Passing dataclass default_factory as positional argument causes internal error.

[jonathanslenders]

Crash Report

The detaclasses field is missing a default_factory keyword argument. I'm passing it as a positional argument.

Traceback

Traceback (most recent call last):
  File "/home/jonathan/.pyenv/versions/lazy-maestro-3.9/bin/mypy", line 33, in <module>
    sys.exit(load_entry_point('mypy', 'console_scripts', 'mypy')())
  File "/home/jonathan/git/mypy/mypy/__main__.py", line 11, in console_entry
    main(None, sys.stdout, sys.stderr)
  File "/home/jonathan/git/mypy/mypy/main.py", line 98, in main
    res = build.build(sources, options, None, flush_errors, fscache, stdout, stderr)
  File "/home/jonathan/git/mypy/mypy/build.py", line 179, in build
    result = _build(
  File "/home/jonathan/git/mypy/mypy/build.py", line 253, in _build
    graph = dispatch(sources, manager, stdout)
  File "/home/jonathan/git/mypy/mypy/build.py", line 2688, in dispatch
    process_graph(graph, manager)
  File "/home/jonathan/git/mypy/mypy/build.py", line 3012, in process_graph
    process_stale_scc(graph, scc, manager)
  File "/home/jonathan/git/mypy/mypy/build.py", line 3104, in process_stale_scc
    mypy.semanal_main.semantic_analysis_for_scc(graph, scc, manager.errors)
  File "/home/jonathan/git/mypy/mypy/semanal_main.py", line 78, in semantic_analysis_for_scc
    process_top_levels(graph, scc, patches)
  File "/home/jonathan/git/mypy/mypy/semanal_main.py", line 199, in process_top_levels
    deferred, incomplete, progress = semantic_analyze_target(next_id, state,
  File "/home/jonathan/git/mypy/mypy/semanal_main.py", line 326, in semantic_analyze_target
    analyzer.refresh_partial(refresh_node,
  File "/home/jonathan/git/mypy/mypy/semanal.py", line 394, in refresh_partial
    self.refresh_top_level(node)
  File "/home/jonathan/git/mypy/mypy/semanal.py", line 405, in refresh_top_level
    self.accept(d)
  File "/home/jonathan/git/mypy/mypy/semanal.py", line 4835, in accept
    node.accept(self)
  File "/home/jonathan/git/mypy/mypy/nodes.py", line 950, in accept
    return visitor.visit_class_def(self)
  File "/home/jonathan/git/mypy/mypy/semanal.py", line 1048, in visit_class_def
    self.analyze_class(defn)
  File "/home/jonathan/git/mypy/mypy/semanal.py", line 1125, in analyze_class
    self.analyze_class_body_common(defn)
  File "/home/jonathan/git/mypy/mypy/semanal.py", line 1134, in analyze_class_body_common
    self.apply_class_plugin_hooks(defn)
  File "/home/jonathan/git/mypy/mypy/semanal.py", line 1180, in apply_class_plugin_hooks
    hook(ClassDefContext(defn, decorator, self))
  File "/home/jonathan/git/mypy/mypy/plugins/dataclasses.py", line 361, in dataclass_class_maker_callback
    transformer.transform()
  File "/home/jonathan/git/mypy/mypy/plugins/dataclasses.py", line 100, in transform
    attributes = self.collect_attributes()
  File "/home/jonathan/git/mypy/mypy/plugins/dataclasses.py", line 249, in collect_attributes
    has_field_call, field_args = _collect_field_args(stmt.rvalue)
  File "/home/jonathan/git/mypy/mypy/plugins/dataclasses.py", line 378, in _collect_field_args
    assert name is not None.)

To Reproduce

from dataclasses import dataclass, field

@dataclass
class C:
    attribute: int = field(int)

Your Environment

• Mypy version used: mypy 0.820+dev.797544d8f97c478770eb178ba966cc7b1d0e6020.dirty
• Mypy command-line flags: none
• Mypy configuration options from mypy.ini (and other config files): /
• Python version used: 3.9.2
• Operating system and version: Linux

[syheliel]

The definition of field disable the positional arguments, see PEP-3102:

def field(*,
          default: _MISSING_TYPE = MISSING,
          default_factory: _MISSING_TYPE = MISSING,
          init: bool = True,
          repr: bool = True,
          hash: Any = None,
          compare: bool = True,
          metadata: Any = None) -> Field

If you want to use default_factory, you must specify it as the keyword.

[jonathanslenders]

@syheliel: Of course, but forgetting to do so should not crash Mypy. That's what this issue is about.

[syheliel]

@jonathanslenders: I tried the following codes and doesn't cause the crash. It seems that when using decorator @dataclass, check_overload_call are not called, which leads to wrong argument's underreporting and causes the crash.

from dataclasses import dataclass, field
class A:
    a:int = field(int)
a = field(int)

result:

example.py:9: error: No overload variant of "field" matches argument type "Type[int]"
example.py:9: note: Possible overload variants:
example.py:9: note:     def [_T] field(*, default: _T, init: bool = ..., repr: bool = ..., hash: Optional[bool] = ..., compare: bool = ..., metadata: Optional[Mapping[str, Any]] = ...) -> _T
example.py:9: note:     def [_T] field(*, default_factory: Callable[[], _T], init: bool = ..., repr: bool = ..., hash: Optional[bool] = ..., compare: bool = ..., metadata: Optional[Mapping[str, Any]] = ...) -> _T
example.py:9: note:     def field(*, init: bool = ..., repr: bool = ..., hash: Optional[bool] = ..., compare: bool = ..., metadata: Optional[Mapping[str, Any]] = ...) -> Any
example.py:10: error: No overload variant of "field" matches argument type "Type[int]"
example.py:10: note: Possible overload variants:
example.py:10: note:     def [_T] field(*, default: _T, init: bool = ..., repr: bool = ..., hash: Optional[bool] = ..., compare: bool = ..., metadata: Optional[Mapping[str, Any]] = ...) -> _T
example.py:10: note:     def [_T] field(*, default_factory: Callable[[], _T], init: bool = ..., repr: bool = ..., hash: Optional[bool] = ..., compare: bool = ..., metadata: Optional[Mapping[str, Any]] = ...) -> _T
example.py:10: note:     def field(*, init: bool = ..., repr: bool = ..., hash: Optional[bool] = ..., compare: bool = ..., metadata: Optional[Mapping[str, Any]] = ...) -> Any
Found 2 errors in 1 file (checked 1 source file)

[hi-ogawa]

I was investigating this issue and I just noticed that this crash is also "fixed" by the latest PR #11137, but the error message looks a bit odd now.

Here is an example as a unit test:

[case testDataclassFieldDoesNotFailOnPositionalArguments]
# flags: --python-version 3.7
from dataclasses import dataclass, field

@dataclass
class C:
    x: int = field(0)  # E: Unpacking **kwargs in "field()" is not supported \
                       # E: No overload variant of "field" matches argument type "int" \
                       # N: Possible overload variants: \
                       # N:     def [_T] field(*, default: _T, init: bool = ..., repr: bool = ..., hash: Optional[bool] = ..., compare: bool = ..., metadata: Optional[Mapping[str, Any]] = ..., kw_only: bool = ...) -> _T \
                       # N:     def [_T] field(*, default_factory: Callable[[], _T], init: bool = ..., repr: bool = ..., hash: Optional[bool] = ..., compare: bool = ..., metadata: Optional[Mapping[str, Any]] = ..., kw_only: bool = ...) -> _T \
                       # N:     def field(*, init: bool = ..., repr: bool = ..., hash: Optional[bool] = ..., compare: bool = ..., metadata: Optional[Mapping[str, Any]] = ..., kw_only: bool = ...) -> Any
[builtins fixtures/dataclasses.pyi]

This is an obvious type error anyway, so maybe mypy devs don't have to really care about this kind of issue though.

[JelleZijlstra]

Thanks for pointing this out! It's still useful to make the error as clear as possible. cc @sobolevn in case you're interested in tweaking this code a bit more.

[sobolevn]

Yes, sure. Any suggestions for a better message? 🤔

[JelleZijlstra]

"field()" does not accept positional arguments maybe?