mirror of
https://github.com/trezor/trezor-firmware.git
synced 2024-11-22 15:38:11 +00:00
88 lines
3.3 KiB
Python
88 lines
3.3 KiB
Python
"""
|
|
Taken from https://gist.github.com/patrickfuller/e2ea8a94badc5b6967ef3ca0a9452a43 and modified.
|
|
|
|
Currently writes all issues that have some Weight.
|
|
"""
|
|
import argparse
|
|
import csv
|
|
from getpass import getpass
|
|
import requests
|
|
|
|
auth_id = None
|
|
auth_secret = None
|
|
|
|
PRIORITIES = ("P1", "P2", "P3", "P4")
|
|
SEVERITIES = ("S1", "S2", "S3", "S4")
|
|
WEIGHTS = ("W0", "W1/2", "W1", "W2", "W3", "W5", "W8", "W13", "W20", "W40", "W100")
|
|
|
|
def write_issues(r, csvout):
|
|
"""Parses JSON response and writes to CSV."""
|
|
if r.status_code != 200:
|
|
raise Exception(r.status_code)
|
|
for issue in r.json():
|
|
if 'pull_request' not in issue:
|
|
priority = ""
|
|
severity = ""
|
|
weight = ""
|
|
labels = []
|
|
for l in issue['labels']:
|
|
if l["name"][:2] in PRIORITIES:
|
|
priority = l["name"]
|
|
elif l["name"][:2] in SEVERITIES:
|
|
severity = l["name"]
|
|
elif l["name"] in WEIGHTS:
|
|
weight = l["name"][1:]
|
|
if weight == "1/2":
|
|
weight = "0.5"
|
|
else:
|
|
labels.append(l["name"])
|
|
if not weight:
|
|
continue
|
|
labels = ", ".join(labels)
|
|
date = issue['created_at'].split('T')[0]
|
|
milestone = issue['milestone']['title'] if issue['milestone'] else ""
|
|
|
|
csvout.writerow([issue['title'], issue['number'], issue['html_url'], issue['state'],
|
|
milestone, priority, severity, weight, labels])
|
|
|
|
|
|
def get_issues(name):
|
|
"""Requests issues from GitHub API and writes to CSV file."""
|
|
if auth_secret:
|
|
url = 'https://api.github.com/repos/{}/issues?state=all&client_id={}&client_secret={}'.format(name, auth_id, auth_secret)
|
|
else:
|
|
url = 'https://api.github.com/repos/{}/issues?state=all'.format(name)
|
|
r = requests.get(url)
|
|
|
|
csvfilename = '{}-issues.csv'.format(name.replace('/', '-'))
|
|
with open(csvfilename, 'w', newline='') as csvfile:
|
|
csvout = csv.writer(csvfile)
|
|
csvout.writerow(['Title', 'Number', 'URL', 'State', 'Milestone', 'Priority', 'Severity', 'Weight', 'Labels'])
|
|
write_issues(r, csvout)
|
|
|
|
# Multiple requests are required if response is paged
|
|
if 'link' in r.headers:
|
|
pages = {rel[6:-1]: url[url.index('<')+1:-1] for url, rel in
|
|
(link.split(';') for link in
|
|
r.headers['link'].split(','))}
|
|
while 'last' in pages and 'next' in pages:
|
|
pages = {rel[6:-1]: url[url.index('<')+1:-1] for url, rel in
|
|
(link.split(';') for link in
|
|
r.headers['link'].split(','))}
|
|
r = requests.get(pages['next'])
|
|
write_issues(r, csvout)
|
|
if pages['next'] == pages['last']:
|
|
break
|
|
|
|
|
|
parser = argparse.ArgumentParser(description="Write GitHub repository issues "
|
|
"to CSV file.")
|
|
parser.add_argument('repositories', nargs='+', help="Repository names, "
|
|
"formatted as 'username/repo'")
|
|
parser.add_argument('--all', action='store_true', help="Returns both open "
|
|
"and closed issues.")
|
|
args = parser.parse_args()
|
|
|
|
for repository in args.repositories:
|
|
get_issues(repository)
|