FREEZE.unindexed

tools/binary_size/libsupersize/models.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.
 """Classes that comprise the data model for binary size analysis.
 
 The primary classes are Symbol, and SymbolGroup.
 
 Description of common properties:
   * address: The start address of the symbol.
         May be 0 (e.g. for .bss or for SymbolGroups).
   * size: The number of bytes this symbol takes up, including padding that comes
        before |address|.
+  * num_aliases: The number of symbols with the same address (including self).
+  * pss: size / num_aliases.
   * padding: The number of bytes of padding before |address| due to this symbol.
   * name: Symbol names with parameter list removed.
         Never None, but will be '' for anonymous symbols.
   * full_name: Symbols names with parameter list left in.
        Never None, but will be '' for anonymous symbols, and for symbols that do
        not contain a parameter list.
   * is_anonymous: True when the symbol exists in an anonymous namespace (which
        are removed from both full_name and name during normalization).
   * section_name: E.g. ".text", ".rodata", ".data.rel.local"
   * section: The second character of |section_name|. E.g. "t", "r", "d".
 """
 
 import collections
 import copy
+import logging
 import os
 import re
 
 import match_util
 
 
 METADATA_GIT_REVISION = 'git_revision'
 METADATA_APK_FILENAME = 'apk_file_name'  # Path relative to output_directory.
 METADATA_MAP_FILENAME = 'map_file_name'  # Path relative to output_directory.
 METADATA_ELF_ARCHITECTURE = 'elf_arch'  # "Machine" field from readelf -h
@@ skipping 115 matching lines @@
 
   Refer to module docs for field descriptions.
   """
 
   __slots__ = (
       'address',
       'full_name',
       'is_anonymous',
       'object_path',
       'name',
+      'num_aliases',
       'padding',
       'section_name',
       'source_path',
       'size',
   )
 
   def __init__(self, section_name, size_without_padding, address=None,
                name=None, source_path=None, object_path=None,
-               full_name=None, is_anonymous=False):
+               full_name=None, is_anonymous=False, num_aliases=1):
     self.section_name = section_name
     self.address = address or 0
     self.name = name or ''
     self.full_name = full_name or ''
     self.source_path = source_path or ''
     self.object_path = object_path or ''
     self.size = size_without_padding
-    # Change this to be a bitfield of flags if ever there is a need to add
-    # another similar thing.
     self.is_anonymous = is_anonymous
+    self.num_aliases = num_aliases
     self.padding = 0
 
   def __repr__(self):
     return ('%s@%x(size_without_padding=%d,padding=%d,name=%s,path=%s,anon=%d)'
             % (self.section_name, self.address, self.size_without_padding,
                self.padding, self.name, self.source_path or self.object_path,
                int(self.is_anonymous)))
 
+  @property
+  def pss(self):
+    return float(self.size) / self.num_aliases
+
+  @property
+  def pss_without_padding(self):
+    return float(self.size_without_padding) / self.num_aliases
+
 
 class SymbolGroup(BaseSymbol):
   """Represents a group of symbols using the same interface as Symbol.
 
   SymbolGroups are immutable. All filtering / sorting will return new
   SymbolGroups objects.
 
   Overrides many __functions__. E.g. the following are all valid:
   * len(group)
   * iter(group)
   * group[0]
   * group['0x1234']  # By symbol address
   * without_group2 = group1 - group2
   * unioned = group1 + group2
   """
 
   __slots__ = (
       '_padding',
       '_size',
+      '_pss',
       '_symbols',
       '_filtered_symbols',
       'full_name',
       'name',
       'section_name',
       'is_sorted',
   )
 
   def __init__(self, symbols, filtered_symbols=None, name=None,
                full_name=None, section_name=None, is_sorted=False):
     self._padding = None
     self._size = None
+    self._pss = None
     self._symbols = symbols
     self._filtered_symbols = filtered_symbols or []
     self.name = name or ''
     self.full_name = full_name
     self.section_name = section_name or '.*'
     self.is_sorted = is_sorted
 
   def __repr__(self):
     return 'Group(name=%s,count=%d,size=%d)' % (
         self.name, len(self), self.size)
@@ skipping 40 matching lines @@
 
   @property
   def is_anonymous(self):
     first = self._symbols[0].is_anonymous
     return first if all(
         s.is_anonymous == first for s in self._symbols) else False
 
   @property
   def object_path(self):
     first = self._symbols[0].object_path
-    return first if all(s.object_path == first for s in self._symbols) else None
+    return first if all(s.object_path == first for s in self._symbols) else ''
 
   @property
   def source_path(self):
     first = self._symbols[0].source_path
-    return first if all(s.source_path == first for s in self._symbols) else None
+    return first if all(s.source_path == first for s in self._symbols) else ''
+
+  def IterUniqueSymbols(self):
+    seen_addresses = set()
+    for s in self:
+      if s.num_aliases == 1:
+        yield s
+      elif s.address not in seen_addresses:
+        seen_addresses.add(s.address)
+        yield s
 
   @property
   def size(self):
     if self._size is None:
       if self.IsBss():
         self._size = sum(s.size for s in self)
-      self._size = sum(s.size for s in self if not s.IsBss())
+      else:
+        self._size = sum(s.size for s in self.IterUniqueSymbols())
     return self._size
 
   @property
+  def pss(self):
+    if self._pss is None:
+      if self.IsBss():
+        self._pss = self.size
+      else:
+        self._pss = sum(s.pss for s in self)
+    return self._pss
+
+  @property
   def padding(self):
     if self._padding is None:
-      self._padding = sum(s.padding for s in self)
+      self._padding = sum(s.padding for s in self.IterUniqueSymbols())
     return self._padding
 
+  @property
+  def num_aliases(self):
+    return 1
+
   def IsGroup(self):
     return True
 
   def _CreateTransformed(self, symbols, filtered_symbols=None, name=None,
                          section_name=None, is_sorted=None):
     if is_sorted is None:
       is_sorted = self.is_sorted
     return SymbolGroup(symbols, filtered_symbols=filtered_symbols, name=name,
                        section_name=section_name, is_sorted=is_sorted)
 
@@ skipping 13 matching lines @@
 
   def SortedByAddress(self, reverse=False):
     return self.Sorted(key=(lambda s:s.address), reverse=reverse)
 
   def SortedByCount(self, reverse=False):
     return self.Sorted(key=(lambda s:len(s) if s.IsGroup() else 1),
                        reverse=not reverse)
 
   def Filter(self, func):
     filtered_and_kept = ([], [])
-    for symbol in self:
-      filtered_and_kept[int(bool(func(symbol)))].append(symbol)
+    symbol = None
+    try:
+      for symbol in self:
+        filtered_and_kept[int(bool(func(symbol)))].append(symbol)
+    except:
+      logging.warning('Filter failed on symbol %r', symbol)
+      raise
+
     return self._CreateTransformed(filtered_and_kept[1],
                                    filtered_symbols=filtered_and_kept[0],
                                    section_name=self.section_name)
 
   def WhereBiggerThan(self, min_size):
     return self.Filter(lambda s: s.size >= min_size)
 
   def WhereInSection(self, section):
     if len(section) == 1:
       ret = self.Filter(lambda s: s.section == section)
@@ skipping 335 matching lines @@
 
 def _ExtractPrefixBeforeSeparator(string, separator, count=1):
   idx = -len(separator)
   prev_idx = None
   for _ in xrange(count):
     idx = string.find(separator, idx + len(separator))
     if idx < 0:
       break
     prev_idx = idx
   return string[:prev_idx]