Fix freeing non-allocated memory when copy shape_type

[KHLee529]

This PR fixed the bug described in Issue #437.

Bug Description

Memory crash happened when constructing a SimpleArray from np.ndarray. This was due to freeing the memory of non-allocated memory in the copy assignment operator of shape_type aka. small_vector<size_t>.

This bug only happens when the size of the shape of the source np.ndarray is larger then 3

Bug Backtrace

In File <cpp/modmesh/buffer/small_vector.hpp> line 146: delete[] m_head;
In File <cpp/modmesh/buffer/SimpleArray.hpp> line 351: m_shape = shape;
In File <cpp/modmesh/buffer/SimpleArray.hpp> line 317: SimpleArray(shape, stride, buffer)
In File <cpp/modmesh/buffer/pymod/wrap_SimpleArray.hpp> line 94: return wrapped_type(shape, stride, buffer, is_c_contiguous, is_f_contiguous);

Solution

The delete statement where the error happens are intended to remove the allocated buffer when the source vector has larger capacity than the target vector. However when deleting the array, it wasn't check whether the buffer was allocated or not. Therefore, this PR adds this check. Also I went through all the constructors and assignment operators of small_vector and confirmed that this issue doesn't happen in other place.

[ThreeMonth03]

@yungyuc Should we write the unittest to check the functionality of the small vector?

[yungyuc]

Thanks a lot for catching this! Mistake from me: 2a1c689#diff-d2ce5a8e02d44440a1f3a13051001c758fb6fdbd1bb01f2754a024b3812a657fR112

@ThreeMonth03 It is hard to make dedicated tests for errors like this, but the testing code in issue #437 should be added to unit tests. Testing high-level behaviors is sufficient for preventing regression.

[yungyuc]

@KHLee529 @ThreeMonth03 could (one of) you please add a unit test in a followup PR?

[yungyuc]

@KHLee529 Next time please use a unique branch to create PR. Using master or any existing branch for PR makes it hard to pull your change without disrupting existing code development.

(Avoid what you did in this PR:)

[KHLee529]

@yungyuc Got it. I'll make a new branch next time.
I thought both master will somehow be different when pull to local but I realized that is not the way git does.
Thank you for the advice!

[yungyuc]

No problem. Don't forget to add the unit test.