File: data_quality.py

package info (click to toggle)
chromium 139.0.7258.127-1
  • links: PTS, VCS
  • area: main
  • in suites:
  • size: 6,122,068 kB
  • sloc: cpp: 35,100,771; ansic: 7,163,530; javascript: 4,103,002; python: 1,436,920; asm: 946,517; xml: 746,709; pascal: 187,653; perl: 88,691; sh: 88,436; objc: 79,953; sql: 51,488; cs: 44,583; fortran: 24,137; makefile: 22,147; tcl: 15,277; php: 13,980; yacc: 8,984; ruby: 7,485; awk: 3,720; lisp: 3,096; lex: 1,327; ada: 727; jsp: 228; sed: 36
file content (265 lines) | stat: -rw-r--r-- 10,485 bytes parent folder | download | duplicates (8)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
# Copyright 2021 The Chromium Authors
# Use of this source code is governed by a BSD-style license that can be
# found in the LICENSE file.
"""Checks that collected symbols are not missing things."""

import logging
import models
import os


class QualityCheckError(Exception):
  pass


def _Divide(a, b):
  return float(a) / b if b else 0


def CheckDataQuality(size_info, track_string_literals):
  logging.debug('Grouping symbols')
  grouped = size_info.raw_symbols.GroupedByContainerAndSectionName()
  section_sizes = size_info.section_sizes
  logging.debug('computing')
  errors = []
  for symbols in grouped:
    segment_has_error = []  # List so can be mutated from nested function.
    container = symbols[0].container
    section_name = symbols[0].section_name
    segment_size = container.section_sizes[section_name]
    is_other = section_name == models.SECTION_OTHER
    is_native = section_name in models.NATIVE_SECTIONS
    is_dex = section_name in models.DEX_SECTIONS

    logging.debug('checking section %s<%s>', section_name, container.name)

    actual_size = 0.0
    actual_padding = 0.0
    placeholder_size = 0.0
    no_name_size = 0.0
    no_source_path_size = 0.0
    no_attribution_size = 0.0
    no_component_size = 0.0
    string_literal_size = 0.0
    alias_count = 0
    generated_count = 0
    unlikely_count = 0
    startup_count = 0

    for sym in symbols:
      pss = sym.pss
      actual_size += pss
      actual_padding += sym.padding_pss
      if sym.full_name.startswith('**'):
        placeholder_size += pss
      if not sym.full_name:
        no_name_size += pss
      if not sym.source_path:
        no_source_path_size += pss
      if not (sym.full_name or sym.source_path or sym.object_path):
        no_attribution_size += pss
      if not sym.component:
        no_component_size += pss
      if sym.IsStringLiteral():
        string_literal_size += pss
      alias_count += int(bool(sym.aliases and sym is sym.aliases[0]))
      generated_count += int(bool(sym.flags & models.FLAG_GENERATED_SOURCE))
      unlikely_count += int(bool(sym.flags & models.FLAG_UNLIKELY))
      startup_count += int(bool(sym.flags & models.FLAG_STARTUP))

      if os.path.isabs(sym.source_path):
        errors.append('Abs path found in source_path: ' + repr(sym))
      if os.path.isabs(sym.object_path):
        errors.append('Abs path found in object_path: ' + repr(sym))

    def report_error(msg, *args):
      if not segment_has_error:
        segment_has_error.append(True)
        errors.append(('Error(s) found in container "{}", section "{}", '
                       'which has {} symbols totalling {} bytes: ').format(
                           container.name, section_name, len(symbols),
                           segment_size))
      full_msg = msg.format(*args)
      errors.append('    ' + full_msg)

    def report_size_error(kind, size, limit_fraction):
      report_error(
          'Abnormally high number of bytes attributed to {}: {:.0f} '
          '({:.0%}, limit was {:.0%}).', kind, size,
          _Divide(size, segment_size), limit_fraction)

    def check_size(kind, size, limit_fraction):
      limit = limit_fraction * segment_size
      if size > limit:
        report_size_error(kind, size, limit_fraction)

    def check_some_exist(kind, count, limit=1):
      if count < limit:
        report_error(
            'Expected at least {} {} to exist. '
            'Found only {} out of {} symbols.', limit, kind, count,
            len(symbols))

    if not isinstance(segment_size, int):
      report_error('Section size should be a whole number.')
      continue
    if segment_size < 1:
      report_error('Section size less than one.')
      continue
    if round(actual_size) != segment_size:
      report_error('Sum of symbols sizes do not match section size. Sum={}',
                   round(actual_size))
      continue

    check_size('padding', actual_padding, (0.05 if is_other else 0.01))

    # One bad symbol can mess up small containers.
    is_small_section = (len(symbols) < 10 or
                        _Divide(segment_size, section_sizes[section_name]) < .1)
    if not is_small_section:
      # Dex string tables show up as placeholders.
      check_size('placeholders', placeholder_size, (0.2 if is_dex else 0.01))

      check_size('symbols without names', no_name_size, 0.01)
      check_size('symbols without source paths', no_source_path_size, 0.1)
      check_size('symbols without name or path', no_attribution_size, 0.01)
      check_size('symbols without component', no_component_size, 0.20)

      if track_string_literals and section_name == models.SECTION_RODATA:
        if _Divide(string_literal_size, segment_size) < .05:
          report_error(
              'Expected more size from string literals. Found only {} ({:.1%})',
              string_literal_size, _Divide(string_literal_size, segment_size))

      if is_native:
        check_some_exist('symbol aliases', alias_count)
      if is_native or is_dex:
        check_some_exist('generated symbols', generated_count)
      if section_name == models.SECTION_TEXT:
        check_some_exist('symbols annotated by AFDO profile', unlikely_count)
        check_some_exist('static initializers', startup_count)

  if errors:
    # Cap the number of log messages.
    MAX_ERRORS = 40
    logging.error('--check-data-quality Found %d errors:', len(errors))
    for msg in errors[:MAX_ERRORS]:
      logging.error('Failed: %s', msg)
    if len(errors) > MAX_ERRORS:
      logging.error('... and %d more.', len(errors) - MAX_ERRORS)
    raise QualityCheckError()


# TODO(agrieve): Have this utilize the stats collected by CheckDataQuality().
def _DescribeSizeInfoContainerCoverage(raw_symbols, container):
  """Yields lines describing how accurate |size_info| is."""
  for section, section_name in models.SECTION_TO_SECTION_NAME.items():
    expected_size = container.section_sizes.get(section_name)
    in_section = raw_symbols.WhereInSection(section_name, container=container)
    actual_size = in_section.size

    if expected_size is None:
      yield 'Section {}: {} bytes from {} symbols.'.format(
          section_name, actual_size, len(in_section))
    else:
      size_fraction = _Divide(actual_size, expected_size)
      yield ('Section {}: has {:.1%} of {} bytes accounted for from '
             '{} symbols. {} bytes are unaccounted for.').format(
                 section_name, size_fraction, actual_size, len(in_section),
                 expected_size - actual_size)

    padding = in_section.padding
    yield '* Padding accounts for {} bytes ({:.1%})'.format(
        padding, _Divide(padding, actual_size))

    def size_msg(syms, show_padding=False):
      size = syms.size if not show_padding else syms.size_without_padding
      size_msg = 'Accounts for {} bytes ({:.1%}).'.format(
          size, _Divide(size, actual_size))
      if show_padding:
        size_msg = size_msg[:-1] + ' padding is {} bytes.'.format(syms.padding)
      return size_msg

    syms = in_section.Filter(lambda s: s.source_path)
    yield '* {} have source paths. {}'.format(len(syms), size_msg(syms))
    syms = in_section.WhereHasComponent()
    yield '* {} have a component assigned. {}'.format(len(syms), size_msg(syms))

    syms = in_section.WhereIsPlaceholder()
    if syms:
      yield '* {} placeholders exist (symbols that start with **). {}'.format(
          len(syms), size_msg(syms))

    syms = syms.Inverted().WhereHasAnyAttribution().Inverted()
    if syms:
      yield '* {} symbols have no name or path. {}'.format(
          len(syms), size_msg(syms))

    if section == 'r':
      syms = in_section.Filter(lambda s: s.IsStringLiteral())
      yield '* {} string literals exist. {}'.format(
          len(syms), size_msg(syms, show_padding=True))

    syms = in_section.Filter(lambda s: s.aliases)
    if syms:
      uniques = sum(1 for s in syms.IterUniqueSymbols())
      saved = sum(s.size_without_padding * (s.num_aliases - 1)
                  for s in syms.IterUniqueSymbols())
      yield ('* {} aliases exist, mapped to {} unique addresses '
             '({} bytes saved)').format(len(syms), uniques, saved)

    syms = in_section.WhereObjectPathMatches('{shared}')
    if syms:
      yield '* {} symbols have shared ownership. {}'.format(
          len(syms), size_msg(syms))
    else:
      yield '* 0 symbols have shared ownership.'

    for flag, desc in ((models.FLAG_HOT, 'marked as "hot"'),
                       (models.FLAG_UNLIKELY, 'marked as "unlikely"'),
                       (models.FLAG_STARTUP,
                        'marked as "startup"'), (models.FLAG_CLONE, 'clones'),
                       (models.FLAG_GENERATED_SOURCE,
                        'from generated sources')):
      syms = in_section.WhereHasFlag(flag)
      if syms:
        yield '* {} symbols are {}. {}'.format(len(syms), desc, size_msg(syms))

    spam_counter = 0
    i = 1
    count = len(in_section)
    while i < count:
      prev_sym = in_section[i - 1]
      sym = in_section[i]
      if (not sym.full_name.startswith('*')
          # Assembly symbol are iffy.
          and not prev_sym.source_path.endswith('.S') and
          not sym.source_path.endswith('.S')
          # String literal symbol creation is imperfect.
          and not prev_sym.IsStringLiteral() and not sym.IsStringLiteral()
          # Thresholds found by experimenting with arm32 Chrome.
          # E.g.: Set to 0 and see what warnings appear, then take max value.
          and ((sym.section in 'rd' and sym.padding >= 256) or
               (sym.section in 't' and sym.padding >= 64))):
        # TODO(crbug.com/40626114): We should synthesize symbols for these gaps
        #     rather than attribute them as padding.
        spam_counter += 1
        if spam_counter > 5:
          break
        yield 'Large padding of {} between:'.format(sym.padding)
        yield '  A) ' + repr(in_section[i - 1])
        yield '  B) ' + repr(sym)
      # All aliases will have the same padding.
      i += sym.num_aliases


def DescribeSizeInfoCoverage(size_info):
  for i, container in enumerate(size_info.containers):
    if i > 0:
      yield ''
    if container.name:
      yield 'Container <%s>' % container.name
    # TODO(huangs): Change to use "yield from" once linters allow this.
    for line in _DescribeSizeInfoContainerCoverage(size_info.raw_symbols,
                                                   container):
      yield line