Supersize: Include string literal prefix in name

Instead of naming string literals 'string literal', name ASCII-encodable
string literals using the first few characters of the literal (e.g.
"string contents" or "long string contents[...]").

Non-ASCII-encodable string literals are still called 'string literal',
and this change should be backwards-compatible.

Sample (filtered) output at
https://storage.googleapis.com/chrome-supersize/viewer.html?load_url=oneoffs%2Fsample-string-literals2.ndjson&byteunit=B&include=%22

This increased a sample .size filesize by 640K, about a 5% increase.

Bug: 939221
Change-Id: Iaf6a0200b1bf5ed9420f506377eb7b4d422d56cc
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1815881
Commit-Queue: Jasper Chapman-Black <jaspercb@chromium.org>
Reviewed-by: Andrew Grieve <agrieve@chromium.org>
Cr-Commit-Position: refs/heads/master@{#699097}

tools/binary_size/libsupersize/archive.py

@@ -15,6 +15,7 @@ import logging
 import os
 import posixpath
 import re
+import string
 import subprocess
 import sys
 import tempfile
@@ -34,6 +35,7 @@ import ninja_parser
 import nm
 import obj_analyzer
 import path_util
+import string_extract
 
 sys.path.insert(1, os.path.join(path_util.SRC_ROOT, 'tools', 'grit'))
 from grit.format import data_pack
@@ -159,6 +161,10 @@ def _NormalizeNames(raw_symbols):
     elif symbol.IsDex():
       symbol.full_name, symbol.template_name, symbol.name = (
           function_signature.ParseJava(full_name))
+    elif symbol.IsStringLiteral():
+      symbol.full_name = full_name
+      symbol.template_name = full_name
+      symbol.name = full_name
     elif symbol.IsNative():
       # Remove [clone] suffix, and set flag accordingly.
       # Search from left-to-right, as multiple [clone]s can exist.
@@ -844,6 +850,24 @@ def _DeduceObjectPathForSwitchTables(raw_symbols, object_paths_by_name):
         num_deduced, num_arbitrations, num_unassigned)
 
 
+def _NameStringLiterals(raw_symbols, elf_path, tool_prefix):
+  # Assign ASCII-readable string literals names like "string contents".
+  STRING_LENGTH_CUTOFF = 30
+
+  for sym, name in string_extract.ReadStringLiterals(raw_symbols, elf_path,
+                                                     tool_prefix):
+    # Newlines and tabs are used as delimiters in file_format.py
+    # At this point, names still have a terminating null byte.
+    name = name.translate(None, '\t\n').strip('\00')
+    is_printable = all(c in string.printable for c in name)
+    if not is_printable:
+      sym.full_name = models.STRING_LITERAL_NAME
+    elif len(name) > STRING_LENGTH_CUTOFF:
+      sym.full_name = '"{}[...]"'.format(name[:STRING_LENGTH_CUTOFF])
+    else:
+      sym.full_name = '"{}"'.format(name)
+
+
 def _ParseElfInfo(map_path, elf_path, tool_prefix, track_string_literals,
                   outdir_context=None, linker_name=None):
   """Adds ELF section sizes and symbols."""
@@ -956,6 +980,9 @@ def _ParseElfInfo(map_path, elf_path, tool_prefix, track_string_literals,
 
   linker_map_parser.DeduceObjectPathsFromThinMap(raw_symbols, linker_map_extras)
 
+  if elf_path:
+    _NameStringLiterals(raw_symbols, elf_path, tool_prefix)
+
   # If we have an ELF file, use its sizes as the source of truth, since some
   # sections can differ from the .map.
   return (elf_section_sizes if elf_path else map_section_sizes, raw_symbols,
@@ -1521,7 +1548,7 @@ def CreateSizeInfo(
 
   # Do not call _NormalizeNames() during archive since that method tends to need
   # tweaks over time. Calling it only when loading .size files allows for more
-  # flexability.
+  # flexibility.
   if normalize_names:
     _NormalizeNames(raw_symbols)
 

tools/binary_size/libsupersize/console.py

@@ -126,26 +126,8 @@ class _Session(object):
     elf_path = self._ElfPathForSymbol(
         size_info, tool_prefix, elf_path)
 
