pi-mega-compact 0.8.6 → 0.8.7
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
|
@@ -0,0 +1,189 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* mega-cache-replay.test.ts — locks the v0.8.7 cache-stability fix.
|
|
3
|
+
*
|
|
4
|
+
* Two tests (reuses the mega-teamrun.test.ts harness shape: mock pi + the REAL
|
|
5
|
+
* compiled extension at extensions/mega-compact.js):
|
|
6
|
+
* a. REPLAY: drive >=2 gated context events past the debounce within ONE epoch
|
|
7
|
+
* (same lastCheckpointId) and assert diagLiveTrimReplays > 0 AND the returned
|
|
8
|
+
* messages array is byte-identical (deepEqual) across replays (stable prefix).
|
|
9
|
+
* b. DEDUP-ON-DIFFERENT-CHECKPOINT: after a fresh trim, simulate a re-compact
|
|
10
|
+
* (context grew on the token basis) that DEDUPS onto a DIFFERENT existing
|
|
11
|
+
* checkpoint id (L0 contentHash match against an OLDER checkpoint, so
|
|
12
|
+
* result.checkpointId != rt.lastCheckpointId), then assert the NEXT gated
|
|
13
|
+
* event STILL replays (diagLiveTrimReplays increments) — i.e. the cache key
|
|
14
|
+
* trimCache.checkpointId === rt.lastCheckpointId holds. This is the P2 gap the
|
|
15
|
+
* v0.8.6 audit found: keying on the dedup-volatile result.checkpointId
|
|
16
|
+
* disabled replay for the rest of the epoch after such a dedup fire.
|
|
17
|
+
*
|
|
18
|
+
* MEGACOMPACT_PGLITE_DISABLED keeps the run fast (no WASM index init).
|
|
19
|
+
*/
|
|
20
|
+
import { test } from "node:test";
|
|
21
|
+
import assert from "node:assert/strict";
|
|
22
|
+
import { mkdtempSync, rmSync } from "node:fs";
|
|
23
|
+
import { tmpdir } from "node:os";
|
|
24
|
+
import { join } from "node:path";
|
|
25
|
+
import { createRequire } from "node:module";
|
|
26
|
+
import { closeVectorIndex } from "../src/store/vectorIndex.js";
|
|
27
|
+
const require = createRequire(import.meta.url);
|
|
28
|
+
const baseTmp = mkdtempSync(join(tmpdir(), "mc-cache-"));
|
|
29
|
+
process.env.MEGACOMPACT_INDEX_DIR = join(baseTmp, "index");
|
|
30
|
+
process.env.MEGACOMPACT_PGLITE_DISABLED = "true"; // fast: skip WASM index
|
|
31
|
+
let counter = 0;
|
|
32
|
+
function harness() {
|
|
33
|
+
const stateDir = join(baseTmp, `run-${counter++}`);
|
|
34
|
+
process.env.MEGACOMPACT_STATE_DIR = stateDir;
|
|
35
|
+
process.env.MEGACOMPACT_DEBUG = "true";
|
|
36
|
+
process.env.MEGACOMPACT_THRESHOLD_TOKENS = "50";
|
|
37
|
+
process.env.MEGACOMPACT_FAST_GATE_PCT = "1";
|
|
38
|
+
process.env.MEGACOMPACT_ANCHOR_USER_MESSAGES = "1";
|
|
39
|
+
process.env.MEGACOMPACT_DURABLE_TRIM_FLOOR = "0";
|
|
40
|
+
process.env.MEGACOMPACT_MEMORY_AUTO_REVIEW = "false";
|
|
41
|
+
process.env.MEGACOMPACT_RAPTOR_ENABLED = "false";
|
|
42
|
+
// Disable the FUZZY dedup tiers (L1 MinHash/LSH + L2 cosine) so the DEDUP test
|
|
43
|
+
// is controlled by L0 contentHash only: setB (different vocabulary) then
|
|
44
|
+
// creates a genuinely NEW checkpoint instead of fuzzy-matching setA's, and
|
|
45
|
+
// setA-re-again still L0-contentHash-dedups onto the first setA checkpoint.
|
|
46
|
+
// The TrigramEmbedder otherwise matches on shared structural trigrams
|
|
47
|
+
// ("— step N" / "Edit"), collapsing setB onto setA. Harmless for the REPLAY
|
|
48
|
+
// test (pure replay, no dedup reliance).
|
|
49
|
+
process.env.MEGACOMPACT_L1_ENABLED = "false";
|
|
50
|
+
process.env.MEGACOMPACT_L2_ENABLED = "false";
|
|
51
|
+
delete process.env.MEGACOMPACT_LEGACY_DURABLE_TRIM;
|
|
52
|
+
// Mutable context-usage so tests can drive re-compact on the TOKEN basis by
|
|
53
|
+
// raising `tokens` while keeping `percent` null (→ token gate + token
|
|
54
|
+
// grewEnough path in context-handler.ts). The REPLAY test keeps percent=100 so
|
|
55
|
+
// the percent-basis grewEnough (>=10) never trips (no re-compact → pure replay).
|
|
56
|
+
const usage = { tokens: 200000, contextWindow: 200000, percent: 100 };
|
|
57
|
+
const handlers = {};
|
|
58
|
+
const compactCalls = [];
|
|
59
|
+
function msg(role, text, toolName) {
|
|
60
|
+
if (role === "assistant" && toolName) {
|
|
61
|
+
return { role: "assistant", content: [{ type: "toolCall", name: toolName, id: "c1", arguments: {} }], api: "anthropic-messages", provider: "anthropic", model: "m", usage: { inputTokens: 1, outputTokens: 1, cacheReadTokens: 0, cacheWriteTokens: 0 }, stopReason: "tool_use", timestamp: 0 };
|
|
62
|
+
}
|
|
63
|
+
if (role === "toolResult" && toolName) {
|
|
64
|
+
return { role: "toolResult", content: [{ type: "text", text }], toolCallId: "c1", toolName, isError: false, timestamp: 0 };
|
|
65
|
+
}
|
|
66
|
+
return { role: "user", content: text, timestamp: 0 };
|
|
67
|
+
}
|
|
68
|
+
// Build a session of `n` tool-call triples tagged `tag`. Set A and B differ in
|
|
69
|
+
// content (so B never dedups against A) but A === A reproduces the same
|
|
70
|
+
// regionText → same L0 contentHash → dedup onto the first A checkpoint.
|
|
71
|
+
function buildSession(tag, n) {
|
|
72
|
+
const s = [];
|
|
73
|
+
for (let i = 0; i < n; i++) {
|
|
74
|
+
s.push(msg("user", `[${tag}] we decided to use approach ${i} for module ${i}`));
|
|
75
|
+
s.push(msg("assistant", `[${tag}] edited module ${i}`, "Edit"));
|
|
76
|
+
s.push(msg("toolResult", `[${tag}] edited module ${i}`, "Edit"));
|
|
77
|
+
}
|
|
78
|
+
return s;
|
|
79
|
+
}
|
|
80
|
+
const toEntry = (m, i) => ({ type: "message", id: `e${i}`, parentId: null, timestamp: String(i), message: m });
|
|
81
|
+
const sessionManager = {
|
|
82
|
+
getSessionId: () => "sess_cache_001",
|
|
83
|
+
getEntries: () => buildSession("A", 14).map(toEntry),
|
|
84
|
+
getBranch: () => buildSession("A", 14).map(toEntry),
|
|
85
|
+
};
|
|
86
|
+
function makeCtx(over = {}) {
|
|
87
|
+
return {
|
|
88
|
+
ui: { setStatus: () => { }, notify: () => { }, select: () => { }, confirm: async () => true, input: async () => "", setWidget: () => { } },
|
|
89
|
+
mode: "tui", hasUI: true, cwd: stateDir, sessionManager,
|
|
90
|
+
modelRegistry: {}, model: undefined, isIdle: () => true, isProjectTrusted: () => true,
|
|
91
|
+
signal: undefined, abort: () => { }, hasPendingMessages: () => false, shutdown: () => { },
|
|
92
|
+
getContextUsage: () => ({ ...usage }),
|
|
93
|
+
compact: (opts) => { compactCalls.push(opts); return undefined; },
|
|
94
|
+
getSystemPrompt: () => "system base",
|
|
95
|
+
...over,
|
|
96
|
+
};
|
|
97
|
+
}
|
|
98
|
+
const pi = {
|
|
99
|
+
on: (ev, h) => { handlers[ev] = h; },
|
|
100
|
+
registerCommand: () => { }, registerTool: () => { }, registerShortcut: () => { },
|
|
101
|
+
registerFlag: () => { }, getFlag: () => undefined, registerMessageRenderer: () => { },
|
|
102
|
+
registerEntryRenderer: () => { }, sendMessage: () => { }, sendUserMessage: () => { },
|
|
103
|
+
appendEntry: () => { }, setSessionName: () => { }, getSessionName: () => undefined,
|
|
104
|
+
setLabel: () => { }, exec: async () => ({ stdout: "", stderr: "", code: 0 }),
|
|
105
|
+
getActiveTools: () => [], getAllTools: () => [], setActiveTools: () => { },
|
|
106
|
+
getCommands: () => [], setModel: async () => false, getThinkingLevel: () => "off",
|
|
107
|
+
setThinkingLevel: () => { },
|
|
108
|
+
};
|
|
109
|
+
const mod = require("./mega-compact.js");
|
|
110
|
+
mod.default(pi);
|
|
111
|
+
const { lastRuntime } = require("./mega-events.js");
|
|
112
|
+
const fire = (ev, event, ctx) => handlers[ev](event, ctx);
|
|
113
|
+
return {
|
|
114
|
+
stateDir, handlers, compactCalls, fire, ctx: makeCtx, usage, buildSession,
|
|
115
|
+
runtime: lastRuntime, // MegaRuntime with diag* counters + rt + trimCache
|
|
116
|
+
// Bypass the 2s debounce so each fire proceeds without real waiting.
|
|
117
|
+
clearDebounce: () => { if (lastRuntime)
|
|
118
|
+
lastRuntime.debounceUntil = 0; },
|
|
119
|
+
};
|
|
120
|
+
}
|
|
121
|
+
test("REPLAY: >=2 gated context events within one epoch replay verbatim (byte-identical)", async () => {
|
|
122
|
+
const h = harness();
|
|
123
|
+
const ctx = h.ctx();
|
|
124
|
+
const session = h.buildSession("A", 14);
|
|
125
|
+
// Fire 3 gated context events; clearDebounce between so each passes the gate.
|
|
126
|
+
// percent stays 100 → percent-basis grewEnough (>=10) never trips → pure replay.
|
|
127
|
+
h.clearDebounce();
|
|
128
|
+
const r1 = await h.fire("context", { type: "context", messages: session }, ctx);
|
|
129
|
+
h.clearDebounce();
|
|
130
|
+
const r2 = await h.fire("context", { type: "context", messages: session }, ctx);
|
|
131
|
+
h.clearDebounce();
|
|
132
|
+
const r3 = await h.fire("context", { type: "context", messages: session }, ctx);
|
|
133
|
+
const rt = h.runtime;
|
|
134
|
+
assert.ok(rt.diagLiveTrimFires >= 1, "fresh trim fired on first event");
|
|
135
|
+
assert.ok(rt.diagLiveTrimReplays >= 2, `replay fired >=2 (got ${rt.diagLiveTrimReplays})`);
|
|
136
|
+
// byte-identical (stable KV-cache prefix) across replays
|
|
137
|
+
assert.deepEqual(r2?.messages, r3?.messages, "replay messages byte-identical across replays");
|
|
138
|
+
// replay matches the fresh-trim view (shallow-copy preserves content)
|
|
139
|
+
assert.deepEqual(r1?.messages, r2?.messages, "replay matches fresh-trim view (stable prefix)");
|
|
140
|
+
});
|
|
141
|
+
test("DEDUP: re-compact that dedups onto a DIFFERENT checkpoint still replays next (P2 fix)", async () => {
|
|
142
|
+
const h = harness();
|
|
143
|
+
// Token-basis growth path: percent null → token gate + token grewEnough
|
|
144
|
+
// (currentTokens - trimCache.ctxTokens >= effectiveThreshold * 0.5 = 25).
|
|
145
|
+
h.usage.percent = null;
|
|
146
|
+
h.usage.tokens = 200000;
|
|
147
|
+
const ctx = h.ctx();
|
|
148
|
+
const setA = h.buildSession("A", 14);
|
|
149
|
+
const setB = h.buildSession("B", 14); // different content, same length
|
|
150
|
+
const rt = h.runtime;
|
|
151
|
+
// 1) Fresh trim on setA → genuinely new checkpoint cp_A. lastCheckpointId = cp_A.
|
|
152
|
+
h.clearDebounce();
|
|
153
|
+
await h.fire("context", { type: "context", messages: setA }, ctx);
|
|
154
|
+
const cpA = rt.rt.lastCheckpointId;
|
|
155
|
+
assert.ok(cpA, "cp_A created on fresh trim");
|
|
156
|
+
assert.equal(rt.diagLiveTrimFires, 1, "first fire was a fresh trim");
|
|
157
|
+
// 2) Re-compact on setB (grew tokens) → genuinely new checkpoint cp_B (not deduped).
|
|
158
|
+
h.usage.tokens = 200100; // grew 100 >= 25
|
|
159
|
+
h.clearDebounce();
|
|
160
|
+
await h.fire("context", { type: "context", messages: setB }, ctx);
|
|
161
|
+
const cpB = rt.rt.lastCheckpointId;
|
|
162
|
+
assert.notEqual(cpB, cpA, "cp_B is a different, genuinely new checkpoint");
|
|
163
|
+
assert.equal(rt.rt.dedupSkips, 0, "setB did not dedup (different vocabulary, fuzzy tiers off)");
|
|
164
|
+
// 3) Re-compact on setA AGAIN (grew tokens) → L0 contentHash dedup onto cp_A.
|
|
165
|
+
// result.checkpointId = cp_A (!= lastCheckpointId cp_B); lastCheckpointId is
|
|
166
|
+
// NOT updated on a dedup (compact.ts:100-104), so it stays cp_B. With the
|
|
167
|
+
// fix, trimCache.checkpointId is keyed on lastCheckpointId (cp_B), NOT the
|
|
168
|
+
// dedup-volatile result.checkpointId (cp_A).
|
|
169
|
+
h.usage.tokens = 200200; // grew 100 >= 25
|
|
170
|
+
h.clearDebounce();
|
|
171
|
+
await h.fire("context", { type: "context", messages: setA }, ctx);
|
|
172
|
+
assert.equal(rt.rt.lastCheckpointId, cpB, "dedup did NOT bump lastCheckpointId (still cp_B)");
|
|
173
|
+
assert.ok(rt.rt.dedupSkips >= 1, "setA re-compact deduped onto an existing checkpoint");
|
|
174
|
+
// The P2 invariant: the cache key must equal the stable epoch signal.
|
|
175
|
+
assert.equal(rt.trimCache?.checkpointId, rt.rt.lastCheckpointId, "trimCache.checkpointId keyed on lastCheckpointId (P2 fix), not dedup-volatile result.checkpointId");
|
|
176
|
+
// 4) Next gated event (no growth) MUST replay instead of re-running runCompact.
|
|
177
|
+
// Without the fix, trimCache.checkpointId (cp_A) != lastCheckpointId (cp_B)
|
|
178
|
+
// → the replay condition is false → runCompact re-runs every fire → the
|
|
179
|
+
// thrash silently persists in that path (the audit's finding).
|
|
180
|
+
const replaysBefore = rt.diagLiveTrimReplays;
|
|
181
|
+
h.usage.tokens = 200200; // no growth → replay
|
|
182
|
+
h.clearDebounce();
|
|
183
|
+
await h.fire("context", { type: "context", messages: setA }, ctx);
|
|
184
|
+
assert.ok(rt.diagLiveTrimReplays > replaysBefore, `replay fired after dedup-onto-different-checkpoint (got ${rt.diagLiveTrimReplays}, was ${replaysBefore})`);
|
|
185
|
+
});
|
|
186
|
+
test("cleanup", async () => {
|
|
187
|
+
await closeVectorIndex();
|
|
188
|
+
rmSync(baseTmp, { recursive: true, force: true });
|
|
189
|
+
});
|
|
@@ -146,7 +146,9 @@ export function registerContextHandler(pi, runtime, config) {
|
|
|
146
146
|
runtime.diagLiveTrimFires++; // trim view returned this call (replay counts as a fire)
|
|
147
147
|
runtime.diagLiveTrimReplays++;
|
|
148
148
|
runtime.snapshot(ctx);
|
|
149
|
-
|
|
149
|
+
// v0.8.7: shallow-copy the cached summary so pi's transformContext can't
|
|
150
|
+
// mutate the shared reference across replays (audit P3).
|
|
151
|
+
return { messages: [{ ...runtime.trimCache.summaryAgentMsg }, ...recent] };
|
|
150
152
|
}
|
|
151
153
|
// else: context grew enough → fall through to re-compact (cache is stale)
|
|
152
154
|
}
|
|
@@ -262,7 +264,20 @@ export function registerContextHandler(pi, runtime, config) {
|
|
|
262
264
|
// replay it verbatim (stabilizing the KV-cache prefix) instead of
|
|
263
265
|
// regenerating a fresh summary + sentinel every fire.
|
|
264
266
|
runtime.trimCache = {
|
|
265
|
-
|
|
267
|
+
// v0.8.7: key the replay cache on the STABLE epoch signal
|
|
268
|
+
// (rt.lastCheckpointId) instead of ran.result.checkpointId, which is
|
|
269
|
+
// dedup-volatile: on a re-compact that dedups onto a DIFFERENT existing
|
|
270
|
+
// checkpoint, result.checkpointId is the matched id (engine.ts:188) while
|
|
271
|
+
// lastCheckpointId is only updated on a genuinely new checkpoint
|
|
272
|
+
// (compact.ts:100-104). Keying on result.checkpointId would make
|
|
273
|
+
// trimCache.checkpointId != rt.lastCheckpointId forever after that
|
|
274
|
+
// dedup fire, disabling replay for the rest of the epoch (the
|
|
275
|
+
// alternating cache-miss that 0.8.6 meant to fix). Prefer the stable
|
|
276
|
+
// signal; fall back to result.checkpointId then the epoch timestamp
|
|
277
|
+
// only for the no-checkpoint edge case.
|
|
278
|
+
checkpointId: runtime.rt.lastCheckpointId ??
|
|
279
|
+
ran.result.checkpointId ??
|
|
280
|
+
`epoch-${runtime.rt.lastCompactAt ?? Date.now()}`,
|
|
266
281
|
cut,
|
|
267
282
|
summaryAgentMsg,
|
|
268
283
|
ctxPct: pct ?? null,
|
|
@@ -0,0 +1,211 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* mega-cache-replay.test.ts — locks the v0.8.7 cache-stability fix.
|
|
3
|
+
*
|
|
4
|
+
* Two tests (reuses the mega-teamrun.test.ts harness shape: mock pi + the REAL
|
|
5
|
+
* compiled extension at extensions/mega-compact.js):
|
|
6
|
+
* a. REPLAY: drive >=2 gated context events past the debounce within ONE epoch
|
|
7
|
+
* (same lastCheckpointId) and assert diagLiveTrimReplays > 0 AND the returned
|
|
8
|
+
* messages array is byte-identical (deepEqual) across replays (stable prefix).
|
|
9
|
+
* b. DEDUP-ON-DIFFERENT-CHECKPOINT: after a fresh trim, simulate a re-compact
|
|
10
|
+
* (context grew on the token basis) that DEDUPS onto a DIFFERENT existing
|
|
11
|
+
* checkpoint id (L0 contentHash match against an OLDER checkpoint, so
|
|
12
|
+
* result.checkpointId != rt.lastCheckpointId), then assert the NEXT gated
|
|
13
|
+
* event STILL replays (diagLiveTrimReplays increments) — i.e. the cache key
|
|
14
|
+
* trimCache.checkpointId === rt.lastCheckpointId holds. This is the P2 gap the
|
|
15
|
+
* v0.8.6 audit found: keying on the dedup-volatile result.checkpointId
|
|
16
|
+
* disabled replay for the rest of the epoch after such a dedup fire.
|
|
17
|
+
*
|
|
18
|
+
* MEGACOMPACT_PGLITE_DISABLED keeps the run fast (no WASM index init).
|
|
19
|
+
*/
|
|
20
|
+
import { test } from "node:test";
|
|
21
|
+
import assert from "node:assert/strict";
|
|
22
|
+
import { mkdtempSync, rmSync } from "node:fs";
|
|
23
|
+
import { tmpdir } from "node:os";
|
|
24
|
+
import { join } from "node:path";
|
|
25
|
+
import { createRequire } from "node:module";
|
|
26
|
+
import { closeVectorIndex } from "../src/store/vectorIndex.js";
|
|
27
|
+
import type { AgentMessage } from "@earendil-works/pi-agent-core";
|
|
28
|
+
|
|
29
|
+
const require = createRequire(import.meta.url);
|
|
30
|
+
const baseTmp = mkdtempSync(join(tmpdir(), "mc-cache-"));
|
|
31
|
+
process.env.MEGACOMPACT_INDEX_DIR = join(baseTmp, "index");
|
|
32
|
+
process.env.MEGACOMPACT_PGLITE_DISABLED = "true"; // fast: skip WASM index
|
|
33
|
+
let counter = 0;
|
|
34
|
+
|
|
35
|
+
function harness() {
|
|
36
|
+
const stateDir = join(baseTmp, `run-${counter++}`);
|
|
37
|
+
process.env.MEGACOMPACT_STATE_DIR = stateDir;
|
|
38
|
+
process.env.MEGACOMPACT_DEBUG = "true";
|
|
39
|
+
process.env.MEGACOMPACT_THRESHOLD_TOKENS = "50";
|
|
40
|
+
process.env.MEGACOMPACT_FAST_GATE_PCT = "1";
|
|
41
|
+
process.env.MEGACOMPACT_ANCHOR_USER_MESSAGES = "1";
|
|
42
|
+
process.env.MEGACOMPACT_DURABLE_TRIM_FLOOR = "0";
|
|
43
|
+
process.env.MEGACOMPACT_MEMORY_AUTO_REVIEW = "false";
|
|
44
|
+
process.env.MEGACOMPACT_RAPTOR_ENABLED = "false";
|
|
45
|
+
// Disable the FUZZY dedup tiers (L1 MinHash/LSH + L2 cosine) so the DEDUP test
|
|
46
|
+
// is controlled by L0 contentHash only: setB (different vocabulary) then
|
|
47
|
+
// creates a genuinely NEW checkpoint instead of fuzzy-matching setA's, and
|
|
48
|
+
// setA-re-again still L0-contentHash-dedups onto the first setA checkpoint.
|
|
49
|
+
// The TrigramEmbedder otherwise matches on shared structural trigrams
|
|
50
|
+
// ("— step N" / "Edit"), collapsing setB onto setA. Harmless for the REPLAY
|
|
51
|
+
// test (pure replay, no dedup reliance).
|
|
52
|
+
process.env.MEGACOMPACT_L1_ENABLED = "false";
|
|
53
|
+
process.env.MEGACOMPACT_L2_ENABLED = "false";
|
|
54
|
+
delete process.env.MEGACOMPACT_LEGACY_DURABLE_TRIM;
|
|
55
|
+
|
|
56
|
+
// Mutable context-usage so tests can drive re-compact on the TOKEN basis by
|
|
57
|
+
// raising `tokens` while keeping `percent` null (→ token gate + token
|
|
58
|
+
// grewEnough path in context-handler.ts). The REPLAY test keeps percent=100 so
|
|
59
|
+
// the percent-basis grewEnough (>=10) never trips (no re-compact → pure replay).
|
|
60
|
+
const usage = { tokens: 200000, contextWindow: 200000, percent: 100 as number | null };
|
|
61
|
+
|
|
62
|
+
const handlers: Record<string, Function> = {};
|
|
63
|
+
const compactCalls: any[] = [];
|
|
64
|
+
|
|
65
|
+
function msg(role: string, text: string, toolName?: string): AgentMessage {
|
|
66
|
+
if (role === "assistant" && toolName) {
|
|
67
|
+
return { role: "assistant", content: [{ type: "toolCall", name: toolName, id: "c1", arguments: {} }], api: "anthropic-messages", provider: "anthropic", model: "m", usage: { inputTokens: 1, outputTokens: 1, cacheReadTokens: 0, cacheWriteTokens: 0 }, stopReason: "tool_use", timestamp: 0 } as unknown as AgentMessage;
|
|
68
|
+
}
|
|
69
|
+
if (role === "toolResult" && toolName) {
|
|
70
|
+
return { role: "toolResult", content: [{ type: "text", text }], toolCallId: "c1", toolName, isError: false, timestamp: 0 } as unknown as AgentMessage;
|
|
71
|
+
}
|
|
72
|
+
return { role: "user", content: text, timestamp: 0 } as unknown as AgentMessage;
|
|
73
|
+
}
|
|
74
|
+
|
|
75
|
+
// Build a session of `n` tool-call triples tagged `tag`. Set A and B differ in
|
|
76
|
+
// content (so B never dedups against A) but A === A reproduces the same
|
|
77
|
+
// regionText → same L0 contentHash → dedup onto the first A checkpoint.
|
|
78
|
+
function buildSession(tag: string, n: number): AgentMessage[] {
|
|
79
|
+
const s: AgentMessage[] = [];
|
|
80
|
+
for (let i = 0; i < n; i++) {
|
|
81
|
+
s.push(msg("user", `[${tag}] we decided to use approach ${i} for module ${i}`));
|
|
82
|
+
s.push(msg("assistant", `[${tag}] edited module ${i}`, "Edit"));
|
|
83
|
+
s.push(msg("toolResult", `[${tag}] edited module ${i}`, "Edit"));
|
|
84
|
+
}
|
|
85
|
+
return s;
|
|
86
|
+
}
|
|
87
|
+
|
|
88
|
+
const toEntry = (m: AgentMessage, i: number): any => ({ type: "message", id: `e${i}`, parentId: null, timestamp: String(i), message: m });
|
|
89
|
+
const sessionManager = {
|
|
90
|
+
getSessionId: () => "sess_cache_001",
|
|
91
|
+
getEntries: () => buildSession("A", 14).map(toEntry),
|
|
92
|
+
getBranch: () => buildSession("A", 14).map(toEntry),
|
|
93
|
+
};
|
|
94
|
+
|
|
95
|
+
function makeCtx(over: Partial<any> = {}) {
|
|
96
|
+
return {
|
|
97
|
+
ui: { setStatus: () => {}, notify: () => {}, select: () => {}, confirm: async () => true, input: async () => "", setWidget: () => {} },
|
|
98
|
+
mode: "tui" as any, hasUI: true, cwd: stateDir, sessionManager,
|
|
99
|
+
modelRegistry: {} as any, model: undefined, isIdle: () => true, isProjectTrusted: () => true,
|
|
100
|
+
signal: undefined, abort: () => {}, hasPendingMessages: () => false, shutdown: () => {},
|
|
101
|
+
getContextUsage: () => ({ ...usage }),
|
|
102
|
+
compact: (opts?: any) => { compactCalls.push(opts); return undefined; },
|
|
103
|
+
getSystemPrompt: () => "system base",
|
|
104
|
+
...over,
|
|
105
|
+
} as any;
|
|
106
|
+
}
|
|
107
|
+
|
|
108
|
+
const pi = {
|
|
109
|
+
on: (ev: string, h: Function) => { handlers[ev] = h; },
|
|
110
|
+
registerCommand: () => {}, registerTool: () => {}, registerShortcut: () => {},
|
|
111
|
+
registerFlag: () => {}, getFlag: () => undefined, registerMessageRenderer: () => {},
|
|
112
|
+
registerEntryRenderer: () => {}, sendMessage: () => {}, sendUserMessage: () => {},
|
|
113
|
+
appendEntry: () => {}, setSessionName: () => {}, getSessionName: () => undefined,
|
|
114
|
+
setLabel: () => {}, exec: async () => ({ stdout: "", stderr: "", code: 0 }),
|
|
115
|
+
getActiveTools: () => [], getAllTools: () => [], setActiveTools: () => {},
|
|
116
|
+
getCommands: () => [], setModel: async () => false, getThinkingLevel: () => "off" as any,
|
|
117
|
+
setThinkingLevel: () => {},
|
|
118
|
+
} as any;
|
|
119
|
+
|
|
120
|
+
const mod = require("./mega-compact.js") as { default: (p: any) => void };
|
|
121
|
+
mod.default(pi);
|
|
122
|
+
const { lastRuntime } = require("./mega-events.js") as { lastRuntime: any };
|
|
123
|
+
|
|
124
|
+
const fire = (ev: string, event: any, ctx: any) => handlers[ev](event, ctx);
|
|
125
|
+
return {
|
|
126
|
+
stateDir, handlers, compactCalls, fire, ctx: makeCtx, usage, buildSession,
|
|
127
|
+
runtime: lastRuntime, // MegaRuntime with diag* counters + rt + trimCache
|
|
128
|
+
// Bypass the 2s debounce so each fire proceeds without real waiting.
