Writing a C compiler in 500 lines of Python

In C, nothing you have not parsed yet (what it to the right) is necessary for what you've already parsed (what lies to the left). (Necessary for type checking it or translating it.)

E.g. to call a function later in the file, you need a prior declaration. Or else, an implicit one (possibly wrong) will be assumed from the call itself.

This is not true in some C++ situations, like class declarations. In a class definition there can be functions with bodies. Those are inline functions. The functons can freely refer to each other in either direction. Type checking a class declaration therefore requires all of it to be parsed.

A one pass language is advantageous even if you're building a serious multi-pass compiler with optimization. This is because that exercise doesn't require an AST! Multi-pass doesn't mean AST.

Building an AST doesn't require just more memory, but more code and development work: more complexity in the code to build the abstraction and to traverse it. It's useful if you need to manipulate or analyze the program in ways that are closely related to the source language. If the language cannot be checked in one pass, you might need one; you wouldn't want to be doing the checking on an intermediate representation, where you've lost the relationship to the code. AST building can be reused for other purposes, like code formatting, refactoring, communicating with an IDE for code completion and whatnot.

If the only thing you're going to do with an AST is walk it up and down to do some checks, and then to generate code, and you do all that in an order that could have been done without the AST (like a bottom-up, left to right traversal), then it was kind of a waste to construct it; those checks and generation could have been done as the phrase structure rules were parsed.