-    address, offset, _ = string_extract.LookupElfRodataInfo(
-        elf_path, tool_prefix)
-    adjust = offset - address
-    ret = []
-    with open(elf_path, 'rb') as f:
-      for symbol in thing:
-        if symbol.section != 'r' or (
-            not all_rodata and not symbol.IsStringLiteral()):
-          continue
-        f.seek(symbol.address + adjust)
-        data = f.read(symbol.size_without_padding)
-        # As of Oct 2017, there are ~90 symbols name .L.str(.##). These appear
-        # in the linker map file explicitly, and there doesn't seem to be a
-        # pattern as to which variables lose their kConstant name (the more
-        # common case), or which string literals don't get moved to
-        # ** merge strings (less common).
-        if symbol.IsStringLiteral() or (
-            all_rodata and data and data[-1] == '\0'):
-          ret.append((symbol, data))
-    return ret
+    return string_extract.ReadStringLiterals(
+        thing, elf_path, tool_prefix, all_rodata=all_rodata)
 
   def _DiffFunc(self, before=None, after=None, sort=True):
     """Diffs two SizeInfo objects. Returns a DeltaSizeInfo.

tools/binary_size/libsupersize/models.py

@@ -168,7 +168,6 @@ DIFF_COUNT_DELTA = [0, 0, 1, -1]
 
 STRING_LITERAL_NAME = 'string literal'
 
-
 class BaseSizeInfo(object):
   """Base class for SizeInfo and DeltaSizeInfo.
 
@@ -359,7 +358,10 @@ class BaseSymbol(object):
         self.name.endswith(']') and not self.name.endswith('[]'))
 
   def IsStringLiteral(self):
-    return self.full_name == STRING_LITERAL_NAME
+    # String literals have names like "string" or "very_long_str[...]", while
+    # non-ASCII strings are named STRING_LITERAL_NAME.
+    return self.full_name.startswith(
+        '"') or self.full_name == STRING_LITERAL_NAME
 
   # Used for diffs to know whether or not it is accurate to consider two symbols
   # with the same name as being the same.

tools/binary_size/libsupersize/string_extract.py

@@ -10,6 +10,9 @@ LookupElfRodataInfo():
 ReadFileChunks():
   Reads raw data from a file, given a list of ranges in the file.
 
+ReadStringLiterals():
+  Reads the ELF file to find the string contents of a list of string literals.
+
 ResolveStringPiecesIndirect():
   BulkForkAndCall() target: Given {path: [string addresses]} and
   [raw_string_data for each string_section]:
@@ -290,3 +293,29 @@ def ResolveStringPieces(encoded_strings_by_path, string_data):
 
   ret = _AnnotateStringData(string_data, GeneratePathAndValues())
   return [concurrent.EncodeDictOfLists(x) for x in ret]
+
+
+def ReadStringLiterals(symbols, elf_path, tool_prefix, all_rodata=False):
+  """Returns an iterable of (symbol, string) for all string literal symbols.
+
+  Args:
+    symbols: An iterable of Symbols
+    elf_path: Path to the executable containing the symbols.
+    all_rodata: Assume every symbol within .rodata that ends with a \0 is a
+         string literal.
+  """
+  address, offset, _ = LookupElfRodataInfo(elf_path, tool_prefix)
+  adjust = offset - address
+  with open(elf_path, 'rb') as f:
+    for symbol in symbols:
+      if symbol.section != 'r':
+        continue
+      f.seek(symbol.address + adjust)
+      data = f.read(symbol.size_without_padding)
+      # As of Oct 2017, there are ~90 symbols name .L.str(.##). These appear
+      # in the linker map file explicitly, and there doesn't seem to be a
+      # pattern as to which variables lose their kConstant name (the more
+      # common case), or which string literals don't get moved to
+      # ** merge strings (less common).
+      if symbol.IsStringLiteral() or (all_rodata and data and data[-1] == '\0'):
+        yield ((symbol, data))

tools/binary_size/libsupersize/testdata/Archive_Apk.golden

@@ -288,9 +288,9 @@ Section .other: has 100.0% of 39228839 bytes accounted for from 5 symbols. 0 byt
 .other@0(size_without_padding=1024,padding=0,full_name=res/drawable-v13/test.xml,object_path=,source_path=chrome/android/res/drawable/test.xml,flags={},num_aliases=1,component=)
 .other@0(size_without_padding=0,padding=764,full_name=Overhead: APK file,object_path=,source_path=,flags={},num_aliases=1,component=)
 .other@0(size_without_padding=0,padding=33984171,full_name=Overhead: ELF file,object_path=,source_path=,flags={},num_aliases=1,component=)
-.rodata@266e600(size_without_padding=5,padding=0,full_name=string literal,object_path=base/base/page_allocator.o,source_path=base/page_allocator.cc,flags={},num_aliases=2,component=Blink>Internal)
-.rodata@266e600(size_without_padding=5,padding=0,full_name=string literal,object_path=third_party/icu/icuuc/ucnv_ext.o,source_path=third_party/icu/ucnv_ext.c,flags={gen},num_aliases=2,component=Internal>Android)
-.rodata@266e605(size_without_padding=16,padding=0,full_name=string literal,object_path=third_party/icu/icuuc/ucnv_ext.o,source_path=third_party/icu/ucnv_ext.c,flags={gen},num_aliases=1,component=Internal>Android)
+.rodata@266e600(size_without_padding=5,padding=0,full_name="Str1",object_path=base/base/page_allocator.o,source_path=base/page_allocator.cc,flags={},num_aliases=2,component=Blink>Internal)
+.rodata@266e600(size_without_padding=5,padding=0,full_name="Str1",object_path=third_party/icu/icuuc/ucnv_ext.o,source_path=third_party/icu/ucnv_ext.c,flags={gen},num_aliases=2,component=Internal>Android)
+.rodata@266e605(size_without_padding=16,padding=0,full_name="String literal2",object_path=third_party/icu/icuuc/ucnv_ext.o,source_path=third_party/icu/ucnv_ext.c,flags={gen},num_aliases=1,component=Internal>Android)
 .rodata@266e630(size_without_padding=16,padding=27,full_name=** merge strings,object_path=,source_path=,flags={},num_aliases=1,component=)
 .rodata@284d600(size_without_padding=3425,padding=1961920,full_name=** merge constants,object_path=,source_path=,flags={},num_aliases=1,component=)
 .rodata@284e364(size_without_padding=0,padding=3,full_name=** symbol gap 0,object_path=,source_path=,flags={},num_aliases=1,component=)

tools/binary_size/libsupersize/testdata/Archive_Elf.golden

@@ -85,9 +85,9 @@ Section .other: has 100.0% of 33984171 bytes accounted for from 1 symbols. 0 byt
 .data.rel.ro.local@2cd84e0(size_without_padding=16,padding=16,full_name=.Lswitch.table.45,object_path=third_party/gvr-android-sdk/libgvr_shim_static_arm.a/libcontroller_api_impl.a_controller_api_impl.o,source_path=,flags={},num_aliases=1,component=)
 .data.rel.ro.local@2cd84f0(size_without_padding=8,padding=0,full_name=kSystemClassPrefixes,object_path=third_party/gvr-android-sdk/libgvr_shim_static_arm.a/libport_android_jni.a_jni_utils.o,source_path=,flags={anon},num_aliases=1,component=)
 .other@0(size_without_padding=0,padding=33984171,full_name=Overhead: ELF file,object_path=,source_path=,flags={},num_aliases=1,component=)
-.rodata@266e600(size_without_padding=5,padding=0,full_name=string literal,object_path=base/base/page_allocator.o,source_path=base/page_allocator.cc,flags={},num_aliases=2,component=Blink>Internal)
-.rodata@266e600(size_without_padding=5,padding=0,full_name=string literal,object_path=third_party/icu/icuuc/ucnv_ext.o,source_path=third_party/icu/ucnv_ext.c,flags={gen},num_aliases=2,component=Internal>Android)
-.rodata@266e605(size_without_padding=16,padding=0,full_name=string literal,object_path=third_party/icu/icuuc/ucnv_ext.o,source_path=third_party/icu/ucnv_ext.c,flags={gen},num_aliases=1,component=Internal>Android)
+.rodata@266e600(size_without_padding=5,padding=0,full_name="Str1",object_path=base/base/page_allocator.o,source_path=base/page_allocator.cc,flags={},num_aliases=2,component=Blink>Internal)
+.rodata@266e600(size_without_padding=5,padding=0,full_name="Str1",object_path=third_party/icu/icuuc/ucnv_ext.o,source_path=third_party/icu/ucnv_ext.c,flags={gen},num_aliases=2,component=Internal>Android)
+.rodata@266e605(size_without_padding=16,padding=0,full_name="String literal2",object_path=third_party/icu/icuuc/ucnv_ext.o,source_path=third_party/icu/ucnv_ext.c,flags={gen},num_aliases=1,component=Internal>Android)
 .rodata@266e630(size_without_padding=16,padding=27,full_name=** merge strings,object_path=,source_path=,flags={},num_aliases=1,component=)
 .rodata@284d600(size_without_padding=3425,padding=1961920,full_name=** merge constants,object_path=,source_path=,flags={},num_aliases=1,component=)
 .rodata@284e364(size_without_padding=0,padding=3,full_name=** symbol gap 0,object_path=,source_path=,flags={},num_aliases=1,component=)

