Track opcode sample counts in flamegraph collector

Stores per-node opcode counts in the tree structure. Exports opcode
mapping (names and deopt relationships) in JSON so the JS renderer can
show instruction names and distinguish specialized variants.

Author: pablogsal

Lib/profiling/sampling/_flamegraph_assets/flamegraph.js

@@ -8,6 +8,32 @@ let currentThreadFilter = 'all';
 // Heat colors are now defined in CSS variables (--heat-1 through --heat-8)
 // and automatically switch with theme changes - no JS color arrays needed!
 
+// Opcode mappings - loaded from embedded data (generated by Python)
+let OPCODE_NAMES = {};
+let DEOPT_MAP = {};
+
+// Initialize opcode mappings from embedded data
+function initOpcodeMapping(data) {
+    if (data && data.opcode_mapping) {
+        OPCODE_NAMES = data.opcode_mapping.names || {};
+        DEOPT_MAP = data.opcode_mapping.deopt || {};
+    }
+}
+
+// Get opcode info from opcode number
+function getOpcodeInfo(opcode) {
+    const opname = OPCODE_NAMES[opcode] || `<${opcode}>`;
+    const baseOpcode = DEOPT_MAP[opcode];
+    const isSpecialized = baseOpcode !== undefined;
+    const baseOpname = isSpecialized ? (OPCODE_NAMES[baseOpcode] || `<${baseOpcode}>`) : opname;
+
+    return {
+        opname: opname,
+        baseOpname: baseOpname,
+        isSpecialized: isSpecialized
+    };
+}
+
 // ============================================================================
 // String Resolution
 // ============================================================================
@@ -249,6 +275,55 @@ function createPythonTooltip(data) {
         </div>`;
     }
 
+    // Create bytecode/opcode section if available
+    let opcodeSection = "";
+    const opcodes = d.data.opcodes;
+    if (opcodes && typeof opcodes === 'object' && Object.keys(opcodes).length > 0) {
+      // Sort opcodes by sample count (descending)
+      const sortedOpcodes = Object.entries(opcodes)
+        .sort((a, b) => b[1] - a[1])
+        .slice(0, 8); // Limit to top 8
+
+      const totalOpcodeSamples = sortedOpcodes.reduce((sum, [, count]) => sum + count, 0);
+      const maxCount = sortedOpcodes[0][1] || 1;
+
+      const opcodeLines = sortedOpcodes.map(([opcode, count]) => {
+        const opcodeInfo = getOpcodeInfo(parseInt(opcode, 10));
+        const pct = ((count / totalOpcodeSamples) * 100).toFixed(1);
+        const barWidth = (count / maxCount) * 100;
+        const specializedBadge = opcodeInfo.isSpecialized
+          ? '<span style="background: #2e7d32; color: white; font-size: 9px; padding: 1px 4px; border-radius: 3px; margin-left: 4px;">SPECIALIZED</span>'
+          : '';
+        const baseOpHint = opcodeInfo.isSpecialized
+          ? `<span style="color: #888; font-size: 11px; margin-left: 4px;">(${opcodeInfo.baseOpname})</span>`
+          : '';
+
+        return `
+          <div style="display: grid; grid-template-columns: 1fr 60px 60px; gap: 8px; align-items: center; padding: 3px 0;">
+            <div style="font-family: monospace; font-size: 11px; color: ${opcodeInfo.isSpecialized ? '#2e7d32' : '#333'}; overflow: hidden; text-overflow: ellipsis; white-space: nowrap;">
+              ${opcodeInfo.opname}${baseOpHint}${specializedBadge}
+            </div>
+            <div style="text-align: right; font-size: 11px; color: #666;">${count.toLocaleString()}</div>
+            <div style="background: #e9ecef; border-radius: 2px; height: 8px; overflow: hidden;">
+              <div style="background: linear-gradient(90deg, #3776ab, #5a9bd5); height: 100%; width: ${barWidth}%;"></div>
+            </div>
+          </div>`;
+      }).join('');
+
+      opcodeSection = `
+        <div style="margin-top: 16px; padding-top: 12px;
+                    border-top: 1px solid #e9ecef;">
+          <div style="color: #3776ab; font-size: 13px;
+                      margin-bottom: 8px; font-weight: 600;">
+            Bytecode Instructions:
+          </div>
+          <div style="background: #f8f9fa; border: 1px solid #e9ecef;
+                      border-radius: 6px; padding: 10px;">
+            ${opcodeLines}
+          </div>
+        </div>`;
+    }
+
     const fileLocationHTML = isSpecialFrame ? "" : `
       <div class="tooltip-location">${filename}${d.data.lineno ? ":" + d.data.lineno : ""}</div>`;
 
@@ -275,6 +350,7 @@ function createPythonTooltip(data) {
         ` : ''}
       </div>
       ${sourceSection}
+      ${opcodeSection}
       <div class="tooltip-hint">
         ${childCount > 0 ? "Click to zoom into this function" : "Leaf function - no children"}
       </div>
@@ -994,6 +1070,9 @@ function initFlamegraph() {
     processedData = resolveStringIndices(EMBEDDED_DATA);
   }
 
+  // Initialize opcode mapping from embedded data
+  initOpcodeMapping(EMBEDDED_DATA);
+
   originalData = processedData;
   initThreadFilter(processedData);
 

Lib/profiling/sampling/stack_collector.py

@@ -7,7 +7,8 @@
 import os
 
 from ._css_utils import get_combined_css
-from .collector import Collector
+from .collector import Collector, extract_lineno
+from .opcode_utils import get_opcode_mapping
 from .string_table import StringTable
 
 
@@ -32,7 +33,11 @@ def __init__(self, *args, **kwargs):
         self.stack_counter = collections.Counter()
 
     def process_frames(self, frames, thread_id):
-        call_tree = tuple(reversed(frames))
+        # Extract only (filename, lineno, funcname) - opcode not needed for collapsed stacks
+        # frame is (filename, location, funcname, opcode)
+        call_tree = tuple(
+            (f[0], extract_lineno(f[1]), f[2]) for f in reversed(frames)
+        )
         self.stack_counter[(call_tree, thread_id)] += 1
 
     def export(self, filename):
@@ -205,6 +210,11 @@ def convert_children(children, min_samples):
                     source_indices = [self._string_table.intern(line) for line in source]
                     child_entry["source"] = source_indices
 
+                # Include opcode data if available
+                opcodes = node.get("opcodes", {})
+                if opcodes:
+                    child_entry["opcodes"] = dict(opcodes)
+
                 # Recurse
                 child_entry["children"] = convert_children(
                     node["children"], min_samples
@@ -251,6 +261,9 @@ def convert_children(children, min_samples):
                 **stats
             }
 
+        # Build opcode mapping for JS
+        opcode_mapping = get_opcode_mapping()
+
         # If we only have one root child, make it the root to avoid redundant level
         if len(root_children) == 1:
             main_child = root_children[0]
@@ -265,6 +278,7 @@ def convert_children(children, min_samples):
             }
             main_child["threads"] = sorted(list(self._all_threads))
             main_child["strings"] = self._string_table.get_strings()
+            main_child["opcode_mapping"] = opcode_mapping
             return main_child
 
         return {
@@ -277,27 +291,41 @@ def convert_children(children, min_samples):
                 "per_thread_stats": per_thread_stats_with_pct
             },
             "threads": sorted(list(self._all_threads)),
-            "strings": self._string_table.get_strings()
+            "strings": self._string_table.get_strings(),
+            "opcode_mapping": opcode_mapping
         }
 
     def process_frames(self, frames, thread_id):
-        # Reverse to root->leaf
-        call_tree = reversed(frames)
+        """Process stack frames into flamegraph tree structure.
+
+        Args:
+            frames: List of (filename, location, funcname, opcode) tuples in
+                    leaf-to-root order. location is (lineno, end_lineno, col_offset, end_col_offset).
+                    opcode is None if not gathered.
+            thread_id: Thread ID for this stack trace
+        """
+        # Reverse to root->leaf order for tree building
         self._root["samples"] += 1
         self._total_samples += 1
         self._root["threads"].add(thread_id)
         self._all_threads.add(thread_id)
 
         current = self._root
-        for func in call_tree:
+        for filename, location, funcname, opcode in reversed(frames):
+            lineno = extract_lineno(location)
+            func = (filename, lineno, funcname)
             func = self._func_intern.setdefault(func, func)
-            children = current["children"]
-            node = children.get(func)
+
+            node = current["children"].get(func)
             if node is None:
-                node = {"samples": 0, "children": {}, "threads": set()}
-                children[func] = node
+                node = {"samples": 0, "children": {}, "threads": set(), "opcodes": collections.Counter()}
+                current["children"][func] = node
             node["samples"] += 1
             node["threads"].add(thread_id)
+
+            if opcode is not None:
+                node["opcodes"][opcode] += 1
+
             current = node
 
     def _get_source_lines(self, func):