Walk me through what actually happens when you call a generator function.

Calling a generator function runs no body code at all; it returns a generator object. The body executes only when you call next on it, running until the first yield, which suspends the frame with its local variables intact. Each subsequent next resumes exactly where it stopped. When the body returns, the generator raises StopIteration, which a for loop catches for you.

Calling it does almost nothing. Python sees the yield keyword at compile time, marks the function as a generator, and the call just hands back a generator object with a frozen frame. Nothing in the body has run yet, which trips people up when they expect their validation at the top to fire immediately. The first next runs until it hits a yield, passes that value out, and freezes the frame with every local still alive. Call next again and execution picks up on the line after the yield. When the function falls off the end or hits a bare return, Python raises StopIteration, and a for loop treats that as the signal to stop. Under the hood a generator satisfies the iterator protocol because it has both dunder iter and dunder next, so it works anywhere an iterable is expected. I lean on this for streaming pipelines. On one project we had four transformation stages chained as generators, and no stage ever held more than a single record.