Regression: dict | None or dict fallback breaks type inference for nested dicts after PR #19249

Bug Report

After upgrading mypy past PR #19249, we observed new errors related to dictionary fallback using or expressions when value types are unions of nested dicts and optional types.

This appears to be a regression in type inference, as the same code passed in previous versions of mypy.

To Reproduce

from typing import Union, Optional, Any

P = dict[str, Union[Optional[str], dict[str, Optional[str]]]]

def f(x: P) -> None:
    pass

def g(x: dict[str, Any] | None, s: str | None) -> None:
    f(x or {'x': s})  # ❌ Error in master branch, passes in latest mypy release (1.17.0)

mypy playground Gist: https://mypy-play.net/?mypy=master&python=3.12&gist=c4a5711c815d342425983850913896fe

Actual Behavior

main.py:9: error: Argument 1 to "f" has incompatible type "dict[str, Any] | dict[str, str | None]"; expected "dict[str, str | dict[str, str | None] | None]"  [arg-type]
Found 1 error in 1 file (checked 1 source file)

Your Environment

Mypy version used: mypy 1.18.0+dev.abb1cc787455501e43f344c8af2de7d0d262e5ef
Mypy command-line flags:
Mypy configuration options from mypy.ini (and other config files):
Python version used: 3.11

Thanks for bisecting!

This only used to work because of Any, {'x': s} was inferred as dict[str, Any] to match the LHS (so not an inference regression, strictly speaking, but still somewhat annoying).

OK, the context is applied correctly. The problem is that we're failing to apply our "ignore trivial (only Any) constraints when or-ing them" fallback when constraints for multiple targets coexist.

We're trying to or the following constraints:

[
    [1 <: builtins.str, 1 :> builtins.str, 2 <: Any, 2 :> Any], 
    None, 
    [1 <: builtins.str, 1 :> builtins.str, 2 <: builtins.str | None | builtins.dict[builtins.str, builtins.str | None], 2 :> builtins.str | None | builtins.dict[builtins.str, builtins.str | None]],
]

None is ignored. We can or subsets of those constraints for every variable, but can't or whole groups.

mypy/mypy/constraints.py

Lines 539 to 544 in d528707

    
           # All options have same structure. In this case we can merge-in trivial 
        
           # options (i.e. those that only have Any) and try again. 
        
           # TODO: More generally, if a given (variable, direction) pair appears in 
        
           # every option, combine the bounds with meet/join always, not just for Any. 
        
           trivial_options = select_trivial(valid_options) 
        
           if trivial_options and len(trivial_options) < len(valid_options):

I'll try to tackle that TODO sooner or later (it should be enough to resolve this ticket), but feel free to help:)

@ilevkivskyi

@ilevkivskyi Would you have time to have a look at this, since this is potentially blocking the next release?

I can't give any promises, but I will try to look at this in the next few days.

Okay, let's see the primer first, but it looks like I was overcomplicating things again.

@sterliakov

@sterliakov Is my help still needed here, or you are working on this? (I see there are some problems in the PR for type variables with values.)

@ilevkivskyi

@ilevkivskyi I'm looking into them, yes:)

Do not use outer context for `or` expr inference if the LHS has Any (p…

Navigation Menu

Repository navigation

Regression: dict | None or dict fallback breaks type inference for nested dicts after PR #19249 #19492

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

	# All options have same structure. In this case we can merge-in trivial
	# options (i.e. those that only have Any) and try again.
	# TODO: More generally, if a given (variable, direction) pair appears in
	# every option, combine the bounds with meet/join always, not just for Any.
	trivial_options = select_trivial(valid_options)
	if trivial_options and len(trivial_options) < len(valid_options):

Regression: dict | None or dict fallback breaks type inference for nested dicts after PR #19249 #19492

Description

Activity

sterliakov commented on Jul 26, 2025

sterliakov commented on Jul 26, 2025

JukkaL commented on Aug 19, 2025

ilevkivskyi commented on Aug 19, 2025

sterliakov commented on Aug 20, 2025

ilevkivskyi commented on Aug 21, 2025

sterliakov commented on Aug 22, 2025

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions