Add a constant that's False at runtime but True when type checking #230

Closed
gvanrossum opened this Issue Jun 4, 2016 · 11 comments

Projects

None yet

6 participants

@gvanrossum
Member

This would provide a standard idiom for including imports that are only necessary while type checking, see e.g. python/mypy#1646 (which currently recommends if False: import ... because that's the only way to do it).

@gvanrossum gvanrossum added this to the 3.5.2 milestone Jun 4, 2016
@gvanrossum
Member

There's one case where I think the if False: idiom may still be needed: to import the typing module itself. Some code has the constraint that it shouldn't depend on any 3rd party packages. For Python 2.7 and for 3.2--3.4 the only reasonable ways to do that are either

if False:
    from typing import ...

or

try:
    from typing import ...
except ImportError:
    ...

To survive the non-existence of the typing module, some idioms can't be used (especially type aliases, type variables, and generic classes) but many others are still available (especially # type comments for function signatures and variables, and function annotations in string quotes).

We probably should add some language to the PEP to direct type checkers to support at least one of these (which would then become the preferred way). Currently I've mostly been recommending if False: for this, because it's easiest to write.

@gnprice
gnprice commented Jun 6, 2016

This would be nice to get into the 3.5.2 release if we can.

Some name ideas:

typing.in_typechecker
typing.in_type_system
not typing.at_runtime
typing.typechecker is not None  # else 'mypy', 'pytype', 'semmle', etc.

Out of those I think I'd go for in_typechecker (or in_type_checker), but I don't have a strong view.

@gnprice
gnprice commented Jun 6, 2016

/cc @markshannon , since this came up in discussion at PyCon last week

@JukkaL
Contributor
JukkaL commented Jun 7, 2016

Here are some more ideas (with an usage example to better illustrate what it would look like):

if typing.type_checker:   # could be similar to Greg's typechecker, but with _
    from model import Person

if typing.type_check:
    from model import Person

if typing.TYPE_CHECK:   # this is probably my favorite
    from model import Person

if typing.CHECKER:
    from model import Person

if typing.STATIC:
    from model import Person

if typing.ANALYZE:
    from model import Person

if not typing.RUNNING:
    from model import Person

if not typing.RUNTIME:
    from model import Person

Also, should this be okay if I don't want to use string literals in annotations:

if TYPE_CHECK:
    from model import Person
else:
    Person = 'Person'

def show(p: Person) -> None: ...
@kirbyfan64
Contributor

IMO it really should be some kind of present tense verb to emphasize "if currently type checking".

Maybe typing.TYPE_CHECKING?

@JukkaL
Contributor
JukkaL commented Jun 7, 2016

TYPE_CHECKING sounds pretty good to me. This (along with many of the other suggestions) has the minor drawback that it implies a particular sort of tool, even though other tools such as pure type inference tools, refactoring tools or IDEs that don't do type checking can also use these imports.

@gvanrossum
Member
@markshannon

We could generali(s|z)e TYPE_CHECKING to STATIC_CHECKING, but TYPE_CHECKING is fine (I don't want to bike shed too much).

@gvanrossum
Member

@matthiaskramm any objection to TYPE_CHECKING?

--Guido (mobile)
On Jun 7, 2016 11:24 AM, "Mark Shannon" [email protected] wrote:

We could generali(s|z)e TYPE_CHECKING to STATIC_CHECKING, but
TYPE_CHECKING is fine (I don't want to bike shed too much).


You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
#230 (comment), or mute
the thread
https://github.com/notifications/unsubscribe/ACwrMhol27dTYRoZwmWAO4dekhcy3jg4ks5qJbddgaJpZM4IuLP7
.

@matthiaskramm
Contributor

I'm happy with typing.TYPE_CHECKING.

@gvanrossum
Member
@gvanrossum gvanrossum closed this in #234 Jun 8, 2016
@JukkaL JukkaL referenced this issue in python/mypy Jun 8, 2016
Open

Support typing.TYPE_CHECKING #1680

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment