fyi: signatures with addresses are changing

[William Kahn-Greene]

(cc: stability, crash-reporting-wg)

Hi!

In bug 1895527, we opted to make a couple of fundamental changes to signature generation:

1. stop including module offsets in frame signatures

   • libxul.so@0x8ecff becomes libxul.so
   • libgtk-3.so.0@0x3ad8c6 becomes libgtk-3.so.0
   • (unloaded SnagAA64.dll@0x2e92) becomes (unloaded SnagAA64.dll)

2. collapse repeated XUL, .dll, .so, <name omitted>, and <unknown in > frames

This has a couple of consequences for signatures where we're missing module symbols:

1. signatures based on modules with missing symbols will fall into the same bucket now so we don't have to have bugs with twenty different variations of libGLESv2_adreno.so@0xNNNNN which require ongoing upkeep

2. some kinds of consecutive frames where we don't have a function symbol will now collapse down to a single frame reducing the likelihood the signature is truncated before an interesting bit

Examples of changes:

Crash id: 35af4662-3adc-4591-8322-c3a1e0240507
Original: libxul.so@0x54e5bd2 | libxul.so@0x6c6b461 | libxul.so@0x1e282d4 | libxul.so@0x1e38c88 | libxul.so@0x1c98d7e | libxul.so@0xdb8fff | libxul.so@0xd8ecff | libxul.so@0x9760a8f | libxul.so@0x1c9ee25 | libxul.so@0x9760a8f | libxul.so@0xd8ec2b | libxul.so@0xd...
New:      libxul.so | __clock_gettime_2
Same?:    False

Crash id: 787d2f23-1065-4853-bb2c-8e03c0240507
Original: libxul.so@0x4e5acd8 | libxul.so@0x7f79ba3 | _pt_root
New:      libxul.so | _pt_root
Same?:    False

Crash id: fd518273-515a-4ef5-8423-3f7950240507
Original: libxul.so@0x652b380 | libxul.so@0x652c08c | libxul.so@0x652c5c9 | libxul.so@0x67512fd | libxul.so@0xb65427 | <name omitted> | <name omitted> | <name omitted> | malloc | libxul.so@0x17a2a4a | libxul.so@0x17a2a4a | <name omitted> | <name omitted> | arena...
New:      libxul.so | <name omitted> | malloc | libxul.so | <name omitted> | arena_dalloc | <name omitted> | arena_t::RallocSmallOrLarge | <name omitted> | BaseAllocator::realloc | <name omitted> | BaseAllocator::realloc | <name omitted> | realloc | libxul.so | ...
Same?:    False
Notes:    (1)
          SigTruncate: SigTrunc: signature truncated due to length

Crash id: 4f975e81-68e6-44f3-af65-6fc740240507
Original: libxul.so@0x503dbd4 | libxul.so@0x9466eff | libxul.so@0x50a137a | libxul.so@0x947107f | libxul.so@0x56c8508 | libxul.so@0x505be2f | libxul.so@0x27b1caf | libxul.so@0x950329f | libxul.so@0x95032af | libxul.so@0x950329f | libxul.so@0x50a693e | libxul.so@...
New:      libxul.so | update_get_addr
Same?:    False

Crash id: a4ed6532-136c-44f8-936d-482a70240507
Original: libxul.so@0x6d17d80 | libxul.so@0x8958ce9 | libxul.so@0xa8d1fff | libxul.so@0x5872e33 | RedBlackTree<T>::Remove
New:      libxul.so | RedBlackTree<T>::Remove
Same?:    False

Crash id: c6db49a9-75df-4fc1-8553-2587f0240507
Original: (unloaded SnagAA64.dll@0x2e92) | LdrpCallInitRoutine
New:      (unloaded SnagAA64.dll) | LdrpCallInitRoutine
Same?:    False

Crash id: 117fdf74-8980-4ca9-904e-4ff7b0240507
Original: XUL@0x6b4e60 | XUL@0x6c54bc | XUL@0x49a8550 | XUL@0x497cf74 | XUL@0x4952820 | XUL@0x1246988 | XUL@0x1243eb4 | XUL@0x124ee20 | XUL@0x3e27b0 | XUL@0x3eab1c | XUL@0x3f9324 | XUL@0x6bbbcc | XUL@0x67d5dc | XUL@0x119b1ec | XUL@0x11aee34 | XUL@0x16b537c | XUL...
New:      XUL | content_process_main | IPC_Message_Name=PDocAccessible::Msg_Cache
Same?:    False

Crash id: cd7a425f-38cc-4649-bc8b-a36ef0240507
Original: libcryptoki.4.1.4.dylib@0x14580
New:      libcryptoki.4.1.4.dylib
Same?:    False

If you have any concerns, let us know on #crashreporting on Matrix or #obs-help on Slack.

Thanks!

/will

[Gabriele Svelto]

Hello everybody,
several new signatures have appeared due to this change but probably
more than we'd usually get. This is because we've been missing symbols
on Ubuntu & Debian for a certain number of packages (see bug 1894678
[1]). As the symbol scrapers catch up I expect several of these
signatures to go away or at least significantly drop in volume.

Gabriele

[1] Scrape Debian and Ubuntu t64 packages
https://bugzilla.mozilla.org/show_bug.cgi?id=1894678

On 09/05/24 21:57, William Kahn-Greene wrote:
> (cc: stability, crash-reporting-wg)
>
> Hi!
>

> In bug 1895527 <https://bugzilla.mozilla.org/show_bug.cgi?id=1895527>,

> we opted to make a couple of fundamental changes to signature generation:
>

> 1.

>
> stop including module offsets in frame signatures
>

> * |libxul.so@0x8ecff| becomes |libxul.so|
> * |libgtk-3.so.0@0x3ad8c6| becomes |libgtk-3.so.0|
> * |(unloaded SnagAA64.dll@0x2e92)| becomes |(unloaded SnagAA64.dll)|
> 2.
>
> collapse repeated |XUL|, |.dll|, |.so|, |<name omitted>|, and

> |<unknown in >| frames
>
> This has a couple of consequences for signatures where we're missing
> module symbols:
>

> 1.

>
> signatures based on modules with missing symbols will fall into the
> same bucket now so we don't have to have bugs with twenty different
> variations of |libGLESv2_adreno.so@0xNNNNN| which require ongoing upkeep
>

> 2.

> --
> You received this message because you are subscribed to the Google
> Groups "crash-reporting-wg" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to crash-reporting...@mozilla.com
> <mailto:crash-reporting...@mozilla.com>.
> To view this discussion on the web visit
> https://groups.google.com/a/mozilla.com/d/msgid/crash-reporting-wg/CAKnh9qgWOwQH6_69%2BY3t2qukLqQ9V%3DPh11ybHTLpo4a0n%2BK3xA%40mail.gmail.com <https://groups.google.com/a/mozilla.com/d/msgid/crash-reporting-wg/CAKnh9qgWOwQH6_69%2BY3t2qukLqQ9V%3DPh11ybHTLpo4a0n%2BK3xA%40mail.gmail.com?utm_medium=email&utm_source=footer>.
> For more options, visit https://groups.google.com/a/mozilla.com/d/optout
> <https://groups.google.com/a/mozilla.com/d/optout>.