|
|
129
|
+
clearDebounce: () => { if (lastRuntime) lastRuntime.debounceUntil = 0; },
|
|
130
|
+
};
|
|
131
|
+
}
|
|
132
|
+
|
|
133
|
+
test("REPLAY: >=2 gated context events within one epoch replay verbatim (byte-identical)", async () => {
|
|
134
|
+
const h = harness();
|
|
135
|
+
const ctx = h.ctx();
|
|
136
|
+
const session = h.buildSession("A", 14);
|
|
137
|
+
|
|
138
|
+
// Fire 3 gated context events; clearDebounce between so each passes the gate.
|
|
139
|
+
// percent stays 100 → percent-basis grewEnough (>=10) never trips → pure replay.
|
|
140
|
+
h.clearDebounce();
|
|
141
|
+
const r1 = await h.fire("context", { type: "context", messages: session }, ctx);
|
|
142
|
+
h.clearDebounce();
|
|
143
|
+
const r2 = await h.fire("context", { type: "context", messages: session }, ctx);
|
|
144
|
+
h.clearDebounce();
|
|
145
|
+
const r3 = await h.fire("context", { type: "context", messages: session }, ctx);
|
|
146
|
+
|
|
147
|
+
const rt = h.runtime;
|
|
148
|
+
assert.ok(rt.diagLiveTrimFires >= 1, "fresh trim fired on first event");
|
|
149
|
+
assert.ok(rt.diagLiveTrimReplays >= 2, `replay fired >=2 (got ${rt.diagLiveTrimReplays})`);
|
|
150
|
+
// byte-identical (stable KV-cache prefix) across replays
|
|
151
|
+
assert.deepEqual(r2?.messages, r3?.messages, "replay messages byte-identical across replays");
|
|
152
|
+
// replay matches the fresh-trim view (shallow-copy preserves content)
|
|
153
|
+
assert.deepEqual(r1?.messages, r2?.messages, "replay matches fresh-trim view (stable prefix)");
|
|
154
|
+
});
|
|
155
|
+
|
|
156
|
+
test("DEDUP: re-compact that dedups onto a DIFFERENT checkpoint still replays next (P2 fix)", async () => {
|
|
157
|
+
const h = harness();
|
|
158
|
+
// Token-basis growth path: percent null → token gate + token grewEnough
|
|
159
|
+
// (currentTokens - trimCache.ctxTokens >= effectiveThreshold * 0.5 = 25).
|
|
160
|
+
h.usage.percent = null;
|
|
161
|
+
h.usage.tokens = 200000;
|
|
162
|
+
const ctx = h.ctx();
|
|
163
|
+
const setA = h.buildSession("A", 14);
|
|
164
|
+
const setB = h.buildSession("B", 14); // different content, same length
|
|
165
|
+
const rt = h.runtime;
|
|
166
|
+
|
|
167
|
+
// 1) Fresh trim on setA → genuinely new checkpoint cp_A. lastCheckpointId = cp_A.
|
|
168
|
+
h.clearDebounce();
|
|
169
|
+
await h.fire("context", { type: "context", messages: setA }, ctx);
|
|
170
|
+
const cpA = rt.rt.lastCheckpointId;
|
|
171
|
+
assert.ok(cpA, "cp_A created on fresh trim");
|
|
172
|
+
assert.equal(rt.diagLiveTrimFires, 1, "first fire was a fresh trim");
|
|
173
|
+
|
|
174
|
+
// 2) Re-compact on setB (grew tokens) → genuinely new checkpoint cp_B (not deduped).
|
|
175
|
+
h.usage.tokens = 200100; // grew 100 >= 25
|
|
176
|
+
h.clearDebounce();
|
|
177
|
+
await h.fire("context", { type: "context", messages: setB }, ctx);
|
|
178
|
+
const cpB = rt.rt.lastCheckpointId;
|
|
179
|
+
assert.notEqual(cpB, cpA, "cp_B is a different, genuinely new checkpoint");
|
|
180
|
+
assert.equal(rt.rt.dedupSkips, 0, "setB did not dedup (different vocabulary, fuzzy tiers off)");
|
|
181
|
+
|
|
182
|
+
// 3) Re-compact on setA AGAIN (grew tokens) → L0 contentHash dedup onto cp_A.
