Support multi-line # type: comments for functions

Some linters get mad about long lines, even comments, so we would like to be able to break long # type: comments into multiple lines. This is especially desirable for function annotations if the function has lots of arguments, since you can't shorten those using type aliases. (And claiming that's just bad code doesn't do anyone any good -- I know, but sometimes I can't afford to fix it.)

Yes. I've actually been waiting for somebody to complain about this :-)

But what should it look like? # type: on every line, or just on the first line? If the latter, we could get in trouble trying to parse a # type: comment followed by an ordinary comment. So I propose the former:

def whatever(arg1, arg2, arg3,
             arg5, arg5, arg6):
    # type: (int, int, int,
    # type:  str, str, str) -> float
    # this is not a type comment
    return (arg1 + arg2 + arg3 +
            float(arg4) + float(arg5) + float(arg6))

What about only considering the next line to be type comment if it simply isn't complete? e.g. (a, isn't finished (no closing paren), so the next line would be considered the rest. Same this for a ->, a[b, etc.

I'm not sure how to do that with Jukka's parser. (If it was Python's parser we could use the strategy for recognizing unfinished input used by code.py (based on codeop.py). But mypy's syntax for function annotations in comments is not valid Python. Also, this might mask subtle syntax errors (a line with an unclosed paren followed by a line meant as a non-type comment). I guess this should be a low priority feature. And we should carefully consider how to do it so that other tools interested in supporting this syntax can also parse it (e.g. PyCharm in 2.7 mode).

I always assumed that we'd keep track of open parens and square brackets to determine when the type annotation ends. Example:

def whatever(arg1, arg2, arg3,
             arg5, arg5, arg6):
    # type: (int, int, int,
    #        str, str, str) -> float
    # this is not a type comment
    return (arg1 + arg2 + arg3 +
            float(arg4) + float(arg5) + float(arg6))

Reporting parse errors correctly might be a little tricky. Mypy should perhaps explicitly mention in a parse error message whether the error is within an (assumed) type annotation.

However, repeating # type: would also work and would probably be a little easier to implement. I don't have a strong opinion on this, though the single # type: alternative looks a little cleaner to me.

Should we discuss this in https://github.com/ambv/typehinting as well?

Yeah, we probably need to augment PEP 484 with the comment-based function annotations anyways.

Another idea is to support optionally including argument names in comments:

def whatever(arg1, arg2, arg3,
             arg4, arg5, arg6):
    # type: (arg1: int, arg2: int, arg3: int,
    #        arg4: str, arg5: str, arg6: str) -> float
    # this is not a type comment
    return (arg1 + arg2 + arg3 +
            float(arg4) + float(arg5) + float(arg6))

This way it would be easier to map argument names to types, especially if there are a ton of arguments. The obvious tradeoff is that there would be more redundancy, but mypy could check that the argument names match, and we could have a tool that automatically inserts dummy annotations with the correct argument names.

Another idea for specifying signatures of functions with many arguments:

def whatever(
         arg1,  # type: int 
         arg2,  # type: int
         arg3,  # type: int
         arg4,  # type: str
         arg5,  # type: str
         arg6,  # type: str
         ):
    # type: (...) -> float
    # this is not a type comment (note that "..." above is literal)
    return (arg1 + arg2 + arg3 +
            float(arg4) + float(arg5) + float(arg6))

This way argument name and type would closer to each other, a bit like in the Python 3 annotation syntax. The tradeoff is that there can only be a single argument per line. For functions with only a few arguments the original annotation syntax is fine and we'll want to preserve it as the common case option.

@vlasovskikh

I'd like to hear from the PyCharm developers -- @vlasovskikh do you have any preference? How easy would any of these be to add to PyCharm?

Either form would be relatively easy to support in PyCharm.

There are a couple of minor issues with the first named-arguments-in-comments form. Our code formatter can automatically reformat the second form (with #type: ...) according to the project code style, while the first one would require a special formatter for this custom multi-line syntax. For the first form we would have update named references as a part of the rename parameter refactoring while the second form doesn't require that.

The second form seems preferable to me. My understanding is that good style generally dictates having one arg per line once you need to wrap arguments anyway, so it's a pretty natural fit.

OK, let's go with 2.

It's looking like the new parser will be in 0.3.2 now. I'll implement this after it lands (see #1353).

Support per-argument type comment syntax (#1738)

Navigation Menu

Repository navigation

Support multi-line # type: comments for functions #1102

11 remaining items

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Support multi-line # type: comments for functions #1102

Description

Activity

JukkaL commented on Jan 8, 2016

gvanrossum commented on Jan 8, 2016

refi64 commented on Jan 8, 2016

gvanrossum commented on Jan 8, 2016

JukkaL commented on Jan 8, 2016

gvanrossum commented on Jan 8, 2016

JukkaL commented on Feb 8, 2016

gvanrossum commented on Feb 8, 2016

vlasovskikh commented on Feb 9, 2016

ddfisher commented on Mar 2, 2016

gvanrossum commented on Mar 2, 2016

11 remaining items

ddfisher commented on Apr 8, 2016

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions