本文整理汇总了Python中util.Util类的典型用法代码示例。如果您正苦于以下问题:Python Util类的具体用法?Python Util怎么用?Python Util使用的例子?那么恭喜您, 这里精选的类代码示例或许可以为您提供帮助。
在下文中一共展示了Util类的15个代码示例,这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞,您的评价将有助于系统推荐出更棒的Python代码示例。
示例1: optimizeSubdirectory
def optimizeSubdirectory(self, path, callback, new_path, extension = "", minimize = False):
"""optimizes a subdirectory within a directory being optimized
Arguments:
path -- path to directory
callback -- function to run the file through
new_path -- path to optimized parent directory
extension -- extension to search for in the directory
minimize -- whether or not we should minimize the file contents (html)
Returns:
void
"""
subdir_path = new_path + "/" + path.split("/").pop()
skip = self.prepareDirectory(subdir_path)
if skip is True:
return
for dir_file in Util.getFilesFromDir(path, extension):
if Util.isDir(dir_file):
self.optimizeSubdirectory(dir_file, callback, subdir_path, extension, minimize)
continue
new_file_path = subdir_path + "/" + Util.getFileName(dir_file)
self.optimizeFile(dir_file, callback, minimize, new_file_path)
示例2: callback
def callback(self, cmd, action, target, msg):
if msg is None:
self._home.publish_msg(cmd, u"时间格式错误")
return False, None
if msg.endswith(u'点') or \
msg.endswith(u'分'):
t = Util.gap_for_timestring(msg)
elif msg.endswith(u"秒"):
t = int(Util.cn2dig(msg[:-1]))
elif msg.endswith(u"分钟"):
t = int(Util.cn2dig(msg[:-2]))*60
elif msg.endswith(u"小时"):
t = int(Util.cn2dig(msg[:-2]))*60*60
else:
self._home.publish_msg(cmd, u"时间格式错误")
return False
if t is None:
self._home.publish_msg(cmd, u"时间格式错误")
return False, None
DEBUG("thread wait for %d sec" % (t, ))
self._home.publish_msg(cmd, action + target + msg)
threading.current_thread().waitUtil(t)
if threading.current_thread().stopped():
return False
self._home.setResume(True)
count = 7
Sound.play( Res.get_res_path("sound/com_bell") , True, count)
self._home.setResume(False)
return True
示例3: parse_args
def parse_args(args_list):
log.info("Stetl version = %s" % __version__)
argparser = argparse.ArgumentParser(description='Invoke Stetl')
argparser.add_argument('-c ', '--config', type=str, help='ETL config file in .ini format', dest='config_file',
required=False)
argparser.add_argument('-s ', '--section', type=str, help='Section in the config file to execute, default is [etl]',
dest='config_section', required=False)
argparser.add_argument('-a ', '--args', type=str,
help='Arguments or .properties files to be substituted for symbolic {argN}s in Stetl config file,\
as -a "arg1=foo arg2=bar" and/or -a args.properties, multiple -a options are possible',
dest='config_args', required=False, action='append')
argparser.add_argument('-d ', '--doc', type=str,
help='Get component documentation like its configuration parameters, e.g. stetl doc stetl.inputs.fileinput.FileInput',
dest='doc_args', required=False)
args = argparser.parse_args(args_list)
if args.config_args:
args_total = dict()
for arg in args.config_args:
if os.path.isfile(arg):
log.info('Found args file at: %s' % arg)
args_total = Util.merge_two_dicts(args_total, Util.propsfile_to_dict(arg))
else:
# Convert string to dict: http://stackoverflow.com/a/1248990
args_total = Util.merge_two_dicts(args_total, Util.string_to_dict(arg))
args.config_args = args_total
return args
示例4: processMaps
def processMaps(self):
"""loops through classes and ids to process to determine shorter names to use for them
and creates a dictionary with these mappings
Returns:
void
"""
# reverse sort so we can figure out the biggest savings
classes = self.class_counter.items()
classes.sort(key = itemgetter(1), reverse=True)
for class_name, savings in classes:
small_class = "." + VarFactory.getNext("class")
# adblock extensions may block class "ad" so we should never generate it
# also if the generated class already exists as a class to be processed
# we can't use it or bad things will happen
while small_class == ".ad" or Util.keyInTupleList(small_class, classes):
small_class = "." + VarFactory.getNext("class")
self.class_map[class_name] = small_class
ids = self.id_counter.items()
ids.sort(key = itemgetter(1), reverse=True)
for id, savings in ids:
small_id = "#" + VarFactory.getNext("id")
# same holds true for ids as classes
while small_id == "#ad" or Util.keyInTupleList(small_id, ids):
small_id = "#" + VarFactory.getNext("id")
self.id_map[id] = small_id
示例5: __init__
def __init__(self, name, ip, iface, hostname="", loginfo=""):
"""
Create a new host object, with the given name, IP specification and interface
name -- Nickname for the host
ip -- IP specification for the host or subnet (e.g. "127.0.0.1 10.0.0.0/24")
iface -- Interface nickname this is connected to (only one!)
