fix(ui-trajectory): timing provenance on sub-span lanes; assembled snapshot for both views

Responding to ds-review-bot on #664:

- SubSpanLane gains a 'timing' discriminant (measured | running | unknown).
  A settle-only replay entry (callTime null, start outside the window) was
  previously indistinguishable from a measured 0 ms span; it now renders
  hollow with a 'duration unknown' hover title, and durationMs stays null
  for anything unmeasured. Pairs with the client-runtime fix that stopped
  fabricating callTime = settle time (826c3696a on the live-parallel PR).

- The built-client Code Mode fixture snapshot now switches to the
  Trajectory and Waterfall tabs and pins the assembled rendering: three
  Sub cells with real +0.8s durations and three measured lanes with their
  hover titles — product-visible coverage through the real bundle graph,
  not just package-level jsdom fixtures.

Agent Note (both languages) updated for the timing contract; pairing
re-recorded.
This commit is contained in:
Tianyi Cui
2026-07-26 14:20:59 +08:00
parent c5e1f53d7e
commit 835156b038
8 changed files with 133 additions and 14 deletions

View File

@@ -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
2026-07-26-code-mode-trajectory-waterfall-spans.md: 54449bcf8612a39461a769173d7f60c742f67ad8
2026-07-26-code-mode-trajectory-waterfall-spans.zh.md: fbfb26c3a62554d60a6cb561ead78e10cd4115cd
2026-07-26-code-mode-trajectory-waterfall-spans.md: fe4dcc25dbf211cf69e0d33937cf87a7482852e2
2026-07-26-code-mode-trajectory-waterfall-spans.zh.md: aaae06b1fca1b5587d06aa7704adec421d2b2c27

View File

@@ -15,7 +15,7 @@ Trajectory and waterfall still rendered a `run_code` turn as one opaque Tool cel
**Trajectory: `subtool` cells interleaved after their parent Tool cell. Waterfall: real-time sub-lanes under the owning turn row.**
- **Trajectory**: the layout fold takes the snapshot's `codeDispatches` index; after each Tool cell whose `callId` has dispatches (assistant-block calls, orphan results, and running calls alike), it interleaves one `subtool` cell per sub-dispatch in start order — indexes stay sequential across the interleave. A settled sub-call's duration is its start/settle pair (`durationSeconds(sub.time, sub.callTime)`); a running one shows the em dash, exactly the native in-flight convention. The new cell kind wears a `Sub` tag (business tint) and a 28px indent so nesting reads at a glance.
- **Waterfall**: `deriveSubSpans` folds the dispatch index into per-turn lanes with REAL timing — each parent's dispatch window is first start → last settle, and every lane's offset/width is its fraction of that window, so parallel sub-calls (PR3) visibly overlap. Running lanes extend to the window end at reduced opacity with a null duration. Lanes draw under the owning turn's bar row, scaled into a fixed lane budget.
- **Waterfall**: `deriveSubSpans` folds the dispatch index into per-turn lanes with REAL timing — each parent's dispatch window is first start → last settle, and every lane's offset/width is its fraction of that window, so parallel sub-calls (PR3) visibly overlap. Each lane carries a `timing` provenance tag: `measured` (pair observed), `running` (settle pending — extends to the window end at reduced opacity), or `unknown` (settle-only replay window, `callTime: null` — drawn hollow and titled "duration unknown", never a fabricated 0 ms). Lanes draw under the owning turn's bar row, scaled into a fixed lane budget.
- Both views read `codeDispatches` through the standard snapshot hook — no new wire data, no new stores; replay renders identically to live by construction.
## Alternatives considered
@@ -28,4 +28,4 @@ Trajectory and waterfall still rendered a `run_code` turn as one opaque Tool cel
## Consequences
The waterfall carries the first REAL wall-time rendering in the client (turn bars remain node-count stand-ins — the contrast is deliberate and labeled by hover titles). Trajectory cell indexes now count sub-calls, so `#N` totals grow on Code Mode turns. Specs pin the interleave order and durations, the running em-dash arm, window fractions (offsets/widths), the running-lane extension, and the rendered lane under the turn row.
The waterfall carries the first REAL wall-time rendering in the client (turn bars remain node-count stand-ins — the contrast is deliberate and labeled by hover titles). Trajectory cell indexes now count sub-calls, so `#N` totals grow on Code Mode turns. Specs pin the interleave order and durations, the running em-dash arm, window fractions (offsets/widths), the running-lane extension, the unknown-timing (settle-only) lane, and the rendered lane under the turn row; the built-client Code Mode fixture snapshot additionally pins both tabs' assembled rendering (sub-cells with real +0.8s durations, measured lanes).

