compact, but they cannot decide whether the core preparation step itself should reject the request. Without this guard in core, every extension and the upstream fallback compact() call would have to repeat the same emptiness check.

Modified upstream files

  • compaction.tsprepareCompaction() returns undefined when there is nothing to summarize.

Expected merge conflict zones

  • LOW: compaction.ts prepareCompaction() is rarely changed upstream. The guard is a small additive check immediately before the final return; conflict resolution is to keep the guard and apply it after upstream's preparation logic computes messagesToSummarize / turnPrefixMessages.

Migration notes

If upstream changes prepareCompaction() to compute additional summary inputs (for example a separate "trailing reminders" array), extend the emptiness guard to include them. The invariant: never return a defined CompactionPreparation whose total summarizable content is empty.

Branch Summarization Routes Through Compaction Hook (2026-04-27)

What changed

  • branch-summarization.ts: generateBranchSummary() now emits session_before_compact with reason: "branch" before the default branch prompt path when an extension runner is provided.
  • branch-summarization.ts: Branch entries are converted into an equivalent CompactionPreparation object for extensions.
  • branch-summarization.ts: Extension { compaction: CompactionResult } responses override the branch summary; { cancel: true } aborts branch summarization.

Why

  • Branch summary was a separate route with a different prompt and no Critical Context section, causing the 9 inconsistencies the user listed.
  • Routing through session_before_compact lets the builtin extension provide one canonical 9-section prompt across all 6 routes.
  • The existing BRANCH_SUMMARY_PROMPT remains the fallback when no extension overrides.

Why extension system couldn't handle this

The branch summarization path did not emit a compaction event before building its default prompt. Extensions can only replace branch summary content after this seam exists in core.

Modified upstream files

  • branch-summarization.ts — emits session_before_compact for branch summaries and accepts extension-provided compaction summaries.

Expected merge conflict zones

  • LOW: branch-summarization.ts is rarely touched upstream. If upstream changes branch summary preparation, keep the hook emission before default prompt construction and update the CompactionPreparation mapping to match the new data flow.

Migration notes

If upstream changes branch summary preparation or adds new branch summary data sources, keep the session_before_compact hook emission before default prompt construction and update the CompactionPreparation mapping to match the new data flow. The BRANCH_SUMMARY_PROMPT fallback must remain intact for sessions without the compaction extension.