| OLD | NEW |
| (Empty) | |
| 1 # Copyright 2017 The Chromium Authors. All rights reserved. |
| 2 # Use of this source code is governed by a BSD-style license that can be |
| 3 # found in the LICENSE file. |
| 4 """Classes that comprise the data model for binary size analysis.""" |
| 5 |
| 6 import collections |
| 7 import copy |
| 8 import re |
| 9 |
| 10 |
| 11 SECTION_TO_SECTION_NAME = { |
| 12 'b': '.bss', |
| 13 'd': '.data', |
| 14 'r': '.rodata', |
| 15 't': '.text', |
| 16 } |
| 17 |
| 18 |
| 19 class SizeInfo(object): |
| 20 """Represents all size information for a single binary. |
| 21 |
| 22 Fields: |
| 23 section_sizes: A dict of section_name -> size. |
| 24 symbols: A SymbolGroup (or SymbolDiff) with all symbols in it. |
| 25 """ |
| 26 __slots__ = ( |
| 27 'symbols', |
| 28 'section_sizes', |
| 29 ) |
| 30 |
| 31 """Root size information.""" |
| 32 def __init__(self, symbols, section_sizes): |
| 33 self.symbols = symbols |
| 34 self.section_sizes = section_sizes # E.g. {'.text': 0} |
| 35 |
| 36 |
| 37 class BaseSymbol(object): |
| 38 """Base class for Symbol and SymbolGroup.""" |
| 39 __slots__ = () |
| 40 |
| 41 @property |
| 42 def section(self): |
| 43 """Returns the one-letter section. |
| 44 |
| 45 E.g. If section_name == '.rodata', then section == 'r'. |
| 46 """ |
| 47 return self.section_name[1] |
| 48 |
| 49 @property |
| 50 def size_without_padding(self): |
| 51 return self.size - self.padding |
| 52 |
| 53 @property |
| 54 def end_address(self): |
| 55 return self.address + self.size_without_padding |
| 56 |
| 57 def IsBss(self): |
| 58 return self.section_name == '.bss' |
| 59 |
| 60 def IsGroup(self): |
| 61 return False |
| 62 |
| 63 def IsGenerated(self): |
| 64 # TODO(agrieve): Also match generated functions such as: |
| 65 # startup._GLOBAL__sub_I_page_allocator.cc |
| 66 return self.name.endswith(']') and not self.name.endswith('[]') |
| 67 |
| 68 def _Key(self): |
| 69 """Returns a tuple that can be used to see if two Symbol are the same. |
| 70 |
| 71 Keys are not guaranteed to be unique within a SymbolGroup. For example, it |
| 72 is common to have multiple "** merge strings" symbols, which will have a |
| 73 common key.""" |
| 74 return (self.section_name, self.function_signature or self.name) |
| 75 |
| 76 |
| 77 class Symbol(BaseSymbol): |
| 78 """Represents a single symbol within a binary.""" |
| 79 |
| 80 __slots__ = ( |
| 81 'section_name', |
| 82 'address', |
| 83 'size', |
| 84 'padding', |
| 85 'name', |
| 86 'function_signature', |
| 87 'path', |
| 88 ) |
| 89 |
| 90 def __init__(self, section_name, size_without_padding, address=None, |
| 91 name=None, path=None, function_signature=None): |
| 92 self.section_name = section_name |
| 93 self.address = address or 0 |
| 94 self.name = name or '' |
| 95 self.function_signature = function_signature or '' |
| 96 self.path = path or '' |
| 97 self.size = size_without_padding |
| 98 self.padding = 0 |
| 99 |
| 100 def __repr__(self): |
| 101 return '%s@%x(size=%d,padding=%d,name=%s,path=%s)' % ( |
| 102 self.section_name, self.address, self.size_without_padding, |
| 103 self.padding, self.name, self.path) |
| 104 |
| 105 |
| 106 class SymbolGroup(BaseSymbol): |
| 107 """Represents a group of symbols using the same interface as Symbol. |
| 108 |
| 109 SymbolGroups are immutable. All filtering / sorting will return new |
| 110 SymbolGroups objects. |
| 111 """ |
| 112 |
| 113 __slots__ = ( |
| 114 'symbols', |
| 115 'filtered_symbols', |
| 116 'name', |
| 117 'section_name', |
| 118 ) |
| 119 |
| 120 def __init__(self, symbols, filtered_symbols=None, name=None, |
| 121 section_name=None): |
| 122 self.symbols = symbols |
| 123 self.filtered_symbols = filtered_symbols or [] |
| 124 self.name = name or '' |
| 125 self.section_name = section_name or '.*' |
| 126 |
| 127 def __repr__(self): |
| 128 return 'Group(name=%s,count=%d,size=%d)' % ( |
| 129 self.name, len(self), self.size) |
| 130 |
| 131 def __iter__(self): |
| 132 return iter(self.symbols) |
| 133 |
| 134 def __len__(self): |
| 135 return len(self.symbols) |
| 136 |
| 137 def __getitem__(self, index): |
| 138 return self.symbols[index] |
| 139 |
| 140 def __sub__(self, other): |
| 141 other_ids = set(id(s) for s in other) |
| 142 new_symbols = [s for s in self if id(s) not in other_ids] |
| 143 return self._CreateTransformed(new_symbols, section_name=self.section_name) |
| 144 |
| 145 def __add__(self, other): |
| 146 self_ids = set(id(s) for s in self) |
| 147 new_symbols = self.symbols + [s for s in other if id(s) not in self_ids] |
| 148 return self._CreateTransformed(new_symbols, section_name=self.section_name) |
| 149 |
| 150 @property |
| 151 def address(self): |
| 152 return 0 |
| 153 |
| 154 @property |
| 155 def function_signature(self): |
| 156 return None |
| 157 |
| 158 @property |
| 159 def path(self): |
| 160 return None |
| 161 |
| 162 @property |
| 163 def size(self): |
| 164 if self.IsBss(): |
| 165 return sum(s.size for s in self) |
| 166 return sum(s.size for s in self if not s.IsBss()) |
| 167 |
| 168 @property |
| 169 def padding(self): |
| 170 return sum(s.padding for s in self) |
| 171 |
| 172 def IsGroup(self): |
| 173 return True |
| 174 |
| 175 def _CreateTransformed(self, symbols, filtered_symbols=None, name=None, |
| 176 section_name=None): |
| 177 return SymbolGroup(symbols, filtered_symbols=filtered_symbols, name=name, |
| 178 section_name=section_name) |
| 179 |
| 180 def Sorted(self, cmp_func=None, key=None, reverse=False): |
| 181 # Default to sorting by abs(size) then name. |
| 182 if cmp_func is None and key is None: |
| 183 cmp_func = lambda a, b: cmp((a.IsBss(), abs(b.size), a.name), |
| 184 (b.IsBss(), abs(a.size), b.name)) |
| 185 |
| 186 new_symbols = sorted(self.symbols, cmp_func, key, reverse) |
| 187 return self._CreateTransformed(new_symbols, |
| 188 filtered_symbols=self.filtered_symbols, |
| 189 section_name=self.section_name) |
| 190 |
| 191 def Filter(self, func): |
| 192 filtered_and_kept = ([], []) |
| 193 for symbol in self: |
| 194 filtered_and_kept[int(bool(func(symbol)))].append(symbol) |
| 195 return self._CreateTransformed(filtered_and_kept[1], |
| 196 filtered_symbols=filtered_and_kept[0], |
| 197 section_name=self.section_name) |
| 198 |
| 199 def WhereBiggerThan(self, min_size): |
| 200 return self.Filter(lambda s: s.size >= min_size) |
| 201 |
| 202 def WhereInSection(self, section): |
| 203 if len(section) == 1: |
| 204 ret = self.Filter(lambda s: s.section == section) |
| 205 ret.section_name = SECTION_TO_SECTION_NAME[section] |
| 206 else: |
| 207 ret = self.Filter(lambda s: s.section_name == section) |
| 208 ret.section_name = section |
| 209 return ret |
| 210 |
| 211 def WhereIsGenerated(self): |
| 212 return self.Filter(lambda s: s.IsGenerated()) |
| 213 |
| 214 def WhereNameMatches(self, pattern): |
| 215 regex = re.compile(pattern) |
| 216 return self.Filter(lambda s: regex.search(s.name)) |
| 217 |
| 218 def WherePathMatches(self, pattern): |
| 219 regex = re.compile(pattern) |
| 220 return self.Filter(lambda s: s.path and regex.search(s.path)) |
| 221 |
| 222 def WhereAddressInRange(self, start, end): |
| 223 return self.Filter(lambda s: s.address >= start and s.address <= end) |
| 224 |
| 225 def WhereHasAnyAttribution(self): |
| 226 return self.Filter(lambda s: s.name or s.path) |
| 227 |
| 228 def Inverted(self): |
| 229 return self._CreateTransformed(self.filtered_symbols, |
| 230 filtered_symbols=self.symbols) |
| 231 |
| 232 def GroupBy(self, func): |
| 233 new_syms = [] |
| 234 filtered_symbols = [] |
| 235 symbols_by_token = collections.defaultdict(list) |
| 236 for symbol in self: |
| 237 token = func(symbol) |
| 238 if not token: |
| 239 filtered_symbols.append(symbol) |
| 240 continue |
| 241 symbols_by_token[token].append(symbol) |
| 242 for token, symbols in symbols_by_token.iteritems(): |
| 243 new_syms.append(self._CreateTransformed(symbols, name=token, |
| 244 section_name=self.section_name)) |
| 245 return self._CreateTransformed(new_syms, filtered_symbols=filtered_symbols, |
| 246 section_name=self.section_name) |
| 247 |
| 248 def GroupByNamespace(self, depth=1): |
| 249 def extract_namespace(symbol): |
| 250 # Does not distinguish between classes and namespaces. |
| 251 idx = -2 |
| 252 for _ in xrange(depth): |
| 253 idx = symbol.name.find('::', idx + 2) |
| 254 if idx != -1: |
| 255 ret = symbol.name[:idx] |
| 256 if '<' not in ret: |
| 257 return ret |
| 258 return '{global}' |
| 259 return self.GroupBy(extract_namespace) |
| 260 |
| 261 def GroupByPath(self, depth=1): |
| 262 def extract_path(symbol): |
| 263 idx = -1 |
| 264 for _ in xrange(depth): |
| 265 idx = symbol.path.find('/', idx + 1) |
| 266 if idx != -1: |
| 267 return symbol.path[:idx] |
| 268 return '{path unknown}' |
| 269 return self.GroupBy(extract_path) |
| 270 |
| 271 |
| 272 class SymbolDiff(SymbolGroup): |
| 273 """A SymbolGroup subclass representing a diff of two other SymbolGroups. |
| 274 |
| 275 All Symbols contained within have a |size| which is actually the size delta. |
| 276 Additionally, metadata is kept about which symbols were added / removed / |
| 277 changed. |
| 278 """ |
| 279 __slots__ = ( |
| 280 '_added_ids', |
| 281 '_removed_ids', |
| 282 ) |
| 283 |
| 284 def __init__(self, added, removed, similar): |
| 285 self._added_ids = set(id(s) for s in added) |
| 286 self._removed_ids = set(id(s) for s in removed) |
| 287 symbols = [] |
| 288 symbols.extend(added) |
| 289 symbols.extend(removed) |
| 290 symbols.extend(similar) |
| 291 super(SymbolDiff, self).__init__(symbols) |
| 292 |
| 293 def __repr__(self): |
| 294 return '%s(%d added, %d removed, %d changed, %d unchanged, size=%d)' % ( |
| 295 'SymbolGroup', self.added_count, self.removed_count, self.changed_count, |
| 296 self.unchanged_count, self.size) |
| 297 |
| 298 def _CreateTransformed(self, symbols, filtered_symbols=None, name=None, |
| 299 section_name=None): |
| 300 ret = SymbolDiff.__new__(SymbolDiff) |
| 301 # Printing sorts, so fast-path the same symbols case. |
| 302 if len(symbols) == len(self.symbols): |
| 303 ret._added_ids = self._added_ids |
| 304 ret._removed_ids = self._removed_ids |
| 305 else: |
| 306 ret._added_ids = set(id(s) for s in symbols if self.IsAdded(s)) |
| 307 ret._removed_ids = set(id(s) for s in symbols if self.IsRemoved(s)) |
| 308 super(SymbolDiff, ret).__init__(symbols, filtered_symbols=filtered_symbols, |
| 309 name=name, section_name=section_name) |
| 310 |
| 311 return ret |
| 312 |
| 313 @property |
| 314 def added_count(self): |
| 315 return len(self._added_ids) |
| 316 |
| 317 @property |
| 318 def removed_count(self): |
| 319 return len(self._removed_ids) |
| 320 |
| 321 @property |
| 322 def changed_count(self): |
| 323 not_changed = self.unchanged_count + self.added_count + self.removed_count |
| 324 return len(self) - not_changed |
| 325 |
| 326 @property |
| 327 def unchanged_count(self): |
| 328 return sum(1 for s in self if self.IsSimilar(s) and s.size == 0) |
| 329 |
| 330 def IsAdded(self, sym): |
| 331 return id(sym) in self._added_ids |
| 332 |
| 333 def IsSimilar(self, sym): |
| 334 key = id(sym) |
| 335 return key not in self._added_ids and key not in self._removed_ids |
| 336 |
| 337 def IsRemoved(self, sym): |
| 338 return id(sym) in self._removed_ids |
| 339 |
| 340 def WhereNotUnchanged(self): |
| 341 return self.Filter(lambda s: not self.IsSimilar(s) or s.size) |
| 342 |
| 343 |
| 344 def Diff(new, old): |
| 345 """Diffs two SizeInfo or SymbolGroup objects. |
| 346 |
| 347 When diffing SizeInfos, ret.section_sizes are the result of |new| - |old|, and |
| 348 ret.symbols will be a SymbolDiff. |
| 349 |
| 350 When diffing SymbolGroups, a SymbolDiff is returned. |
| 351 |
| 352 Returns: |
| 353 Returns a SizeInfo when args are of type SizeInfo. |
| 354 Returns a SymbolDiff when args are of type SymbolGroup. |
| 355 """ |
| 356 if isinstance(new, SizeInfo): |
| 357 assert isinstance(old, SizeInfo) |
| 358 section_sizes = { |
| 359 k:new.section_sizes[k] - v for k, v in old.section_sizes.iteritems()} |
| 360 symbol_diff = Diff(new.symbols, old.symbols) |
| 361 return SizeInfo(symbol_diff, section_sizes) |
| 362 |
| 363 assert isinstance(new, SymbolGroup) and isinstance(old, SymbolGroup) |
| 364 symbols_by_key = collections.defaultdict(list) |
| 365 for s in old: |
| 366 symbols_by_key[s._Key()].append(s) |
| 367 |
| 368 added = [] |
| 369 removed = [] |
| 370 similar = [] |
| 371 # For similar symbols, padding is zeroed out. In order to not lose the |
| 372 # information entirely, store it in aggregate. |
| 373 padding_by_section_name = collections.defaultdict(int) |
| 374 for new_sym in new: |
| 375 matching_syms = symbols_by_key.get(new_sym._Key()) |
| 376 if matching_syms: |
| 377 old_sym = matching_syms.pop(0) |
| 378 # More stable/useful to compare size without padding. |
| 379 size_diff = (new_sym.size_without_padding - |
| 380 old_sym.size_without_padding) |
| 381 merged_sym = Symbol(old_sym.section_name, size_diff, |
| 382 address=old_sym.address, name=old_sym.name, |
| 383 path=old_sym.path, |
| 384 function_signature=old_sym.function_signature) |
| 385 similar.append(merged_sym) |
| 386 padding_by_section_name[new_sym.section_name] += ( |
| 387 new_sym.padding - old_sym.padding) |
| 388 else: |
| 389 added.append(new_sym) |
| 390 |
| 391 for remaining_syms in symbols_by_key.itervalues(): |
| 392 for old_sym in remaining_syms: |
| 393 duped = copy.copy(old_sym) |
| 394 duped.size = -duped.size |
| 395 duped.padding = -duped.padding |
| 396 removed.append(duped) |
| 397 |
| 398 for section_name, padding in padding_by_section_name.iteritems(): |
| 399 similar.append(Symbol(section_name, padding, |
| 400 name='** aggregate padding of delta symbols')) |
| 401 return SymbolDiff(added, removed, similar) |
| OLD | NEW |