View File

@@ -15,7 +15,7 @@ trajectory 过去仍把一个 `run_code` 轮次渲染为单个不透明的 Tool
**trajectory`subtool` 单元格穿插在其父 Tool 单元格之后。waterfall所属轮次行之下、带真实计时的子泳道sub-lane**
- **trajectory**:布局 fold 接收快照的 `codeDispatches` 索引;凡某个 Tool 单元格的 `callId` 名下存在分发assistant 块内的调用、孤儿结果与运行中的调用一视同仁fold 就在该单元格之后按启动顺序为每个子分发穿插一个 `subtool` 单元格,索引在整个穿插序列中保持连续编号。已结算子调用的耗时来自其 start/settle 事件对(`durationSeconds(sub.time, sub.callTime)`);运行中的子调用则显示破折号,与原生的进行中约定完全一致。新增的单元格类型带有 `Sub` 标签business 色调)与 28px 缩进,嵌套关系一眼可辨。
- **waterfall**`deriveSubSpans` 把分发索引折叠成带真实计时的逐轮次泳道:每个父调用的分发窗口为首个 start → 最后一个 settle每条泳道的偏移/宽度即其在该窗口中的占比因此并行的子调用PR3会肉眼可见地重叠。运行中的泳道以较低不透明度延伸至窗口末端,耗时为 null。泳道绘制在所属轮次的条形行之下,并缩放进固定的泳道预算。
- **waterfall**`deriveSubSpans` 把分发索引折叠成带真实计时的逐轮次泳道:每个父调用的分发窗口为首个 start → 最后一个 settle每条泳道的偏移/宽度即其在该窗口中的占比因此并行的子调用PR3会肉眼可见地重叠。每条泳道带有 `timing` 来源标记:`measured`(观察到了成对事件)、`running`settle 未到 — 以较低不透明度延伸至窗口末端)或 `unknown`(回放窗口只含 settle、`callTime: null` — 画成空心并以「duration unknown」为悬停标题绝不伪造 0 ms。泳道绘制在所属轮次的条形行之下,并缩放进固定的泳道预算。
- 两个视图都经由标准的快照 hook 读取 `codeDispatches`:没有新的 wire 数据,也没有新的 store回放的渲染由构造保证与实时完全一致。
## 曾考虑的替代方案
@@ -28,4 +28,4 @@ trajectory 过去仍把一个 `run_code` 轮次渲染为单个不透明的 Tool
## 后果
waterfall 承载了 client 中第一处真实的墙钟时间渲染轮次条仍是节点计数的占位这一反差是有意为之并由悬停标题标注。trajectory 的单元格索引现在会把子调用计入,因此 Code Mode 轮次上的 `#N` 总数会随之增大。spec 锁定穿插顺序与耗时、运行中的破折号分支、窗口占比(偏移/宽度)、运行中泳道的延伸,以及轮次行之下实际渲染出的泳道
waterfall 承载了 client 中第一处真实的墙钟时间渲染轮次条仍是节点计数的占位这一反差是有意为之并由悬停标题标注。trajectory 的单元格索引现在会把子调用计入,因此 Code Mode 轮次上的 `#N` 总数会随之增大。spec 锁定穿插顺序与耗时、运行中的破折号分支、窗口占比(偏移/宽度)、运行中泳道的延伸、unknown 计时(仅 settle泳道以及轮次行之下实际渲染出的泳道构建产物级的 Code Mode fixture 快照另行锁定两个标签页的组装后渲染(带真实 +0.8s 耗时的子单元格、measured 泳道)

