supersize: Add symbol.template_name, and strip <>s from symbol.name

tools/binary_size/libsupersize/cluster_symbols.py

Index: tools/binary_size/libsupersize/cluster_symbols.py
diff --git a/tools/binary_size/libsupersize/cluster_symbols.py b/tools/binary_size/libsupersize/cluster_symbols.py
index 2c1f407da43b3c9b4e301100373f07c2bb026c2d..672c777fed6ef9416ef3beb8b5d9850994a74db7 100644
--- a/tools/binary_size/libsupersize/cluster_symbols.py
+++ b/tools/binary_size/libsupersize/cluster_symbols.py
@@ -8,8 +8,18 @@ import collections
 import logging
 import re
 
+import function_signature
 
-# Refer to models.SymbolGroup.Cluster() for pydoc
+
+def _StripCloneSuffix(name):
+  # Multiple attributes could exist, so search from left-to-right.
+  idx = name.find(' [clone ')
+  if idx != -1:
+    return name[:idx]
+  return name
+
+
+# Refer to models.SymbolGroup.Clustered() for pydoc
 def ClusterSymbols(symbols):
   # http://unix.stackexchange.com/questions/223013/function-symbol-gets-part-suffix-after-compilation
   # Example name suffixes:
@@ -22,40 +32,40 @@ def ClusterSymbols(symbols):
   logging.debug('Creating name -> symbol map')
   clone_indices = []
   indices_by_full_name = {}
-  # (section_name, name, full_name) -> [(index, sym),...]
+  # (section_name, full_name_no_attr) -> [(index, sym),...]
   replacements_by_tup = collections.defaultdict(list)
   for i, symbol in enumerate(symbols):
-    if symbol.name.startswith('**'):
+    name = symbol.full_name
+    if not name:
+      continue
+    if name.startswith('*'):
       # "symbol gap 3" -> "symbol gaps"
-      name = re.sub(r'\s+\d+( \(.*\))?$', 's', symbol.name)
-      replacements_by_tup[(symbol.section_name, name, None)].append((i, symbol))
-    elif symbol.full_name:
-      if symbol.full_name.endswith(']') and ' [clone ' in symbol.full_name:
-        clone_indices.append(i)
-      else:
-        indices_by_full_name[symbol.full_name] = i
+      name = re.sub(r'\s+\d+( \(.*\))?$', 's', name)
+      replacements_by_tup[(symbol.section_name, name)].append((i, symbol))
+    elif name.endswith(']') and ' [clone ' in name:
+      clone_indices.append(i)
+    else:
+      indices_by_full_name[name] = i
 
   # Step 2: Collect same-named clone symbols.
   logging.debug('Grouping all clones')
   group_names_by_index = {}
   for i in clone_indices:
     symbol = symbols[i]
-    # Multiple attributes could exist, so search from left-to-right.
-    stripped_name = symbol.name[:symbol.name.index(' [clone ')]
-    stripped_full_name = symbol.full_name[:symbol.full_name.index(' [clone ')]
-    name_tup = (symbol.section_name, stripped_name, stripped_full_name)
+    stripped_full_name = _StripCloneSuffix(symbol.full_name)
+    name_tup = (symbol.section_name, stripped_full_name)
     replacement_list = replacements_by_tup[name_tup]
 
     if not replacement_list:
       # First occurance, check for non-clone symbol.
-      non_clone_idx = indices_by_full_name.get(stripped_name)
+      non_clone_idx = indices_by_full_name.get(stripped_full_name)
       if non_clone_idx is not None:
         non_clone_symbol = symbols[non_clone_idx]
         replacement_list.append((non_clone_idx, non_clone_symbol))
-        group_names_by_index[non_clone_idx] = stripped_name
+        group_names_by_index[non_clone_idx] = stripped_full_name
 
     replacement_list.append((i, symbol))
-    group_names_by_index[i] = stripped_name
+    group_names_by_index[i] = stripped_full_name
 
   # Step 3: Undo clustering when length=1.
   # Removing these groups means Diff() logic must know about [clone] suffix.
@@ -77,14 +87,13 @@ def ClusterSymbols(symbols):
   dest_index = 0
   src_index = 0
   seen_tups = set()
-  replacement_tup_by_index = {}
+  index_and_name_tups = []

[estevenson, 2017/05/09 22:15:12]

I find this function hard to grok, basically had to try to figure out what it
does all over again. Might be worth it to add some more detailed comments, your
call.

[agrieve, 2017/05/10 00:51:38]

likewise. For now I've just added a TODO comment to the top saying that it's not
actually even conceptually different from GroupedBy(). Maybe could just be built
ontop of that?

   for name_tup, replacement_list in replacements_by_tup.iteritems():
-    for tup in replacement_list:
-      replacement_tup_by_index[tup[0]] = name_tup
+    for symbol_tup in replacement_list:
+      index_and_name_tups.append((symbol_tup[0], name_tup))
 
-  sorted_items = replacement_tup_by_index.items()
-  sorted_items.sort(key=lambda tup: tup[0])
-  for index, name_tup in sorted_items:
+  index_and_name_tups.sort(key=lambda tup: tup[0])
+  for index, name_tup in index_and_name_tups:
     count = index - src_index
     grouped_symbols[dest_index:dest_index + count] = (
         symbols[src_index:src_index + count])
@@ -93,9 +102,20 @@ def ClusterSymbols(symbols):
     if name_tup not in seen_tups:
       seen_tups.add(name_tup)
       group_symbols = [tup[1] for tup in replacements_by_tup[name_tup]]
-      grouped_symbols[dest_index] = symbols._CreateTransformed(
-          group_symbols, name=name_tup[1], full_name=name_tup[2],
-          section_name=name_tup[0])
+      section_name, stripped_full_name = name_tup
+      if stripped_full_name.startswith('*'):
+        stripped_template_name = stripped_full_name
+        stripped_name = stripped_full_name
+      else:
+        stripped_template_name = _StripCloneSuffix(
+            group_symbols[0].template_name)
+        stripped_name = _StripCloneSuffix(group_symbols[0].name)
+      cluster = symbols._CreateTransformed(
+          group_symbols, full_name=stripped_full_name,
+          template_name=stripped_template_name, name=stripped_name,
+          section_name=section_name)
+      function_signature.InternSameNames(cluster)
+      grouped_symbols[dest_index] = cluster
       dest_index += 1
 
   assert len(grouped_symbols[dest_index:None]) == len(symbols[src_index:None])