长话短说，我的Gist。
给定一个没有requirements.txt的Python项目，如果想知道需要安装哪些包才能满足这个项目的依赖需求，一个容易想到的方法就是对每一个.py文件，用模式匹配（如正则表达式）找import xxx，并记录xxx为需要的包。然而import语句有很多形式，如：import xxx、import xxx as aaa、import xxx as aaa, yyy as bbb、from xxx.yyy import fff as ccc、from .zzz import ggg。因此，更好的方法是利用抽象语法树ast模块来找出所有import语句。
Python的import语句对应ast的两种节点：ast.import和ast.importFrom。要从ast.import获取导入包的列表，可用：
[a.name for a in node.names] # 其中node是ast.import类型的
要从ast.importFrom获取导入的包，可用：
node.module # 其中node是ast.importFrom类型的
值得注意的是如果当前import语句是from . import xxx，node.module将会是None，此时node.level > 0，意味着相对导入。因此，要想获得所有导入的包（除了相对导入外，因为相对导入的包绝不会是需要安装的依赖），可以这样：
import ast
# 假设source包含待解析源码
root = ast.parse(source)
result = []
for node in ast.walk(root):
if isinstance(node, ast.import):
for a in node.names:
result.append(a.name.split('.', maxsplit=1)[0])
elif isinstance(node, ast.importFrom):
if node.level == 0:
result.append(node.module.split('.', maxsplit=1)[0])
然而绝对导入的包也有可能是工作目录中已存在的模块或包啊，此时我们就可以根据导入路径判断它是不是指工作目录下的包：
def exists_local(path, rootpkg):
filepath = os.path.join(rootpkg, path.replace('.', os.path.sep))
# see if path is a local package
if os.path.isdir(filepath) and os.path.isfile(
os.path.join(filepath, '__init__.py')):
return True
# see if path is a local module
if os.path.isfile(filepath + '.py'):
return True
return False
其中path是导入路径，rootpkg是根包所在目录（定义见这里）。
把这个核心功能稍作包装，便可写出下面的完整可执行代码：
from __future__ import print_function
import argparse
import os
import ast
import sys
import pkgutil
import itertools
import logging
def make_parser():
parser = argparse.ArgumentParser(
description=('List all root imports. The *root* import of '
'`import pkg1.mod1` is "pkg1".'))
parser.add_argument(
'-g',
'--greedy',
action='store_true',
help=('find also import statements within try block, '
'if block, while block, function definition, '
'etc.'))
parser.add_argument(
'-i',
'--include-installed',
action='store_true',
help='include installed/built-in modules/packages')
parser.add_argument(
'-T',
'--files-from',
metavar='LIST_FILE',
help=('if specified, the files to process '
'will be read one per line from '
'LIST_FILE; if specified as `-', '
'stdin will be expected to contain '
'the files to process. Note that '
'SOURCE_FILEs, if exist, take '
'precedence (see below)'))
parser.add_argument(
'rootpkg',
metavar='ROOTPKG_DIR',
type=dir_type,
help=
('the directory of the root package. See '
'https://docs.python.org/3.7/distutils/setupscript.html#listing-whole-packages '
'about *root package*. Local packages/modules will be '
'excluded from the results. For example, if '
'there are "mod1.py" and "mod2.py", and in '
'"mod2.py" there is `import mod1`, then "mod1" '
'won't be listed in the result.'))
parser.add_argument(
'filenames',
metavar='SOURCE_FILE',
nargs='*',
help=('if specified one or more files, '
'only these SOURCE_FILEs will get '
'processed regardless of `-T' '
'option; if no SOURCE_FILE is '
'specified, `-T', if exists, is '
'processed. In both cases, the '
'final results will be joined'))
return parser
def dir_type(string):
if not os.path.isdir(string):
raise argparse.ArgumentTypeError('must be a directory')
return string
# Reference: https://stackoverflow.com/a/9049549/7881370
def yield_imports(root, greedy):
"""
Yield all absolute imports.
"""
traverse = ast.walk if greedy else ast.iter_child_nodes
for node in traverse(root):
if isinstance(node, ast.import):
for a in node.names:
yield a.name
elif isinstance(node, ast.importFrom):
# if node.level > 0, the import is relative
if node.level == 0:
yield node.module
def exists_local(path, rootpkg):
"""
Returns ``True`` if the absolute import ``path`` refers to a package or
a module residing under the working directory, else ``False``.
"""
filepath = os.path.join(rootpkg, path.replace('.', os.path.sep))
# see if path is a local package
if os.path.isdir(filepath) and os.path.isfile(
os.path.join(filepath, '__init__.py')):
return True
# see if path is a local module
if os.path.isfile(filepath + '.py'):
return True
return False
def filter_local(imports_iterable, rootpkg):
"""
Remove modules and packages in the working directory, and yield root
imports.
"""
for path in imports_iterable:
if not exists_local(path, rootpkg):
yield path.split('.', 1)[0]
def filter_installed(imports_iterable):
"""
Remove modules and packages already installed, which include built-in
modules and packages and those already installed (e.g. via ``pip``).
"""
installed = set(
itertools.chain(sys.builtin_module_names,
(x[1] for x in pkgutil.iter_modules())))
for name in imports_iterable:
if name not in installed:
yield name
def collect_sources(filenames, files_from):
if filenames:
for filename in filenames:
yield filename
elif files_from == '-':
try:
for line in sys.stdin:
yield line.rstrip('n')
except KeyboardInterrupt:
pass
elif files_from:
try:
with open(files_from) as infile:
for line in infile:
yield line.rstrip('n')
except OSError:
logging.exception('failed to read from "{}"'.format(files_from))
def main():
logging.basicConfig(format='%(levelname)s: %(message)s')
args = make_parser().parse_args()
all_imports = []
for filename in collect_sources(args.filenames, args.files_from):
try:
with open(filename) as infile:
root = ast.parse(infile.read(), filename)
except OSError:
logging.exception('skipped')
except SyntaxError:
logging.exception('failed to parse "{}"; skipped'.format(filename))
else:
all_imports.append(yield_imports(root, args.greedy))
all_imports = itertools.chain.from_iterable(all_imports)
all_imports = filter_local(all_imports, args.rootpkg)
if not args.include_installed:
all_imports = filter_installed(all_imports)
all_imports = sorted(set(all_imports))
if all_imports:
print('n'.join(all_imports))
logging.shutdown()
if __name__ == '__main__':
main()
需要注意的是，程序的输出并不一定是PyPI上包的名字（例如，import bs4然而pip install beautifulsoup4）。
类似项目：pipreqs。核心代码是几乎一样的，但包装得不同。
