From 51adfbf52c8a447fe0dbbe5ea1a97ff512edf3b7 Mon Sep 17 00:00:00 2001
From: Tianyi Cui <53024+tianyicui@users.noreply.github.com>
Date: Thu, 3 Sep 2026 17:50:09 +0800
Subject: [PATCH] chore(session): retire the v1-to-v2 benchmark script
The repository-only benchmark, its unit-suite spec, the root script, and the
README section are removed; the accepted acceptance run's figures stay in the
embedded-assistant-streams note as recorded facts. The note also records why
the migration edge reuses the `dsh-llm` stream helpers instead of frozen
copies and why its target validation re-checks message/stream agreement
itself. A validation test that recomputed `RELEASED_V2_EVENT_TYPES` from the
same expression as its definition is dropped.
---
...01-v2-embedded-assistant-streams.i18n.yaml | 4 +-
...026-09-01-v2-embedded-assistant-streams.md | 4 +-
...-09-01-v2-embedded-assistant-streams.zh.md | 4 +-
package.json | 1 -
.../session-format-v1-to-v2/README.i18n.yaml | 4 +-
.../session/session-format-v1-to-v2/README.md | 9 -
.../session-format-v1-to-v2/README.zh.md | 9 -
.../tests/validation.spec.ts | 9 -
.../benchmark-session-format-v1-to-v2.spec.ts | 67 --
scripts/benchmark-session-format-v1-to-v2.ts | 774 ------------------
10 files changed, 8 insertions(+), 877 deletions(-)
delete mode 100644 scripts/benchmark-session-format-v1-to-v2.spec.ts
delete mode 100644 scripts/benchmark-session-format-v1-to-v2.ts
diff --git a/.agents/notes/implemented/architecture/2026-09-01-v2-embedded-assistant-streams.i18n.yaml b/.agents/notes/implemented/architecture/2026-09-01-v2-embedded-assistant-streams.i18n.yaml
index be7f7c5e58..4dc0cc872d 100644
--- a/.agents/notes/implemented/architecture/2026-09-01-v2-embedded-assistant-streams.i18n.yaml
+++ b/.agents/notes/implemented/architecture/2026-09-01-v2-embedded-assistant-streams.i18n.yaml
@@ -2,5 +2,5 @@
# side as of the last confirmed-consistent state. Both languages carry equal authority;
# after editing either side, bring the other along and re-record with:
# pnpm run verify-translation-pairing --write .agents/notes/implemented/architecture/2026-09-01-v2-embedded-assistant-streams.md
-2026-09-01-v2-embedded-assistant-streams.md: 805e52579be250ed372585c48fad434f7e15aadc
-2026-09-01-v2-embedded-assistant-streams.zh.md: 05e1a76f149a92c35af1ede28babfd2513773352
+2026-09-01-v2-embedded-assistant-streams.md: 5475f467db7de7f0dfdec29764adf2fdc5c9961b
+2026-09-01-v2-embedded-assistant-streams.zh.md: 60e983e48358bdc5ba25f1e2c4b35cc866b8a1f3
diff --git a/.agents/notes/implemented/architecture/2026-09-01-v2-embedded-assistant-streams.md b/.agents/notes/implemented/architecture/2026-09-01-v2-embedded-assistant-streams.md
index 805e52579b..5475f467db 100644
--- a/.agents/notes/implemented/architecture/2026-09-01-v2-embedded-assistant-streams.md
+++ b/.agents/notes/implemented/architecture/2026-09-01-v2-embedded-assistant-streams.md
@@ -31,7 +31,7 @@ The Web follow adapter opts into these process-local frames and adds the last du
### Released v1 to v2 migration
-The adjacent migration validates the complete frozen v1 artifact, groups chunks by turn, step, terminal boundary, and exact message provenance, and then substitutes one settlement per attempt. A successful group's chunks move into its message. An unclaimed group becomes `assistant/attempt` at the last consumed chunk's position. Unrelated interleaved events retain their relative order, and survivors receive dense v2 sequence numbers.
+The adjacent migration validates the complete frozen v1 artifact, groups chunks by turn, step, terminal boundary, and exact message provenance, and then substitutes one settlement per attempt. A successful group's chunks move into its message. An unclaimed group becomes `assistant/attempt` at the last consumed chunk's position. Unrelated interleaved events retain their relative order, and survivors receive dense v2 sequence numbers. The edge compacts, expands, and re-assembles embedded streams through the runtime `AssistantStreamAccumulator`, `expandAssistantStream`, and `BlockAssembler` from `dsh-llm` instead of frozen copies, because that package owns the v2 stream encoding. Target validation re-checks agreement between each migrated `assistant/message` and its embedded stream itself, so a disagreeing v1 log is refused as an unsupported migration with its source artifact retained instead of surfacing as corruption from the installed Session restoration. A later format that changes the stream encoding must freeze copies of these helpers into this edge.
The edge remaps the finite declared reference inventory: envelope provenance, surface replacement endpoints, command source events, compaction ranges and shadowed lists, and title message lists. The model-visible text of a validated `session/title-llm-request` remains byte-identical in the source sequence namespace while its `messageSeqs` field moves to the v2 namespace; target validation therefore does not reconstruct that text from remapped sequences. A reference to a consumed chunk refuses migration; it is never redirected to a settlement with different meaning. The edge also refuses an inherited cut that splits an attempt.
@@ -47,7 +47,7 @@ Generation selection and publication follow the [released Session migration deci
The compact-stream tests pin exact accumulation and expansion for text, reasoning, tool arguments, raw chunks, timestamp gaps, malformed records, and detached snapshots. The v1-to-v2 tests cover successful and failed attempts, interleaving, dense sequence and reference remapping, source-sequence title framing, seed-cut insertion and split refusal, strict source and target validation, one-row v2 encoding, backend-compatible provenance ranges, raw and Zstandard publication, and no-write current reads.
-The manual performance acceptance measures static catalog-routing overhead against direct released-v2 restoration of the same already parsed physical rows across three runs, 100 warmup pairs, and 600 measured pairs. It does not compare v1 with v2 or time backend I/O. Every pooled median and p95 regression must remain within 5%; the accepted run's worst p95 regression was 3.150%. `--smoke` reports a non-gating diagnostic sample.
+The pre-merge performance acceptance measured static catalog-routing overhead against direct released-v2 restoration of the same already parsed physical rows across three runs, 100 warmup pairs, and 600 measured pairs; it did not compare v1 with v2 or time backend I/O. Every pooled median and p95 regression stayed within the 5% budget, with a worst p95 regression of 3.150%.
