Infinite loop in osmium::memory::Buffer::reserve_space

[botanegg]

I catch an inf loop by simple code using Buffer(size_t capacity, auto_grow auto_grow = auto_grow::yes) ctor

    // reproducing code
    osmium::memory::Buffer node_buffer{0, osmium::memory::Buffer::auto_grow::yes};
    osmium::builder::NodeBuilder builder{node_buffer};

Buffer(size_t capacity, auto_grow auto_grow = auto_grow::yes) uses new unsigned char[capacity] and set m_memory to non-null (that is "valid buffer"), however m_capacity is 0;
Because m_capacity is 0 reserve_space is infinite:

                        size_t new_capacity = m_capacity * 2;
                        while (m_written + size > new_capacity) {
                            new_capacity *= 2;
                        }

Simple fixing m_memory(capacity ? new unsigned char[capacity] : nullptr) leads to assertion

[joto]

Looks like I didn't expect anybody to choose a zero capacity. This is fixed in 7685449 by using a minimum capacity for the buffer. If you don't want to switch, you can use a non-zero capacity yourself.

[botanegg]

Were there any reason to use a growth factor = 2 rather than 1.5 as an example here FBVector.md?

[joto]

@botanegg I didn't think too much about it. The reasoning in FBVector.md is interesting, but I don't think it is particularly important. You don't just allocate one vector/buffer and keep it around a long time, but a typical program will have many of them being allocated and deallocated, etc. So memory can just be reused for something else. Maybe somebody wants to benchmark this?