@lll9p/pi-better-compaction 0.2.1 → 0.4.0
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.
- package/README.md +83 -132
- package/README.zh-CN.md +142 -0
- package/package.json +6 -2
- package/src/compact-client-v2.ts +428 -0
- package/src/compact-client.ts +1 -63
- package/src/config.ts +15 -0
- package/src/extension-runtime.ts +196 -7
- package/src/native-fallback.ts +21 -3
- package/src/retained-messages.ts +98 -0
- package/src/runtime.ts +41 -2
- package/src/shared-headers.ts +103 -0
- package/src/types.ts +20 -4
package/src/extension-runtime.ts
CHANGED
|
@@ -6,6 +6,7 @@ import type {
|
|
|
6
6
|
SessionBeforeCompactEvent,
|
|
7
7
|
} from "@earendil-works/pi-coding-agent";
|
|
8
8
|
import { executeNativeCompaction } from "./compact-client";
|
|
9
|
+
import { executeV2Compaction } from "./compact-client-v2";
|
|
9
10
|
import { loadExtensionConfig } from "./config";
|
|
10
11
|
import { writeDebugArtifact } from "./debug";
|
|
11
12
|
import { resolveLatestNativeCompactionEntry } from "./details-store";
|
|
@@ -15,6 +16,7 @@ import {
|
|
|
15
16
|
serializeLiveTailToResponsesInput,
|
|
16
17
|
} from "./payload-rewrite";
|
|
17
18
|
import { getCompactionRequestExtras, rememberRequestContext } from "./request-context-cache";
|
|
19
|
+
import { buildRetainedMessages } from "./retained-messages";
|
|
18
20
|
import {
|
|
19
21
|
isResponsesCompatiblePayload,
|
|
20
22
|
resolveNativeCompactionEnvironment,
|
|
@@ -26,6 +28,8 @@ import {
|
|
|
26
28
|
createNativeCompactionResult,
|
|
27
29
|
EXTENSION_ID,
|
|
28
30
|
isNativeCompactionDetails,
|
|
31
|
+
NATIVE_COMPACTION_STRATEGY,
|
|
32
|
+
NATIVE_COMPACTION_STRATEGY_V2,
|
|
29
33
|
type ExtensionConfig,
|
|
30
34
|
type NativeCompactionDetails,
|
|
31
35
|
type NativeCompactionRequestMeta,
|
|
@@ -90,7 +94,7 @@ function buildCompactionInstructions(systemPrompt: string, customInstructions?:
|
|
|
90
94
|
return `${systemPrompt}\n\nAdditional user guidance for this manual /compact request:\n${guidance}`;
|
|
91
95
|
}
|
|
92
96
|
|
|
93
|
-
async function
|
|
97
|
+
async function runResponsesV1Compact(
|
|
94
98
|
event: SessionBeforeCompactEvent,
|
|
95
99
|
ctx: ExtensionContext,
|
|
96
100
|
config: ExtensionConfig,
|
|
@@ -135,7 +139,7 @@ async function runResponsesNativeCompact(
|
|
|
135
139
|
writeDebugArtifact(
|
|
136
140
|
"compaction-event",
|
|
137
141
|
{
|
|
138
|
-
event: "session_before_compact.
|
|
142
|
+
event: "session_before_compact.v1-compact-skip",
|
|
139
143
|
reason: latestNativeCompaction.reason,
|
|
140
144
|
provider: runtime.provider,
|
|
141
145
|
api: runtime.api,
|
|
@@ -169,7 +173,7 @@ async function runResponsesNativeCompact(
|
|
|
169
173
|
writeDebugArtifact(
|
|
170
174
|
"compaction-event",
|
|
171
175
|
{
|
|
172
|
-
event: "session_before_compact.
|
|
176
|
+
event: "session_before_compact.v1-compact-failure",
|
|
173
177
|
reason: compactResult.reason,
|
|
174
178
|
status: compactResult.status,
|
|
175
179
|
errorMessage: compactResult.errorMessage,
|
|
@@ -196,7 +200,7 @@ async function runResponsesNativeCompact(
|
|
|
196
200
|
writeDebugArtifact(
|
|
197
201
|
"compaction-event",
|
|
198
202
|
{
|
|
199
|
-
event: "session_before_compact.invalid-native-details",
|
|
203
|
+
event: "session_before_compact.v1-invalid-native-details",
|
|
200
204
|
reason: error instanceof Error ? error.message : String(error),
|
|
201
205
|
provider: runtime.provider,
|
|
202
206
|
api: runtime.api,
|
|
@@ -219,7 +223,7 @@ async function runResponsesNativeCompact(
|
|
|
219
223
|
writeDebugArtifact(
|
|
220
224
|
"compaction-event",
|
|
221
225
|
{
|
|
222
|
-
event: "session_before_compact.
|
|
226
|
+
event: "session_before_compact.v1-compact-success",
|
|
223
227
|
provider: runtime.provider,
|
|
224
228
|
api: runtime.api,
|
|
225
229
|
model: runtime.model,
|
|
@@ -238,6 +242,164 @@ async function runResponsesNativeCompact(
|
|
|
238
242
|
return { outcome: "success", compaction };
|
|
239
243
|
}
|
|
240
244
|
|
|
245
|
+
/**
|
|
246
|
+
* V2 compaction: stream a Responses request with compaction_trigger appended.
|
|
247
|
+
* On success, returns retained messages + encrypted compaction blob.
|
|
248
|
+
*/
|
|
249
|
+
async function runResponsesV2Compact(
|
|
250
|
+
event: SessionBeforeCompactEvent,
|
|
251
|
+
ctx: ExtensionContext,
|
|
252
|
+
config: ExtensionConfig,
|
|
253
|
+
runtime: NativeCompactionRuntime,
|
|
254
|
+
): Promise<ResponsesCompactOutcome> {
|
|
255
|
+
const instructions = buildCompactionInstructions(ctx.getSystemPrompt(), event.customInstructions);
|
|
256
|
+
const branchEntries = ctx.sessionManager.getBranch();
|
|
257
|
+
const latestNativeCompaction = resolveLatestNativeCompactionEntry(branchEntries, {
|
|
258
|
+
provider: runtime.provider,
|
|
259
|
+
api: runtime.api,
|
|
260
|
+
model: runtime.model,
|
|
261
|
+
baseUrl: runtime.baseUrl,
|
|
262
|
+
});
|
|
263
|
+
|
|
264
|
+
let requestSource: "session-context" | "non-native-session-context" | "latest-native-replay";
|
|
265
|
+
let request: NativeCompactionRequestBody;
|
|
266
|
+
if (latestNativeCompaction.ok) {
|
|
267
|
+
const liveTailEntries = branchEntries.slice(latestNativeCompaction.index + 1);
|
|
268
|
+
requestSource = "latest-native-replay";
|
|
269
|
+
const input: ResponsesInputItem[] = [
|
|
270
|
+
...(cloneOpaqueWindow(latestNativeCompaction.entry.details.compactedWindow) as ResponsesInputItem[]),
|
|
271
|
+
...serializeLiveTailToResponsesInput({ model: runtime.currentModel, entries: liveTailEntries }),
|
|
272
|
+
];
|
|
273
|
+
request = {
|
|
274
|
+
model: runtime.currentModel.id,
|
|
275
|
+
input,
|
|
276
|
+
instructions,
|
|
277
|
+
};
|
|
278
|
+
} else if (
|
|
279
|
+
latestNativeCompaction.reason === "no-compaction" ||
|
|
280
|
+
(latestNativeCompaction.reason === "latest-compaction-not-native" &&
|
|
281
|
+
config.allowCompactionContinuityBreak)
|
|
282
|
+
) {
|
|
283
|
+
requestSource =
|
|
284
|
+
latestNativeCompaction.reason === "no-compaction" ? "session-context" : "non-native-session-context";
|
|
285
|
+
request = serializeMessagesToCompactRequest({
|
|
286
|
+
model: runtime.currentModel,
|
|
287
|
+
messages: ctx.sessionManager.buildSessionContext().messages,
|
|
288
|
+
instructions,
|
|
289
|
+
});
|
|
290
|
+
} else {
|
|
291
|
+
writeDebugArtifact(
|
|
292
|
+
"compaction-event",
|
|
293
|
+
{
|
|
294
|
+
event: "session_before_compact.v2-compact-skip",
|
|
295
|
+
reason: latestNativeCompaction.reason,
|
|
296
|
+
provider: runtime.provider,
|
|
297
|
+
api: runtime.api,
|
|
298
|
+
model: runtime.model,
|
|
299
|
+
baseUrl: runtime.baseUrl,
|
|
300
|
+
latestCompactionIndex: latestNativeCompaction.latestCompactionIndex,
|
|
301
|
+
latestCompactionIdentity: getCompactionIdentityDebugInfo(latestNativeCompaction.latestCompaction),
|
|
302
|
+
},
|
|
303
|
+
config,
|
|
304
|
+
ctx,
|
|
305
|
+
);
|
|
306
|
+
return { outcome: "failed" };
|
|
307
|
+
}
|
|
308
|
+
|
|
309
|
+
const extras = getCompactionRequestExtras(runtime.model, getSessionId(ctx));
|
|
310
|
+
if (extras) {
|
|
311
|
+
request = { ...request, ...extras };
|
|
312
|
+
}
|
|
313
|
+
|
|
314
|
+
const v2Result = await executeV2Compaction({
|
|
315
|
+
runtime,
|
|
316
|
+
request,
|
|
317
|
+
signal: event.signal,
|
|
318
|
+
settings: config,
|
|
319
|
+
context: ctx,
|
|
320
|
+
});
|
|
321
|
+
|
|
322
|
+
if (!v2Result.ok) {
|
|
323
|
+
writeDebugArtifact(
|
|
324
|
+
"compaction-event",
|
|
325
|
+
{
|
|
326
|
+
event: "session_before_compact.v2-compact-failure",
|
|
327
|
+
reason: v2Result.reason,
|
|
328
|
+
status: v2Result.status,
|
|
329
|
+
errorMessage: v2Result.errorMessage,
|
|
330
|
+
},
|
|
331
|
+
config,
|
|
332
|
+
ctx,
|
|
333
|
+
);
|
|
334
|
+
return v2Result.reason === "aborted" ? { outcome: "aborted" } : { outcome: "failed" };
|
|
335
|
+
}
|
|
336
|
+
|
|
337
|
+
// Build compacted window: retained messages + compaction blob.
|
|
338
|
+
const retainedMessages = buildRetainedMessages(request.input);
|
|
339
|
+
const compactedWindow = [...retainedMessages, v2Result.compactionItem];
|
|
340
|
+
|
|
341
|
+
let details: NativeCompactionDetails;
|
|
342
|
+
try {
|
|
343
|
+
details = createNativeCompactionDetails(
|
|
344
|
+
{
|
|
345
|
+
provider: runtime.provider,
|
|
346
|
+
api: runtime.api,
|
|
347
|
+
model: runtime.model,
|
|
348
|
+
baseUrl: runtime.baseUrl,
|
|
349
|
+
compactedWindow,
|
|
350
|
+
compactResponseId: v2Result.responseId,
|
|
351
|
+
createdAt: v2Result.createdAt,
|
|
352
|
+
requestMeta: buildCompactionRequestMeta(event),
|
|
353
|
+
},
|
|
354
|
+
NATIVE_COMPACTION_STRATEGY_V2,
|
|
355
|
+
);
|
|
356
|
+
} catch (error) {
|
|
357
|
+
writeDebugArtifact(
|
|
358
|
+
"compaction-event",
|
|
359
|
+
{
|
|
360
|
+
event: "session_before_compact.v2-invalid-native-details",
|
|
361
|
+
reason: error instanceof Error ? error.message : String(error),
|
|
362
|
+
provider: runtime.provider,
|
|
363
|
+
api: runtime.api,
|
|
364
|
+
model: runtime.model,
|
|
365
|
+
baseUrl: runtime.baseUrl,
|
|
366
|
+
},
|
|
367
|
+
config,
|
|
368
|
+
ctx,
|
|
369
|
+
);
|
|
370
|
+
return { outcome: "failed" };
|
|
371
|
+
}
|
|
372
|
+
|
|
373
|
+
// V2 blob is encrypted; no summary text can be extracted.
|
|
374
|
+
const compaction = createNativeCompactionResult({
|
|
375
|
+
firstKeptEntryId: event.preparation.firstKeptEntryId,
|
|
376
|
+
tokensBefore: event.preparation.tokensBefore,
|
|
377
|
+
details,
|
|
378
|
+
});
|
|
379
|
+
|
|
380
|
+
writeDebugArtifact(
|
|
381
|
+
"compaction-event",
|
|
382
|
+
{
|
|
383
|
+
event: "session_before_compact.v2-compact-success",
|
|
384
|
+
provider: runtime.provider,
|
|
385
|
+
api: runtime.api,
|
|
386
|
+
model: runtime.model,
|
|
387
|
+
requestSource,
|
|
388
|
+
requestInputItems: request.input.length,
|
|
389
|
+
requestExtras: extras ? Object.keys(extras) : [],
|
|
390
|
+
compactResponseId: v2Result.responseId,
|
|
391
|
+
retainedMessageCount: retainedMessages.length,
|
|
392
|
+
compactedItems: compactedWindow.length,
|
|
393
|
+
usage: v2Result.usage,
|
|
394
|
+
firstKeptEntryId: event.preparation.firstKeptEntryId,
|
|
395
|
+
},
|
|
396
|
+
config,
|
|
397
|
+
ctx,
|
|
398
|
+
);
|
|
399
|
+
|
|
400
|
+
return { outcome: "success", compaction };
|
|
401
|
+
}
|
|
402
|
+
|
|
241
403
|
async function handleSessionBeforeCompact(event: SessionBeforeCompactEvent, ctx: ExtensionContext) {
|
|
242
404
|
const { config } = loadExtensionConfig();
|
|
243
405
|
if (!config.enabled) {
|
|
@@ -271,7 +433,14 @@ async function handleSessionBeforeCompact(event: SessionBeforeCompactEvent, ctx:
|
|
|
271
433
|
responsesCompactApis: config.responsesCompactApis,
|
|
272
434
|
});
|
|
273
435
|
if (resolution.ok) {
|
|
274
|
-
|
|
436
|
+
let responsesOutcome: ResponsesCompactOutcome;
|
|
437
|
+
|
|
438
|
+
if (config.compactionVersion === "v2") {
|
|
439
|
+
responsesOutcome = await runResponsesV2Compact(event, ctx, config, resolution.runtime);
|
|
440
|
+
} else {
|
|
441
|
+
responsesOutcome = await runResponsesV1Compact(event, ctx, config, resolution.runtime);
|
|
442
|
+
}
|
|
443
|
+
|
|
275
444
|
if (responsesOutcome.outcome === "success") {
|
|
276
445
|
return { compaction: responsesOutcome.compaction };
|
|
277
446
|
}
|
|
@@ -296,7 +465,7 @@ async function handleSessionBeforeCompact(event: SessionBeforeCompactEvent, ctx:
|
|
|
296
465
|
}
|
|
297
466
|
|
|
298
467
|
// Branch 2: run pi's native compaction method with the configured model.
|
|
299
|
-
const fallback = await runNativeFallbackCompaction({ ctx, event, config });
|
|
468
|
+
const fallback = await runNativeFallbackCompaction({ ctx, event, config, sessionId: getSessionId(ctx) });
|
|
300
469
|
if (fallback.ok) {
|
|
301
470
|
if (ctx.hasUI) {
|
|
302
471
|
ctx.ui.notify(
|
|
@@ -309,6 +478,7 @@ async function handleSessionBeforeCompact(event: SessionBeforeCompactEvent, ctx:
|
|
|
309
478
|
{
|
|
310
479
|
event: "session_before_compact.fallback-success",
|
|
311
480
|
model: fallback.model,
|
|
481
|
+
usage: fallback.usage,
|
|
312
482
|
},
|
|
313
483
|
config,
|
|
314
484
|
ctx,
|
|
@@ -501,4 +671,23 @@ export default function (pi: ExtensionAPI) {
|
|
|
501
671
|
|
|
502
672
|
pi.on("session_before_compact", handleSessionBeforeCompact);
|
|
503
673
|
pi.on("before_provider_request", handleBeforeProviderRequest);
|
|
674
|
+
|
|
675
|
+
pi.on("session_compact_failed", (event, ctx) => {
|
|
676
|
+
const { config } = loadExtensionConfig();
|
|
677
|
+
if (!config.enabled) return;
|
|
678
|
+
|
|
679
|
+
writeDebugArtifact(
|
|
680
|
+
"compaction-event",
|
|
681
|
+
{
|
|
682
|
+
event: "session_compact_failed",
|
|
683
|
+
reason: event.reason,
|
|
684
|
+
errorMessage: event.errorMessage,
|
|
685
|
+
aborted: event.aborted,
|
|
686
|
+
willRetry: event.willRetry,
|
|
687
|
+
fromExtension: event.fromExtension,
|
|
688
|
+
},
|
|
689
|
+
config,
|
|
690
|
+
ctx,
|
|
691
|
+
);
|
|
692
|
+
});
|
|
504
693
|
}
|
package/src/native-fallback.ts
CHANGED
|
@@ -26,6 +26,7 @@ export type NativeFallbackResult =
|
|
|
26
26
|
ok: true;
|
|
27
27
|
result: CompactionResult;
|
|
28
28
|
model: { provider: string; id: string };
|
|
29
|
+
usage?: CompactionResult["usage"];
|
|
29
30
|
}
|
|
30
31
|
| {
|
|
31
32
|
ok: false;
|
|
@@ -38,9 +39,21 @@ export type NativeFallbackResult =
|
|
|
38
39
|
export type NativeCompactFn = typeof compact;
|
|
39
40
|
|
|
40
41
|
type ResolvedAuth =
|
|
41
|
-
| { ok: true; apiKey?: string; headers?: Record<string, string>; env?: Record<string, string> }
|
|
42
|
+
| { ok: true; apiKey?: string; headers?: Record<string, string | null>; env?: Record<string, string> }
|
|
42
43
|
| { ok: false; error: string };
|
|
43
44
|
|
|
45
|
+
/** Strip null-valued entries so downstream consumers receive a clean Record<string, string>. */
|
|
46
|
+
function filterNullHeaders(headers: Record<string, string | null> | undefined): Record<string, string> | undefined {
|
|
47
|
+
if (!headers) return undefined;
|
|
48
|
+
const filtered: Record<string, string> = {};
|
|
49
|
+
for (const [key, value] of Object.entries(headers)) {
|
|
50
|
+
if (value !== null) {
|
|
51
|
+
filtered[key] = value;
|
|
52
|
+
}
|
|
53
|
+
}
|
|
54
|
+
return Object.keys(filtered).length > 0 ? filtered : undefined;
|
|
55
|
+
}
|
|
56
|
+
|
|
44
57
|
/** Parse "provider/model-id" (model ids may themselves contain slashes). */
|
|
45
58
|
export function parseModelSpec(spec: string): ParsedModelSpec | undefined {
|
|
46
59
|
const trimmed = spec.trim();
|
|
@@ -82,6 +95,7 @@ export async function runNativeFallbackCompaction(args: {
|
|
|
82
95
|
event: SessionBeforeCompactEvent;
|
|
83
96
|
config: ExtensionConfig;
|
|
84
97
|
compactFn?: NativeCompactFn;
|
|
98
|
+
sessionId?: string;
