Check dangling pointers: Add runtime check for `c_p_pointer` behaviour

[Smit-create]

I was testing pointers in lpython and used the following python code:

from ltypes import c_p_pointer, CPtr, pointer, i16, Pointer

def f():
    yq: CPtr
    yptr1: Pointer[i16[:]]
    y: i16[2]
    y[0] = 1
    y[1] = 2
    yptr1 = pointer(y)
    # This works
    print(yptr1[0], yptr1[1])
    assert yptr1[0] == 1
    assert yptr1[1] == 2
    c_p_pointer(yq, yptr1)

    # Shouldn't this work the same way as earlier?
    print(yptr1[0], yptr1[1])
f()

Running this I get:

% lpython expr.py && ./a.out
1 2
zsh: segmentation fault  ./a.out

[certik]

Can you specify what you currently get, and what is the expected behavior?

[Smit-create]

Can you specify what you currently get, and what is the expected behavior?

Yeah, updated the OP

[certik]

My understanding is that c_p_pointer assigns yq to yptr1. But yq points to random memory, so I would expect it to segfault, which it does.

[czgdp1807]

# Shouldn't this work the same way as earlier?
    print(yptr1[0], yptr1[1])

No it shouldn't. yptr1's data element now points to the same location where yq points. yptr1 is no longer linked to y. So segmentation fault or any random behaviour is completely normal here.

[Smit-create]

Understood. Thanks, @certik and @czgdp1807.

[certik]

Yes, it's an explanation why it segfaults. However, I would like to figure out how not to segfault in Debug mode. To do that, we should add runtime checks for dangling pointers (yq above), and if used (in print(yptr1[0], yptr1[1]) above), it should give a runtime stacktrace, saying that the pointer is dangling, so it cannot be used. In Release mode there would be no checks, so it could segfault (in which case the user recompiles in Debug mode and gets a nice stacktrace).

This is not urgent, and I think we might want to tackle this a little bit later, but it's something I definitely want to do.

[czgdp1807]

I see. I will look into it though at the bottom of my priority list right now. Let me know if it should be dealt with on an urgent basis. Thanks.

[certik]

This is not urgent at all. I am thinking perhaps in 6 months we can start adding these runtime checks (for bounds checking, stack overflow, dangling pointers, etc.). Right now let's get LPython into alpha and LFortran into beta.