cppcheck/tools/donate-cpu-server.py

953 lines
33 KiB
Python
Raw Normal View History

# Server for 'donate-cpu.py'
import glob
import json
import os
import socket
import re
2018-08-24 13:04:25 +02:00
import datetime
import time
from threading import Thread
import sys
import urllib
import logging
import logging.handlers
import operator
2018-08-24 13:04:25 +02:00
# Version scheme (MAJOR.MINOR.PATCH) should orientate on "Semantic Versioning" https://semver.org/
# Every change in this script should result in increasing the version number accordingly (exceptions may be cosmetic
# changes)
SERVER_VERSION = "1.1.2"
OLD_VERSION = '1.88'
2018-12-08 11:39:44 +01:00
# Set up logging
logger = logging.getLogger()
logger.setLevel(logging.INFO)
# Logging to console
handler_stream = logging.StreamHandler()
logger.addHandler(handler_stream)
# Log errors to a rotating file
logfile = sys.path[0]
if logfile:
logfile += '/'
logfile += 'donate-cpu-server.log'
handler_file = logging.handlers.RotatingFileHandler(filename=logfile, maxBytes=100*1024, backupCount=1)
handler_file.setLevel(logging.ERROR)
logger.addHandler(handler_file)
# Set up an exception hook for all uncaught exceptions so they can be logged
def handle_uncaught_exception(exc_type, exc_value, exc_traceback):
if issubclass(exc_type, KeyboardInterrupt):
sys.__excepthook__(exc_type, exc_value, exc_traceback)
return
logging.error("Uncaught exception", exc_info=(exc_type, exc_value, exc_traceback))
sys.excepthook = handle_uncaught_exception
2018-08-24 13:04:25 +02:00
def strDateTime():
return datetime.datetime.now().strftime('%Y-%m-%d %H:%M')
2018-08-24 13:04:25 +02:00
def dateTimeFromStr(datestr):
return datetime.datetime.strptime(datestr, '%Y-%m-%d %H:%M')
def overviewReport():
html = '<html><head><title>daca@home</title></head><body>\n'
html += '<h1>daca@home</h1>\n'
html += '<a href="crash.html">Crash report</a><br>\n'
html += '<a href="stale.html">Stale report</a><br>\n'
html += '<a href="diff.html">Diff report</a><br>\n'
html += '<a href="head.html">HEAD report</a><br>\n'
2018-09-02 09:35:38 +02:00
html += '<a href="latest.html">Latest results</a><br>\n'
html += '<a href="time.html">Time report</a><br>\n'
html += '<a href="check_library_function_report.html">checkLibraryFunction report</a><br>\n'
html += '<a href="check_library_noreturn_report.html">checkLibraryNoReturn report</a><br>\n'
html += '<a href="check_library_use_ignore_report.html">checkLibraryUseIgnore report</a><br>\n'
html += '<br>\n'
html += 'version ' + SERVER_VERSION + '\n'
html += '</body></html>'
return html
def fmt(a, b, c, d, e):
column_width = [40, 10, 5, 6, 6, 8]
ret = a
while len(ret) < column_width[0]:
ret += ' '
if len(ret) == column_width[0]:
ret += ' ' + b[:10]
while len(ret) < (column_width[0] + 1 + column_width[1]):
ret += ' '
ret += ' '
ret += b[-5:].rjust(column_width[2]) + ' '
if not c is None:
ret += c.rjust(column_width[3]) + ' '
if not d is None:
ret += d.rjust(column_width[4]) + ' '
if not e is None:
ret += e.rjust(column_width[5])
2018-08-25 10:59:49 +02:00
if a != 'Package':
pos = ret.find(' ')
ret = '<a href="' + a + '">' + a + '</a>' + ret[pos:]
return ret
2018-08-25 09:06:15 +02:00
def latestReport(latestResults):
2018-08-26 16:47:20 +02:00
html = '<html><head><title>Latest daca@home results</title></head><body>\n'
html += '<h1>Latest daca@home results</h1>\n'
html += '<pre>\n<b>' + fmt('Package', 'Date Time', OLD_VERSION, 'Head', 'Diff') + '</b>\n'
# Write report for latest results
2018-08-25 09:06:15 +02:00
for filename in latestResults:
if not os.path.isfile(filename):
continue
package = filename[filename.rfind('/')+1:]
current_year = datetime.date.today().year
datestr = ''
count = ['0', '0']
2018-08-26 16:47:20 +02:00
lost = 0
added = 0
for line in open(filename, 'rt'):
line = line.strip()
if line.startswith(str(current_year) + '-') or line.startswith(str(current_year - 1) + '-'):
datestr = line
2018-08-26 16:47:20 +02:00
#elif line.startswith('cppcheck:'):
# cppcheck = line[9:]
elif line.startswith('count: '):
count = line.split(' ')[1:]
elif line.startswith('head ') and not line.startswith('head results:'):
2018-08-26 16:47:20 +02:00
added += 1
2018-12-08 11:39:44 +01:00
elif line.startswith(OLD_VERSION + ' '):
lost += 1
2018-08-26 16:47:20 +02:00
diff = ''
if lost > 0:
diff += '-' + str(lost)
if added > 0:
diff += '+' + str(added)
html += fmt(package, datestr, count[1], count[0], diff) + '\n'
html += '</pre></body></html>\n'
return html
2018-08-31 14:28:01 +02:00
def crashReport():
html = '<html><head><title>Crash report</title></head><body>\n'
html += '<h1>Crash report</h1>\n'
html += '<pre>\n'
html += '<b>' + fmt('Package', 'Date Time', OLD_VERSION, 'Head', None) + '</b>\n'
current_year = datetime.date.today().year
2018-08-31 14:28:01 +02:00
for filename in sorted(glob.glob(os.path.expanduser('~/daca@home/donated-results/*'))):
if not os.path.isfile(filename):
continue
datestr = ''
2018-08-31 14:28:01 +02:00
for line in open(filename, 'rt'):
line = line.strip()
if line.startswith(str(current_year) + '-') or line.startswith(str(current_year - 1) + '-'):
datestr = line
2018-08-31 14:28:01 +02:00
if not line.startswith('count:'):
continue
if line.find('Crash') < 0:
break
package = filename[filename.rfind('/')+1:]
2018-08-31 14:28:01 +02:00
counts = line.strip().split(' ')
c2 = ''
2018-08-31 14:28:01 +02:00
if counts[2] == 'Crash!':
c2 = 'Crash'
c1 = ''
2018-08-31 14:28:01 +02:00
if counts[1] == 'Crash!':
c1 = 'Crash'
html += fmt(package, datestr, c2, c1, None) + '\n'
break
html += '</pre>\n'
html += '</body></html>\n'
return html
def staleReport():
html = '<html><head><title>Stale report</title></head><body>\n'
html += '<h1>Stale report</h1>\n'
html += '<pre>\n'
html += '<b>' + fmt('Package', 'Date Time', None, None, None) + '</b>\n'
current_year = datetime.date.today().year
for filename in sorted(glob.glob(os.path.expanduser('~/daca@home/donated-results/*'))):
if not os.path.isfile(filename):
continue
for line in open(filename, 'rt'):
line = line.strip()
if line.startswith(str(current_year) + '-') or line.startswith(str(current_year - 1) + '-'):
datestr = line
else:
continue
dt = dateTimeFromStr(datestr)
diff = datetime.datetime.now() - dt
if diff.days < 30:
continue
package = filename[filename.rfind('/')+1:]
html += fmt(package, datestr, None, None, None) + '\n'
2018-08-31 14:28:01 +02:00
break
html += '</pre>\n'
2018-11-28 20:36:19 +01:00
2018-08-31 14:28:01 +02:00
html += '</body></html>\n'
return html
def diffReportFromDict(out, today):
html = '<pre>\n'
2018-12-08 11:39:44 +01:00
html += '<b>MessageID ' + OLD_VERSION + ' Head</b>\n'
sum0 = 0
sum1 = 0
for messageId in sorted(out.keys()):
line = messageId + ' '
counts = out[messageId]
sum0 += counts[0]
sum1 += counts[1]
if counts[0] > 0:
c = str(counts[0])
while len(line) < 40 - len(c):
line += ' '
line += c + ' '
if counts[1] > 0:
c = str(counts[1])
while len(line) < 48 - len(c):
line += ' '
line += c
line = '<a href="diff' + today + '-' + messageId + '">' + messageId + '</a>' + line[line.find(' '):]
html += line + '\n'
# Sum
html += '================================================\n'
line = ''
while len(line) < 40 - len(str(sum0)):
line += ' '
line += str(sum0) + ' '
while len(line) < 48 - len(str(sum1)):
line += ' '
line += str(sum1)
html += line + '\n'
html += '</pre>\n'
return html
def diffReport(resultsPath):
out = {}
outToday = {}
today = strDateTime()[:10]
for filename in sorted(glob.glob(resultsPath + '/*.diff')):
if not os.path.isfile(filename):
continue
with open(filename, 'rt') as f:
data = json.loads(f.read())
uploadedToday = data['date'] == today
2019-06-30 14:14:02 +02:00
for messageId in data['sums']:
sums = data['sums'][messageId]
if messageId not in out:
out[messageId] = [0, 0]
out[messageId][0] += sums[OLD_VERSION]
out[messageId][1] += sums['head']
if uploadedToday:
if messageId not in outToday:
outToday[messageId] = [0, 0]
outToday[messageId][0] += sums[OLD_VERSION]
outToday[messageId][1] += sums['head']
html = '<html><head><title>Diff report</title></head><body>\n'
html += '<h1>Diff report</h1>\n'
html += '<h2>Uploaded today</h2>'
html += diffReportFromDict(outToday, 'today')
html += '<h2>All</h2>'
html += diffReportFromDict(out, '')
return html
def generate_package_diff_statistics(filename):
is_diff = False
sums = {}
for line in open(filename, 'rt'):
line = line.strip()
if line == 'diff:':
is_diff = True
continue
elif not is_diff:
continue
if not line.endswith(']'):
continue
version = None
if line.startswith(OLD_VERSION + ' '):
version = OLD_VERSION
elif line.startswith('head '):
version = 'head'
else:
continue
messageId = line[line.rfind('[')+1:len(line)-1]
if messageId not in sums:
sums[messageId] = { OLD_VERSION: 0, 'head': 0 }
sums[messageId][version] += 1
output = { 'date': strDateTime()[:10], 'sums': sums }
filename_diff = filename + '.diff'
if sums:
with open(filename_diff, 'wt') as f:
f.write(json.dumps(output))
elif os.path.isfile(filename_diff):
os.remove(filename_diff)
2018-08-27 18:44:17 +02:00
def diffMessageIdReport(resultPath, messageId):
text = messageId + '\n'
e = '[' + messageId + ']\n'
for filename in sorted(glob.glob(resultPath + '/*.diff')):
if not os.path.isfile(filename):
continue
with open(filename, 'rt') as f:
diff_stats = f.read()
if not messageId in diff_stats:
continue
url = None
diff = False
for line in open(filename[:-5], 'rt'):
if line.startswith('ftp://'):
url = line
elif line == 'diff:\n':
diff = True
elif not diff:
continue
elif line.endswith(e):
if url:
text += url
url = None
text += line
return text
def diffMessageIdTodayReport(resultPath, messageId):
text = messageId + '\n'
e = '[' + messageId + ']\n'
today = strDateTime()[:10]
for filename in sorted(glob.glob(resultPath + '/*.diff')):
if not os.path.isfile(filename):
continue
with open(filename, 'rt') as f:
diff_stats = f.read()
if not messageId in diff_stats:
continue
if not today in diff_stats:
continue
url = None
diff = False
firstLine = True
for line in open(filename[:-5], 'rt'):
if firstLine:
firstLine = False
if not line.startswith(today):
break
if line.startswith('ftp://'):
url = line
elif line == 'diff:\n':
diff = True
elif not diff:
continue
elif line.endswith(e):
if url:
text += url
url = None
text += line
return text
def headReportFromDict(out, today):
html = '<pre>\n'
html += '<b>MessageID Count</b>\n'
sumTotal = 0
for messageId in sorted(out.keys()):
line = messageId + ' '
counts = out[messageId]
sumTotal += counts
if counts > 0:
c = str(counts)
while len(line) < 48 - len(c):
line += ' '
line += c + ' '
line = '<a href="head' + today + '-' + messageId + '">' + messageId + '</a>' + line[line.find(' '):]
html += line + '\n'
# Sum
html += '================================================\n'
line = ''
while len(line) < 48 - len(str(sumTotal)):
line += ' '
line += str(sumTotal) + ' '
html += line + '\n'
html += '</pre>\n'
return html
def headReport(resultsPath):
out = {}
outToday = {}
today = strDateTime()[:10]
for filename in sorted(glob.glob(resultsPath + '/*')):
if not os.path.isfile(filename):
continue
uploadedToday = False
firstLine = True
headResults = False
for line in open(filename, 'rt'):
if firstLine:
if line.startswith(today):
uploadedToday = True
firstLine = False
continue
line = line.strip()
if line.startswith('head results:'):
headResults = True
continue
if line.startswith('diff:'):
if headResults:
break
if not headResults:
continue
if not line.endswith(']'):
continue
if ': note: ' in line:
# notes normally do not contain message ids but can end with ']'
continue
message_id_start_pos = line.rfind('[')
if message_id_start_pos <= 0:
continue
messageId = line[message_id_start_pos+1:len(line)-1]
if ' ' in messageId:
# skip invalid messageIds
continue
if messageId not in out:
out[messageId] = 0
out[messageId] += 1
if uploadedToday:
if messageId not in outToday:
outToday[messageId] = 0
outToday[messageId] += 1
html = '<html><head><title>HEAD report</title></head><body>\n'
html += '<h1>HEAD report</h1>\n'
html += '<h2>Uploaded today</h2>'
html += headReportFromDict(outToday, 'today')
html += '<h2>All</h2>'
html += headReportFromDict(out, '')
return html
def headMessageIdReport(resultPath, messageId):
text = messageId + '\n'
e = '[' + messageId + ']\n'
for filename in sorted(glob.glob(resultPath + '/*')):
if not os.path.isfile(filename):
continue
url = None
headResults = False
for line in open(filename, 'rt'):
if line.startswith('ftp://'):
url = line
elif line.startswith('head results:'):
headResults = True
elif not headResults:
continue
elif headResults and line.startswith('diff:'):
break
elif line.endswith(e):
if url:
text += url
url = None
text += line
return text
def headMessageIdTodayReport(resultPath, messageId):
text = messageId + '\n'
e = '[' + messageId + ']\n'
today = strDateTime()[:10]
for filename in sorted(glob.glob(resultPath + '/*')):
if not os.path.isfile(filename):
continue
url = None
headResults = False
firstLine = True
for line in open(filename, 'rt'):
if firstLine:
firstLine = False
if not line.startswith(today):
break
if line.startswith('ftp://'):
url = line
elif line.startswith('head results:'):
headResults = True
elif not headResults:
continue
elif headResults and line.startswith('diff:'):
break
elif line.endswith(e):
if url:
text += url
url = None
text += line
return text
2018-09-02 09:35:38 +02:00
def timeReport(resultPath):
html = '<html><head><title>Time report</title></head><body>\n'
html += '<h1>Time report</h1>\n'
html += '<pre>\n'
column_widths = [25, 10, 10, 10]
html += '<b>'
html += 'Package '.ljust(column_widths[0]) + ' ' + \
OLD_VERSION.rjust(column_widths[1]) + ' ' + \
'Head'.rjust(column_widths[2]) + ' ' + \
'Factor'.rjust(column_widths[3])
html += '</b>\n'
total_time_base = 0.0
total_time_head = 0.0
2018-09-02 09:35:38 +02:00
for filename in glob.glob(resultPath + '/*'):
if not os.path.isfile(filename):
continue
for line in open(filename, 'rt'):
2018-09-02 09:35:38 +02:00
if not line.startswith('elapsed-time:'):
continue
split_line = line.strip().split()
time_base = float(split_line[2])
time_head = float(split_line[1])
total_time_base += time_base
total_time_head += time_head
suspicious_time_difference = False
if time_base > 1 and time_base*2 < time_head:
suspicious_time_difference = True
elif time_head > 1 and time_head*2 < time_base:
suspicious_time_difference = True
if suspicious_time_difference:
if time_base > 0.0:
time_factor = time_head / time_base
else:
time_factor = 0.0
html += filename[len(resultPath)+1:].ljust(column_widths[0]) + ' ' + \
split_line[2].rjust(column_widths[1]) + ' ' + \
split_line[1].rjust(column_widths[2]) + ' ' + \
'{:.2f}'.format(time_factor).rjust(column_widths[3]) + '\n'
2018-09-02 09:35:38 +02:00
break
html += '\n'
if total_time_base > 0.0:
total_time_factor = total_time_head / total_time_base
else:
total_time_factor = 0.0
html += 'Time for all packages (not just the ones listed above):\n'
html += 'Total time: '.ljust(column_widths[0]) + ' ' + \
'{:.1f}'.format(total_time_base).rjust(column_widths[1]) + ' ' + \
'{:.1f}'.format(total_time_head).rjust(column_widths[2]) + ' ' + \
'{:.2f}'.format(total_time_factor).rjust(column_widths[3])
html += '\n'
html += '</pre>\n'
html += '</body></html>\n'
return html
2018-09-02 09:35:38 +02:00
def check_library_report(result_path, message_id):
if message_id not in ('checkLibraryNoReturn', 'checkLibraryFunction', 'checkLibraryUseIgnore'):
error_message = 'Invalid value ' + message_id + ' for message_id parameter.'
print(error_message)
return error_message
functions_shown_max = 50000
html = '<html><head><title>' + message_id + ' report</title></head><body>\n'
html += '<h1>' + message_id + ' report</h1>\n'
html += 'Top ' + str(functions_shown_max) + ' functions are shown.'
html += '<pre>\n'
column_widths = [10, 100]
html += '<b>'
html += 'Count'.rjust(column_widths[0]) + ' ' + \
'Function'
html += '</b>\n'
function_counts = {}
for filename in glob.glob(result_path + '/*'):
if not os.path.isfile(filename):
continue
info_messages = False
for line in open(filename, 'rt'):
if line == 'info messages:\n':
info_messages = True
if not info_messages:
continue
if line.endswith('[' + message_id + ']\n'):
if message_id is 'checkLibraryFunction':
function_name = line[(line.find('for function ') + len('for function ')):line.rfind('[') - 1]
else:
function_name = line[(line.find(': Function ') + len(': Function ')):line.rfind('should have') - 1]
function_counts[function_name] = function_counts.setdefault(function_name, 0) + 1
function_details_list = []
for function_name, count in sorted(function_counts.items(), key=operator.itemgetter(1), reverse=True):
if len(function_details_list) >= functions_shown_max:
break
function_details_list.append(str(count).rjust(column_widths[0]) + ' ' +
'<a href="check_library-' + urllib.quote_plus(function_name) + '">' + function_name + '</a>\n')
html += ''.join(function_details_list)
html += '</pre>\n'
html += '</body></html>\n'
return html
# Lists all checkLibrary* messages regarding the given function name
def check_library_function_name(result_path, function_name):
print('check_library_function_name')
function_name = urllib.unquote_plus(function_name)
output_lines_list = []
for filename in glob.glob(result_path + '/*'):
if not os.path.isfile(filename):
continue
info_messages = False
url = None
cppcheck_options = None
for line in open(filename, 'rt'):
if line.startswith('ftp://'):
url = line
elif line.startswith('cppcheck-options:'):
cppcheck_options = line
elif line == 'info messages:\n':
info_messages = True
if not info_messages:
continue
if '[checkLibrary' in line:
if (' ' + function_name) in line:
if url:
output_lines_list.append(url)
url = None
if cppcheck_options:
output_lines_list.append(cppcheck_options)
cppcheck_options = None
output_lines_list.append(line)
return ''.join(output_lines_list)
2018-08-25 10:25:05 +02:00
def sendAll(connection, data):
while data:
num = connection.send(data)
if num < len(data):
data = data[num:]
2018-08-25 10:25:05 +02:00
else:
data = None
def httpGetResponse(connection, data, contentType):
2018-08-26 11:17:18 +02:00
resp = 'HTTP/1.1 200 OK\r\n'
resp += 'Connection: close\r\n'
resp += 'Content-length: ' + str(len(data)) + '\r\n'
resp += 'Content-type: ' + contentType + '\r\n\r\n'
2018-08-25 10:59:49 +02:00
resp += data
sendAll(connection, resp)
class HttpClientThread(Thread):
def __init__(self, connection, cmd, resultPath, latestResults):
Thread.__init__(self)
self.connection = connection
self.cmd = cmd[:cmd.find('\n')]
self.resultPath = resultPath
self.latestResults = latestResults
def run(self):
try:
cmd = self.cmd
print('[' + strDateTime() + '] ' + cmd)
res = re.match(r'GET /([a-zA-Z0-9_\-\.\+%]*) HTTP', cmd)
if res is None:
self.connection.close()
return
url = res.group(1)
if url == '':
html = overviewReport()
httpGetResponse(self.connection, html, 'text/html')
elif url == 'latest.html':
html = latestReport(self.latestResults)
httpGetResponse(self.connection, html, 'text/html')
elif url == 'crash.html':
2018-08-31 14:28:01 +02:00
html = crashReport()
httpGetResponse(self.connection, html, 'text/html')
elif url == 'stale.html':
html = staleReport()
httpGetResponse(self.connection, html, 'text/html')
elif url == 'diff.html':
html = diffReport(self.resultPath)
httpGetResponse(self.connection, html, 'text/html')
elif url.startswith('difftoday-'):
messageId = url[10:]
text = diffMessageIdTodayReport(self.resultPath, messageId)
httpGetResponse(self.connection, text, 'text/plain')
elif url.startswith('diff-'):
messageId = url[5:]
2018-08-27 18:44:17 +02:00
text = diffMessageIdReport(self.resultPath, messageId)
httpGetResponse(self.connection, text, 'text/plain')
elif url == 'head.html':
html = headReport(self.resultPath)
httpGetResponse(self.connection, html, 'text/html')
elif url.startswith('headtoday-'):
messageId = url[10:]
text = headMessageIdTodayReport(self.resultPath, messageId)
httpGetResponse(self.connection, text, 'text/plain')
elif url.startswith('head-'):
messageId = url[5:]
text = headMessageIdReport(self.resultPath, messageId)
httpGetResponse(self.connection, text, 'text/plain')
elif url == 'time.html':
2018-09-02 09:35:38 +02:00
text = timeReport(self.resultPath)
httpGetResponse(self.connection, text, 'text/html')
elif url == 'check_library_function_report.html':
text = check_library_report(self.resultPath + '/' + 'info_output', message_id='checkLibraryFunction')
httpGetResponse(self.connection, text, 'text/html')
elif url == 'check_library_noreturn_report.html':
text = check_library_report(self.resultPath + '/' + 'info_output', message_id='checkLibraryNoReturn')
httpGetResponse(self.connection, text, 'text/html')
elif url == 'check_library_use_ignore_report.html':
text = check_library_report(self.resultPath + '/' + 'info_output', message_id='checkLibraryUseIgnore')
httpGetResponse(self.connection, text, 'text/html')
elif url.startswith('check_library-'):
print('check library function !')
function_name = url[len('check_library-'):]
text = check_library_function_name(self.resultPath + '/' + 'info_output', function_name)
httpGetResponse(self.connection, text, 'text/plain')
else:
filename = resultPath + '/' + url
if not os.path.isfile(filename):
print('HTTP/1.1 404 Not Found')
self.connection.send('HTTP/1.1 404 Not Found\r\n\r\n')
else:
f = open(filename, 'rt')
data = f.read()
f.close()
httpGetResponse(self.connection, data, 'text/plain')
2018-08-26 11:17:18 +02:00
finally:
time.sleep(1)
self.connection.close()
def server(server_address_port, packages, packageIndex, resultPath):
socket.setdefaulttimeout(30)
sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
sock.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
server_address = ('', server_address_port)
sock.bind(server_address)
sock.listen(1)
latestResults = []
if os.path.isfile('latest.txt'):
with open('latest.txt', 'rt') as f:
latestResults = f.read().strip().split(' ')
print('[' + strDateTime() + '] version ' + SERVER_VERSION)
print('[' + strDateTime() + '] listening on port ' + str(server_address_port))
while True:
# wait for a connection
print('[' + strDateTime() + '] waiting for a connection')
connection, client_address = sock.accept()
2018-08-25 20:00:04 +02:00
try:
cmd = connection.recv(128)
2018-08-25 20:43:20 +02:00
except socket.error:
connection.close()
2018-08-25 20:00:04 +02:00
continue
if cmd.find('\n') < 1:
continue
firstLine = cmd[:cmd.find('\n')]
if re.match('[a-zA-Z0-9./ ]+', firstLine) is None:
connection.close()
continue
if cmd.startswith('GET /'):
newThread = HttpClientThread(connection, cmd, resultPath, latestResults)
newThread.start()
elif cmd == 'GetCppcheckVersions\n':
2018-12-08 11:39:44 +01:00
reply = 'head ' + OLD_VERSION
print('[' + strDateTime() + '] GetCppcheckVersions: ' + reply)
connection.send(reply)
connection.close()
elif cmd == 'get\n':
pkg = packages[packageIndex].strip()
packages[packageIndex] = pkg
packageIndex += 1
if packageIndex >= len(packages):
packageIndex = 0
f = open('package-index.txt', 'wt')
f.write(str(packageIndex) + '\n')
f.close()
print('[' + strDateTime() + '] get:' + pkg)
connection.send(pkg)
connection.close()
elif cmd.startswith('write\nftp://'):
# read data
data = cmd[cmd.find('ftp'):]
try:
t = 0
max_data_size = 2 * 1024 * 1024
while (len(data) < max_data_size) and (not data.endswith('\nDONE')) and (t < 10):
d = connection.recv(1024)
if d:
t = 0
data += d
else:
time.sleep(0.2)
t += 0.2
connection.close()
except socket.error as e:
pass
pos = data.find('\n')
if pos < 10:
continue
url = data[:pos]
2018-08-26 16:47:20 +02:00
print('[' + strDateTime() + '] write:' + url)
# save data
res = re.match(r'ftp://.*pool/main/[^/]+/([^/]+)/[^/]*tar.(gz|bz2)', url)
if res is None:
print('results not written. res is None.')
