Rare App (semi-)crash: Assertion failed when trying to add new face

adrianklein

Hi,
when I try to add a new face on an image, in like 1 in 20 cases, it results in the following error message:

I’ve tried activating the debug logs, but since it’s still scanning my photo library in the background, that is quite noisy - especially since it continues scanning, if I don’t click on any of the buttons.
As soon as I click on cancel / retry / ignore, the app shuts down.
The log ends on the images it scanned at that moment, but nothing about a reason for the crash.

I’ve also looked around and I guess I found the place in the logs, where I issued the add command:
+0ms | Request: {"version":"1.6.0","instance":"3f65191d-1773-411b-8d38-1783c13bdefe","events":[{"name":"MenuCommand","value":"context_add_face"}]}
but again, nothing hints at a reason for a crash.

I’m on version 1.6, windows x64, amd cpu and gpu, and I have a mini crashdump available. Let me know if you need that for debugging.

Andrey

adrianklein thank you for reporting it. Unfortunately, this is know issue and I am still trying to figure out how to fight it. The challenge is that it randomly happens to some users, but to majority it just don’t. According to code, this is caused by failure to allocate small chunk of memory, but what are the reasons behind that - I’m still trying to figure this out. Sorry to say, but I don’t have solution for this at this moment, but I am still working on it.

Unfortunately, debug logging is useless for this kind of errors.

adrianklein

Andrey ohh I see! Yeah low level problems are the worst …

Do you have a way to reproduce the issue? Or does it not happen for you?

What I was wondering … the app seems to continue running / scanning, as long as I don’t react to that message.
I guess they use separate threads? Is there a way to catch that failed allocation with some kind of restart / retry?

Sorry, and it itches me to find out more 🙈
I’ve tried to open my minidump in visual studio, that gave some more insights:

Exception Code: 0xc0000409
Unhandled exception at 0x00007FF8C183968D (tonfotos.node)

but not enough, so I’ve now switched the DumpType to Full. Waiting for the next crash now 😄

Andrey

adrianklein Do you have a way to reproduce the issue? Or does it not happen for you?

Unfortunately, I can’t reproduce it. And most users never face it, but some face it regularly.

adrianklein What I was wondering … the app seems to continue running / scanning, as long as I don’t react to that message.
I guess they use separate threads? Is there a way to catch that failed allocation with some kind of restart / retry?

Yes, that is a separate thread. And I assume it is OS who kills the app, so I don’t know how to catch it. Though I would not dare to assume it is not possible, I just don’t know how to do that. Try…Catch mechanism does not seem to help.

adrianklein but not enough, so I’ve now switched the DumpType to Full. Waiting for the next crash now 😄

Oh, that part I have, I know the exact line where it crashes. It is deep inside dlib (open source neural network library). It tries to allocate small chunk of memory from the heap and fails. Some other libraries would just do access violation, but this one is more sophisticated and has assert that throws. But it is not assert that is a problem, problem is this failure to allocate memory, and it is so deep that is just not possible to do any workaround or graceful handling. So this is where I stuck so far.

adrianklein

Andrey

Oh, that part I have, I know the exact line where it crashes. It is deep inside dlib (open source neural network library)

Would you share the line number with me? 😇

adrianklein

@Andrey I might have another clue 😃

I’m not totally sure if it’s really related, but it seems that the app has finally finished scanning all my images - and since then I haven’t had a single crash when manually adding faces!

Could there be some sort of racecondition when the inital scan adds a face while I manually add a face?

Andrey

adrianklein technically it could be a race condition thought I don’t really have any evidences. It only happens during face recognition, so that is totally normal that it stopped when indexing is done.