supersize: Strip linker-added prefixes from symbol names

tools/binary_size/libsupersize/archive.py

 # Copyright 2017 The Chromium Authors. All rights reserved.
 # Use of this source code is governed by a BSD-style license that can be
 # found in the LICENSE file.
 
 """Main Python API for analyzing binary size."""
 
 import argparse
 import calendar
 import collections
 import datetime
@@ skipping 20 matching lines @@
 
 def _OpenMaybeGz(path, mode=None):
   """Calls `gzip.open()` if |path| ends in ".gz", otherwise calls `open()`."""
   if path.endswith('.gz'):
     if mode and 'w' in mode:
       return gzip.GzipFile(path, mode, 1)
     return gzip.open(path, mode)
   return open(path, mode or 'r')
 
 
+def _StripLinkerAddedSymbolPrefixes(symbols):
+  """Removes prefixes sometimes added to symbol names during link
+
+  Removing prefixes make symbol names match up with those found in .o files.
+  """
+  for symbol in symbols:
+    name = symbol.name
+    if name.startswith('startup.'):
+      symbol.flags |= models.FLAG_STARTUP
+      symbol.name = name[8:]
+    elif name.startswith('unlikely.'):
+      symbol.flags |= models.FLAG_UNLIKELY
+      symbol.name = name[9:]
+    elif name.startswith('rel.local.'):
+      symbol.flags |= models.FLAG_REL_LOCAL
+      symbol.name = name[10:]
+    elif name.startswith('rel.'):
+      symbol.flags |= models.FLAG_REL
+      symbol.name = name[4:]
+
+
 def _UnmangleRemainingSymbols(symbols, tool_prefix):
   """Uses c++filt to unmangle any symbols that need it."""
   to_process = [s for s in symbols if s.name.startswith('_Z')]
   if not to_process:
     return
 
   logging.info('Unmangling %d names', len(to_process))
   proc = subprocess.Popen([tool_prefix + 'c++filt'], stdin=subprocess.PIPE,
                           stdout=subprocess.PIPE)
   stdout = proc.communicate('\n'.join(s.name for s in to_process))[0]
@@ skipping 31 matching lines @@
       found_prefixes.add(symbol.name[:idx + 3])
       symbol.name = symbol.name[idx + 4:] + ' [' + symbol.name[:idx] + ']'
 
     # Strip out return type, and identify where parameter list starts.
     if symbol.section == 't':
       symbol.full_name, symbol.name = function_signature.Parse(symbol.name)
 
     # Remove anonymous namespaces (they just harm clustering).
     non_anonymous = symbol.name.replace('(anonymous namespace)::', '')
     if symbol.name != non_anonymous:
-      symbol.is_anonymous = True
+      symbol.flags |= models.FLAG_ANONYMOUS
       symbol.name = non_anonymous
       symbol.full_name = symbol.full_name.replace(
           '(anonymous namespace)::', '')
 
     if symbol.section != 't' and '(' in symbol.name:
       # Pretty rare. Example:
       # blink::CSSValueKeywordsHash::findValueImpl(char const*)::value_word_list
       symbol.full_name = symbol.name
       symbol.name = re.sub(r'\(.*\)', '', symbol.full_name)
 
@@ skipping 234 matching lines @@
   with _OpenMaybeGz(map_path) as map_file:
     section_sizes, raw_symbols = (
         linker_map_parser.MapFileParser().Parse(map_file))
 
   if not no_source_paths:
     logging.info('Extracting source paths from .ninja files')
     all_found = _ExtractSourcePaths(raw_symbols, lazy_paths.output_directory)
     assert all_found, (
         'One or more source file paths could not be found. Likely caused by '
         '.ninja files being generated at a different time than the .map file.')
+
+  logging.info('Stripping linker prefixes from symbol names')
+  _StripLinkerAddedSymbolPrefixes(raw_symbols)
   # Map file for some reason doesn't unmangle all names.
   # Unmangle prints its own log statement.
   _UnmangleRemainingSymbols(raw_symbols, lazy_paths.tool_prefix)
   logging.info('Normalizing object paths')
   _NormalizeObjectPaths(raw_symbols)
   size_info = models.SizeInfo(section_sizes, raw_symbols)
 
   # Name normalization not strictly required, but makes for smaller files.
   if raw_only:
     logging.info('Normalizing symbol names')
@@ skipping 193 matching lines @@
           logging.warning('Packed section not present: %s', packed_section_name)
         else:
           size_info.section_sizes['%s (unpacked)' % packed_section_name] = (
               unstripped_section_sizes.get(packed_section_name))
 
   logging.info('Recording metadata: \n  %s',
                '\n  '.join(describe.DescribeMetadata(size_info.metadata)))
   logging.info('Saving result to %s', args.size_file)
   file_format.SaveSizeInfo(size_info, args.size_file)
   logging.info('Done')