Should a misplaced # type: comment be an error?

[gvanrossum]

Discuss. This becomes somewhat important since we're introducing a new, faster parser in mypy that treats them as syntax errors. @ddfisher

[gvanrossum]

There's a special case that I'd like to discuss too: what if the misplaced # type: comment is part of a longer comment? Maybe something like

def foobar():
    # We can't turn on this type comment yet
    # type: () -> List[int]
    return []

[ddfisher]

I think misplaced # type: ... comments should be syntax errors. Internally, there have been several misplaced type comments that were ignored (a few mistakenly placed after a docstring, and one that had code inserted above it in a later commit), and I'd expect to see this continue to happen if we don't explicitly guard against it. It's pretty bad for these errors to be silent, because it means you think certain code is getting type checked when it really isn't. (And this can happen at a distance: e.g. you think calls to a certain function are being type checked, but they aren't anymore because some line was added before the type comment.)

I don't think the special case should be treated any differently, because I'd expect the typical case to look like this instead:

def foobar():
    # Some comment describing foobar or the type of foobar
    # type: () -> List[int]
    return []

In this case, we'd want to throw an error rather than be silent. Also, type comments are now (statically) semantically meaningful, so it doesn't feel too weird to me that they should have to be commented when used in a comment context, e.g.:

def foobar():
    # We can't turn on this type comment yet
    # # type: () -> List[int]
    return []

There's already precedent for double-comments like this:

x = 1  # type: int  # some comment about the type

In my mind, the existence of this construct furthers the feel of type comments as a distinct syntactic construct from normal comments (at least where PEP 484 type checkers are concerned).

[gvanrossum]

My worry is mainly about things that look like type comments but aren't meant to be, and which occur in portions of the code that were meant to be untyped. Admittedly I haven't found any of these. :-) I think I'm not yet comfortable with changing the PEP to require this behavior, but I'm fine with introducing it in mypy and see if anyone reports that they are inconvenienced by it (in real code).

[ddfisher]

That sounds like a good approach to me!