perf: add build stage analysis dimension and simplify project name inference (#2559)

Author: xusd320

.github/workflows/pack-perf.yml

@@ -71,7 +71,7 @@ jobs:
 
       - name: Analyze Traces
         run: |
-          python3 agents/tools/analyze_trace.py .trace/trace_antd.json .trace/report_antd.md
+          TRACE_PROJECT=examples/with-antd python3 agents/tools/analyze_trace.py .trace/trace_antd.json .trace/report_antd.md
           # Prepare PR comment content
           echo "## 📊 Performance Benchmark Report (with-antd)" > .trace/pr_comment.md
           cat .trace/report_antd.md >> .trace/pr_comment.md

agents/tools/analyze_trace.py

@@ -12,7 +12,7 @@
 from collections import defaultdict
 from datetime import datetime
 
-def analyze_trace(trace_path, output_path):
+def analyze_trace(trace_path, output_path, override_project_name=None):
     """Analyze Chrome Trace and generate performance report following the protocol"""
 
     print(f"📊 Loading trace: {trace_path}")
@@ -21,6 +21,9 @@ def analyze_trace(trace_path, output_path):
 
     events = data if isinstance(data, list) else data.get("traceEvents", [])
     print(f"✅ Analyzing {len(events):,} events...")
+
+    # Try to infer project name
+    project_name = override_project_name or os.environ.get("TRACE_PROJECT") or "Unknown Project"
 
     # Initialize metrics
     min_ts, max_ts = float('inf'), float('-inf')
@@ -130,18 +133,6 @@ def analyze_trace(trace_path, output_path):
     trace_file = os.path.basename(trace_path)
     trace_size_gb = os.path.getsize(trace_path) / (1024**3)
 
-    # Try to infer project name
-    project_name = "Unknown Project"
-    abs_path = os.path.abspath(trace_path)
-    path_parts = abs_path.split(os.sep)
-    if "examples" in path_parts:
-        try:
-            idx = path_parts.index("examples")
-            if idx + 1 < len(path_parts):
-                project_name = f"examples/{path_parts[idx+1]}"
-        except ValueError:
-            pass
-    
     # Pre-calculate derived metrics
     max_thread_work = max(thread_work, 1) # Avoid division by zero
     total_tasks_safe = max(total_tasks, 1)
@@ -336,14 +327,15 @@ def analyze_trace(trace_path, output_path):
 
 if __name__ == "__main__":
     if len(sys.argv) < 3:
-        print("Usage: python3 analyze_trace.py <trace_file> <output_report>")
+        print("Usage: python3 analyze_trace.py <trace_file> <output_report> [project_name]")
         sys.exit(1)
 
     trace_file = sys.argv[1]
     report_file = sys.argv[2]
+    override_project_name = sys.argv[3] if len(sys.argv) > 3 else None
 
     if not os.path.exists(trace_file):
         print(f"Error: Trace file not found: {trace_file}")
         sys.exit(1)
 
-    analyze_trace(trace_file, report_file)
+    analyze_trace(trace_file, report_file, override_project_name)

agents/utoopack-performance-agent.md

@@ -97,7 +97,8 @@ Follow this tiered hierarchy. Solve P0 before descending to P1, as high-level sc
 ---
 
 ## 🚀 Step 3: Actionable Diagnostic Workflow
-1. **Quantitative Baseline**: Run the summary script `python3 agents/tools/analyze_trace.py <trace_file> <output_report>` to list Top 20 tasks by `count` and `sum(duration)`.
+1. **Quantitative Baseline**: Run the summary script with the **`TRACE_PROJECT`** environment variable to identify the build target.
+   - *Command*: `TRACE_PROJECT=examples/with-antd python3 agents/tools/analyze_trace.py <trace_file> <output_report>`
 2. **Qualitative Timeline Scan**: Open the trace in `chrome://tracing` or `edge://tracing`. Look for "Wall-like" structures (Parallelism) vs "Staircase" structures (Serialism).
 3. **Causal Attribution**: Identify the `Parent Span` of the top bottlenecks to understand *why* they were invoked.
 4. **Final Reporting**: Summarize findings and save the report to `./agents/reports/utoopack_performance_report_YYYYMMDD_HHMMSS.md`. Include specific Tiered signals and recommended actions.