Fix addCallback binding

[cnheitman]

Fix PR fixes a small memory leak in the addCallback function. This issue was spotted than to LIEF and their new nanobind implementation of its Python bindings (ref here).

[JonathanSalwan]

Mmmh, I remember adding these INC_REF due to a bad counting. I think that's why unit tests segfault =/

[cnheitman]

I'm checking the tests to understand better what's going on.

[JonathanSalwan]

Here is the commit that added these INC. Maybe it can help you. 23f5c21

[cnheitman]

I updated the PR. Now tests pass (at least locally), and I'm not getting the nanobind leak messages when using Triton with LIEF v0.14.1. It does not seem necessary to increase the refs in addCallback as the are "borrowed" refs and, in a similar way, they don't need to be decreased in removeCallback.

Let's wait until all tests run and see what happens.

[jordan9001]

These changes brought back this SEGFAULT

#1035

[JonathanSalwan]

Erf, you're right and I don't know why we didn't have this unittest... @cnheitman can you backport the patch and try another fix for the memleak?

[jordan9001]

I think it makes sense to me that we require the addition of a reference to any function object sent to addCallback (and subtraction on removeCallback).

Could the cause of the memory leak be that removeCallback isn't always being called when the context is destroyed? Do we need to add logic to TritonContext_dealloc to remove any remaining references?

[JonathanSalwan]

I've reverted #1322 and added an unit test for #1035. Now, let's see how we can fix the memory leak.

[cnheitman]

Hey! A bit late... I'll try to check this again (and how it interacts with #1035) later this week.