For usage, whether something is a macro or not should either be irrelevant (when they're used as a more powerful function call or decorator), or else entirely obvious from the way you use it (when they're defining a new pseudo-statement), so it doesn't make sense to emphasize the syntactic marker too much. Putting the marker gives the compiler the info and reader the info they need without being too obtrusive (and also matches the way Rust macros are used). Using "!" as a prefix operator is also effectively already claimed by IPython for shell command execution, so I'd be surprised if we ever used that spelling for anything else.

For the PEP itself, I'd like to see the bijection macro presented in the abstract to give context for the specification section. While I've had my doubts about macros in the past due to the inevitable evolution of different "dialects" of Python, I think we already hit that point years ago through sophisticated use of metaclasses and other features (Django code looks very different from NumPy code, for example). Judicious use of macros should offer opportunities to make domain specific dialects *easier* for new users to pick up, rather than harder.

For the specification section, my main question/suggestion would be around sibling macros: how about calling those "decorator macros" and making the invocation syntax "@name!" rather than the bare "name!"?

Cheers,

Nick.