Avoid calling memcpy with NULL pointers

[pah]

According to the C/C++ standards, calling memcpy(NULL, NULL, 0) is undefined behaviour. Recent GCC versions may rely on this by optimizing NULL pointer checks more aggressively, see [1].

This patch tries to avoid calling std::memcpy with zero elements. As a side effect, explicitly return NULL when requesting an empty block from MemoryPoolAllocator::Malloc.

This may be related to #301. Opening the pull-request as requested by @miloyip.

[1] https://gcc.gnu.org/gcc-4.9/porting_to.html

[miloyip]

I think adding a branch in SetObjectRaw() and SetArrayRaw() is not a good idea.
They are only called by EndObject() and EndArray() respectively. And the parameter should never been NULL. Adding assertions should be enough.

[pah]

Are you concerned about the performance? I have not seen any difference caused by this change.
But in case of empty arrays/objects (count == 0, why wouldn't the pointers be NULL, too?

[miloyip]

If the objective is to prevent calling memcpy() when count == 0, I think it should be:

if (count) {
    data_.a.elements = (GenericValue*)allocator.Malloc(count * sizeof(GenericValue));
    std::memcpy(data_.a.elements, values, count * sizeof(GenericValue));
}
else
    data_.a.elements = 0;
data_.a.size = data_.a.capacity = count;

values is always non-null.

[pah]

Oh, you're right. values comes from the stack. Will adjust the PR tonight.

[pah]

PR updated, see 0c5c153.

[coveralls]

Coverage decreased (-0.04%) to 99.96% when pulling 0c5c153 on pah:fix/strict-memcpy into 30ace6f on miloyip:master.

[coveralls]

Coverage decreased (-0.05%) to 99.952% when pulling 0c5c153 on pah:fix/strict-memcpy into 30ace6f on miloyip:master.

[coveralls]

Coverage decreased (-0.04%) to 99.96% when pulling 0c5c153 on pah:fix/strict-memcpy into 30ace6f on miloyip:master.

[coveralls]

Coverage decreased (-0.04%) to 99.96% when pulling 0c5c153 on pah:fix/strict-memcpy into 30ace6f on miloyip:master.

[miloyip]

I added a test case and also standardize CrtAllocator::Malloc() in 0e8bbe5.

[pah]

Thanks! 👍