Dart generates broken debug information

[ueman]

This issue was originally reported in getsentry/sentry-dart#591

flutter build apk --split-per-abi --obfuscate --split-debug-info myfolder --release --verbose generates broken debug information since this commit 2362cd5 in Dart and since this commit flutter/flutter@f190bfb in Flutter.

I will let others reply to this but the current state of affairs is that what flutter generates is definitely a broken file. This is me summarzing what @relaxolotl found out:

The generated .dynamic section includes bad DT_HASH, DT_STRTAB and DT_SYMTAB records that point to virtual addresses which are not in a valid segment. Unfortunately our ELF library (goblin) treats these as fatal whereas some other tools can skip over this.

You can however see that llvm-readelf will also complain about this:

/usr/local/Cellar/llvm/12.0.1/bin/llvm-readelf: warning: 'app.android-arm(flutter2.5).symbols': unable to parse DT_HASH: virtual address is not in any segment: 0x17a420
/usr/local/Cellar/llvm/12.0.1/bin/llvm-readelf: warning: 'app.android-arm(flutter2.5).symbols': unable to parse DT_STRTAB: virtual address is not in any segment: 0x17a338
/usr/local/Cellar/llvm/12.0.1/bin/llvm-readelf: warning: 'app.android-arm(flutter2.5).symbols': unable to parse DT_SYMTAB: virtual address is not in any segment: 0x17a3c0

The commit that introduces the reordering is this one: dart-lang/sdk@2362cd5

Source: getsentry/sentry-dart#591 (comment)

Edit: Also see getsentry/sentry-dart#591 (comment) for even more information.

[mraleph]

@sstrickl could you take a look at this?

[sstrickl]

Looking at it now.

[sstrickl]

So, looking at an example separately written debugging file:

$ llvm-readelf-11 -SldW /tmp/save-debug-info-flag-testULIWJG/debug.so |& tee output.txt
llvm-readelf-11: warning: '/tmp/save-debug-info-flag-testULIWJG/debug.so': Unable to parse DT_HASH: virtual address is not in any segment: 0x98408
llvm-readelf-11: warning: '/tmp/save-debug-info-flag-testULIWJG/debug.so': Unable to parse DT_STRTAB: virtual address is not in any segment: 0x982f0
llvm-readelf-11: warning: '/tmp/save-debug-info-flag-testULIWJG/debug.so': Unable to parse DT_SYMTAB: virtual address is not in any segment: 0x98378
There are 15 section headers, starting at offset 0xe9dc8:

Section Headers:
  [Nr] Name              Type            Address          Off    Size   ES Flg Lk Inf Al
  [ 0]                   NULL            0000000000000000 000000 000000 00      0   0  0
  [ 1] .note.gnu.build-id NOTE           0000000000000190 000190 000020 00   A  0   0  4
  [ 2] .rodata           NOBITS          00000000000001b0 0001b0 000000 00   A  0   0 16
  [ 3] .dynstr           STRTAB          00000000000982f0 0001b0 000085 00   A  0   0  8
  [ 4] .dynsym           DYNSYM          0000000000098378 000238 000090 18   A  3   1  8
  [ 5] .hash             HASH            0000000000098408 0002c8 000038 04   A  4   0  8
  [ 6] .text             NOBITS          000000000009c000 004000 000000 00  AX  0   0 16
  [ 7] .bss              NOBITS          0000000000168000 004000 000000 00  WA  0   0  8
  [ 8] .dynamic          DYNAMIC         0000000000168018 004000 000060 10  WA  3   0  8
  [ 9] .shstrtab         STRTAB          0000000000000000 004060 000086 00      0   0  8
  [10] .debug_abbrev     PROGBITS        0000000000000000 0040e6 00003b 00      0   0  1
  [11] .debug_info       PROGBITS        0000000000000000 004121 03d1a2 00      0   0  1
  [12] .debug_line       PROGBITS        0000000000000000 0412c3 05ea0e 00      0   0  1
  [13] .strtab           STRTAB          0000000000000000 09fcd8 0342b0 00      0   0  8
  [14] .symtab           SYMTAB          0000000000000000 0d3f88 015e40 18     13 3731  8
Key to Flags:
  W (write), A (alloc), X (execute), M (merge), S (strings), I (info),
  L (link order), O (extra OS processing required), G (group), T (TLS),
  C (compressed), x (unknown), o (OS specific), E (exclude),
  l (large), p (processor specific)

Elf file type is DYN (Shared object file)
Entry point 0x0
There are 6 program headers, starting at offset 64

Program Headers:
  Type           Offset   VirtAddr           PhysAddr           FileSiz  MemSiz   Flg Align
  PHDR           0x000040 0x0000000000000040 0x0000000000000040 0x000150 0x000150 R   0x8
  LOAD           0x000000 0x0000000000000000 0x0000000000000000 0x000300 0x098440 R   0x4000
  LOAD           0x004000 0x000000000009c000 0x000000000009c000 0x000000 0x0cb6c0 R E 0x4000
  LOAD           0x004000 0x0000000000168000 0x0000000000168000 0x000060 0x000078 RW  0x4000
  NOTE           0x000190 0x0000000000000190 0x0000000000000190 0x000020 0x000020 R   0x4
  DYNAMIC        0x004000 0x0000000000168018 0x0000000000168018 0x000060 0x000060 RW  0x8

 Section to Segment mapping:
  Segment Sections...
   00     
   01     .note.gnu.build-id .rodata .dynstr .dynsym .hash 
   02     .text 
   03     .bss .dynamic 
   04     .note.gnu.build-id 
   05     .dynamic 
   None   .shstrtab .debug_abbrev .debug_info .debug_line .strtab .symtab 
Dynamic section at offset 0x4000 contains 6 entries:
  Tag                Type     Name/Value
  0x0000000000000004 (HASH)   0x98408
  0x0000000000000005 (STRTAB) 0x982f0
  0x000000000000000a (STRSZ)  133 (bytes)
  0x0000000000000006 (SYMTAB) 0x98378
  0x000000000000000b (SYMENT) 24 (bytes)
  0x0000000000000000 (NULL)   0x0

While it looks like there's a segment with those addresses in there, the segment starts with the NOBITS .rodata section, and so the later non-NOBITS sections in the segment have addresses which don't correspond to the loaded file contents for the segment. Will have a CL fixing this shortly.

[sstrickl]

Fix is up for review in CL 214800.

[ueman]

Does this qualify for a hotfix release?

[mraleph]

@ueman yes, we will request a cherry-pick

[dnfield]

/cc @devoncarew

[devoncarew]

If this has high impact - something we want to try and cherry pick sooner rather than later - we'll want a cp request for this asap (see the issue template at the bottom of https://github.com/dart-lang/sdk/wiki/Cherry-picks-to-a-release-channel).