Did these systems really not learn to recognize a run of four spaces as a single token, despite all the Python source code fed to them?

For one tokenizer, a couple years ago.