feat: inspect/enrich event in crashpad backend #843
Conversation
|
Codecov Report
Additional details and impacted files@@ Coverage Diff @@
## master #843 +/- ##
==========================================
+ Coverage 82.08% 82.57% +0.48%
==========================================
Files 52 53 +1
Lines 7141 7352 +211
Branches 1149 1185 +36
==========================================
+ Hits 5862 6071 +209
+ Misses 1181 1172 -9
- Partials 98 109 +11 |
| @@ -183,6 +191,7 @@ sentry__crashpad_handler(int signum, siginfo_t *info, ucontext_t *user_context) | |||
| } else { | |||
| SENTRY_TRACE("event was discarded"); | |||
| } | |||
| sentry_value_decref(event); | |||
There was a problem hiding this comment.
I believe there is a leak and double-free here, with the one in crashpad_backend_free
In particular, setting data->crash_event above does not decref any value that was previously there, and it transfers ownership, so it should probably incref before doing so.
There was a problem hiding this comment.
You're right, but wouldn't the preferable option be to only decref the event in the "discarded" path and leave the reference count given that we only transfer ownership from one ref to the other (and event isn't accessed any longer)? Or do you mean in case someone access event later on?
There was a problem hiding this comment.
In particular, setting
data->crash_eventabove does notdecrefany value that was previously there, and it transfers ownership, so it should probably incref before doing so.
@markushi asked the same thing in the last PR. We can defensively decref the data->crash_event, but there is no allocation happening before this at the moment.
There was a problem hiding this comment.
This is all very academic, since you are in the crash path here, so you will not ever call into this function again. Though I believe its still worth doing
There was a problem hiding this comment.
Yeah I understand. Even if the effect of fixing this is inconsequential to the execution of the crash-handler, it should still be correct as documentation for future code additions. I just got rid of the local event and only use data->crash_event in the crash-handler (which is now defensifely decrefed).
Since no one commented on my suggestion here: #840 (comment), here is the PR that implements that :P