1"""Utilities related to attribute docstring extraction."""
2
3from __future__ import annotations
4
5import ast
6import inspect
7import sys
8import textwrap
9from typing import Any
10
11
12class DocstringVisitor(ast.NodeVisitor):
13 def __init__(self) -> None:
14 super().__init__()
15
16 self.target: str | None = None
17 self.attrs: dict[str, str] = {}
18 self.previous_node_type: type[ast.AST] | None = None
19
20 def visit(self, node: ast.AST) -> Any:
21 node_result = super().visit(node)
22 self.previous_node_type = type(node)
23 return node_result
24
25 def visit_AnnAssign(self, node: ast.AnnAssign) -> Any:
26 if isinstance(node.target, ast.Name):
27 self.target = node.target.id
28
29 def visit_Expr(self, node: ast.Expr) -> Any:
30 if (
31 isinstance(node.value, ast.Constant)
32 and isinstance(node.value.value, str)
33 and self.previous_node_type is ast.AnnAssign
34 ):
35 docstring = inspect.cleandoc(node.value.value)
36 if self.target:
37 self.attrs[self.target] = docstring
38 self.target = None
39
40
41def _dedent_source_lines(source: list[str]) -> str:
42 # Required for nested class definitions, e.g. in a function block
43 dedent_source = textwrap.dedent(''.join(source))
44 if dedent_source.startswith((' ', '\t')):
45 # We are in the case where there's a dedented (usually multiline) string
46 # at a lower indentation level than the class itself. We wrap our class
47 # in a function as a workaround.
48 dedent_source = f'def dedent_workaround():\n{dedent_source}'
49 return dedent_source
50
51
52def _extract_source_from_frame(cls: type[Any]) -> list[str] | None:
53 frame = inspect.currentframe()
54
55 while frame:
56 if inspect.getmodule(frame) is inspect.getmodule(cls):
57 lnum = frame.f_lineno
58 try:
59 lines, _ = inspect.findsource(frame)
60 except OSError: # pragma: no cover
61 # Source can't be retrieved (maybe because running in an interactive terminal),
62 # we don't want to error here.
63 pass
64 else:
65 block_lines = inspect.getblock(lines[lnum - 1 :])
66 dedent_source = _dedent_source_lines(block_lines)
67 try:
68 block_tree = ast.parse(dedent_source)
69 except SyntaxError:
70 pass
71 else:
72 stmt = block_tree.body[0]
73 if isinstance(stmt, ast.FunctionDef) and stmt.name == 'dedent_workaround':
74 # `_dedent_source_lines` wrapped the class around the workaround function
75 stmt = stmt.body[0]
76 if isinstance(stmt, ast.ClassDef) and stmt.name == cls.__name__:
77 return block_lines
78
79 frame = frame.f_back
80
81
82def extract_docstrings_from_cls(cls: type[Any], use_inspect: bool = False) -> dict[str, str]:
83 """Map model attributes and their corresponding docstring.
84
85 Args:
86 cls: The class of the Pydantic model to inspect.
87 use_inspect: Whether to skip usage of frames to find the object and use
88 the `inspect` module instead.
89
90 Returns:
91 A mapping containing attribute names and their corresponding docstring.
92 """
93 if use_inspect or sys.version_info >= (3, 13):
94 # On Python < 3.13, `inspect.getsourcelines()` might not work as expected
95 # if two classes have the same name in the same source file.
96 # On Python 3.13+, it will use the new `__firstlineno__` class attribute,
97 # making it way more robust.
98 try:
99 source, _ = inspect.getsourcelines(cls)
100 except OSError: # pragma: no cover
101 return {}
102 else:
103 # TODO remove this implementation when we drop support for Python 3.12:
104 source = _extract_source_from_frame(cls)
105
106 if not source:
107 return {}
108
109 dedent_source = _dedent_source_lines(source)
110
111 visitor = DocstringVisitor()
112 visitor.visit(ast.parse(dedent_source))
113 return visitor.attrs