mirror of
https://github.com/tahoe-lafs/tahoe-lafs.git
synced 2025-01-18 10:46:24 +00:00
script to fetch PR information from GitHub
This commit is contained in:
parent
43c0c0b3ec
commit
ae2f6f6ef2
2
NEWS.rst
2
NEWS.rst
@ -7,8 +7,6 @@ User-Visible Changes in Tahoe-LAFS
|
|||||||
Release ?? (??)
|
Release ?? (??)
|
||||||
'''''''''''''''
|
'''''''''''''''
|
||||||
|
|
||||||
TOP
|
|
||||||
|
|
||||||
unedited list of changes since last release, needs cleanup, explanation,
|
unedited list of changes since last release, needs cleanup, explanation,
|
||||||
credit authors, limit to user-visible things
|
credit authors, limit to user-visible things
|
||||||
|
|
||||||
|
125
fetch-pr.py
Normal file
125
fetch-pr.py
Normal file
@ -0,0 +1,125 @@
|
|||||||
|
# given a PR number, get all contributers and the summary from
|
||||||
|
# GitHub's API
|
||||||
|
|
||||||
|
import sys
|
||||||
|
import json
|
||||||
|
import base64
|
||||||
|
|
||||||
|
from twisted.internet.task import react
|
||||||
|
from twisted.internet.defer import inlineCallbacks, returnValue
|
||||||
|
|
||||||
|
import treq
|
||||||
|
|
||||||
|
base_pr_url = "https://api.github.com/repos/tahoe-lafs/tahoe-lafs/pulls/{}"
|
||||||
|
|
||||||
|
#async def main(reactor):
|
||||||
|
@inlineCallbacks
|
||||||
|
def main(reactor):
|
||||||
|
"""
|
||||||
|
Fetch Pull Request (PR) information from GitHub.
|
||||||
|
|
||||||
|
Either pass a list of PR numbers on the command-line, or pipe text
|
||||||
|
containing references like: "There is a PR123 somewhere" from
|
||||||
|
which instances of "PRxxx" are extrated. From GitHub's API we get
|
||||||
|
all author information and anyone who disucced the PR and print a
|
||||||
|
summary afterwards.
|
||||||
|
|
||||||
|
You need a 'token' file containing two lines: your username, and
|
||||||
|
access token (get this from the GitHub Web UI).
|
||||||
|
"""
|
||||||
|
if len(sys.argv) < 2:
|
||||||
|
data = sys.stdin.read()
|
||||||
|
if not len(data):
|
||||||
|
print("put some PR numbers on the command-line")
|
||||||
|
returnValue(1)
|
||||||
|
else:
|
||||||
|
all_prs = set()
|
||||||
|
for word in data.split():
|
||||||
|
word = word.strip()
|
||||||
|
if word.startswith('PR'):
|
||||||
|
all_prs.add(word[2:])
|
||||||
|
all_prs = list(all_prs)
|
||||||
|
print("Found {} PRs in stdin text".format(len(all_prs)))
|
||||||
|
else:
|
||||||
|
all_prs = sys.argv[1:]
|
||||||
|
|
||||||
|
# a 'token' file contains two lines: username, github token
|
||||||
|
try:
|
||||||
|
with open('token', 'r') as f:
|
||||||
|
data = f.read().strip()
|
||||||
|
username, token = data.split('\n', 1)
|
||||||
|
except (IOError, EnvironmentError) as e:
|
||||||
|
print("Couldn't open or parse 'token' file: {}".format(e))
|
||||||
|
returnValue(1)
|
||||||
|
except ValueError:
|
||||||
|
print("'token' should contain two lines: username, github token")
|
||||||
|
returnValue(1)
|
||||||
|
|
||||||
|
headers = {
|
||||||
|
"User-Agent": "treq",
|
||||||
|
"Authorization": "Basic {}".format(base64.b64encode("{}:{}".format(username, token))),
|
||||||
|
}
|
||||||
|
|
||||||
|
pr_info = {}
|
||||||
|
|
||||||
|
for pr in all_prs:
|
||||||
|
print("Fetching PR{}".format(pr))
|
||||||
|
resp = yield treq.get(
|
||||||
|
base_pr_url.format(pr),
|
||||||
|
headers=headers,
|
||||||
|
)
|
||||||
|
raw_data = yield resp.text()
|
||||||
|
data = json.loads(raw_data)
|
||||||
|
|
||||||
|
code_handles = set()
|
||||||
|
help_handles = set()
|
||||||
|
ignore = ('codecov-io', )
|
||||||
|
|
||||||
|
print("Commits:")
|
||||||
|
commits_resp = yield treq.get(data['commits_url'], headers=headers)
|
||||||
|
commits_data = yield commits_resp.text()
|
||||||
|
commits = json.loads(commits_data)
|
||||||
|
for commit in commits:
|
||||||
|
if commit['author'] is None:
|
||||||
|
print(" {}: no author!".format(commit['sha']))
|
||||||
|
else:
|
||||||
|
author = commit['author']['login']
|
||||||
|
print(" {}: {}".format(commit['sha'], author))
|
||||||
|
if author not in ignore:
|
||||||
|
code_handles.add(author)
|
||||||
|
|
||||||
|
print("Comments:")
|
||||||
|
comments_resp = yield treq.get(data['comments_url'], headers=headers)
|
||||||
|
comments_data = yield comments_resp.text()
|
||||||
|
comments = json.loads(comments_data)
|
||||||
|
for comment in comments:
|
||||||
|
author = comment['user']['login']
|
||||||
|
if author not in ignore:
|
||||||
|
help_handles.add(author)
|
||||||
|
print(" {}: {}".format(author, comment['body'].replace('\n', ' ')[:60]))
|
||||||
|
|
||||||
|
pr_info[pr] = (
|
||||||
|
code_handles,
|
||||||
|
help_handles - help_handles.intersection(code_handles),
|
||||||
|
)
|
||||||
|
|
||||||
|
unique_handles = set()
|
||||||
|
for pr, (code_handles, help_handles) in sorted(pr_info.items()):
|
||||||
|
coders = ', '.join('`{}`_'.format(c) for c in code_handles)
|
||||||
|
helpers = ', '.join('`{}`_'.format(c) for c in help_handles)
|
||||||
|
if helpers:
|
||||||
|
print("`PR{}`_: {} (with {})".format(pr, coders, helpers))
|
||||||
|
else:
|
||||||
|
print("`PR{}`_: {}".format(pr, coders))
|
||||||
|
for h in code_handles.union(help_handles):
|
||||||
|
unique_handles.add(h)
|
||||||
|
|
||||||
|
for pr in sorted(pr_info.keys()):
|
||||||
|
print(".. _PR{}: https://github.com/tahoe-lafs/tahoe-lafs/pull/{}".format(pr, pr))
|
||||||
|
for h in sorted(unique_handles):
|
||||||
|
print(".. _{}: https://github.com/{}".format(h, h))
|
||||||
|
|
||||||
|
|
||||||
|
if __name__ == "__main__":
|
||||||
|
#react(lambda r: ensureDeferred(main(r)))
|
||||||
|
react(main)
|
Loading…
Reference in New Issue
Block a user