track `if typing.TYPE_CHECKING` to warn about non runtime bindings by terencehonles · Pull Request #622 · PyCQA/pyflakes

terencehonles · 2021-03-23T20:45:27Z

When importing or defining values in if typing.TYPE_CHECKING blocks the bound names will not be available at runtime and may cause errors when used in the following way:

importtypingiftyping.TYPE_CHECKING:
frommoduleimportType# some slow import or circular referencedefmethod(value) ->Type: # the import is needed by the type checkerassertisinstance(value, Type) # this is a runtime error

This change allows pyflakes to track what names are bound for runtime use, and allows it to warn when a non runtime name is used in a runtime context.

asottile · 2021-03-23T21:49:31Z

#530 predates this and has similar goals / problems -- perhaps collaborate with that PR?

terencehonles · 2021-03-23T22:25:19Z

#530 predates this and has similar goals / problems -- perhaps collaborate with that PR?

@asottile It looks like I actually ended up writing about the same code 😅 however there are a few things I have which the other one doesn't:

The logic handles the non runtime bindings in either the body or the else of the if/else expression
The runtime attribute defaults to True (it generally will be) which needs less changes in other parts of the codebase.
The lines https://github.com/PyCQA/pyflakes/pull/530/files#diff-5df44d79406311387e0056faae7a8092d1ce9cc85c7d15420186e45a0c712283R1178-R1181 continues looking up the stack for a runtime value, but it's more likely that the non runtime value will be declared at a higher scope and therefore the code I wrote warns and aborts immediately.

It also looks like the code was waiting review for almost a year. Is there something you wanted changed from that PR? I'm willing to work with @PetterS, but I didn't realize there was an PR that old for this already opened.

asottile · 2021-03-23T23:54:12Z

ah shoot, that's probably on me -- I might've missed the last round of reviews on there and then it lapsed (and now conflicts)

PetterS · 2021-03-24T08:39:53Z

I can revive that PR if we agree that it it is likely to be merged.

This is a problem that has been observed for real code, so would be good to fix.

terencehonles · 2021-03-24T17:17:58Z

I can revive that PR if we agree that it it is likely to be merged.
This is a problem that has been observed for real code, so would be good to fix.

It should be pretty easy to xref this PR in order to see what might also need to change since I based my PR off the most recent release (I was not aware of your PR when starting). We're 99% inline with each other's implementations.

PetterS · 2021-03-25T08:01:40Z

ah shoot, that's probably on me -- I might've missed the last round of reviews on there and then it lapsed (and now conflicts)

Merged it with master again now.

terencehonles · 2023-09-27T11:52:33Z

#530 predates this and has similar goals / problems -- perhaps collaborate with that PR?

@asottile since the PR in comment above is not going to be updated by its author

#530 (comment)

any chance you'd be willing to review this PR?

As mentioned above #622 (comment) it had used negated logic compared to the other PR in order to minimize the number of changes to the code base, and I've continued to both use and update this PR for changes on the development branch.

I can seek a 2nd approver if we can move forward with this PR.

asottile

I don't think the tests cover all the cases you intend to demonstrate here

asottile · 2023-09-27T13:14:49Z

 nodeDepth = 0
 offset = None
 _in_annotation = AnnotationState.NONE
+ _in_type_check_guard = False


these are actually all incorrect -- should be assigned in __init__ since they are not class vars

This is a pretty common pattern, and as you can see it already existed here. I've moved all of these into the __init__ to match your request.

asottile · 2023-09-27T13:16:12Z

@@ -1073,12 +1078,18 @@ def handleNodeLoad(self, node, parent):
 self.report(messages.InvalidPrintSyntax, node)

 try:


this try is now way too broad -- it originally only guarded the name lookup but now has a whole bunch of unrelated code in it

I only added a little bit here, and the exception was already broader than it needed to be. I didn't realize this was a problem, and I moved all the unrelated code (including the existing code) out of the exception to satisfy this request.

asottile · 2023-09-27T13:20:02Z

 LIST = TUPLE

 def IMPORT(self, node):
+ runtime = not self._in_type_check_guard


if we're going to invert this maybe it should just be _runtime ? but then that's not a great name so maybe runtime is not a great name?

terencehonles · 2023-10-06T20:17:22Z

Thanks for the review, I'll see when I have time to address it.

