Skip to content

Instantly share code, notes, and snippets.

@timfel

timfel/package_c_api_overview.py

Created August 25, 2022 18:29
Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save timfel/3fc3b1f339917ff48b69289e79a39f8d to your computer and use it in GitHub Desktop.
Save timfel/3fc3b1f339917ff48b69289e79a39f8d to your computer and use it in GitHub Desktop.
Download ZIP
Raw
package_c_api_overview.py
from collections import namedtuple
import os
import pprint
import re
import requests
import shutil
import subprocess
import tarfile
import traceback
import zipfile
session = requests.Session()
EXTRA_URLS = {
# source urls for some projects that have no sdist in PyPI
"torch": "https://github.com/pytorch/pytorch/archive/refs/tags/v1.10.2.zip",
"tensorflow": "https://github.com/tensorflow/tensorflow/archive/refs/tags/v2.8.0.zip",
"caffe": "https://github.com/BVLC/caffe/archive/refs/heads/master.zip",
"torchvision": "https://github.com/pytorch/vision/archive/refs/tags/v0.11.3.zip",
"keras": "https://github.com/keras-team/keras/archive/refs/tags/v2.8.0.zip",
"cv2": "https://github.com/opencv/opencv-python/archive/refs/heads/3.4.zip",
"nimbusml": "https://github.com/microsoft/NimbusML/archive/refs/heads/master.zip",
"mxnet": "https://github.com/apache/incubator-mxnet/archive/refs/tags/v2.0.0.beta0.rc1.zip",
"onnxruntime": "https://github.com/microsoft/onnxruntime/archive/refs/tags/v1.10.0.zip",
"python-util": "https://github.com/MisterL2/python-util/archive/refs/heads/master.zip",
}
DS_IMPORTANT_PACKAGES = list(
dict.fromkeys(
[
# top 10 in Data Science Through the Looking Glass
"numpy",
"matplotlib",
"pandas",
"scikit-learn",
"scipy",
"seaborn",
"tensorflow",
# "pylab", # pylab is just matplotlib.pylab
"requests",
"statsmodels",
# top 10 rank change in Data Science Through the Looking Glass
"torch",
"keras",
"xgboost",
"Pillow", # called PIL in the paper
"python-util",
"cv2", # really python-opencv
"tqdm",
"sqlalchemy",
"gensim",
"tensorflow",
# top 10 pct change in Data Science Through the Looking Glass
"pandas",
"matplotlib",
"scikit-learn",
"seaborn",
"keras",
"torch",
"numpy",
"tensorflow",
"Pillow", # called PIL in the paper
"cv2", # really python-opencv
# top 5 deep learning imports in Data Science Through the Looking Glass
"tensorflow",
"keras",
"theano",
"caffe",
"torch",
# top 10 imports in Data Science Through the Looking Glass
"scikit-learn",
"numpy",
"matplotlib",
"pandas",
"scipy",
"keras",
"seaborn",
"tensorflow",
"nltk",
"statsmodels",
# extras from correlation
"bs4",
"torchvision",
# "selenium", # pure Python
# release analysis packages from Data Science Through the Looking Glass
"keras",
"lasagne",
"matplotlib",
"nolearn",
"numpy",
"pandas",
"scikit-learn",
"scipy",
"seaborn",
"nimbusml",
"mxnet",
# extra packages
"category-encoders",
"dask",
"imbalanced-learn",
"lightgbm",
"onnx",
"onnxmltools",
"onnxruntime",
"orbit-ml",
"psutil",
"pyod",
"skl2onnx",
"sktime",
"plotly",
# needed
"cython",
]
)
)
SLOCCOUNT_LANGUAGE_PATTERN = re.compile(r"([a-z]+)=(\d+)")
TARFILE_PATTERN = re.compile(r"\.(tar\.gz|tgz|bz2)$")
NATIVE_FILE_PATTERN = re.compile(r"\.(cc|cxx|cpp|C|CC|c\+\+|c|h|hpp|pyx|rs)$")
NATIVE_API_PATTERN = re.compile(
r"""
pyo3 # Rust package
|
\#include\s+[<"]cppy # cppy
|
\#include\s+[<"]pybind11 # pybind11
|
\#include\s+[<"]Python\.h[>"] # includes Python API
|
\#include\s+[<"]numpy # dependency on numpy API
|
Py_VISIT # macro used in tp_traverse
|
PyStructSequence_Desc # tuple subclass that uses a hack to hide some fields from Python code
|
PyCapsule_Destructor # destructor for capsules
|
PyObject_GC_U?n?Track # potentially (?) problematic GC access
|
PyUnicode_(?:DATA|WRITE|READ) # direct access to unicode void*
