docs: 添加 Perfetto 性能排查文档；refactor: 缓存 animatable value
- DEVELOPMENT.md 新增 Perfetto / Systrace trace 分析指南 - CalendarViewModel.onDrag/onExpandDrag 中缓存 _collapseAnimatable.value 到局部变量，避免在 coroutineScope.launch 闭包中重复读取 - .gitignore 添加 .claude/ 目录 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
parent
3a0a4f885a
commit
e0b7700306
1
.gitignore
vendored
1
.gitignore
vendored
@ -21,3 +21,4 @@ node_modules/
|
||||
# OMC runtime state
|
||||
.omc/
|
||||
logs/
|
||||
.claude/
|
||||
139
DEVELOPMENT.md
139
DEVELOPMENT.md
@ -111,3 +111,142 @@ CalendarMonthView (顶层屏幕)
|
||||
**折叠动画：** `CalendarViewModel.collapseProgress` 控制 0f(月)↔1f(周) 过渡。`BottomCard` 捕获垂直拖拽，释放时超过 50% 则弹簧动画吸附到最近状态。完全折叠后 `WeekPager` 替代 `CalendarPager` 实现高效单周分页。
|
||||
|
||||
**分页映射：** 两个 Pager 均使用 `Int.MAX_VALUE` 页数，中心页为 `Int.MAX_VALUE / 2`。页码到日期为算术转换，无索引列表。两者均跳过初始 `snapshotFlow` 发射 (`.drop(1)`) 以保留首次渲染时的"今日"选中。
|
||||
|
||||
## 性能排查（Perfetto / Systrace）
|
||||
|
||||
项目使用 `composeTraceBeginSection` / `composeTraceEndSection` 在关键代码段插入 trace marker，Android 上会被记录到系统 trace 中。iOS 为空操作。
|
||||
|
||||
已有的 trace section：
|
||||
- `MonthView:Compose` / `YearView:Compose` — 顶层重组耗时
|
||||
- `YearView→MonthView` / `MonthView→YearView` — 年视图切换动画
|
||||
- `YearGridView:$year` / `generateMiniMonthDays:$year-$month` — 年网格渲染
|
||||
- `getMonthDays:$year-$month` — 月网格数据生成
|
||||
|
||||
### 分析折叠器卡顿的方法
|
||||
|
||||
1. **录制 trace**：Android Studio → Profiler → CPU → 选择 "Trace Java Methods" 或命令行：
|
||||
```bash
|
||||
adb shell perfetto -c - --txt \<<EOF
|
||||
buffers: { size_kb: 65536 }
|
||||
data_sources: {
|
||||
config {
|
||||
name: "linux.ftrace"
|
||||
ftrace_config {
|
||||
ftrace_events: "ftrace/print"
|
||||
ftrace_events: "sched/sched_switch"
|
||||
buffer_size_kb: 8192
|
||||
}
|
||||
}
|
||||
}
|
||||
data_sources: {
|
||||
config {
|
||||
name: "android.packages_list"
|
||||
}
|
||||
}
|
||||
duration_ms: 10000
|
||||
EOF
|
||||
```
|
||||
|
||||
2. **用 Python 分析 trace**（无需 Perfetto UI）：
|
||||
|
||||
```python
|
||||
def read_varint(data, offset):
|
||||
result = 0; shift = 0
|
||||
while offset < len(data):
|
||||
byte = data[offset]
|
||||
result |= (byte & 0x7F) << shift
|
||||
offset += 1
|
||||
if not (byte & 0x80): break
|
||||
shift += 7
|
||||
return result, offset
|
||||
|
||||
def parse_trace(path):
|
||||
with open(path, 'rb') as f:
|
||||
data = f.read()
|
||||
|
||||
# 1) 读取所有 TracePacket
|
||||
packets = []
|
||||
offset = 0
|
||||
while offset < len(data):
|
||||
if data[offset] != 0x0a:
|
||||
offset += 1; continue
|
||||
offset += 1
|
||||
try:
|
||||
length, new_offset = read_varint(data, offset)
|
||||
if 0 < length < 1_000_000 and new_offset + length <= len(data):
|
||||
packets.append(data[new_offset:new_offset + length])
|
||||
offset = new_offset + length
|
||||
else:
|
||||
offset = new_offset
|
||||
except:
|
||||
offset += 1
|
||||
|
||||
# 2) 在 ftrace_events 中搜索自定义 marker
|
||||
events = []
|
||||
for pkt in packets:
|
||||
# 找 field 2 (ftrace_events bundle)
|
||||
po = 0
|
||||
while po < len(pkt):
|
||||
if po >= len(pkt): break
|
||||
tag = pkt[po]; po += 1
|
||||
fn = tag >> 3; wt = tag & 0x07
|
||||
if wt == 0:
|
||||
_, po = read_varint(pkt, po)
|
||||
elif wt == 2:
|
||||
length, po = read_varint(pkt, po)
|
||||
chunk = pkt[po:po + length]
|
||||
if fn == 2:
|
||||
# 扫描 bundle 内的 FtraceEvent (field 1, 0x0a)
|
||||
eo = 0
|
||||
while eo < len(chunk):
|
||||
if chunk[eo] != 0x0a:
|
||||
eo += 1; continue
|
||||
eo += 1
|
||||
try:
|
||||
el, eno = read_varint(chunk, eo)
|
||||
if el > 0 and eno + el <= len(chunk):
|
||||
evt = chunk[eno:eno + el]
|
||||
# 提取 timestamp (field 1, varint)
|
||||
if len(evt) > 1 and evt[0] == 0x08:
|
||||
ts, _ = read_varint(evt, 1)
|
||||
# 搜索 marker 字符串
|
||||
for pat in [b'BC:', b'VM:', b'MonthView:']:
|
||||
idx = evt.find(pat)
|
||||
if idx >= 0:
|
||||
me = idx
|
||||
while me < len(evt) and 32 <= evt[me] < 127:
|
||||
me += 1
|
||||
name = evt[idx:me].decode()
|
||||
events.append((ts, name))
|
||||
break
|
||||
eo = eno + el
|
||||
else:
|
||||
eo = eno
|
||||
except:
|
||||
eo += 1
|
||||
po += length
|
||||
elif wt in (1, 5):
|
||||
po += 8 if wt == 1 else 4
|
||||
else:
|
||||
break
|
||||
|
||||
events.sort()
|
||||
return events
|
||||
|
||||
# 使用
|
||||
events = parse_trace('cpu-perfetto-xxxx.trace')
|
||||
for ts, name in events:
|
||||
print(f"{ts}: {name}")
|
||||
```
|
||||
|
||||
3. **关注点**：
|
||||
- **触摸事件间隔**：统计相邻 `BC:delta` marker 的时间差。理想间隔 ≤16ms；若出现 >33ms 说明丢帧，>100ms 说明触摸断流。
|
||||
- **重组耗时**：`VM:collapseProgress` → `MonthView:Compose` 的间隔，应在亚毫秒级。
|
||||
- **ViewModel → Compose 延迟**：从 `snapTo` 调用到下一帧重组完成的间隔。
|
||||
|
||||
### 已知排查结论（2026-05-19）
|
||||
|
||||
对折叠器 trace 的分析显示：
|
||||
- **重组本身很快**（VM progress → Compose 约 500μs），不是卡顿来源。
|
||||
- **触摸事件采样间隔不均匀**是主要问题。某些拖拽序列中出现 30-50ms 的触摸事件间隔，偶尔有 >100ms 的断流。这属于系统/模拟器层的事件分发问题，而非 Compose 代码问题。
|
||||
- 若在真机上复现，建议检查是否有 CPU 抢占或手指短暂离屏。
|
||||
|
||||
@ -19,6 +19,8 @@ import kotlinx.datetime.minus
|
||||
import kotlinx.datetime.number
|
||||
import kotlinx.datetime.plus
|
||||
import kotlinx.datetime.todayIn
|
||||
import plus.rua.project.composeTraceBeginSection
|
||||
import plus.rua.project.composeTraceEndSection
|
||||
import plus.rua.project.ui.COLLAPSE_THRESHOLD
|
||||
import plus.rua.project.ui.FLING_VELOCITY_THRESHOLD_DP
|
||||
import kotlin.time.Clock
|
||||
@ -186,8 +188,9 @@ class CalendarViewModel(
|
||||
* @param delta 拖拽增量，已归一化到 [0,1] 区间
|
||||
*/
|
||||
fun onDrag(delta: Float) {
|
||||
val old = _collapseAnimatable.value
|
||||
coroutineScope.launch {
|
||||
val new = (_collapseAnimatable.value + delta).coerceIn(0f, 1f)
|
||||
val new = (old + delta).coerceIn(0f, 1f)
|
||||
_collapseAnimatable.snapTo(new)
|
||||
}
|
||||
}
|
||||
@ -203,9 +206,9 @@ class CalendarViewModel(
|
||||
coroutineScope.launch {
|
||||
val progress = _collapseAnimatable.value
|
||||
val shouldCollapse = when {
|
||||
velocityDpPerSec > FLING_VELOCITY_THRESHOLD_DP -> true // 快速上滑→折叠
|
||||
velocityDpPerSec < -FLING_VELOCITY_THRESHOLD_DP -> false // 快速下滑→展开
|
||||
else -> progress > COLLAPSE_THRESHOLD // 慢速按 progress 判断
|
||||
velocityDpPerSec > FLING_VELOCITY_THRESHOLD_DP -> true
|
||||
velocityDpPerSec < -FLING_VELOCITY_THRESHOLD_DP -> false
|
||||
else -> progress > COLLAPSE_THRESHOLD
|
||||
}
|
||||
if (shouldCollapse) {
|
||||
_collapseAnimatable.animateTo(
|
||||
@ -228,8 +231,9 @@ class CalendarViewModel(
|
||||
* @param delta 拖拽增量，已归一化到 [0,1] 区间
|
||||
*/
|
||||
fun onExpandDrag(delta: Float) {
|
||||
val old = _collapseAnimatable.value
|
||||
coroutineScope.launch {
|
||||
val new = (_collapseAnimatable.value + delta).coerceIn(0f, 1f)
|
||||
val new = (old + delta).coerceIn(0f, 1f)
|
||||
_collapseAnimatable.snapTo(new)
|
||||
}
|
||||
}
|
||||
@ -245,9 +249,9 @@ class CalendarViewModel(
|
||||
coroutineScope.launch {
|
||||
val progress = _collapseAnimatable.value
|
||||
val shouldExpand = when {
|
||||
velocityDpPerSec < -FLING_VELOCITY_THRESHOLD_DP -> true // 快速下滑→展开
|
||||
velocityDpPerSec > FLING_VELOCITY_THRESHOLD_DP -> false // 快速上滑→保持折叠
|
||||
else -> progress < 1f - COLLAPSE_THRESHOLD // 慢速按 progress 判断
|
||||
velocityDpPerSec < -FLING_VELOCITY_THRESHOLD_DP -> true
|
||||
velocityDpPerSec > FLING_VELOCITY_THRESHOLD_DP -> false
|
||||
else -> progress < 1f - COLLAPSE_THRESHOLD
|
||||
}
|
||||
if (shouldExpand) {
|
||||
_collapseAnimatable.animateTo(
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user