|
|
85
99
|
}): Promise<NativeFallbackResult> {
|
|
86
100
|
const { ctx, event, config } = args;
|
|
87
101
|
const compactFn = args.compactFn ?? compact;
|
|
@@ -120,12 +134,15 @@ export async function runNativeFallbackCompaction(args: {
|
|
|
120
134
|
event.preparation,
|
|
121
135
|
model,
|
|
122
136
|
auth.apiKey,
|
|
123
|
-
auth.headers,
|
|
137
|
+
filterNullHeaders(auth.headers),
|
|
124
138
|
event.customInstructions,
|
|
125
139
|
event.signal,
|
|
126
140
|
config.compactionThinkingLevel,
|
|
127
|
-
undefined,
|
|
141
|
+
undefined, // streamFn
|
|
128
142
|
auth.env,
|
|
143
|
+
undefined, // retry (use pi defaults)
|
|
144
|
+
undefined, // callbacks
|
|
145
|
+
args.sessionId,
|
|
129
146
|
);
|
|
130
147
|
|
|
131
148
|
if (event.signal.aborted) {
|
|
@@ -139,6 +156,7 @@ export async function runNativeFallbackCompaction(args: {
|
|
|
139
156
|
ok: true,
|
|
140
157
|
result,
|
|
141
158
|
model: { provider: model.provider, id: model.id },
|
|
159
|
+
usage: result.usage,
|
|
142
160
|
};
|
|
143
161
|
} catch (error) {
|
|
144
162
|
if (event.signal.aborted || isAbortError(error)) {
|
|
@@ -0,0 +1,98 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* V2 compaction retained-message filtering.
|
|
3
|
+
*
|
|
4
|
+
* After the API returns an encrypted compaction blob, V2 keeps a window of recent
|
|
5
|
+
* user/developer/system messages alongside the blob so the model retains explicit
|
|
6
|
+
* user instructions and context anchors. This module mirrors the filtering and
|
|
7
|
+
* budget logic from codex-rs `compact_remote_v2.rs`.
|
|
8
|
+
*/
|
|
9
|
+
|
|
10
|
+
/** Token budget for retained messages (matches codex-rs RETAINED_MESSAGE_TOKEN_BUDGET). */
|
|
11
|
+
export const RETAINED_MESSAGE_TOKEN_BUDGET = 65_536;
|
|
12
|
+
|
|
13
|
+
/** Messages larger than this are excluded even if they are of a retained role. */
|
|
14
|
+
const MAX_SINGLE_ITEM_TOKENS = 10_000;
|
|
15
|
+
|
|
16
|
+
/** Rough chars-per-token ratio for budget estimation. */
|
|
17
|
+
const CHARS_PER_TOKEN = 4;
|
|
18
|
+
|
|
19
|
+
function isRecord(value: unknown): value is Record<string, unknown> {
|
|
20
|
+
return !!value && typeof value === "object" && !Array.isArray(value);
|
|
21
|
+
}
|
|
22
|
+
|
|
23
|
+
/**
|
|
24
|
+
* Estimate the token count of an opaque input item by serializing to JSON and
|
|
25
|
+
* dividing by the chars-per-token ratio. This is intentionally rough — codex-rs
|
|
26
|
+
* uses a real tokenizer but we don't have one in the extension runtime.
|
|
27
|
+
*/
|
|
28
|
+
export function estimateItemTokens(item: unknown): number {
|
|
29
|
+
try {
|
|
30
|
+
const length = JSON.stringify(item).length;
|
|
31
|
+
return Math.ceil(length / CHARS_PER_TOKEN);
|
|
32
|
+
} catch {
|
|
33
|
+
return 0;
|
|
34
|
+
}
|
|
35
|
+
}
|
|
36
|
+
|
|
37
|
+
/**
|
|
38
|
+
* Whether an input item should be retained alongside the compaction blob.
|
|
39
|
+
*
|
|
40
|
+
* Retained roles (matching codex-rs `is_retained_for_remote_compaction_v2`):
|
|
41
|
+
* - `user` messages
|
|
42
|
+
* - `developer` messages
|
|
43
|
+
* - `system` messages
|
|
44
|
+
*
|
|
45
|
+
* Everything else (assistant, function_call, function_call_output, reasoning,
|
|
46
|
+
* compaction, etc.) is excluded.
