Emit INLINE and INLINE_ORIGIN directives

Question

Emit INLINE and INLINE_ORIGIN directives

gabrielesvelto opened this issue 3 years ago · comments

Breakpad added support for inlined functions in revision 4f5b8147, we should support it too. Note that in breakpad's dump_syms INLINE and INLINE_ORIGIN directives are emitted conditionally because they increase the size of the output significantly (they use the -d flag for it).

Markus Stange · Answer 1 · Sat Sep 18 2021 01:01:17 GMT+0800 (China Standard Time)

If I wanted to implement this today, what would be the blockers? Which downstream consumers of .sym files exist and which of them would we need to make compatible with the new format first (e.g. by teaching them to ignore the new lines)?

Markus Stange · Answer 2 · Sat Sep 18 2021 04:04:01 GMT+0800 (China Standard Time)

This is the list:

The current stackwalker, mozilla-services/minidump-stackwalk#49 (Edit: This is now unused.)
The new stackwalker, rust-minidump/rust-minidump#239
Tecken, https://bugzilla.mozilla.org/show_bug.cgi?id=1731368
Eliot, https://bugzilla.mozilla.org/show_bug.cgi?id=1731367

Once these workarounds are in place, this issue will be unblocked.

To actually make use of the data, we have getsentry/symbolic#432 and rust-minidump/rust-minidump#239.

Will Kahn-Greene · Answer 3 · Tue Sep 21 2021 00:56:38 GMT+0800 (China Standard Time)

The Tecken bug (1731368) is straight-forward: ignore all the INLINE lines. (Edit: turned out it already ignores INLINE lines--so we'rd good here.)

The other two are more complicated because we effectively have to convert from the sym-with-inlines to sym-without-inlines format to maintain the "we've got the line at the outermost level" thing.

I can fix Eliot, but probably won't get to it until after I've got it in prod.

I don't think I can fix the current stackwalker any time soon. I'm hoping that we can switch to rust-minidump and then not do it. But that means we'd have to implement ignoring INLINE lines in rust-minidump until it's supported in symbolic.

Gabriele Svelto · Answer 4 · Tue Sep 21 2021 15:20:03 GMT+0800 (China Standard Time)

If we implement support here behind a flag like Google did we can start testing it without having to immediately introduce the functionality in our crash reporting pipeline.

Gabriele Svelto · Answer 5 · Fri Oct 22 2021 17:02:57 GMT+0800 (China Standard Time)

Note: upstream breakpad just changed the format again. I had the feeling that the feature had been rushed and this somewhat confirms my feeling. I'll update the description of the format above but let's wait for this to cook up a bit before we jump on it.

Gabriele Svelto · Answer 6 · Fri Oct 22 2021 17:11:41 GMT+0800 (China Standard Time)

Alright, I've updated getsentry/symbolic#432 to reflect the changes

Markus Stange · Answer 7 · Sat Jun 25 2022 02:09:23 GMT+0800 (China Standard Time)

I've started working on this.

Will Kahn-Greene · Answer 8 · Sat Jun 25 2022 02:10:37 GMT+0800 (China Standard Time)

We can't use this in production until we know that all the downstream users of sym files can handle these lines.

Markus Stange · Answer 9 · Sat Jun 25 2022 02:11:33 GMT+0800 (China Standard Time)

It will be hidden behind a non-default --inlines command line argument.

We can't know that all downstream users of sym files can handle these lines until we have example .sym files with them.

Will Kahn-Greene · Answer 10 · Sat Jun 25 2022 02:39:27 GMT+0800 (China Standard Time)

That works for me! I just panicked because you said "I've started working on this." and I had no idea what that entailed I know I've got downstream stuff that will die and I can't get to fixing it any time soon.