continue
if url not in packages:
url2 = url + '\n'
if url2 not in packages:
print('results not written. url is not in packages.')
continue
# Verify that head was compared to correct OLD_VERSION
versions_found = False
old_version_wrong = False
for line in data.split('\n', 20):
if line.startswith('cppcheck: '):
versions_found = True
if OLD_VERSION not in line.split():
print('Compared to wrong old version. Should be ' + OLD_VERSION + '. Versions compared: ' +
line)
print('Ignoring data.')
old_version_wrong = True
break
if not versions_found:
print('Cppcheck versions missing in result data. Ignoring data.')
continue
if old_version_wrong:
continue
print('results added for package ' + res.group(1))
filename = os.path.join(resultPath, res.group(1))
with open(filename, 'wt') as f:
f.write(strDateTime() + '\n' + data)
# track latest added results..
if len(latestResults) >= 20:
latestResults = latestResults[1:]
latestResults.append(filename)
with open('latest.txt', 'wt') as f:
f.write(' '.join(latestResults))
# generate package.diff..
generate_package_diff_statistics(filename)
elif cmd.startswith('write_info\nftp://'):
# read data
data = cmd[11:]
try:
t = 0
max_data_size = 1024 * 1024
while (len(data) < max_data_size) and (not data.endswith('\nDONE')) and (t < 10):
d = connection.recv(1024)
if d:
t = 0
data += d
else:
time.sleep(0.2)
t += 0.2
connection.close()
except socket.error as e:
pass
pos = data.find('\n')
if pos < 10:
continue
url = data[:pos]
print('[' + strDateTime() + '] write_info:' + url)
# save data
res = re.match(r'ftp://.*pool/main/[^/]+/([^/]+)/[^/]*tar.(gz|bz2)', url)
if res is None:
print('info output not written. res is None.')
continue
if url not in packages:
url2 = url + '\n'
if url2 not in packages:
print('info output not written. url is not in packages.')
continue
print('adding info output for package ' + res.group(1))
info_path = resultPath + '/' + 'info_output'
if not os.path.exists(info_path):
os.mkdir(info_path)
filename = info_path + '/' + res.group(1)
with open(filename, 'wt') as f:
f.write(strDateTime() + '\n' + data)
2018-08-23 22:13:53 +02:00
else:
print('[' + strDateTime() + '] invalid command: ' + firstLine)
connection.close()
if __name__ == "__main__":
workPath = os.path.expanduser('~/daca@home')
os.chdir(workPath)
resultPath = workPath + '/donated-results'
f = open('packages.txt', 'rt')
packages = f.readlines()
f.close()
print('packages: ' + str(len(packages)))
if len(packages) == 0:
print('fatal: there are no packages')
sys.exit(1)
packageIndex = 0
if os.path.isfile('package-index.txt'):
f = open('package-index.txt', 'rt')
packageIndex = int(f.read())
if packageIndex < 0 or packageIndex >= len(packages):
packageIndex = 0
f.close()
server_address_port = 8000
if '--test' in sys.argv[1:]:
server_address_port = 8001
try:
server(server_address_port, packages, packageIndex, resultPath)
except socket.timeout:
print('Timeout!')