Hacker News Reader

Comment by rbonvall | Hacker News Reader

The problem with using void* pointers is that every access to an element costs one extra dereference operation. By using C++ templates, the compiler can create object code specifically compiled for each data type, without the need for pointers. (On the other hand, this can make C++ binaries much larger).

View on HN

In this case, I'm not storing void pointers in the data structure, but I'm using the void pointer as a direct address into "elements". The memory layout is just a sequence of items of length elem_size. vector_push copies the item into that memory (and doubles the size of "elements" with realloc when necessary).

So you have a memcpy() call in vector_push? I wonder if that's less efficient than a straight-up assignment for small objects such as integers. The C++ equivalent can be further optimized during instruction generation by keeping things in the CPU registers as long as possible. In your C case, the int has to be written to memory before memcpy can load it into its buffer.

True, but that can be overcome by another function that just gives you an address to put a new item. You would say, for example:

  *(int *)vector_expand(v) = 2;

This is a super-unclean API, but that's just IMHO.

We'll just have to agree to disagree that what you've shown above is an acceptable API for a vector module.

I'm just making it up as I go along, it's not the prettiest thing. What would you do, in C?

Reply on news.ycombinator.com