|
|
47
|
+
*/
|
|
48
|
+
export function isRetainedItem(item: unknown): boolean {
|
|
49
|
+
if (!isRecord(item)) {
|
|
50
|
+
return false;
|
|
51
|
+
}
|
|
52
|
+
|
|
53
|
+
const role = item.role;
|
|
54
|
+
if (typeof role === "string") {
|
|
55
|
+
return role === "user" || role === "developer" || role === "system";
|
|
56
|
+
}
|
|
57
|
+
|
|
58
|
+
return false;
|
|
59
|
+
}
|
|
60
|
+
|
|
61
|
+
/**
|
|
62
|
+
* From a list of Responses API input items, select the most recent retained
|
|
63
|
+
* messages that fit within the token budget. Items are selected newest-first
|
|
64
|
+
* (reverse order) and returned in their original chronological order.
|
|
65
|
+
*
|
|
66
|
+
* Oversized individual items (> MAX_SINGLE_ITEM_TOKENS) are skipped.
|
|
67
|
+
*/
|
|
68
|
+
export function buildRetainedMessages(
|
|
69
|
+
input: readonly unknown[],
|
|
70
|
+
budget: number = RETAINED_MESSAGE_TOKEN_BUDGET,
|
|
71
|
+
): unknown[] {
|
|
72
|
+
// Filter to retained roles first.
|
|
73
|
+
const retained: Array<{ index: number; item: unknown; tokens: number }> = [];
|
|
74
|
+
for (let i = 0; i < input.length; i++) {
|
|
75
|
+
const item = input[i];
|
|
76
|
+
if (!isRetainedItem(item)) continue;
|
|
77
|
+
|
|
78
|
+
const tokens = estimateItemTokens(item);
|
|
79
|
+
if (tokens > MAX_SINGLE_ITEM_TOKENS) continue;
|
|
80
|
+
|
|
81
|
+
retained.push({ index: i, item, tokens });
|
|
82
|
+
}
|
|
83
|
+
|
|
84
|
+
// Select from newest to oldest within the budget.
|
|
85
|
+
let remaining = budget;
|
|
86
|
+
const selected: typeof retained = [];
|
|
87
|
+
|
|
88
|
+
for (let i = retained.length - 1; i >= 0; i--) {
|
|
89
|
+
const entry = retained[i]!;
|
|
90
|
+
if (entry.tokens > remaining) break;
|
|
91
|
+
remaining -= entry.tokens;
|
|
92
|
+
selected.push(entry);
|
|
93
|
+
}
|
|
94
|
+
|
|
95
|
+
// Return in original chronological order.
|
|
96
|
+
selected.reverse();
|
|
97
|
+
return selected.map((entry) => structuredClone(entry.item));
|
|
98
|
+
}
|
package/src/runtime.ts
CHANGED
|
@@ -4,6 +4,8 @@ import { RESPONSES_COMPACT_CAPABLE_APIS } from "./types";
|
|
|
4
4
|
|
|
5
5
|
const OPENAI_COMPACT_PATH = "responses/compact";
|
|
6
6
|
const CODEX_COMPACT_PATH = "codex/responses/compact";
|
|
7
|
+
const OPENAI_RESPONSES_PATH = "responses";
|
|
8
|
+
const CODEX_RESPONSES_PATH = "codex/responses";
|
|
7
9
|
|
|
8
10
|
type ResponsesCompactApi = (typeof RESPONSES_COMPACT_CAPABLE_APIS)[number];
|
|
9
11
|
|
|
@@ -40,6 +42,7 @@ export type NativeCompactionRuntime = {
|
|
|
40
42
|
headers?: Record<string, string>;
|
|
41
43
|
compactPath: string;
|
|
42
44
|
compactUrl: string;
|
|
45
|
+
responsesUrl: string;
|
|
43
46
|
payload?: ResponsesCompatibleRequestPayload;
|
|
44
47
|
currentModel: RuntimeModel;
|
|
45
48
|
};
|
|
@@ -74,6 +77,29 @@ export function normalizeBaseUrl(baseUrl: string | undefined | null): string | u
|
|
|
74
77
|
return normalized ? normalized : undefined;
|
|
75
78
|
}
|
|
76
79
|
|
|
80
|
+
function buildOpenAIResponsesUrl(baseUrl: string): string {
|
|
81
|
+
const normalized = normalizeBaseUrl(baseUrl) ?? baseUrl;
|
|
82
|
+
if (normalized.endsWith("/responses")) {
|
|
83
|
+
return normalized;
|
|
84
|
+
}
|
|
85
|
+
return `${normalized}/${OPENAI_RESPONSES_PATH}`;
|
|
86
|
+
}
|
|
87
|
+
|
|
88
|
+
function buildCodexResponsesUrl(baseUrl: string): string {
|
|
89
|
+
const normalized = normalizeBaseUrl(baseUrl) ?? baseUrl;