|
|
183
|
+
// result.checkpointId = cp_A (!= lastCheckpointId cp_B); lastCheckpointId is
|
|
184
|
+
// NOT updated on a dedup (compact.ts:100-104), so it stays cp_B. With the
|
|
185
|
+
// fix, trimCache.checkpointId is keyed on lastCheckpointId (cp_B), NOT the
|
|
186
|
+
// dedup-volatile result.checkpointId (cp_A).
|
|
187
|
+
h.usage.tokens = 200200; // grew 100 >= 25
|
|
188
|
+
h.clearDebounce();
|
|
189
|
+
await h.fire("context", { type: "context", messages: setA }, ctx);
|
|
190
|
+
assert.equal(rt.rt.lastCheckpointId, cpB, "dedup did NOT bump lastCheckpointId (still cp_B)");
|
|
191
|
+
assert.ok(rt.rt.dedupSkips >= 1, "setA re-compact deduped onto an existing checkpoint");
|
|
192
|
+
// The P2 invariant: the cache key must equal the stable epoch signal.
|
|
193
|
+
assert.equal(rt.trimCache?.checkpointId, rt.rt.lastCheckpointId,
|
|
194
|
+
"trimCache.checkpointId keyed on lastCheckpointId (P2 fix), not dedup-volatile result.checkpointId");
|
|
195
|
+
|
|
196
|
+
// 4) Next gated event (no growth) MUST replay instead of re-running runCompact.
|
|
197
|
+
// Without the fix, trimCache.checkpointId (cp_A) != lastCheckpointId (cp_B)
|
|
198
|
+
// → the replay condition is false → runCompact re-runs every fire → the
|
|
199
|
+
// thrash silently persists in that path (the audit's finding).
|
|
200
|
+
const replaysBefore = rt.diagLiveTrimReplays;
|
|
201
|
+
h.usage.tokens = 200200; // no growth → replay
|
|
202
|
+
h.clearDebounce();
|
|
203
|
+
await h.fire("context", { type: "context", messages: setA }, ctx);
|
|
204
|
+
assert.ok(rt.diagLiveTrimReplays > replaysBefore,
|
|
205
|
+
`replay fired after dedup-onto-different-checkpoint (got ${rt.diagLiveTrimReplays}, was ${replaysBefore})`);
|
|
206
|
+
});
|
|
207
|
+
|
|
208
|
+
test("cleanup", async () => {
|
|
209
|
+
await closeVectorIndex();
|
|
210
|
+
rmSync(baseTmp, { recursive: true, force: true });
|
|
211
|
+
});
|
|
@@ -184,7 +184,9 @@ export function registerContextHandler(
|
|
|
184
184
|
runtime.diagLiveTrimFires++; // trim view returned this call (replay counts as a fire)
|
|
185
185
|
runtime.diagLiveTrimReplays++;
|
|
186
186
|
runtime.snapshot(ctx);
|
|
187
|
-
|
|
187
|
+
// v0.8.7: shallow-copy the cached summary so pi's transformContext can't
|
|
188
|
+
// mutate the shared reference across replays (audit P3).
|
|
189
|
+
return { messages: [{ ...runtime.trimCache.summaryAgentMsg }, ...recent] };
|
|
188
190
|
}
|
|
189
191
|
// else: context grew enough → fall through to re-compact (cache is stale)
|
|
190
192
|
}
|
|
@@ -308,7 +310,21 @@ export function registerContextHandler(
|
|
|
308
310
|
// replay it verbatim (stabilizing the KV-cache prefix) instead of
|
|
309
311
|
// regenerating a fresh summary + sentinel every fire.
|
|
310
312
|
runtime.trimCache = {
|
|
311
|
-
|
|
313
|
+
// v0.8.7: key the replay cache on the STABLE epoch signal
|
|
314
|
+
// (rt.lastCheckpointId) instead of ran.result.checkpointId, which is
|
|
315
|
+
// dedup-volatile: on a re-compact that dedups onto a DIFFERENT existing
|
|
316
|
+
// checkpoint, result.checkpointId is the matched id (engine.ts:188) while
|
|
317
|
+
// lastCheckpointId is only updated on a genuinely new checkpoint
|
|
318
|
+
// (compact.ts:100-104). Keying on result.checkpointId would make
|
|
319
|
+
// trimCache.checkpointId != rt.lastCheckpointId forever after that
|
|
320
|
+
// dedup fire, disabling replay for the rest of the epoch (the
|
|
321
|
+
// alternating cache-miss that 0.8.6 meant to fix). Prefer the stable
|
|
322
|
+
// signal; fall back to result.checkpointId then the epoch timestamp
|
|
323
|
+
// only for the no-checkpoint edge case.
|
|
324
|
+
checkpointId:
|
|
325
|
+
runtime.rt.lastCheckpointId ??
|
|
326
|
+
ran.result.checkpointId ??
|
|
327
|
+
`epoch-${runtime.rt.lastCompactAt ?? Date.now()}`,
|
|
312
328
|
cut,
|
|
313
329
|
summaryAgentMsg,
|
|
314
330
|
ctxPct: pct ?? null,
|
package/package.json
CHANGED