Agent-loop tests pin durable-before-end ordering, interrupted visible prefixes, failed and retry attempts, abandonment, usage, and replay metadata. Session Controller and Conversation tests pin live transient display, reconnect baselines, committed settlement release, history replay, Chat and Trajectory parity, while TypeScript and Python SDK snapshots pin the external event representation.
diff --git a/.agents/notes/implemented/architecture/2026-09-01-v2-embedded-assistant-streams.zh.md b/.agents/notes/implemented/architecture/2026-09-01-v2-embedded-assistant-streams.zh.md
index 05e1a76f14..60e983e483 100644
--- a/.agents/notes/implemented/architecture/2026-09-01-v2-embedded-assistant-streams.zh.md
+++ b/.agents/notes/implemented/architecture/2026-09-01-v2-embedded-assistant-streams.zh.md
@@ -31,7 +31,7 @@ Web follow adapter 显式选择接收这些进程本地 frame，并为每个 sta
### 已发布 v1 到 v2 迁移
-相邻迁移会校验完整的冻结 v1 产物，按 turn、step、terminal boundary 与精确 message provenance 对 chunk 分组，再为每个 attempt 替换一个 settlement。成功分组的 chunk 移入其 message。未被认领的分组会在最后一个被消费 chunk 的位置变成 `assistant/attempt`。无关的交错事件保持相对顺序，存活事件获得密集 v2 序号。
+相邻迁移会校验完整的冻结 v1 产物，按 turn、step、terminal boundary 与精确 message provenance 对 chunk 分组，再为每个 attempt 替换一个 settlement。成功分组的 chunk 移入其 message。未被认领的分组会在最后一个被消费 chunk 的位置变成 `assistant/attempt`。无关的交错事件保持相对顺序，存活事件获得密集 v2 序号。该迁移边通过 `dsh-llm` 运行时的 `AssistantStreamAccumulator`、`expandAssistantStream` 与 `BlockAssembler` 压缩、展开并重组嵌入 stream，而不持有冻结副本，因为该包拥有 v2 stream 编码。目标校验会自行复核每个迁移后的 `assistant/message` 与其嵌入 stream 是否一致，因此不一致的 v1 日志会作为 unsupported migration 被拒绝并保留源产物，而不是由 installed Session restoration 报告为损坏。日后若某个格式改变 stream 编码，必须把这些 helper 的冻结副本纳入本迁移边。
该迁移边会重映射有限的已声明引用清单：信封 provenance、surface replacement 端点、command source event、compaction range 与 shadowed list，以及 title message list。经过校验的 `session/title-llm-request` 模型可见文本会在源序号命名空间中保持逐字节不变，而它的 `messageSeqs` 字段会迁移到 v2 命名空间；因此目标校验不会根据重映射后的序号重建该文本。指向被消费 chunk 的引用会使迁移失败；它绝不会被重定向到含义不同的 settlement。该迁移边也会拒绝切开 attempt 的继承切点。
@@ -47,7 +47,7 @@ Generation 选择与发布遵循[已发布 Session 迁移决策](2026-08-31-rele
紧凑 stream 测试固定 text、reasoning、tool argument、raw chunk、时间戳间隔、格式错误 record 与分离 snapshot 的精确累积和展开。v1 到 v2 测试覆盖成功与失败 attempt、交错、密集序号与引用重映射、源序号 title framing、seed 切点插入与切分拒绝、严格源与目标校验、每行一个事件的 v2 编码、与 backend 兼容的 provenance range、原始与 Zstandard 发布，以及无写入的当前读取。
-手工 performance acceptance 会在三轮、100 组 warmup pair 与 600 组 measured pair 下，针对同一批已经解析的物理 row，把静态 catalog routing 与直接 released-v2 restoration 比较；它不比较 v1 与 v2，也不计入 backend I/O。每个 pooled median 与 p95 regression 都必须保持在 5% 以内；已接受运行的最差 p95 regression 为 3.150%。`--smoke` 报告不参与 gate 的诊断 sample。
+合并前的 performance acceptance 在三轮、100 组 warmup pair 与 600 组 measured pair 下，针对同一批已经解析的物理 row，把静态 catalog routing 与直接 released-v2 restoration 比较；它不比较 v1 与 v2，也不计入 backend I/O。每个 pooled median 与 p95 regression 都保持在 5% 预算以内，最差 p95 regression 为 3.150%。
Agent-loop 测试固定先持久后 end 的顺序、中断的可见前缀、失败与重试 attempt、abandonment、usage 与 replay metadata。Session Controller 与 Conversation 测试固定实时瞬态显示、重连 baseline、committed settlement 发布、历史回放以及 Chat 与 Trajectory 一致性；TypeScript 与 Python SDK snapshot 固定外部事件表示。
diff --git a/package.json b/package.json
index 38b47d6106..80129383f1 100644
--- a/package.json
+++ b/package.json
@@ -50,7 +50,6 @@
"test:web:perf": "npm run build && npm run test:web:perf:built",
"test:web:perf:built": "DSH_SNAPSHOT=replay vitest run --config vitest.web.perf.config.ts",
"test:web:stress": "npm run build && vitest run --config vitest.web-stress.config.ts",
- "benchmark:session-format-v1-to-v2": "node --expose-gc --import tsx/esm scripts/benchmark-session-format-v1-to-v2.ts",
"benchmark:npm-resolution": "tsx scripts/benchmark-npm-resolution.ts",
"benchmark:npm-resolution:next": "tsx scripts/benchmark-next-package-dependency.ts",
"test:gui": "vitest run packages/client packages/host",
diff --git a/packages/session/session-format-v1-to-v2/README.i18n.yaml b/packages/session/session-format-v1-to-v2/README.i18n.yaml
index 8d6cd606e0..4cc1549168 100644
--- a/packages/session/session-format-v1-to-v2/README.i18n.yaml
+++ b/packages/session/session-format-v1-to-v2/README.i18n.yaml
@@ -2,5 +2,5 @@
# side as of the last confirmed-consistent state. Both languages carry equal authority;
# after editing either side, bring the other along and re-record with:
# pnpm run verify-translation-pairing --write packages/session/session-format-v1-to-v2/README.md
-README.md: 0e2661259b89cdbf4d5e138fe6d7a8fdfeaaea56
-README.zh.md: c066b994a7fe63fa2628fe0d5b4ae950f67da72b
+README.md: 1a61e99d8955ab07cca20bd67cc35c8a48f7a918
+README.zh.md: 3072807faf72f7f62b2342151914c76f62ab247e
diff --git a/packages/session/session-format-v1-to-v2/README.md b/packages/session/session-format-v1-to-v2/README.md
index 0e2661259b..1a61e99d89 100644
--- a/packages/session/session-format-v1-to-v2/README.md
+++ b/packages/session/session-format-v1-to-v2/README.md
@@ -44,14 +44,6 @@ The migration refuses a reference to a consumed chunk instead of redirecting it
The v2 physical header requires `isSeeded` and does not store a numeric cut. The codec derives the cut from the last inherited end-seed marker, writes one event per row, range-encodes only `sourceEventSeqs`, and remains neutral to ordinary event vocabulary and payload growth. Strict migration-target validation freezes the released-v2 inventory and rejects unknown types or members. Current restoration instead admits event types known to the installed Session package plus unknown events carrying `ignorable: true`, then delegates payload and stream semantics to the installed current restorer. All paths retain strict header, event-envelope, sequence, and inherited-cut validation.
-### Measure catalog-dispatch overhead
-
-```text
-pnpm run benchmark:session-format-v1-to-v2
-```
-
-The manual acceptance runs three repetitions with 100 warmup pairs and 600 alternating measured pairs per case. It measures static catalog-routing overhead against direct released-v2 restoration of the same already parsed physical rows, requiring every pooled median and p95 regression to stay within 5%. It neither compares v1 with v2 nor times backend I/O; representation sizes and absolute migration/replay costs are reported separately without a speedup claim. Add `--smoke` only for a short correctness and reporting pass; smoke timing is non-gating and is not an acceptance result.
-
-----
@@ -68,7 +60,6 @@ The edge first groups v1 chunks by turn, step, terminal finish, and explicit mes
| [`src/codec.ts`](src/codec.ts) | Released-v2 header, one-event-per-row encoding, provenance ranges, and recoverable prefix decoding |
| [`src/validation.ts`](src/validation.ts) | Physical v2 envelope/cut validation, exact migration-target policy, and vocabulary-neutral current restoration |
| [`src/dispositions.ts`](src/dispositions.ts) | Frozen released-v2 event and payload-member inventory |
-| [`scripts/benchmark-session-format-v1-to-v2.ts`](../../../scripts/benchmark-session-format-v1-to-v2.ts) | Repository-only catalog-dispatch, migration, token-meter, size, and memory acceptance report |
diff --git a/packages/session/session-format-v1-to-v2/README.zh.md b/packages/session/session-format-v1-to-v2/README.zh.md
index c066b994a7..3072807faf 100644
--- a/packages/session/session-format-v1-to-v2/README.zh.md
+++ b/packages/session/session-format-v1-to-v2/README.zh.md
@@ -44,14 +44,6 @@ const migratedV2 = sessionFormatV1ToV2.migrate(decodedV1)
v2 物理 header 要求 `isSeeded`，且不存储数值切点。编解码器从最后一个 inherited end-seed marker 推导切点，每行写入一个事件，只对 `sourceEventSeqs` 做范围编码，并对普通事件词汇与 payload 扩展保持中立。严格的迁移目标校验会冻结 released-v2 清单并拒绝未知 type 或 member。当前恢复则准入 installed Session package 已知的事件 type，以及携带 `ignorable: true` 的未知事件，再把 payload 与 stream 语义交给 installed current restorer。所有路径仍严格校验 header、event envelope、sequence 与 inherited cut。
-### 测量 catalog dispatch 开销
-
-```text
-pnpm run benchmark:session-format-v1-to-v2
-```
-
-手工 acceptance 会运行三轮，每个 case 使用 100 组 warmup pair 与 600 组交替测量 pair。它针对同一批已经解析的物理 row，把静态 catalog routing 与直接 released-v2 restoration 比较，并要求每个 pooled median 与 p95 regression 都保持在 5% 以内。它既不比较 v1 与 v2，也不计入 backend I/O；representation size 与 migration／replay 绝对成本会单独报告，且不声称加速。只在需要较短的正确性与报告检查时添加 `--smoke`；smoke timing 不参与 gate，也不是 acceptance 结果。
-
-----
@@ -68,7 +60,6 @@ pnpm run benchmark:session-format-v1-to-v2
| [`src/codec.ts`](src/codec.ts) | 已发布 v2 header、每行一个事件的编码、provenance 范围与可恢复前缀解码 |
| [`src/validation.ts`](src/validation.ts) | v2 物理 envelope／cut 校验、精确 migration-target 策略与 vocabulary-neutral current restoration |
| [`src/dispositions.ts`](src/dispositions.ts) | 冻结的已发布 v2 事件与 payload 成员清单 |
-| [`scripts/benchmark-session-format-v1-to-v2.ts`](../../../scripts/benchmark-session-format-v1-to-v2.ts) | 仅用于仓库的 catalog-dispatch、migration、token-meter、size 与 memory acceptance report |
diff --git a/packages/session/session-format-v1-to-v2/tests/validation.spec.ts b/packages/session/session-format-v1-to-v2/tests/validation.spec.ts
index fed915d1d1..891d4788a1 100644
--- a/packages/session/session-format-v1-to-v2/tests/validation.spec.ts
+++ b/packages/session/session-format-v1-to-v2/tests/validation.spec.ts
@@ -7,13 +7,11 @@ import type {
SessionFormatJsonValue,
} from '@deepseek-ai/dsh-session-format'
import {
- RELEASED_V2_EVENT_DISPOSITIONS,
RELEASED_V2_EVENT_TYPES,
assertReleasedV2Artifact,
assertReleasedV2Header,
restoreReleasedV2Artifact,
} from '@deepseek-ai/dsh-session-format-v1-to-v2'
-import { RELEASED_V0_EVENT_TYPES } from '@deepseek-ai/dsh-session-format-v0-to-v1'
const textBlock = { type: 'text', text: 'hello' } as const
const usage = { inputTokens: 3, outputTokens: 2 } as const
@@ -150,13 +148,6 @@ describe('released v2 header validation', () => {
})
describe('released v2 event envelopes and payloads', () => {
- it('freezes the complete event inventory while replacing chunks with attempts', () => {
- const expected = RELEASED_V0_EVENT_TYPES.filter(type => type !== 'assistant/chunk')
- expected.push('assistant/attempt')
- expect(RELEASED_V2_EVENT_TYPES).toStrictEqual([...expected].sort((left, right) => left.localeCompare(right, 'en')))
- expect(Object.keys(RELEASED_V2_EVENT_DISPOSITIONS).sort()).toStrictEqual(expected.sort())
- })
-
it('accepts empty artifacts and ordinary log events with a true ignorable marker', () => {
expect(() => { assertReleasedV2Artifact(artifact([])) }).not.toThrow()
expect(() => { assertReleasedV2Artifact(artifact([
diff --git a/scripts/benchmark-session-format-v1-to-v2.spec.ts b/scripts/benchmark-session-format-v1-to-v2.spec.ts
deleted file mode 100644
index d17fa085a9..0000000000
--- a/scripts/benchmark-session-format-v1-to-v2.spec.ts
+++ /dev/null
@@ -1,67 +0,0 @@
-import { describe, expect, it } from 'vitest'
-import {
- ACCEPTANCE_DEFAULTS,
- SMOKE_DEFAULTS,
- parseOptions,
- percentile,
-} from './benchmark-session-format-v1-to-v2.ts'
-
-describe('v2 performance acceptance options', () => {
- it('pins the full acceptance and non-gating smoke specifications', () => {
- expect(parseOptions([])).toEqual({
- ...ACCEPTANCE_DEFAULTS,
- smoke: false,
- help: false,
- })
- expect(parseOptions(['--smoke'])).toEqual({
- ...SMOKE_DEFAULTS,
- smoke: true,
- help: false,
- })
- })
-
- it('accepts split and equals-form overrides independently of option order', () => {
- expect(parseOptions([
- '--samples=9',
- '--smoke',
- '--runs', '2',
- '--warmups=0',
- '--threshold-percent', '4',
- ])).toEqual({
- runs: 2,
- warmups: 0,
- samples: 9,
- thresholdPercent: 4,
- smoke: true,
- help: false,
- })
- })
-
- it.each([
- [['--unknown'], /unknown benchmark option/],
- [['--runs'], /requires a numeric value/],
- [['--runs', '0'], /runs must be positive/],
- [['--samples', '-1'], /samples must be a non-negative safe integer/],
- [['--threshold-percent', '1.5'], /threshold-percent must be a non-negative safe integer/],
- ] as const)('rejects invalid options %j', (arguments_, expected) => {
- expect(() => parseOptions(arguments_)).toThrow(expected)
- })
-})
-
-describe('v2 performance acceptance statistics', () => {
- it('uses the PR3 discrete percentile estimator without mutating input', () => {
- const values = [4, 1, 3, 2]
- expect(percentile(values, 0)).toBe(1)
- expect(percentile(values, 0.5)).toBe(3)
- expect(percentile(values, 0.95)).toBe(4)
- expect(percentile(values, 1)).toBe(4)
- expect(values).toEqual([4, 1, 3, 2])
- })
-
- it('rejects empty samples and invalid fractions', () => {
- expect(() => percentile([], 0.5)).toThrow(/at least one sample/)
- expect(() => percentile([1], -0.1)).toThrow(/between zero and one/)
- expect(() => percentile([1], 1.1)).toThrow(/between zero and one/)
- expect(() => percentile([1], Number.NaN)).toThrow(/between zero and one/)
- })
-})
diff --git a/scripts/benchmark-session-format-v1-to-v2.ts b/scripts/benchmark-session-format-v1-to-v2.ts
deleted file mode 100644
index 95e5e000e4..0000000000
--- a/scripts/benchmark-session-format-v1-to-v2.ts
+++ /dev/null
@@ -1,774 +0,0 @@
-/**
- * Manual performance acceptance for the released v2 Session format.
- *
- * Run the full gate from the repository root with:
- *
- * pnpm run benchmark:session-format-v1-to-v2
- *
- * Use `--smoke` for a short correctness and reporting pass. The smoke mode
- * reports timing deltas but does not enforce the acceptance ceiling.
- */
-
-import { deepStrictEqual, ok } from 'node:assert'
-import { mkdir, mkdtemp, rm, writeFile } from 'node:fs/promises'
-import { tmpdir } from 'node:os'
-import { dirname, join } from 'node:path'
-import { performance } from 'node:perf_hooks'
-import { pathToFileURL } from 'node:url'
-import { Context } from '@deepseek-ai/cordis'
-import {
- KNOWN_SESSION_EVENT_TYPES,
- Session,
- SessionId,
- SessionLogOffset,
-} from '@deepseek-ai/dsh-session'
-import type {
- SessionEvent,
- SessionHeader,
- SessionId as SessionIdType,
-} from '@deepseek-ai/dsh-session'
-import type { SessionPersistence } from '@deepseek-ai/dsh-session-persistence'
-import {
- snapshotSessionFormatArtifact,
-} from '@deepseek-ai/dsh-session-format'
-import type {
- EncodedSessionFormatArtifact,
- SessionFormatArtifact,
- SessionFormatEvent,
- SessionFormatJsonObject,
-} from '@deepseek-ai/dsh-session-format'
-import {
- assertReleasedV1Artifact,
- releasedV1SessionFormatCodec,
-} from '@deepseek-ai/dsh-session-format-v0-to-v1'
-import {
- assertReleasedV2Artifact,
- releasedV2SessionFormatCodec,
- restoreReleasedV2Artifact,
- sessionFormatV1ToV2,
-} from '../packages/session/session-format-v1-to-v2/src/index.ts'
-import {
- sessionFormatCatalog,
-} from '../packages/session/session-format-catalog/src/generated.ts'
-import {
- validateInstalledCurrentSessionArtifact,
-} from '../packages/session/session-format-catalog/src/current.ts'
-import {
- compressZstdFrame,
- createZstdFrameDecoder,
- scanZstdFrames,
-} from '../packages/session/session-persistence-jsonl/src/zstd.ts'
-import JsonlSessionPersistence from '@deepseek-ai/dsh-session-persistence-jsonl'
-import {
- generationLogPath,
- type JsonlCompression,
-} from '../packages/session/session-persistence-jsonl/src/format.ts'
-import SessionProjectionRegistry from '@deepseek-ai/dsh-session-projection'
-import TokenMeter from '@deepseek-ai/dsh-token-meter'
-
-/** Reproducible full-acceptance sampling specification. */
-export const ACCEPTANCE_DEFAULTS = Object.freeze({
- runs: 3,
- warmups: 100,
- samples: 600,
- thresholdPercent: 5,
-})
-
-/** Short diagnostic sampling specification; it is not an acceptance result. */
-export const SMOKE_DEFAULTS = Object.freeze({
- runs: 1,
- warmups: 3,
- samples: 20,
- thresholdPercent: 5,
-})
-
-interface BenchmarkOptions {
- readonly runs: number
- readonly warmups: number
- readonly samples: number
- readonly thresholdPercent: number
- readonly smoke: boolean
- readonly help: boolean
-}
-
-interface PhysicalInput {
- readonly header: unknown
- readonly rows: readonly unknown[]
-}
-
-interface PhysicalFixture {
- readonly raw: Buffer
- readonly zstd: Buffer
-}
-
-interface Fixture {
- readonly name: 'small' | '100-turn'
- readonly turns: number
- readonly v1: SessionFormatArtifact
- readonly v2: SessionFormatArtifact
- readonly v1Physical: PhysicalFixture
- readonly v2Physical: PhysicalFixture
-}
-
-interface CurrentReadCase {
- readonly label: string
- readonly root: string
- readonly id: SessionIdType
- readonly compression: JsonlCompression
- readonly physical: PhysicalInput
- readonly expected: SessionFormatArtifact
-}
-
-interface Distribution {
- readonly medianUs: number
- readonly p95Us: number
-}
-
-interface PairedResult {
- readonly direct: Distribution
- readonly catalog: Distribution
- readonly medianRegressionPercent: number
- readonly p95RegressionPercent: number
- readonly passed: boolean
-}
-
-type StoredRead = SessionFormatArtifact
-
-interface EventExtras {
- readonly surfaceOp?: SessionFormatEvent['surfaceOp']
- readonly sourceEventSeqs?: SessionFormatEvent['sourceEventSeqs']
-}
-
-let resultSink = 0
-
-/** Parse benchmark-only CLI options without mutating process-global state. */
-export function parseOptions(argv: readonly string[]): BenchmarkOptions {
- const smoke = argv.includes('--smoke')
- const defaults = smoke ? SMOKE_DEFAULTS : ACCEPTANCE_DEFAULTS
- const values: {
- runs: number
- warmups: number
- samples: number
- thresholdPercent: number
- } = { ...defaults }
- let help = false
- for (let index = 0; index < argv.length; index += 1) {
- const argument = argv[index] as string
- if (argument === '--smoke') continue
- if (argument === '--help' || argument === '-h') {
- help = true
- continue
- }
- const match = /^--(runs|warmups|samples|threshold-percent)(?:=(.+))?$/.exec(argument)
- if (match === null) throw new Error(`unknown benchmark option ${JSON.stringify(argument)}`)
- const name = match[1]
- if (name === undefined) throw new Error(`unknown benchmark option ${JSON.stringify(argument)}`)
- const raw = match[2] ?? argv[index + 1]
- if (raw === undefined || (match[2] === undefined && raw.startsWith('--'))) {
- throw new Error(`${name} requires a numeric value`)
- }
- if (match[2] === undefined) index += 1
- const value = Number(raw)
- if (!Number.isSafeInteger(value) || value < 0) {
- throw new Error(`${name} must be a non-negative safe integer`)
- }
- if (name !== 'warmups' && value === 0) throw new Error(`${name} must be positive`)
- switch (name) {
- case 'runs': values.runs = value; break
- case 'warmups': values.warmups = value; break
- case 'samples': values.samples = value; break
- case 'threshold-percent': values.thresholdPercent = value; break
- }
- }
- return { ...values, smoke, help }
-}
-
-/** Calculate the discrete percentile used by the acceptance measurement. */
-export function percentile(values: readonly number[], fraction: number): number {
- if (values.length === 0) throw new Error('percentile requires at least one sample')
- if (!Number.isFinite(fraction) || fraction < 0 || fraction > 1) {
- throw new Error('percentile fraction must be between zero and one')
- }
- const sorted = [...values].sort((left, right) => left - right)
- return sorted[Math.min(sorted.length - 1, Math.floor(sorted.length * fraction))] as number
-}
-
-async function main(): Promise {
- const options = parseOptions(process.argv.slice(2))
- if (options.help) {
- printHelp()
- return
- }
- selfCheckStatistics()
-
- const root = await mkdtemp(join(tmpdir(), 'dsh-v2-format-acceptance-'))
- try {
- const fixtures = await Promise.all([
- createFixture('small', 1),
- createFixture('100-turn', 100),
- ])
- const cases = await materializeCurrentReadCases(root, fixtures)
- await validateFixtureReads(cases)
-
- console.log('Session format v2 catalog-dispatch overhead acceptance')
- console.log(
- 'Direct-current/no-dispatch baseline: the released-v2 codec restores the same parsed physical rows. '
- + 'The candidate routes those rows through the static format catalog; public handles validate each file once.',
- )
- console.log(
- `Sampling: ${options.runs} run(s), ${options.warmups} warmup pair(s), `
- + `${options.samples} alternating measured pair(s), ${options.thresholdPercent}% median/p95 ceiling.`,
- )
- if (options.smoke) console.log('Mode: diagnostic smoke; timing ceiling is reported but not enforced.')
- console.log('')
-
- reportFixtureSizes(fixtures)
- console.log('')
-
- let accepted = true
- console.log('Released-v2 restoration dispatch overhead (pooled hot samples)')
- for (const benchmarkCase of cases) {
- const result = runPairedCurrentRead(benchmarkCase, options)
- accepted &&= result.passed
- printPairedResult(benchmarkCase.label, result)
- }
-
- console.log('')
- console.log('Absolute costs (informational; no speedup claim)')
- for (const fixture of fixtures) {
- const migration = runDistribution(
- () => { consumeArtifact(sessionFormatV1ToV2.migrate(fixture.v1)) },
- options,
- )
- printDistribution(`${fixture.name} v1->v2 migration`, migration)
-
- const session = restoredSession(fixture.v2)
- const tokenMeter = runDistribution(
- () => { consumeMeasurement(measureWithFreshTokenMeter(session)) },
- options,
- )
- printDistribution(`${fixture.name} TokenMeter cold replay+measure`, tokenMeter)
- }
-
- console.log('')
- await reportMemory(fixtures, cases)
- const memory = process.memoryUsage()
- console.log(
- `process memory after run: heapUsed=${formatBytes(memory.heapUsed)}, `
- + `heapTotal=${formatBytes(memory.heapTotal)}, rss=${formatBytes(memory.rss)}`,
- )
-
- if (options.smoke) {
- console.log('')
- console.log('SMOKE COMPLETE (non-acceptance timing sample)')
- } else if (accepted) {
- console.log('')
- console.log('PASS: every pooled catalog-dispatch median and p95 regression is within the 5% ceiling.')
- } else {
- console.error('')
- console.error('FAIL: at least one pooled catalog-dispatch median or p95 regression exceeds the 5% ceiling.')
- process.exitCode = 1
- }
- } finally {
- await rm(root, { recursive: true, force: true })
- }
-}
-
-function printHelp(): void {
- console.log(`Usage: node --expose-gc --import tsx/esm ${process.argv[1] ?? ''} [options]
-
-Options:
- --smoke Use 1 run, 3 warmups, and 20 samples; do not enforce timing.
- --runs N Independent runs (acceptance default: 3).
- --warmups N Alternating warmup pairs per run (acceptance default: 100).
- --samples N Alternating measured pairs per run (acceptance default: 600).
- --threshold-percent N Median and p95 regression ceiling (acceptance default: 5).
- --help Show this help.`)
-}
-
-function selfCheckStatistics(): void {
- deepStrictEqual(parseOptions([]), { ...ACCEPTANCE_DEFAULTS, smoke: false, help: false })
- deepStrictEqual(parseOptions(['--smoke']), { ...SMOKE_DEFAULTS, smoke: true, help: false })
- deepStrictEqual(percentile([4, 1, 3, 2], 0.5), 3)
- deepStrictEqual(percentile([1, 2, 3, 4, 5], 0.95), 5)
-}
-
-async function createFixture(name: Fixture['name'], turns: number): Promise {
- const v1 = buildV1Artifact(name, turns)
- assertReleasedV1Artifact(v1)
- const v2 = sessionFormatV1ToV2.migrate(v1)
- assertReleasedV2Artifact(v2)
- validateInstalledCurrentSessionArtifact(v2)
- ok(v2.events.some(event => event.type === 'assistant/message'
- && Array.isArray((event.data as SessionFormatJsonObject)['stream'])
- && ((event.data as SessionFormatJsonObject)['stream'] as readonly unknown[]).length > 0))
-
- const v1Physical = await encodePhysical(releasedV1SessionFormatCodec.encodeArtifact(v1, { packChunks: true }))
- const v2Physical = await encodePhysical(releasedV2SessionFormatCodec.encodeArtifact(v2))
- const v1Decoded = releasedV1SessionFormatCodec.decodeArtifact(...physicalArguments(parseRaw(v1Physical.raw)))
- const v2Decoded = releasedV2SessionFormatCodec.decodeArtifact(...physicalArguments(parseRaw(v2Physical.raw)))
- deepStrictEqual(v1Decoded, v1)
- deepStrictEqual(v2Decoded, v2)
- deepStrictEqual(
- releasedV1SessionFormatCodec.decodeArtifact(...physicalArguments(parseZstd(v1Physical.zstd))),
- v1,
- )
- deepStrictEqual(
- releasedV2SessionFormatCodec.decodeArtifact(...physicalArguments(parseZstd(v2Physical.zstd))),
- v2,
- )
- return { name, turns, v1, v2, v1Physical, v2Physical }
-}
-
-function buildV1Artifact(name: string, turns: number): SessionFormatArtifact {
- const events: SessionFormatEvent[] = []
- const append = (
- type: string,
- time: number,
- data: SessionFormatEvent['data'],
- extras: EventExtras = {},
- ): number => {
- const seq = events.length
- events.push({ type, seq, time, data, ...extras })
- return seq
- }
-
- for (let turn = 1; turn <= turns; turn += 1) {
- const baseTime = turn * 1_000
- append('turn/start', baseTime, { turn })
- append('user/message', baseTime + 1, {
- id: `user-${name}-${turn}`,
- role: 'user',
- content: [{
- type: 'text',
- text: `Turn ${turn}: inspect the deterministic workspace report and explain the relevant changes clearly.`,
- }],
- source: { kind: 'user' },
- }, { surfaceOp: 'append' })
- append('step/start', baseTime + 2, { turn, step: 1 })
- append('request/header', baseTime + 3, {
- header: {
- config: { provider: 'mock', model: 'benchmark-model', maxTokens: 2_048 },
- system: 'Answer with a concise explanation grounded in the supplied report.',
- tools: [],
- },
- reason: 'initial',
- })
-
- const reasoningParts = [
- `Turn ${turn} establishes the requested scope. `,
- 'The durable facts are checked against the current session state. ',
- 'The response keeps the relevant behavior and omits unrelated details. ',
- 'The final wording records the observable result directly.',
- ]
- const textParts = [
- `For turn ${turn}, the report confirms the requested behavior. `,
- 'The current state is internally consistent, ',
- 'the persisted events retain their required ordering, ',
- 'and the replayed message matches the provider stream. ',
- 'No unrelated setting changes are included. ',
- 'The result remains deterministic across repeated reads. ',
- 'The validation path checks the complete artifact. ',
- 'This completes the requested analysis.',
- ]
- const usage = {
- inputTokens: 800 + turn * 8,
- outputTokens: 160,
- totalTokens: 1_080 + turn * 8,
- cacheReadTokens: 120,
- cacheWriteTokens: 0,
- reasoningTokens: 56,
- }
- const chunkSeqs: number[] = []
- chunkSeqs.push(append('assistant/chunk', baseTime + 10, {
- turn, step: 1, chunk: { type: 'block-start', index: 0, blockType: 'reasoning' },
- }))
- for (const [index, text] of reasoningParts.entries()) {
- chunkSeqs.push(append('assistant/chunk', baseTime + 11 + index, {
- turn, step: 1, chunk: { type: 'reasoning-delta', index: 0, text },
- }))
- }
- chunkSeqs.push(append('assistant/chunk', baseTime + 15, {
- turn,
- step: 1,
- chunk: {
- type: 'block-end',
- index: 0,
- block: { type: 'reasoning', text: reasoningParts.join('') },
- },
- }))
- chunkSeqs.push(append('assistant/chunk', baseTime + 16, {
- turn, step: 1, chunk: { type: 'block-start', index: 1, blockType: 'text' },
- }))
- for (const [index, text] of textParts.entries()) {
- chunkSeqs.push(append('assistant/chunk', baseTime + 17 + index, {
- turn, step: 1, chunk: { type: 'text-delta', index: 1, text },
- }))
- }
- chunkSeqs.push(append('assistant/chunk', baseTime + 25, {
- turn,
- step: 1,
- chunk: { type: 'block-end', index: 1, block: { type: 'text', text: textParts.join('') } },
- }))
- chunkSeqs.push(append('assistant/chunk', baseTime + 26, {
- turn, step: 1, chunk: { type: 'usage', usage },
- }))
- chunkSeqs.push(append('assistant/chunk', baseTime + 27, {
- turn, step: 1, chunk: { type: 'finish', reason: { kind: 'stop' } },
- }))
- append('assistant/message', baseTime + 28, {
- turn,
- step: 1,
- message: {
- id: `assistant-${name}-${turn}`,
- role: 'assistant',
- content: [
- { type: 'reasoning', text: reasoningParts.join('') },
- { type: 'text', text: textParts.join('') },
- ],
- source: { kind: 'model', provider: 'mock', model: 'benchmark-model' },
- },
- usage,
- }, { surfaceOp: 'append', sourceEventSeqs: chunkSeqs })
- append('step/end', baseTime + 29, { turn, step: 1 })
- append('turn/end', baseTime + 30, { turn, reason: { kind: 'completed' } })
- }
-
- return snapshotSessionFormatArtifact({
- header: {
- version: 1,
- id: `v2-performance-${name}`,
- createdAt: 1,
- cwd: '/benchmark',
- isSeeded: false,
- delegationDepth: 0,
- agentPreset: 'benchmark',
- },
- inheritedEventCount: 0,
- events,
- }, `${name} v1 benchmark artifact`)
-}
-
-async function encodePhysical(encoded: EncodedSessionFormatArtifact): Promise {
- const headerLine = `${JSON.stringify(encoded.header)}\n`
- const body = `${encoded.rows.map(row => JSON.stringify(row)).join('\n')}\n`
- const raw = Buffer.from(headerLine + body)
- const zstd = Buffer.concat([
- await compressZstdFrame(headerLine),
- await compressZstdFrame(body),
- ])
- return { raw, zstd }
-}
-
-async function materializeCurrentReadCases(
- root: string,
- fixtures: readonly Fixture[],
-): Promise {
- const cases: CurrentReadCase[] = []
- for (const fixture of fixtures) {
- for (const compression of ['none', 'zstd'] as const) {
- const caseRoot = join(root, `${fixture.name}-${compression}`)
- const id = SessionId(fixture.v2.header.id)
- const path = generationLogPath(caseRoot, fixture.v2.header.cwd, id, 2, compression)
- await mkdir(dirname(path), { recursive: true })
- await writeFile(path, compression === 'none' ? fixture.v2Physical.raw : fixture.v2Physical.zstd)
- cases.push({
- label: `${compression === 'none' ? 'raw' : 'Zstandard'} ${fixture.name}`,
- root: caseRoot,
- id,
- compression,
- physical: compression === 'none'
- ? parseRaw(fixture.v2Physical.raw)
- : parseZstd(fixture.v2Physical.zstd),
- expected: fixture.v2,
- })
- }
- }
- return cases
-}
-
-async function validateFixtureReads(cases: readonly CurrentReadCase[]): Promise {
- for (const benchmarkCase of cases) {
- const mounted = await mountBackend(benchmarkCase)
- try {
- const handle = await mounted.persistence.open(benchmarkCase.id, 'read')
- try {
- deepStrictEqual(handle.header, benchmarkCase.expected.header)
- deepStrictEqual(handle.inheritedEventCount, benchmarkCase.expected.inheritedEventCount)
- deepStrictEqual(await handle.read(), benchmarkCase.expected.events)
- } finally {
- await handle.close()
- }
- const direct = directCurrentRead(benchmarkCase)
- const catalog = catalogCurrentRead(benchmarkCase)
- deepStrictEqual(direct, benchmarkCase.expected)
- deepStrictEqual(catalog, direct)
- } finally {
- await mounted.dispose()
- }
- }
-}
-
-async function mountBackend(benchmarkCase: CurrentReadCase): Promise<{
- readonly persistence: SessionPersistence
- readonly dispose: () => Promise
-}> {
- const context = new Context()
- await context.plugin(JsonlSessionPersistence, {
- root: benchmarkCase.root,
- compression: benchmarkCase.compression,
- })
- return {
- persistence: context.sessionPersistence,
- dispose: async () => context.fiber.dispose(),
- }
-}
-
-function directCurrentRead(benchmarkCase: CurrentReadCase): StoredRead {
- const decoded = snapshotSessionFormatArtifact(
- releasedV2SessionFormatCodec.decodeArtifact(
- benchmarkCase.physical.header,
- benchmarkCase.physical.rows,
- ),
- 'direct released-v2 decoded artifact',
- )
- const source = snapshotSessionFormatArtifact(decoded, 'direct released-v2 source')
- const restored = restoreReleasedV2Artifact(source, KNOWN_SESSION_EVENT_TYPES)
- validateInstalledCurrentSessionArtifact(restored)
- return snapshotSessionFormatArtifact(restored, 'direct current Session restoration')
-}
-
-function catalogCurrentRead(benchmarkCase: CurrentReadCase): StoredRead {
- return sessionFormatCatalog.migrate(sessionFormatCatalog.decodeArtifact(
- benchmarkCase.physical.header,
- benchmarkCase.physical.rows,
- ))
-}
-
-function parseRaw(bytes: Buffer): PhysicalInput {
- return parseJsonl(bytes.toString('utf8'))
-}
-
-function parseZstd(bytes: Buffer): PhysicalInput {
- const scan = scanZstdFrames(bytes)
- if (scan.tornStart !== undefined || scan.frames.length !== 2) {
- throw new Error('benchmark Zstandard artifact must contain two complete frames')
- }
- const decoder = createZstdFrameDecoder()
- let text = ''
- for (const plaintext of decoder.decode(bytes, scan.frames)) text += plaintext.toString('utf8')
- return parseJsonl(text)
-}
-
-function parseJsonl(text: string): PhysicalInput {
- if (!text.endsWith('\n')) throw new Error('benchmark artifact lacks its final newline')
- const lines = text.slice(0, -1).split('\n')
- const header = JSON.parse(lines[0] as string) as unknown
- const rows = lines.slice(1).map(line => JSON.parse(line) as unknown)
- return { header, rows }
-}
-
-function physicalArguments(input: PhysicalInput): [unknown, readonly unknown[]] {
- return [input.header, input.rows]
-}
-
-function runPairedCurrentRead(
- benchmarkCase: CurrentReadCase,
- options: BenchmarkOptions,
-): PairedResult {
- const pooledDirect: number[] = []
- const pooledCatalog: number[] = []
- for (let run = 0; run < options.runs; run += 1) {
- forceGc()
- for (let warmup = 0; warmup < options.warmups; warmup += 1) {
- if ((warmup + run) % 2 === 0) {
- consumeStored(directCurrentRead(benchmarkCase))
- consumeStored(catalogCurrentRead(benchmarkCase))
- } else {
- consumeStored(catalogCurrentRead(benchmarkCase))
- consumeStored(directCurrentRead(benchmarkCase))
- }
- }
- const direct: number[] = []
- const catalog: number[] = []
- for (let sample = 0; sample < options.samples; sample += 1) {
- if ((sample + run) % 2 === 0) {
- direct.push(timedUs(() => { consumeStored(directCurrentRead(benchmarkCase)) }))
- catalog.push(timedUs(() => { consumeStored(catalogCurrentRead(benchmarkCase)) }))
- } else {
- catalog.push(timedUs(() => { consumeStored(catalogCurrentRead(benchmarkCase)) }))
- direct.push(timedUs(() => { consumeStored(directCurrentRead(benchmarkCase)) }))
- }
- }
- pooledDirect.push(...direct)
- pooledCatalog.push(...catalog)
- const directRun = distribution(direct)
- const catalogRun = distribution(catalog)
- console.log(
- ` ${benchmarkCase.label} run ${run + 1}: direct ${formatDistribution(directRun)}; `
- + `catalog ${formatDistribution(catalogRun)}`,
- )
- }
-
- const direct = distribution(pooledDirect)
- const catalog = distribution(pooledCatalog)
- const medianRegressionPercent = percentChange(catalog.medianUs, direct.medianUs)
- const p95RegressionPercent = percentChange(catalog.p95Us, direct.p95Us)
- return {
- direct,
- catalog,
- medianRegressionPercent,
- p95RegressionPercent,
- passed: medianRegressionPercent <= options.thresholdPercent
- && p95RegressionPercent <= options.thresholdPercent,
- }
-}
-
-function runDistribution(operation: () => void, options: BenchmarkOptions): Distribution {
- const samples: number[] = []
- for (let run = 0; run < options.runs; run += 1) {
- forceGc()
- for (let warmup = 0; warmup < options.warmups; warmup += 1) operation()
- for (let sample = 0; sample < options.samples; sample += 1) samples.push(timedUs(operation))
- }
- return distribution(samples)
-}
-
-function distribution(samples: readonly number[]): Distribution {
- return { medianUs: percentile(samples, 0.5), p95Us: percentile(samples, 0.95) }
-}
-
-function timedUs(operation: () => void): number {
- const started = performance.now()
- operation()
- return (performance.now() - started) * 1_000
-}
-
-function consumeArtifact(artifact: SessionFormatArtifact): void {
- resultSink = (resultSink + artifact.events.length + artifact.header.version) | 0
-}
-
-function consumeStored(stored: StoredRead): void {
- resultSink = (resultSink + stored.events.length + stored.header.version) | 0
-}
-
-function consumeMeasurement(measurement: ReturnType): void {
- resultSink = (resultSink + measurement.totalTokens + measurement.nodes.length) | 0
-}
-
-function restoredSession(artifact: SessionFormatArtifact): Session {
- return Session.fromRestore(
- SessionId(artifact.header.id),
- artifact.events as SessionEvent[],
- artifact.header as unknown as SessionHeader,
- SessionLogOffset(artifact.inheritedEventCount),
- )
-}
-
-function measureWithFreshTokenMeter(session: Session): ReturnType {
- const context = new Context()
- new SessionProjectionRegistry(context)
- const meter = new TokenMeter(context)
- return meter.measure(session)
-}
-
-function reportFixtureSizes(fixtures: readonly Fixture[]): void {
- console.log('Logical and physical representation')
- for (const fixture of fixtures) {
- const eventReduction = reductionPercent(fixture.v2.events.length, fixture.v1.events.length)
- const rawReduction = reductionPercent(fixture.v2Physical.raw.length, fixture.v1Physical.raw.length)
- const zstdReduction = reductionPercent(fixture.v2Physical.zstd.length, fixture.v1Physical.zstd.length)
- console.log(
- ` ${fixture.name} (${fixture.turns} turn(s)): logical events ${fixture.v1.events.length} -> ${fixture.v2.events.length} `
- + `(${eventReduction.toFixed(3)}% fewer); raw ${formatBytes(fixture.v1Physical.raw.length)} -> `
- + `${formatBytes(fixture.v2Physical.raw.length)} (change ${formatSigned(-rawReduction)}%); `
- + `Zstandard ${formatBytes(fixture.v1Physical.zstd.length)} -> ${formatBytes(fixture.v2Physical.zstd.length)} `
- + `(change ${formatSigned(-zstdReduction)}%).`,
- )
- }
-}
-
-function printPairedResult(label: string, result: PairedResult): void {
- console.log(
- ` ${label} pooled: direct ${formatDistribution(result.direct)}; catalog ${formatDistribution(result.catalog)}; `
- + `regression median=${formatSigned(result.medianRegressionPercent)}%, `
- + `p95=${formatSigned(result.p95RegressionPercent)}% [${result.passed ? 'within ceiling' : 'exceeds ceiling'}]`,
- )
-}
-
-function printDistribution(label: string, result: Distribution): void {
- console.log(` ${label}: ${formatDistribution(result)}`)
-}
-
-async function reportMemory(
- fixtures: readonly Fixture[],
- cases: readonly CurrentReadCase[],
-): Promise {
- console.log('Retained heap (informational; one forced-GC observation)')
- if (globalThis.gc === undefined) {
- console.log(' unavailable: rerun with --expose-gc for retained-heap observations')
- return
- }
- for (const fixture of fixtures) {
- const rawCase = cases.find(candidate => candidate.label === `raw ${fixture.name}`) as CurrentReadCase
- const direct = await retainedHeap(() => directCurrentRead(rawCase))
- const catalog = await retainedHeap(() => catalogCurrentRead(rawCase))
- const migration = await retainedHeap(() => sessionFormatV1ToV2.migrate(fixture.v1))
- const session = restoredSession(fixture.v2)
- const tokenMeter = await retainedHeap(() => measureWithFreshTokenMeter(session))
- console.log(
- ` ${fixture.name}: direct current=${formatSignedBytes(direct)}, `
- + `catalog current=${formatSignedBytes(catalog)}, migration=${formatSignedBytes(migration)}, `
- + `TokenMeter=${formatSignedBytes(tokenMeter)}`,
- )
- }
-}
-
-async function retainedHeap(operation: () => object | Promise