View File

@@ -5,7 +5,8 @@
// the code-variant parent row titled by the model-authored description, its
// three always-visible nested sub-rows (bash through the sample registration,
// read through GenericToolCard, the failing read wearing the error state),
// the expanded program body, and details-panel resolution of a sub-callId.
// the expanded program body, details-panel resolution of a sub-callId, and
// the trajectory/waterfall tabs' sub-call cells and timing lanes.
import { readFileSync } from 'node:fs'
import { join } from 'node:path'
import { act, cleanup, fireEvent, screen, waitFor, within } from '@testing-library/react'
@@ -177,3 +178,64 @@ it('expands the code row into the program body and resolves a sub-row through th
}
`)
})
it('trajectory and waterfall surface the run_code sub-calls with real timing', async () => {
boot()
await openFixtureSession()
// Switch to the trajectory tab (same slot ring the chat view registers in).
fireEvent.click(await screen.findByRole('tab', { name: 'Trajectory' }))
await waitFor(() => {
expect(document.querySelector('[data-kind="subtool"]')).not.toBeNull()
}, { timeout: 10_000 })
const subCells = [...document.querySelectorAll('[data-kind="subtool"]')]
expect({
// Three Sub cells nested under the run_code Tool cell, in dispatch order,
// each with a real +N.Ns own-duration off the start/settle pair (the
// fixture spaces every event 800ms apart — never the em dash).
subCells: subCells.map(cell => visibleText(cell)),
}).toMatchInlineSnapshot(`
{
"subCells": [
"#53Subbash · {"command":"ls notes","description":"List notes"}+0.8s",
"#54Subread · {"path":"notes/demo.txt"}+0.8s",
"#55Subread · {"path":"notes/missing.txt"}+0.8s",
],
}
`)
// Waterfall: each sub-call draws a measured lane scaled into the parent
// turn's dispatch window.
fireEvent.click(screen.getByRole('tab', { name: 'Waterfall' }))
await waitFor(() => {
expect(document.querySelector('[data-subspan]')).not.toBeNull()
}, { timeout: 10_000 })
const lanes = [...document.querySelectorAll('[data-subspan]')]
expect({
lanes: lanes.map(lane => ({
label: visibleText(lane.querySelector('[class*="subTag"]') ?? lane),
title: lane.querySelector('[data-timing]')?.getAttribute('title'),
timing: lane.querySelector('[data-timing]')?.getAttribute('data-timing'),
})),
}).toMatchInlineSnapshot(`
{
"lanes": [
{
"label": "bash",
"timing": "measured",
"title": "bash · 0.80s",
},
{
"label": "read",
"timing": "measured",
"title": "read · 0.80s",
},
{
"label": "read",
"timing": "measured",
"title": "read · 0.80s",
},
],
}
`)
})

View File

@@ -55,12 +55,15 @@ export function WaterfallView({ useSession, pxPerNode }: ConvViewProps & Waterfa
<span className={css.subTag}>{lane.name}</span>
<span
className={`${css.bar} ${css.barSub}`}
data-running={lane.durationMs === null || undefined}
data-timing={lane.timing}
style={{
marginLeft: Math.round(lane.offsetFraction * SUB_LANE_PX),
width: Math.max(Math.round(lane.widthFraction * SUB_LANE_PX), 4),
}}
title={lane.durationMs === null ? `${lane.name} · running` : `${lane.name} · ${(lane.durationMs / 1000).toFixed(2)}s`}
title={lane.timing === 'measured'
/* durationMs is non-null exactly when timing is measured. */
? `${lane.name} · ${((lane.durationMs ?? 0) / 1000).toFixed(2)}s`
: lane.timing === 'running' ? `${lane.name} · running` : `${lane.name} · duration unknown`}
/>
</div>
))}

View File

@@ -9,8 +9,14 @@ import type { ConversationNode, ConversationSnapshot } from '@deepseek-ai/dsh-cl
export interface SubSpanLane {
callId: string
name: string
/** Wall duration in ms; null while running (start seen, settle not). */
/** Wall duration in ms; null unless both endpoints were observed (`timing: 'measured'`). */
durationMs: number | null
/**
* Timing provenance: `measured` = start/settle pair observed; `running` =
* start seen, settle pending; `unknown` = settle-only replay window (the
* start fell outside), so no duration claim is possible.
*/
timing: 'measured' | 'running' | 'unknown'
/** Start offset as a fraction of the parent turn's dispatch window [0, 1). */
offsetFraction: number
/** Width as a fraction of the window (running lanes extend to the window end). */
@@ -106,6 +112,9 @@ export function deriveSubSpans(
for (const [parent, subs] of codeDispatches) {
if (subs.length === 0) continue
const turn = turnByCall.get(parent) ?? currentTurn
// A settle-only entry (callTime null: its start fell outside the replay
// window) anchors the window by its settle time — a real observation —
// but must never masquerade as a measured zero-duration span.
const starts: number[] = []
const ends: number[] = []
for (const sub of subs) {
@@ -119,12 +128,14 @@ export function deriveSubSpans(
const windowSpan = windowEnd - windowStart
const lanes: SubSpanLane[] = subs.map((sub, i) => {
const settled = 'kind' in sub
const timing = settled ? (sub.callTime === null ? 'unknown' as const : 'measured' as const) : 'running' as const
const start = starts[i] ?? windowStart
const end = settled ? sub.time : windowEnd
return {
callId: sub.callId,
name: settled ? sub.call?.name ?? sub.callId : sub.name,
durationMs: settled ? Math.max(0, sub.time - start) : null,
durationMs: timing === 'measured' ? Math.max(0, end - start) : null,
timing,
offsetFraction: (start - windowStart) / windowSpan,
widthFraction: Math.max((end - start) / windowSpan, 0.02),
}

View File

@@ -71,6 +71,12 @@
background: var(--dsw-alias-state-business-primary);
}
.barSub[data-running] {
.barSub[data-timing='running'] {
opacity: 0.45;
}
/* Settle-only replay entries: no measured span — hollow, not a solid bar. */
.barSub[data-timing='unknown'] {
background: transparent;
border: 1px dashed var(--dsw-alias-state-business-primary);
}

View File

@@ -288,7 +288,7 @@ describe('deriveSubSpans (waterfall lanes)', () => {
const turn3 = lanes.get(3)
expect(turn3).toHaveLength(2)
// Window = 6200..8200 (2000ms). bash: 0..0.4; read: 0.4..1.0.
expect(turn3?.[0]).toMatchObject({ name: 'bash', durationMs: 800, offsetFraction: 0 })
expect(turn3?.[0]).toMatchObject({ name: 'bash', durationMs: 800, timing: 'measured', offsetFraction: 0 })
expect(turn3?.[0]?.widthFraction).toBeCloseTo(0.4)
expect(turn3?.[1]).toMatchObject({ name: 'read', durationMs: 1200 })
expect(turn3?.[1]?.offsetFraction).toBeCloseTo(0.4)
@@ -305,11 +305,23 @@ describe('deriveSubSpans (waterfall lanes)', () => {
]]]) as unknown as ConversationSnapshot['codeDispatches']
const lanes = deriveSubSpans(dispatchNodes, codeDispatches)
const running = lanes.get(3)?.find((lane) => lane.name === 'grep')
expect(running).toMatchObject({ durationMs: null })
expect(running).toMatchObject({ durationMs: null, timing: 'running' })
// Extends from its start to the window end.
expect(running!.offsetFraction + running!.widthFraction).toBeCloseTo(1)
})
it('a settle-only entry (null callTime) is unknown timing, never a measured 0 ms', () => {
const codeDispatches = new Map([['p1', [
{
kind: 'tool-result', seq: 101, time: 8_000, callId: 'p1:code:1',
call: { name: 'bash', argsRaw: '{}' }, callTime: null,
content: [], isError: false, callView: null, resultView: null,
},
]]]) as unknown as ConversationSnapshot['codeDispatches']
const lane = deriveSubSpans(dispatchNodes, codeDispatches).get(3)?.[0]
expect(lane).toMatchObject({ durationMs: null, timing: 'unknown' })
})
it('waterfall renders sub-span lanes under the owning turn row', () => {
const codeDispatches = new Map([['p1', [
{
@@ -333,5 +345,30 @@ describe('deriveSubSpans (waterfall lanes)', () => {
expect(lane).not.toBeNull()
expect(lane!.textContent).toContain('bash')
expect(lane!.querySelector('[title*="1.80s"]')).not.toBeNull()
expect(lane!.querySelector('[data-timing="measured"]')).not.toBeNull()
})
it('waterfall labels a settle-only lane as duration unknown', () => {
const codeDispatches = new Map([['p1', [
{
kind: 'tool-result', seq: 101, time: 8_000, callId: 'p1:code:1',
call: { name: 'read', argsRaw: '{}' }, callTime: null,
content: [], isError: false, callView: null, resultView: null,
},
]]]) as unknown as ConversationSnapshot['codeDispatches']
const store = createSnapshotStore({
nodes: dispatchNodes, partial: null,
runningCalls: [] as ConversationSnapshot['runningCalls'], codeDispatches,
})
const props = {
sessionId: SID,
useSession: bindSnapshotSelector(store) as unknown as UseSession<ConversationSnapshot>,
useSessions: emptySessions(),
useWorkspaces: emptyWorkspaces(),
} as unknown as ConvViewProps
const view = render(createElement(WaterfallView as FC<ConvViewProps>, props))
const bar = view.container.querySelector('[data-timing="unknown"]')
expect(bar).not.toBeNull()
expect(bar!.getAttribute('title')).toContain('duration unknown')
})
})