"""
# verify and store name
if name == "" and not Util.verify_name(name):
raise PyromanException("Host '%s' lacking a valid name at %s" \
% (name, iface, loginfo))
if Firewall.hosts.has_key(name):
raise PyromanException("Duplicate host specification: '%s' at %s" % (name, loginfo))
self.name = name
# verify and store IPs
if ip == "":
raise PyromanException("Host '%s' definition lacking IP address at %s" % (name, loginfo))
self.ip = Util.splitter.split(ip)
for i in self.ip:
if not Util.verify_ipnet(i):
raise PyromanException("IP specification '%s' invalid for host '%s' at %s" \
% (i, name, loginfo))
# verify and store interface
self.iface = iface
if iface == "":
raise PyromanException("Host definition '%s' lacking kernel interfaces at %s" \
% (name, loginfo))
# store "real" hostname (which may be longer than nick)
# this is used for "localhost detection"
self.hostname = hostname
# store loginfo
self.loginfo = loginfo
# register with firewall
Firewall.hosts[name] = self
示例6: __del__
def __del__(self):
Util.trace("destroy DirCheck instance")
streamRef = self._streamRef
# Stop / Invalidate / Release
FSEventStreamStop(streamRef)
FSEventStreamInvalidate(streamRef)
FSEventStreamRelease(streamRef)
示例7: _get_uid
def _get_uid(cls):
user_id = Util.get_user_id()
if not user_id:
print("welcome, creating id for you.")
user_id = requests.get(cls.get_user_id_url).text
Util.set_user_id(user_id)
return user_id
示例8: main
def main():
worm = Worm("http://11.11.0.64:8099/", "http://11.11.0.64:4041/stages/", True, False, "Flint")
# worm = Worm("http://211.69.198.208:8080/", "http://211.69.198.208:8080/history/app-20150722101951-0008/stages/", "Flint")
# finish_sparks = worm.get_finish_sparks()
running_spark = worm.get_finish_spark()
if running_spark != None:
running_stages = running_spark.get_running_stages()
finished_stages = running_spark.get_finished_stages()
stages = []
for finished_stage in finished_stages:
stage_dict = {
"stage_id": finished_stage.get_stage_id(),
"stage_duration": Util.format_time(finished_stage.get_duration()),
"submit_time": finished_stage.get_submit_time(),
"tasks_percent": 100.0,
"gc_time": round(finished_stage.get_gc_time(), 1),
}
stages.append(stage_dict)
for running_stage in running_stages:
stage_dict = {
"stage_id": running_stage.get_stage_id(),
"stage_duration": Util.format_time(running_stage.get_duration()),
"submit_time": running_stage.get_submit_time(),
"tasks_percent": Util.format_tasks_percent(running_stage.get_tasks_percent()),
"gc_time": round(running_stage.get_gc_time(), 1),
}
stages.append(stage_dict)
# print stages
format_spark = {}
format_spark["app_name"] = running_spark.get_app_name()
format_spark["total_time"] = Util.format_time(running_spark.get_total_time())
format_spark["status"] = running_spark.get_status()
format_spark["property"] = running_spark.get_property()
format_spark["stages"] = stages
print format_spark
示例9: on_search_enter_key
def on_search_enter_key(self, entry):
# find /home/shercoder/ \( ! -regex '.*/\..*' \) | grep "soccer"
search_terms = entry.get_text()
is_tag_search = False
if search_terms.startswith('@'):
search_terms = search_terms[1:]
is_tag_search = True
else:
search_terms = search_terms.split(' ')
if entry.get_text():
allfiles = []
if is_tag_search:
results = self._index_manager.search_documents(search_terms)
for hit in results:
allfiles.append((hit['filename'], hit['filepath']))
else:
for root, dirs, files in os.walk(HOME):
files = [f for f in files if not f[0] == '.']
dirs[:] = [d for d in dirs if not d[0] == '.']
for term in search_terms:
for filename in fnmatch.filter(files, "*{}*".format(term)):
allfiles.append((filename, os.path.join(root, filename)))
self._treeview.get_model().generate_search_tree(allfiles)
else:
self._treeview.get_model().generate_tree(HOME)
Util.clear_notebook(self._notebook)
示例10: kernel_version
def kernel_version(min=None, max=None):
"""
Return kernel version or test for a minimum and/or maximum version
min -- minimal kernel version required
max -- maximum kernel version required
"""
if not Firewall._kernelversion:
# query iptables version
kvcmd = subprocess.Popen(Firewall.kernelversioncmd,
stdout=subprocess.PIPE)
result = kvcmd.communicate()[0]
Firewall._kernelversion = result.strip()
# still no version number? - raise PyromanException(an exception)
if not Firewall._kernelversion:
raise Error("Couldn't get kernel version!")
if not min and not max:
return Firewall._kernelversion
if min:
if Util.compare_versions(Firewall._kernelversion, min) < 0:
return False
if max:
if Util.compare_versions(Firewall._kernelversion, max) > 0:
return False
return True
示例11: _dumpVariable
def _dumpVariable( self ):
Util.trace( "---------------------------" )
Util.trace( "index: " + str( self.index_ ) )
Util.trace( "name: " + str( self.name_ ) )
#Util.trace( "ext: " + str( self.ext_ ) )
Util.trace( "mtime: " + str( self.mtime_ ) )
Util.trace( "bgColor: " + str( self.bgColor_ ) )
示例12: build
def build(self, is_main=False, is_method=False, is_class=False):
'''
Copies the base java file, moves the changes into it and compiles it.
If an Exception occurs, the change is rolled back.
'''
shutil.copyfile(START_FILE, MAIN_FILE)
self.parse_file(MAIN_FILE)
# Attempt to compile the new change
try:
Util.compile_java([MAIN_FILE] + self.full_class_names)
except Exception as e:
if verbose_output == True:
print 'Exception: %s' % e
else:
print 'An Exception occurred compiling the program'
# Determine who caused the exception and remove the code
if is_main is True:
self.main_input.pop()
elif is_method is True:
self.methods.pop()
self.method_names.pop()
elif is_class is True:
self.classes.pop()
self.class_names.pop()
filename = self.full_class_names.pop()
Util.remove_file(filename)
return
if is_main is True:
self.run()
示例13: post
def post(self, answerId):
print 'hiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiii'
print answerId
message = self.get_argument("message")
userId = self.get_current_user_id()
answer = Answer.queryByAnswerId(answerId)
# 异常1 回答不存在
if not answer :
self.write(Util.response(None, -1, u'回复不存在'))
return
# 请输入评论内容
if message == "":
self.write(Util.response(None, -1, u'请输入评论内容'))
return
# 不能评论锁定的问题
question = Question.queryById(answer.question_id)
if question.lock == 1 :
self.write(Util.response(None, -1, u'不能评论锁定的问题'))
return
# 你没有发表评论的权限(略)
# 插入一条评论
AnswerComment.addAnswerComment(answerId, userId, message)
rsm = {'item_id':answerId, 'type_name':'answer'}
self.write(Util.response(rsm, 1, None))
示例14: update
def update():
'''
Sets target's siteurl, blogname, and homepage
'''
Util.validate_role()
config = env.config
db_table = 'wp_options'
entries = {
'siteurl': config['site_url'],
'blogname': config['site_name'],
'home': config['site_url']
}
with local_tunnel(config['db_port']):
cnx = mysql.connector.connect(user=config['db_user'],
password=config['db_pass'],
host='127.0.0.1',
port=config['db_port'],
database=config['db_name'])
cnx.start_transaction()
cursor = cnx.cursor()
update_option = ("UPDATE `{db_table}` "
"SET `option_value`=%s "
"WHERE `option_name` LIKE %s".format(db_table=db_table))
for key, value in entries.iteritems():
cursor.execute(update_option, (value, key))
cnx.commit()
cnx.close()
示例15: on_post_save
def on_post_save(self, view):
file = view.file_name()
if not (Util.is_scala(file) or Util.is_java(file)):
return
env = getEnvironment(view.window())
if env and env.is_connected() and env.client.analyzer_ready:
TypeCheckFilesReq([view.file_name()]).run_in(env, async=True)