|
Py_TYPE\([^\)]+\)\s*=\s*(?!__pyx)[_\(\*_a-zA-Z0-9]+ # assignment to type
|
tp_traverse\s*=\s*(?!__pyx)[ \(\*_a-zA-Z0-9]+ # own code that runs on GC
|
ob_type\s*=\s*(?!__pyx)[ \(\*_a-zA-Z0-9]+ # assignment to type
|
tp_bases?\s*=\s*(?!__pyx)[ \(\*_a-zA-Z0-9]+ # assignment to base and bases
|
tp_finalize?\s*=\s*(?!__pyx)[ \(\*_a-zA-Z0-9]+
|
tp_del?\s*=\s*(?!__pyx)[ \(\*_a-zA-Z0-9]+
|
PyTypeObject\s+(?!__pyx)[ \(\*_a-zA-Z0-9_]+\s+=\s+{ # static type definition
""",
re.VERBOSE,
)
PYTHON_API_PATTERN = re.compile("Py_")
def projects():
resp = session.get(
"https://hugovk.github.io/top-pypi-packages/top-pypi-packages-30-days.min.json"
)
resp.raise_for_status()
return [p["project"] for p in resp.json()["rows"]]
def find_url(proj):
resp = session.get(f"https://pypi.org/pypi/{proj}/json")
try:
for u in resp.json()["urls"]:
if u["packagetype"] == "sdist":
return u["url"]
except:
pass
return EXTRA_URLS.get(proj, None)
def download_sdist(idx, proj):
# Download the sdist file for this project. The file is downloaded into a
# generic name, and a stamp file is created. This is done to avoid any
# requests to PyPI even to get the URL. To actually fetch the most recent
# package, the previous download must be deleted manually.
filename = f"{idx:>05}_{proj}"
if os.path.exists(filename):
print(f"Exists: {filename}")
for f in os.scandir():
if f.is_file() and f.name.startswith(filename) and f.name != filename:
return filename, f.name
url = find_url(proj)
if not url:
# Universal wheel only, maybe.
print(f"Cannot find url for {proj}")
return
stamp = f'{filename}_{url[url.rfind("/") + 1 :]}'
print(f"Saving {stamp} to {filename}")
resp = session.get(url)
resp.raise_for_status()
with open(filename, "wb") as f:
f.write(resp.content)
with open(stamp, "wb") as f:
f.write(b"")
return filename, stamp
class PackageInfo(
namedtuple("PackageInfo", "name interesting_files native_apis sloccount")
):
@classmethod
def new(cls, name, interesting_files, native_apis):
return cls(name, interesting_files, native_apis, [[], 0, []])
@property
def api_sloccount(self):
return self.sloccount[0]
@api_sloccount.setter
def api_sloccount(self, value):
self.sloccount[0] = value
@property
def py_sloccount(self):
return self.sloccount[1]
@py_sloccount.setter
def py_sloccount(self, value):
self.sloccount[1] = value
@property
def total_sloccount(self):
return self.sloccount[2]
@total_sloccount.setter
def total_sloccount(self, value):
self.sloccount[2] = value
def __repr__(self):
extensions = set()
for f in self.interesting_files:
extensions.add(os.path.splitext(f)[1])
extensions = ",".join(extensions)
apis = pprint.pformat(list(self.native_apis.keys()), width=120, indent=1)
if "\n" in apis:
apis = ("\n" + apis).replace("\n", "\n ")
sloccount = ["\n "]
for el in self.total_sloccount:
sloccount.append(el)
sloccount.append(sloccount[0])
if self.api_sloccount:
sloccount.append("In files using C API")
sloccount.append(sloccount[0])
for el in self.api_sloccount:
sloccount.append(el)
sloccount.append(sloccount[0])
if self.py_sloccount:
sloccount.append(f"Counting only lines with 'Py' in them: {self.py_sloccount}")
sloccount.append(sloccount[0])
return f"{self.name}: {extensions} with {apis}\n{''.join(sloccount)}"
def main(projs, prefix=""):
infos = []
for idx, p in enumerate(projs):
try:
filename, stamp = download_sdist(f"{prefix}{idx}", p)
except Exception as e:
traceback.print_exc()
print(f"Failed to download {p}")
continue
if stamp.endswith(".zip"):
archive = zipfile.ZipFile(filename)
names = archive.namelist()
members = names
elif TARFILE_PATTERN.search(stamp):
archive = tarfile.open(filename)
members = archive.getmembers()
names = [m.name for m in members]
else:
print(f"Don't know how to extract {stamp}")