tools/binary_size/libsupersize/testdata/Archive_MinimalApks.golden

@@ -289,9 +289,9 @@ Section .other: has 100.0% of 39228839 bytes accounted for from 5 symbols. 0 byt
 .other@0(size_without_padding=1024,padding=0,full_name=res/drawable-v13/test.xml,object_path=,source_path=chrome/android/res/drawable/test.xml,flags={},num_aliases=1,component=)
 .other@0(size_without_padding=0,padding=764,full_name=Overhead: APK file,object_path=,source_path=,flags={},num_aliases=1,component=)
 .other@0(size_without_padding=0,padding=33984171,full_name=Overhead: ELF file,object_path=,source_path=,flags={},num_aliases=1,component=)
-.rodata@266e600(size_without_padding=5,padding=0,full_name=string literal,object_path=base/base/page_allocator.o,source_path=base/page_allocator.cc,flags={},num_aliases=2,component=Blink>Internal)
-.rodata@266e600(size_without_padding=5,padding=0,full_name=string literal,object_path=third_party/icu/icuuc/ucnv_ext.o,source_path=third_party/icu/ucnv_ext.c,flags={gen},num_aliases=2,component=Internal>Android)
-.rodata@266e605(size_without_padding=16,padding=0,full_name=string literal,object_path=third_party/icu/icuuc/ucnv_ext.o,source_path=third_party/icu/ucnv_ext.c,flags={gen},num_aliases=1,component=Internal>Android)
+.rodata@266e600(size_without_padding=5,padding=0,full_name="Str1",object_path=base/base/page_allocator.o,source_path=base/page_allocator.cc,flags={},num_aliases=2,component=Blink>Internal)
+.rodata@266e600(size_without_padding=5,padding=0,full_name="Str1",object_path=third_party/icu/icuuc/ucnv_ext.o,source_path=third_party/icu/ucnv_ext.c,flags={gen},num_aliases=2,component=Internal>Android)
+.rodata@266e605(size_without_padding=16,padding=0,full_name="String literal2",object_path=third_party/icu/icuuc/ucnv_ext.o,source_path=third_party/icu/ucnv_ext.c,flags={gen},num_aliases=1,component=Internal>Android)
 .rodata@266e630(size_without_padding=16,padding=27,full_name=** merge strings,object_path=,source_path=,flags={},num_aliases=1,component=)
 .rodata@284d600(size_without_padding=3425,padding=1961920,full_name=** merge constants,object_path=,source_path=,flags={},num_aliases=1,component=)
 .rodata@284e364(size_without_padding=0,padding=3,full_name=** symbol gap 0,object_path=,source_path=,flags={},num_aliases=1,component=)

tools/binary_size/libsupersize/testdata/Archive_Pak_Files.golden

@@ -87,9 +87,9 @@ Section .other: has 100.0% of 33984171 bytes accounted for from 1 symbols. 0 byt
 .data.rel.ro.local@2cd84e0(size_without_padding=16,padding=16,full_name=.Lswitch.table.45,object_path=third_party/gvr-android-sdk/libgvr_shim_static_arm.a/libcontroller_api_impl.a_controller_api_impl.o,source_path=,flags={},num_aliases=1,component=)
 .data.rel.ro.local@2cd84f0(size_without_padding=8,padding=0,full_name=kSystemClassPrefixes,object_path=third_party/gvr-android-sdk/libgvr_shim_static_arm.a/libport_android_jni.a_jni_utils.o,source_path=,flags={anon},num_aliases=1,component=)
 .other@0(size_without_padding=0,padding=33984171,full_name=Overhead: ELF file,object_path=,source_path=,flags={},num_aliases=1,component=)
-.rodata@266e600(size_without_padding=5,padding=0,full_name=string literal,object_path=base/base/page_allocator.o,source_path=base/page_allocator.cc,flags={},num_aliases=2,component=Blink>Internal)
-.rodata@266e600(size_without_padding=5,padding=0,full_name=string literal,object_path=third_party/icu/icuuc/ucnv_ext.o,source_path=third_party/icu/ucnv_ext.c,flags={gen},num_aliases=2,component=Internal>Android)
-.rodata@266e605(size_without_padding=16,padding=0,full_name=string literal,object_path=third_party/icu/icuuc/ucnv_ext.o,source_path=third_party/icu/ucnv_ext.c,flags={gen},num_aliases=1,component=Internal>Android)
+.rodata@266e600(size_without_padding=5,padding=0,full_name="Str1",object_path=base/base/page_allocator.o,source_path=base/page_allocator.cc,flags={},num_aliases=2,component=Blink>Internal)
+.rodata@266e600(size_without_padding=5,padding=0,full_name="Str1",object_path=third_party/icu/icuuc/ucnv_ext.o,source_path=third_party/icu/ucnv_ext.c,flags={gen},num_aliases=2,component=Internal>Android)
+.rodata@266e605(size_without_padding=16,padding=0,full_name="String literal2",object_path=third_party/icu/icuuc/ucnv_ext.o,source_path=third_party/icu/ucnv_ext.c,flags={gen},num_aliases=1,component=Internal>Android)
 .rodata@266e630(size_without_padding=16,padding=27,full_name=** merge strings,object_path=,source_path=,flags={},num_aliases=1,component=)
 .rodata@284d600(size_without_padding=3425,padding=1961920,full_name=** merge constants,object_path=,source_path=,flags={},num_aliases=1,component=)
 .rodata@284e364(size_without_padding=0,padding=3,full_name=** symbol gap 0,object_path=,source_path=,flags={},num_aliases=1,component=)

tools/binary_size/libsupersize/testdata/Console.golden

@@ -60,8 +60,8 @@ Print(c[-1].GroupedByPath(depth=2).Sorted())
 
 # For even more inspiration, look at canned_queries.py
 # (and feel free to add your own!).
-0: (.rodata@266e600(size_without_padding=5,padding=0,full_name=string literal,object_path=base/base/page_allocator.o,source_path=base/page_allocator.cc,flags={},num_aliases=2,component=Blink>Internal), 'Str1\x00')
-1: (.rodata@266e605(size_without_padding=16,padding=0,full_name=string literal,object_path=third_party/icu/icuuc/ucnv_ext.o,source_path=third_party/icu/ucnv_ext.c,flags={gen},num_aliases=1,component=Internal>Android), 'String literal2\x00')
+0: (.rodata@266e600(size_without_padding=5,padding=0,full_name="Str1",object_path=base/base/page_allocator.o,source_path=base/page_allocator.cc,flags={},num_aliases=2,component=Blink>Internal), 'Str1\x00')
+1: (.rodata@266e605(size_without_padding=16,padding=0,full_name="String literal2",object_path=third_party/icu/icuuc/ucnv_ext.o,source_path=third_party/icu/ucnv_ext.c,flags={gen},num_aliases=1,component=Internal>Android), 'String literal2\x00')
 Metadata:
     elf_arch=arm
     elf_build_id=WhatAnAmazingBuildId
@@ -126,11 +126,11 @@ Index | Running Total | Section@Address | PSS | Path
 13)  34774455 (92.7%) o@0x0        33984171       {no path}
              Overhead: ELF file
 14)  34774457 (92.7%) r@0x266e600  2.5 (size=5)   base/page_allocator.cc
-             string literal (num_aliases=2)
+             "Str1" (num_aliases=2)
 15)  34774460 (92.7%) r@0x266e600  2.5 (size=5)   $root_gen_dir/third_party/icu/ucnv_ext.c
-             string literal (num_aliases=2)
+             "Str1" (num_aliases=2)
 16)  34774476 (92.7%) r@0x266e605  16             $root_gen_dir/third_party/icu/ucnv_ext.c
-             string literal
+             "String literal2"
 17)  34774519 (92.7%) r@0x266e630  43             {no path}
              ** merge strings
 18)  36739864 (98.0%) r@0x284d600  1965345        {no path}

tools/binary_size/libsupersize/testdata/Csv.golden

@@ -25,9 +25,9 @@ GroupCount,Address,SizeWithoutPadding,Padding,NumAliases,PSS,Section,Name
 ,0x2cd84e0,16,16,1,32.0,R,.Lswitch.table.45
 ,0x2cd84f0,8,0,1,8.0,R,kSystemClassPrefixes
 ,0x0,0,33984171,1,33984171.0,o,Overhead: ELF file
-,0x266e600,5,0,2,2.5,r,string literal
-,0x266e600,5,0,2,2.5,r,string literal
-,0x266e605,16,0,1,16.0,r,string literal
+,0x266e600,5,0,2,2.5,r,"""Str1"""
+,0x266e600,5,0,2,2.5,r,"""Str1"""
+,0x266e605,16,0,1,16.0,r,"""String literal2"""
 ,0x266e630,16,27,1,43.0,r,** merge strings
 ,0x284d600,3425,1961920,1,1965345.0,r,** merge constants
 ,0x284e364,0,3,1,3.0,r,** symbol gap 0

tools/binary_size/libsupersize/testdata/FullDescription.golden

@@ -163,13 +163,13 @@ Index | Running Total | Section@Address | ...
              flags={}  name=Overhead: ELF file
 14)  34774457 (92.7%) r@0x266e600  pss=2.5 (size=5)  padding=0  num_aliases=2
              source_path=base/page_allocator.cc 	object_path=base/base/page_allocator.o
-             flags={}  name=string literal
+             flags={}  name="Str1"
 15)  34774460 (92.7%) r@0x266e600  pss=2.5 (size=5)  padding=0  num_aliases=2
              source_path=third_party/icu/ucnv_ext.c 	object_path=third_party/icu/icuuc/ucnv_ext.o
-             flags={gen}  name=string literal
+             flags={gen}  name="Str1"
 16)  34774476 (92.7%) r@0x266e605  pss=16  padding=0  num_aliases=1
              source_path=third_party/icu/ucnv_ext.c 	object_path=third_party/icu/icuuc/ucnv_ext.o
-             flags={gen}  name=string literal
+             flags={gen}  name="String literal2"
 17)  34774519 (92.7%) r@0x266e630  pss=43  padding=27  num_aliases=1
              source_path= 	object_path=
              flags={}  name=** merge strings
@@ -344,13 +344,13 @@ Index | Running Total | Section@Address | ...
              flags={}  name=Overhead: ELF file
 14)  34774457 (92.7%) r@0x266e600  pss=2.5 (size=5)  padding=0  num_aliases=2
              source_path=base/page_allocator.cc 	object_path=base/base/page_allocator.o
-             flags={}  name=string literal
+             flags={}  name="Str1"
 15)  34774460 (92.7%) r@0x266e600  pss=2.5 (size=5)  padding=0  num_aliases=2
              source_path=third_party/icu/ucnv_ext.c 	object_path=third_party/icu/icuuc/ucnv_ext.o
-             flags={gen}  name=string literal
+             flags={gen}  name="Str1"
 16)  34774476 (92.7%) r@0x266e605  pss=16  padding=0  num_aliases=1
              source_path=third_party/icu/ucnv_ext.c 	object_path=third_party/icu/icuuc/ucnv_ext.o
-             flags={gen}  name=string literal
+             flags={gen}  name="String literal2"
 17)  34774519 (92.7%) r@0x266e630  pss=43  padding=27  num_aliases=1
              source_path= 	object_path=
              flags={}  name=** merge strings