|
|
90
|
+
if (normalized.endsWith("/codex/responses")) {
|
|
91
|
+
return normalized;
|
|
92
|
+
}
|
|
93
|
+
if (normalized.endsWith("/codex")) {
|
|
94
|
+
return `${normalized}/responses`;
|
|
95
|
+
}
|
|
96
|
+
return `${normalized}/${CODEX_RESPONSES_PATH}`;
|
|
97
|
+
}
|
|
98
|
+
|
|
99
|
+
export function buildResponsesUrl(baseUrl: string, api: ResponsesCompactApi): string {
|
|
100
|
+
return api === "openai-codex-responses" ? buildCodexResponsesUrl(baseUrl) : buildOpenAIResponsesUrl(baseUrl);
|
|
101
|
+
}
|
|
102
|
+
|
|
77
103
|
function buildOpenAICompactUrl(baseUrl: string): string {
|
|
78
104
|
const normalized = normalizeBaseUrl(baseUrl) ?? baseUrl;
|
|
79
105
|
if (normalized.endsWith("/responses")) {
|
|
@@ -101,13 +127,25 @@ export function buildCompactPath(api: ResponsesCompactApi): string {
|
|
|
101
127
|
return api === "openai-codex-responses" ? CODEX_COMPACT_PATH : OPENAI_COMPACT_PATH;
|
|
102
128
|
}
|
|
103
129
|
|
|
130
|
+
/** Strip null-valued entries so downstream consumers receive a clean Record<string, string>. */
|
|
131
|
+
function filterNullHeaders(headers: Record<string, string | null> | undefined): Record<string, string> | undefined {
|
|
132
|
+
if (!headers) return undefined;
|
|
133
|
+
const filtered: Record<string, string> = {};
|
|
134
|
+
for (const [key, value] of Object.entries(headers)) {
|
|
135
|
+
if (value !== null) {
|
|
136
|
+
filtered[key] = value;
|
|
137
|
+
}
|
|
138
|
+
}
|
|
139
|
+
return Object.keys(filtered).length > 0 ? filtered : undefined;
|
|
140
|
+
}
|
|
141
|
+
|
|
104
142
|
async function resolveRequestAuth(
|
|
105
143
|
ctx: ExtensionContext,
|
|
106
144
|
model: RuntimeModel,
|
|
107
145
|
): Promise<{ apiKey?: string; headers?: Record<string, string> }> {
|
|
108
146
|
const modelRegistry = ctx.modelRegistry as {
|
|
109
147
|
getApiKeyAndHeaders?: (currentModel: RuntimeModel) => Promise<
|
|
110
|
-
| { ok: true; apiKey?: string; headers?: Record<string, string> }
|
|
148
|
+
| { ok: true; apiKey?: string; headers?: Record<string, string | null> }
|
|
111
149
|
| { ok: false; error: string }
|
|
112
150
|
>;
|
|
113
151
|
};
|
|
@@ -117,7 +155,7 @@ async function resolveRequestAuth(
|
|
|
117
155
|
}
|
|
118
156
|
|
|
119
157
|
const auth = await modelRegistry.getApiKeyAndHeaders(model);
|
|
120
|
-
return auth.ok ? { apiKey: auth.apiKey, headers: auth.headers } : {};
|
|
158
|
+
return auth.ok ? { apiKey: auth.apiKey, headers: filterNullHeaders(auth.headers) } : {};
|
|
121
159
|
}
|
|
122
160
|
|
|
123
161
|
export function isSupportedApi(api: string): api is ResponsesCompactApi {
|
|
@@ -234,6 +272,7 @@ export async function resolveNativeCompactionEnvironment(
|
|
|
234
272
|
headers,
|
|
235
273
|
compactPath: buildCompactPath(descriptor.api),
|
|
236
274
|
compactUrl: buildCompactUrl(descriptor.baseUrl, descriptor.api),
|
|
275
|
+
responsesUrl: buildResponsesUrl(descriptor.baseUrl, descriptor.api),
|
|
237
276
|
payload: requestPayload,
|
|
238
277
|
currentModel,
|
|
239
278
|
},
|
|
@@ -0,0 +1,103 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Shared HTTP header construction for V1 and V2 compaction clients.
|
|
3
|
+
*
|
|
4
|
+
* Extracts the header building logic that was previously private in
|
|
5
|
+
* compact-client.ts so both compact-client.ts and compact-client-v2.ts
|
|
6
|
+
* can share it without duplication.
