That's very nice work, and many interesting concepts introduced in the post (for example, arenas, length bounded strings, the Cut struct).

One caveat though:

> If the OBJ source cannot fit in memory, then the model won’t fit in memory.

I don't think that this is true: a (single precision) float textual representation is typically equal or larger than its binary representation (4 bytes), the floating point used in the renderer given later in the post. The numbers given in the cube example are unlikely to occur in real world examples, where one would probably expect more than 2 digits of decimal precision. That being said, for double precision floats, it might be true in many scenarios, but I would not make that a cardinal rule anyway.

This corner cut fits within the objective of the post, which, imho, isn't to make the most efficient program, but provide a great foundation in C to build upon.