When importing or defining values in ``if typing.TYPE_CHECKING`` blocks the bound names will not be available at runtime and may cause errors when used in the following way:: import typing if typing.TYPE_CHECKING: from module import Type # some slow import or circular reference def method(value) -> Type: # the import is needed by the type checker assert isinstance(value, Type) # this is a runtime error This change allows pyflakes to track what names are bound for runtime use, and allows it to warn when a non runtime name is used in a runtime context.

terencehonles · 2025-09-23T15:44:05Z

It has taken me awhile to revisit this PR in order to address the comments (rather than just to update it), but I have addressed most of the comments.

Responding to @jayvdb's comment

However as noted earlier in reviews, runtime isnt a good name. I would prefer to see a more explicit in_type_checking or similar, and have one name consistently used everywhere.

I was avoiding calling it in_type_checking since I thought there might end up being other reasons why a binding wouldn't be in the "runtime" scope and I wouldn't want to add additional flags.

Also I suspect that this feature will be a lot cleaner if TYPE_CHECKING is treated as a special scope, which automatically means items in it are not placed into the normal scope logic. Then items in that special scope can only be considered by explicitly including them, only in the contexts when they are useful.

I agree this might make sense, but it would require a bigger re-work and I believed I was following how _in_annotation was implemented.

I hope these changes are satisfactory.

terencehonles mentioned this pull request Mar 23, 2021
allow turning on run tracing from the command line #623
Closed

sigmavirus24 reviewed Mar 23, 2021
View reviewed changes

Comment threadtox.ini

terencehonles force-pushed the warn-about-using-type-checking-imports-at-runtime branch from ac90967 to a795822Compare March 23, 2021 22:18

terencehonles force-pushed the warn-about-using-type-checking-imports-at-runtime branch from a795822 to 117d5deCompare April 12, 2021 23:21

terencehonles mentioned this pull request Jun 11, 2021
Handle code guarded by 'if TYPE_CHECKING' correctly #530
Closed

terencehonles force-pushed the warn-about-using-type-checking-imports-at-runtime branch from 117d5de to 3bbd41aCompare October 18, 2021 14:43

terencehonles force-pushed the warn-about-using-type-checking-imports-at-runtime branch from 3bbd41a to 53c3070Compare August 11, 2022 11:12

terencehonles force-pushed the warn-about-using-type-checking-imports-at-runtime branch from 53c3070 to 58506b3Compare March 31, 2023 10:58

terencehonles force-pushed the warn-about-using-type-checking-imports-at-runtime branch from 58506b3 to 6dcf6a7Compare September 27, 2023 11:36

asottile reviewed Sep 27, 2023
View reviewed changes

terencehonles force-pushed the warn-about-using-type-checking-imports-at-runtime branch from 6dcf6a7 to 8c4da4aCompare February 9, 2024 09:38

terencehonles force-pushed the warn-about-using-type-checking-imports-at-runtime branch 3 times, most recently from 0480c35 to 0039d2bCompare September 23, 2025 15:31

terencehonles force-pushed the warn-about-using-type-checking-imports-at-runtime branch from 0039d2b to faab0f8Compare September 23, 2025 15:33

		@@ -1073,12 +1078,18 @@ def handleNodeLoad(self, node, parent):
		self.report(messages.InvalidPrintSyntax, node)

		try:

Uh oh!

Conversation

terencehonles commented Mar 23, 2021

Uh oh!

Uh oh!

asottile commented Mar 23, 2021

Uh oh!

terencehonles commented Mar 23, 2021• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

asottile commented Mar 23, 2021

Uh oh!

PetterS commented Mar 24, 2021• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

terencehonles commented Mar 24, 2021

Uh oh!

PetterS commented Mar 25, 2021

Uh oh!

terencehonles commented Sep 27, 2023

Uh oh!

asottile left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

asottileSep 27, 2023

Choose a reason for hiding this comment

Uh oh!

terencehonlesSep 23, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

asottileSep 27, 2023

Choose a reason for hiding this comment

Uh oh!

terencehonlesSep 23, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

asottileSep 27, 2023

Choose a reason for hiding this comment

Uh oh!

terencehonles commented Oct 6, 2023

Uh oh!

terencehonles commented Sep 23, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

terencehonles commented Mar 23, 2021•
edited
Loading

PetterS commented Mar 24, 2021•
edited
Loading