continue
interesting_files = list(
filter(
lambda el: el is not None,
map(
lambda pair: pair[1]
if (
not pair[0].startswith("/")
and ".." not in pair[0]
and NATIVE_FILE_PATTERN.search(pair[0])
)
else None,
zip(names, members),
),
)
)
if interesting_files:
info = PackageInfo.new(name=p, interesting_files=set(), native_apis={})
dirname = f"{filename}.dir"
apidirname = f"{filename}.dir.with_api_usage"
if not os.path.exists(dirname):
os.makedirs(dirname, exist_ok=True)
archive.extractall(path=dirname, members=interesting_files)
chars = list(r"/-\|")
for path in interesting_files:
print("\033[1D", chars[0], sep="", end="", flush=True)
chars = chars[1:] + chars[:1]
name = os.path.join(dirname, getattr(path, "name", path))
with open(name, "rb") as f:
content = f.read().decode("utf-8", errors="replace")
uses_c_api = False
for m in NATIVE_API_PATTERN.finditer(content):
print("\033[1DX\033[1C", end="", flush=True)
info.interesting_files.add(name)
info.native_apis.setdefault(m.group(0), set()).add(name)
uses_c_api = True
uses_c_api = uses_c_api or PYTHON_API_PATTERN.search(content)
if uses_c_api:
sloccountname = os.path.join(apidirname, getattr(path, "name", path))
sloccountnamedir = os.path.dirname(sloccountname)
if not os.path.exists(sloccountnamedir):
os.makedirs(sloccountnamedir, exist_ok=True)
# modify some file endings for sloccount to consider them
if sloccountname.endswith(".pyx"):
sloccountname += ".py"
elif sloccountname.endswith(".rs"):
sloccountname += ".cpp"
shutil.copy(name, sloccountname)
sloccount_sb = []
for line in subprocess.getoutput(f"sloccount {dirname}").split("\n"):
if sloccount_sb:
sloccount_sb.append(line)
break
elif "SLOC-by-Language" in line:
sloccount_sb.append(line)
info.total_sloccount = sloccount_sb
api_sloccount_sb = []
if os.path.exists(apidirname):
for line in subprocess.getoutput(f"sloccount {apidirname}").split("\n"):
if api_sloccount_sb:
api_sloccount_sb.append(line)
break
elif "SLOC-by-Language" in line:
api_sloccount_sb.append(line)
for dirpath,dirnames,filenames in os.walk(apidirname):
print("\033[1D", chars[0], sep="", end="", flush=True)
chars = chars[1:] + chars[:1]
for f in filenames:
if os.path.splitext(f)[1] == ".pyx":
continue
else:
with open(os.path.join(dirpath, f), "r") as file:
for l in file.readlines():
if "Py" in l:
info.py_sloccount = info.py_sloccount + 1
shutil.rmtree(apidirname)
info.api_sloccount = api_sloccount_sb
print()
if info.interesting_files:
infos.append(info)
else:
shutil.rmtree(dirname)
totalSloc = {}
apiSloc = {}
pySloc = 0
print(f"\n{len(infos)} of top {idx} packages found to have C API usage.\n")
for info in infos:
print(info, "\n")
for el in info.total_sloccount:
for m in SLOCCOUNT_LANGUAGE_PATTERN.finditer(el):
totalSloc[m.group(1)] = totalSloc.get(m.group(1), 0) + int(m.group(2))
for el in info.api_sloccount:
for m in SLOCCOUNT_LANGUAGE_PATTERN.finditer(el):
apiSloc[m.group(1)] = apiSloc.get(m.group(1), 0) + int(m.group(2))
pySloc += info.py_sloccount
print()
print("Totals")
print("Package SLOC:", totalSloc)
print("API using files SLOC:", apiSloc)
print("Lines with 'Py':", pySloc)
if __name__ == "__main__":
from argparse import ArgumentParser
import sys
parser = ArgumentParser(
description="Analyse the top packages either of PyPI or for data science. Must choose either or!"
)
parser.add_argument("--top5000", action="store_true")
parser.add_argument("--topDS", action="store_true")
args = parser.parse_args(sys.argv[1:])
if args.top5000 and args.topDS or not args.top5000 and not args.topDS:
parser.print_help()
sys.exit(1)
if args.top5000:
main(projects())
else:
main(DS_IMPORTANT_PACKAGES, prefix="DS_")
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment