memory_block_stack::owns() seems wrong?

[jwdevel]

I haven't written a test to prove this yet, but just looking at the code, I see (comments added):

bool memory_block_stack::owns(const void* ptr) const noexcept
{
    auto address = static_cast<const char*>(ptr);
    for (auto cur = head_; cur; cur = cur->prev)
    {
        auto mem = static_cast<char*>(static_cast<void*>(cur));     // 'mem' is ptr to the 'node' itself (not payload area)
        if (address >= mem && address < mem + cur->usable_size)     // but we check our ptr against the payload size, not full node size!
            return true;
    }
    return false;
}

So, that calculation seems wrong to me.

As I understand it, a given node in the linked list looks like:

+----------------------+
|  node* prev          |     <- "head_" points here
+----------------------+
|  size_t usable_size  |
+----------------------+
|                      |
|  ... payload ...     |
|                      |
+----------------------+

So for instance, if we were checking ownership of the very last byte in the block, I think "owns(ptr)" would wrongly return false?

Should the condition be, instead:

    const char *usable = mem + implementation_offset();
    if (address >= usable && address < usable + cur->usable_size)
        ...

I can test further, but just asking in case there's something obvious I'm missing.

[foonathan]

Good catch, added a test that demonstrates the bug and fixed it.