|
|
7
|
+
*/
|
|
8
|
+
|
|
9
|
+
import type { NativeCompactionRuntime } from "./runtime";
|
|
10
|
+
|
|
11
|
+
const JSON_CONTENT_TYPE = "application/json";
|
|
12
|
+
|
|
13
|
+
function decodeJwtPayload(token: string): Record<string, unknown> | undefined {
|
|
14
|
+
const parts = token.split(".");
|
|
15
|
+
if (parts.length !== 3) {
|
|
16
|
+
return undefined;
|
|
17
|
+
}
|
|
18
|
+
|
|
19
|
+
try {
|
|
20
|
+
const payloadText = Buffer.from(parts[1]!, "base64url").toString("utf8");
|
|
21
|
+
const payload = JSON.parse(payloadText);
|
|
22
|
+
return payload && typeof payload === "object" && !Array.isArray(payload)
|
|
23
|
+
? (payload as Record<string, unknown>)
|
|
24
|
+
: undefined;
|
|
25
|
+
} catch {
|
|
26
|
+
return undefined;
|
|
27
|
+
}
|
|
28
|
+
}
|
|
29
|
+
|
|
30
|
+
function isRecord(value: unknown): value is Record<string, unknown> {
|
|
31
|
+
return !!value && typeof value === "object" && !Array.isArray(value);
|
|
32
|
+
}
|
|
33
|
+
|
|
34
|
+
function extractCodexAccountId(token: string): string | undefined {
|
|
35
|
+
const payload = decodeJwtPayload(token);
|
|
36
|
+
const authClaims = payload?.["https://api.openai.com/auth"];
|
|
37
|
+
if (!isRecord(authClaims)) {
|
|
38
|
+
return undefined;
|
|
39
|
+
}
|
|
40
|
+
|
|
41
|
+
const accountId = authClaims.chatgpt_account_id;
|
|
42
|
+
return typeof accountId === "string" && accountId.trim().length > 0 ? accountId.trim() : undefined;
|
|
43
|
+
}
|
|
44
|
+
|
|
45
|
+
function buildCodexUserAgent(): string {
|
|
46
|
+
const platform = typeof process !== "undefined" ? process.platform : "browser";
|
|
47
|
+
const arch = typeof process !== "undefined" ? process.arch : "unknown";
|
|
48
|
+
return `pi (${platform}; ${arch})`;
|
|
49
|
+
}
|
|
50
|
+
|
|
51
|
+
/**
|
|
52
|
+
* Build HTTP headers for a compaction request from the resolved runtime.
|
|
53
|
+
*
|
|
54
|
+
* Handles model-level headers, extension-resolved headers, authorization,
|
|
55
|
+
* and Codex-specific headers (account ID, originator, user-agent, beta flag).
|
|
56
|
+
*
|
|
57
|
+
* @param accept - The Accept header value. Defaults to `application/json`.
|
|
58
|
+
*/
|
|
59
|
+
export function toHeaders(
|
|
60
|
+
runtime: NativeCompactionRuntime,
|
|
61
|
+
accept: string = JSON_CONTENT_TYPE,
|
|
62
|
+
): Record<string, string> {
|
|
63
|
+
const headers = new Headers();
|
|
64
|
+
// Model-level headers may contain null values (ProviderHeaders); null means "unset".
|
|
65
|
+
for (const [key, value] of Object.entries(runtime.currentModel.headers ?? {})) {
|
|
66
|
+
if (value != null) {
|
|
67
|
+
headers.set(key, String(value));
|
|
68
|
+
}
|
|
69
|
+
}
|
|
70
|
+
// Extension-resolved headers (already filtered by resolveRequestAuth, but defensive).
|
|
71
|
+
for (const [key, value] of Object.entries(runtime.headers ?? {})) {
|
|
72
|
+
if (value == null) {
|
|
73
|
+
headers.delete(key);
|
|
74
|
+
} else {
|
|
75
|
+
headers.set(key, value);
|
|
76
|
+
}
|
|
77
|
+
}
|
|
78
|
+
headers.set("accept", accept);
|
|
79
|
+
headers.set("content-type", JSON_CONTENT_TYPE);
|
|
80
|
+
if (!headers.has("authorization")) {
|
|
81
|
+
headers.set("authorization", `Bearer ${runtime.apiKey}`);
|
|
82
|
+
}
|
|
83
|
+
|
|
84
|
+
if (runtime.api === "openai-codex-responses") {
|
|
85
|
+
const accountId = extractCodexAccountId(runtime.apiKey);
|
|
86
|
+
if (accountId) {
|
|
87
|
+
headers.set("chatgpt-account-id", accountId);
|
|
88
|
+
}
|
|
89
|
+
headers.set("originator", "pi");
|
|
90
|
+
headers.set("user-agent", buildCodexUserAgent());
|
|
91
|
+
headers.set("openai-beta", "responses=experimental");
|
|
92
|
+
}
|
|
93
|
+
|
|
94
|
+
return Object.fromEntries(headers.entries());
|
|
95
|
+
}
|
|
96
|
+
|
|
97
|
+
/** Check whether an error represents an intentional abort (AbortController / AbortSignal). */
|
|
98
|
+
export function isAbortError(error: unknown): boolean {
|
|
99
|
+
return (
|
|
100
|
+
(error instanceof DOMException && error.name === "AbortError") ||
|
|
101
|
+
(error instanceof Error && (error.name === "AbortError" || error.name === "ABORT_ERR"))
|
|
102
|
+
);
|
|
103
|
+
}
|