Handle the effect of kw_only=True in dataclass fields correctly

ChangeLog

@@ -24,6 +24,10 @@ Release date: TBA
 
   Closes PyCQA/pylint#5225
 
+* Handle the effect of ``kw_only=True`` in dataclass fields correctly.
+
+  Closes PyCQA/pylint#7623
+
 * Handle the effect of ``init=False`` in dataclass fields correctly.
 
   Closes PyCQA/pylint#7291

astroid/brain/brain_dataclasses.py

@@ -240,11 +240,12 @@ def _get_previous_field_default(node: nodes.ClassDef, name: str) -> nodes.NodeNG
     return None
 
 
-def _generate_dataclass_init(
+def _generate_dataclass_init(  # pylint: disable=too-many-locals
     node: nodes.ClassDef, assigns: list[nodes.AnnAssign], kw_only_decorated: bool
 ) -> str:
     """Return an init method for a dataclass given the targets."""
     params: list[str] = []
+    kw_only_params: list[str] = []
     assignments: list[str] = []
     assign_names: list[str] = []
 
@@ -323,7 +324,22 @@ def _generate_dataclass_init(
             if previous_default:
                 param_str += f" = {previous_default.as_string()}"
 
-        params.append(param_str)
+        # If the field is a kw_only field, we need to add it to the kw_only_params
+        # This overwrites whether or not the class is kw_only decorated
+        if is_field:
+            kw_only = [k for k in value.keywords if k.arg == "kw_only"]  # type: ignore[union-attr]
+            if kw_only:
+                if kw_only[0].value.bool_value():
+                    kw_only_params.append(param_str)
+                else:
+                    params.append(param_str)
+                continue
+        # If kw_only decorated, we need to add all parameters to the kw_only_params
+        if kw_only_decorated:
+            kw_only_params.append(param_str)
+        else:
+            params.append(param_str)
+
         if not init_var:
             assignments.append(assignment_str)
 
@@ -332,21 +348,16 @@ def _generate_dataclass_init(
     )
 
     # Construct the new init method paramter string
-    params_string = "self, "
-    if prev_pos_only:
-        params_string += prev_pos_only
-    if not kw_only_decorated:
-        params_string += ", ".join(params)
-
+    # First we do the positional only parameters, making sure to add the
+    # the self parameter and the comma to allow adding keyword only parameters
+    params_string = f"self, {prev_pos_only}{', '.join(params)}"
     if not params_string.endswith(", "):
         params_string += ", "
 
-    if prev_kw_only:
-        params_string += "*, " + prev_kw_only
-        if kw_only_decorated:
-            params_string += ", ".join(params) + ", "
-    elif kw_only_decorated:
-        params_string += "*, " + ", ".join(params) + ", "
+    # Then we add the keyword only parameters
+    if prev_kw_only or kw_only_params:
+        params_string += "*, "
+    params_string += f"{prev_kw_only}{', '.join(kw_only_params)}"
 
     assignments_string = "\n    ".join(assignments) if assignments else "pass"
     return f"def __init__({params_string}) -> None:\n    {assignments_string}"

tests/unittest_brain_dataclasses.py

@@ -826,6 +826,49 @@ class Dee(Cee):
         assert [a.name for a in dee_init.args.kwonlyargs] == []
 
 
+def test_kw_only_in_field_call() -> None:
+    """Test that keyword only fields get correctly put at the end of the __init__."""
+
+    first, second, third = astroid.extract_node(
+        """
+    from dataclasses import dataclass, field
+
+    @dataclass
+    class Parent:
+        p1: int = field(kw_only=True, default=0)
+
+    @dataclass
+    class Child(Parent):
+        c1: str
+
+    @dataclass(kw_only=True)
+    class GrandChild(Child):
+        p2: int = field(kw_only=False, default=1)
+        p3: int = field(kw_only=True, default=2)
+
+    Parent.__init__  #@
+    Child.__init__ #@
+    GrandChild.__init__ #@
+    """
+    )
+
+    first_init: bases.UnboundMethod = next(first.infer())
+    assert [a.name for a in first_init.args.args] == ["self"]
+    assert [a.name for a in first_init.args.kwonlyargs] == ["p1"]
+    assert [d.value for d in first_init.args.kw_defaults] == [0]
+
+    second_init: bases.UnboundMethod = next(second.infer())
+    assert [a.name for a in second_init.args.args] == ["self", "c1"]
+    assert [a.name for a in second_init.args.kwonlyargs] == ["p1"]
+    assert [d.value for d in second_init.args.kw_defaults] == [0]
+
+    third_init: bases.UnboundMethod = next(third.infer())
+    assert [a.name for a in third_init.args.args] == ["self", "c1", "p2"]
+    assert [a.name for a in third_init.args.kwonlyargs] == ["p1", "p3"]
+    assert [d.value for d in third_init.args.defaults] == [1]
+    assert [d.value for d in third_init.args.kw_defaults] == [0, 2]
+
+
 def test_dataclass_with_unknown_base() -> None:
     """Regression test for dataclasses with unknown base classes.