Ensure Platform globals are freed on error #2685
Conversation
|
By default, AFAIU, the C language always auto initializes globals with zero. I don't think PR is doing anything, |
|
I'm not so sure of that; in the test failure that inspired me to make this fix, these globals were just addresses and non-NULL, which caused the test to crash when trying to free them. Perhaps it was from a previous run, but the globals should be set to NULL at the end of the run, so I don't think it is that. |
|
Globals are definitely 0 initialized on program start. So I'd be more inclined to look at a bug with a previous run not resetting everything correctly. |
|
How about try writing a test that validates the initial conditions between resets? |
|
I'm first adding some debug asserts to ensure we catch when these are not-NULL. I'll brainstorm some more on how to write a test for this. Seems spinquic might be exercising this scenario quite a bit, and is likely the best place to catch it. |
|
@anrossi there are a couple spinquic failures. Can you double check to see if they are hitting your new asserts? |
anrossi
changed the title
Description
Free platform globals in error path. Fixes #2683
Testing
SpinQuic is how this was found, and the fix was validated against the SpinQuic scenario which found it.
Documentation
N/A.