mirror of
https://github.com/snowykami/mbcp.git
synced 2024-11-22 22:27:38 +08:00
179 lines
4.9 KiB
Python
179 lines
4.9 KiB
Python
"""
|
||
Google docstring parser for Python.
|
||
"""
|
||
from typing import Optional
|
||
|
||
from liteyuki_autodoc.docstring.docstring import Docstring
|
||
|
||
|
||
class Parser:
|
||
...
|
||
|
||
|
||
class GoogleDocstringParser(Parser):
|
||
_tokens = {
|
||
"Args" : "args",
|
||
"Arguments" : "args",
|
||
"参数" : "args",
|
||
|
||
"Return" : "return",
|
||
"Returns" : "return",
|
||
"返回" : "return",
|
||
|
||
"Attribute" : "attribute",
|
||
"Attributes" : "attribute",
|
||
"属性" : "attribute",
|
||
|
||
"Raises" : "raises",
|
||
"Raise" : "raises",
|
||
"引发" : "raises",
|
||
|
||
"Example" : "example",
|
||
"Examples" : "example",
|
||
"示例" : "example",
|
||
|
||
"Yields" : "yields",
|
||
"Yield" : "yields",
|
||
"产出" : "yields",
|
||
|
||
"Requires" : "requires",
|
||
"Require" : "requires",
|
||
"需要" : "requires",
|
||
|
||
"FrontMatter": "front_matter",
|
||
"前言" : "front_matter",
|
||
}
|
||
|
||
def __init__(self, docstring: str, indent: int = 4):
|
||
self.lines = docstring.splitlines()
|
||
self.indent = indent
|
||
self.lineno = 0 # Current line number
|
||
self.char = 0 # Current character position
|
||
|
||
self.docstring = Docstring()
|
||
|
||
def match_token(self) -> Optional[str]:
|
||
for token in self._tokens:
|
||
if self.lines[self.lineno][self.char:].startswith(token):
|
||
self.char += len(token)
|
||
return self._tokens[token]
|
||
return None
|
||
|
||
def parse_args(self):
|
||
"""
|
||
依次解析后面的参数行,直到缩进小于等于当前行的缩进
|
||
"""
|
||
while line := self.match_next_line():
|
||
if ":" in line:
|
||
name, desc = line.split(":", 1)
|
||
self.docstring.add_arg(name.strip(), desc.strip())
|
||
else:
|
||
self.docstring.add_arg(line.strip())
|
||
|
||
def parse_return(self):
|
||
"""
|
||
解析返回值行
|
||
"""
|
||
if line := self.match_next_line():
|
||
self.docstring.add_return(line.strip())
|
||
|
||
def parse_raises(self):
|
||
"""
|
||
解析异常行
|
||
"""
|
||
while line := self.match_next_line():
|
||
if ":" in line:
|
||
name, desc = line.split(":", 1)
|
||
self.docstring.add_raise(name.strip(), desc.strip())
|
||
else:
|
||
self.docstring.add_raise(line.strip())
|
||
|
||
def parse_example(self):
|
||
"""
|
||
解析示例行
|
||
"""
|
||
while line := self.match_next_line():
|
||
if ":" in line:
|
||
name, desc = line.split(":", 1)
|
||
self.docstring.add_example(name.strip(), desc.strip())
|
||
else:
|
||
self.docstring.add_example(line.strip())
|
||
|
||
def parse_attrs(self):
|
||
"""
|
||
解析属性行
|
||
"""
|
||
while line := self.match_next_line():
|
||
if ":" in line:
|
||
name, desc = line.split(":", 1)
|
||
self.docstring.add_attrs(name.strip(), desc.strip())
|
||
else:
|
||
self.docstring.add_attrs(line.strip())
|
||
|
||
def match_next_line(self) -> Optional[str]:
|
||
"""
|
||
在一个子解析器中,解析下一行,直到缩进小于等于当前行的缩进
|
||
Returns:
|
||
"""
|
||
while self.lineno + 1 < len(self.lines):
|
||
line = self.lines[self.lineno + 1]
|
||
|
||
if line.startswith(" " * self.indent):
|
||
line = line[self.indent:]
|
||
else:
|
||
return None
|
||
if not line:
|
||
return None
|
||
self.lineno += 1
|
||
return line
|
||
|
||
def parse(self) -> Docstring:
|
||
"""
|
||
逐行解析,直到遇到EOS
|
||
|
||
最开始未解析到的内容全部加入desc
|
||
|
||
Returns:
|
||
|
||
"""
|
||
add_desc = True
|
||
while self.lineno < len(self.lines):
|
||
token = self.match_token()
|
||
if token is None and add_desc:
|
||
self.docstring.add_desc(self.lines[self.lineno].strip())
|
||
|
||
if token is not None:
|
||
add_desc = False
|
||
|
||
match token:
|
||
|
||
case "args":
|
||
self.parse_args()
|
||
case "return":
|
||
self.parse_return()
|
||
case "attribute":
|
||
self.parse_attrs()
|
||
case "raises":
|
||
self.parse_raises()
|
||
case "example":
|
||
self.parse_example()
|
||
case _:
|
||
self.lineno += 1
|
||
|
||
return self.docstring
|
||
|
||
|
||
class NumpyDocstringParser(Parser):
|
||
...
|
||
|
||
|
||
class ReStructuredParser(Parser):
|
||
...
|
||
|
||
|
||
def parse(docstring: str, parser: str = "google", indent: int = 4) -> Docstring:
|
||
if parser == "google":
|
||
return GoogleDocstringParser(docstring, indent).parse()
|
||
else:
|
||
raise ValueError(f"Unknown parser: {parser}")
|