Ideally .size files would contain entries that cover every byte that makes up Chrome.apk. This bug covers tracking the bytes that go into classes.dex.

I'm currently thinking that the easiest way to get this information is:
Parse dexdump -d:
  - Gives addresses of methods
  - Tells you where strings are used
  - Doesn't really capture all the bytes of a .dex though

We'll need to slurp in proguard .mapping file to record unobfuscated names.

1) We should start by storing one symbol for each method.
2) Then add in symbols for strings (where each usage of a string is a symbol, and identical strings get grouped as aliases)
3) Then think about added entries for types (e.g. per-class overhead not covered by methods / strings)


Question 2: What exactly should we store:
 name = "org.chromium.Foo.barMethod"
 full_name = "org.chromium.Foo.barMethod(int a)"
 size = size_of_instructions
 section_name = "dex.method"
 object_path = "classes.dex"
 source_path = "path/to/file.java"

 name = "org.chromium.Foo.barMethod$str1"
 full_name = "org.chromium.Foo.field(int a)$str1"
 size = size_of_string
 section_name = "dex.class"
 object_path = "classes.dex"
 source_path = "path/to/file.java"
 aliases = [other_symbols_referencing_string]

 name = "org.chromium.Foo"
 full_name = "org.chromium.Foo"
 size = size_of_type_overhead
 section_name = "dex.class"
 object_path = "classes.dex"
 source_path = "path/to/file.java"

Note: there is already some code for doing deobfuscation of .mapping files at:
https://cs.chromium.org/chromium/src/tools/android/dexdiffer/dexdiffer.py
 

Ideally .size files would contain entries that cover every byte that makes up Chrome.apk. This bug covers tracking the bytes that go into classes.dex.

I'm currently thinking that the easiest way to get this information is:
Parse dexdump -d:
  - Gives addresses of methods
  - Tells you where strings are used
  - Doesn't really capture all the bytes of a .dex though

We'll need to slurp in proguard .mapping file to record unobfuscated names.

1) We should start by storing one symbol for each method (or class?).
2) Then add in symbols for strings (where each usage of a string is a symbol, and identical strings get grouped as aliases)
3) Then think about added entries for types (e.g. per-class overhead not covered by methods / strings)


Question 2: What exactly should we store:
 name = "org.chromium.Foo.barMethod"
 full_name = "org.chromium.Foo.barMethod(int a)"
 size = size_of_instructions
 section_name = ".dex.method"
 object_path = "path/to/target_java.jar"
 source_path = "path/to/file.java"

 name = "org.chromium.Foo.barMethod$str1"
 full_name = "org.chromium.Foo.field(int a)$str1"
 size = size_of_string
 section_name = ".dex.str"
 object_path = "path/to/target_java.jar"
 source_path = "path/to/file.java"
 aliases = [other_symbols_referencing_string]

 name = "org.chromium.Foo"
 full_name = "org.chromium.Foo"
 size = size_of_type_overhead
 section_name = ".dex.class"
 object_path = "path/to/target_java.jar"
 source_path = "path/to/file.java"

Note: there is already some code for doing deobfuscation of .mapping files at:
https://cs.chromium.org/chromium/src/tools/android/dexdiffer/dexdiffer.py