import logging
import os
import pytz
+import re
import six
import sys
import subprocess
import tzlocal
DATE_FORMAT = '%Y%m%dT%H%M%SZ'
+DATE_FORMAT_CALC = '%Y-%m-%dT%H:%M:%S'
REPR_OUTPUT_SIZE = 10
PENDING = 'pending'
COMPLETED = 'completed'
VERSION_2_2_0 = six.u('2.2.0')
VERSION_2_3_0 = six.u('2.3.0')
VERSION_2_4_0 = six.u('2.4.0')
+VERSION_2_4_1 = six.u('2.4.1')
+VERSION_2_4_2 = six.u('2.4.2')
+VERSION_2_4_3 = six.u('2.4.3')
logger = logging.getLogger(__name__)
local_zone = tzlocal.get_localzone()
to raise ValueError.
"""
+ def __init__(self, warrior):
+ self.warrior = warrior
+
def _deserialize(self, key, value):
hydrate_func = getattr(self, 'deserialize_{0}'.format(key),
lambda x: x if x != '' else None)
# Convert to local midnight
value_full = datetime.datetime.combine(value, datetime.time.min)
localized = local_zone.localize(value_full)
- elif isinstance(value, datetime.datetime) and value.tzinfo is None:
- # Convert to localized datetime object
- localized = local_zone.localize(value)
+ elif isinstance(value, datetime.datetime):
+ if value.tzinfo is None:
+ # Convert to localized datetime object
+ localized = local_zone.localize(value)
+ else:
+ # If the value is already localized, there is no need to change
+ # time zone at this point. Also None is a valid value too.
+ localized = value
+ elif (isinstance(value, six.string_types)
+ and self.warrior.version >= VERSION_2_4_0):
+ # For strings, use 'task calc' to evaluate the string to datetime
+ # available since TW 2.4.0
+ args = value.split()
+ result = self.warrior.execute_command(['calc'] + args)
+ naive = datetime.datetime.strptime(result[0], DATE_FORMAT_CALC)
+ localized = local_zone.localize(naive)
else:
- # If the value is already localized, there is no need to change
- # time zone at this point. Also None is a valid value too.
- localized = value
+ raise ValueError("Provided value could not be converted to "
+ "datetime, its type is not supported: {}"
+ .format(type(value)))
return localized
# are not propagated.
self._original_data = copy.deepcopy(self._data)
- def _update_data(self, data, update_original=False):
+ def _update_data(self, data, update_original=False, remove_missing=False):
"""
Low level update of the internal _data dict. Data which are coming as
updates should already be serialized. If update_original is True, the
self._data.update(dict((key, self._deserialize(key, value))
for key, value in data.items()))
+ # In certain situations, we want to treat missing keys as removals
+ if remove_missing:
+ for key in set(self._data.keys()) - set(data.keys()):
+ self._data[key] = None
+
if update_original:
self._original_data = copy.deepcopy(self._data)
def __init__(self, task, data={}):
self.task = task
self._load_data(data)
+ super(TaskAnnotation, self).__init__(task.warrior)
def remove(self):
self.task.remove_annotation(self)
"""
pass
+ class ActiveTask(Exception):
+ """
+ Raised when the operation cannot be performed on the active task.
+ """
+ pass
+
+ class InactiveTask(Exception):
+ """
+ Raised when the operation cannot be performed on an inactive task.
+ """
+ pass
+
class NotSaved(Exception):
"""
Raised when the operation cannot be performed on the task, because
# If this is a on-modify event, we are provided with additional
# line of input, which provides updated data
if modify:
- task._update_data(json.loads(input_file.readline().strip()))
+ task._update_data(json.loads(input_file.readline().strip()),
+ remove_missing=True)
return task
def __init__(self, warrior, **kwargs):
- self.warrior = warrior
+ super(Task, self).__init__(warrior)
# Check that user is not able to set read-only value in __init__
for key in kwargs.keys():
def pending(self):
return self['status'] == six.text_type('pending')
+ @property
+ def active(self):
+ return self['start'] is not None
+
@property
def saved(self):
return self['uuid'] is not None or self['id'] is not None
if self.warrior.version < VERSION_2_4_0:
return self._data['description']
else:
- return "description:'{0}'".format(self._data['description'] or '')
+ return six.u("description:'{0}'").format(self._data['description'] or '')
def delete(self):
if not self.saved:
raise Task.CompletedTask("Cannot start a completed task")
elif self.deleted:
raise Task.DeletedTask("Deleted task cannot be started")
+ elif self.active:
+ raise Task.ActiveTask("Task is already active")
self.warrior.execute_command([self['uuid'], 'start'])
# Refresh the status again, so that we have updated info stored
self.refresh(only_fields=['status', 'start'])
+ def stop(self):
+ if not self.saved:
+ raise Task.NotSaved("Task needs to be saved before it can be stopped")
+
+ # Refresh, and raise exception if task is already completed/deleted
+ self.refresh(only_fields=['status'])
+
+ if not self.active:
+ raise Task.InactiveTask("Cannot stop an inactive task")
+
+ self.warrior.execute_command([self['uuid'], 'stop'])
+
+ # Refresh the status again, so that we have updated info stored
+ self.refresh(only_fields=['status', 'start'])
+
def done(self):
if not self.saved:
raise Task.NotSaved("Task needs to be saved before it can be completed")
elif self.deleted:
raise Task.DeletedTask("Deleted task cannot be completed")
+ # Older versions of TW do not stop active task at completion
+ if self.warrior.version < VERSION_2_4_0 and self.active:
+ self.stop()
+
self.warrior.execute_command([self['uuid'], 'done'])
# Refresh the status again, so that we have updated info stored
if serialized_value is '':
escaped_serialized_value = ''
else:
- escaped_serialized_value = "'{0}'".format(serialized_value)
+ escaped_serialized_value = six.u("'{0}'").format(serialized_value)
- format_default = lambda: "{0}:{1}".format(field,
+ format_default = lambda: six.u("{0}:{1}").format(field,
escaped_serialized_value)
format_func = getattr(self, 'format_{0}'.format(field),
A set of parameters to filter the task list with.
"""
- def __init__(self, filter_params=[]):
+ def __init__(self, warrior, filter_params=[]):
self.filter_params = filter_params
+ super(TaskFilter, self).__init__(warrior)
def add_filter(self, filter_str):
self.filter_params.append(filter_str)
attribute_key = key.split('.')[0]
# Since this is user input, we need to normalize before we serialize
- value = self._normalize(key, value)
+ value = self._normalize(attribute_key, value)
value = self._serialize(attribute_key, value)
# If we are filtering by uuid:, do not use uuid keyword
modifier = '.is' if value else '.none'
key = key + modifier if '.' not in key else key
- self.filter_params.append("{0}:{1}".format(key, value))
+ self.filter_params.append(six.u("{0}:{1}").format(key, value))
def get_filter_params(self):
return [f for f in self.filter_params if f]
def clone(self):
- c = self.__class__()
+ c = self.__class__(self.warrior)
c.filter_params = list(self.filter_params)
return c
def __init__(self, warrior=None, filter_obj=None):
self.warrior = warrior
self._result_cache = None
- self.filter_obj = filter_obj or TaskFilter()
+ self.filter_obj = filter_obj or TaskFilter(warrior)
def __deepcopy__(self, memo):
"""
class TaskWarrior(object):
- def __init__(self, data_location='~/.task', create=True):
- data_location = os.path.expanduser(data_location)
- if create and not os.path.exists(data_location):
- os.makedirs(data_location)
+ def __init__(self, data_location=None, create=True, taskrc_location='~/.taskrc'):
+ self.taskrc_location = os.path.expanduser(taskrc_location)
+
+ # If taskrc does not exist, pass / to use defaults and avoid creating
+ # dummy .taskrc file by TaskWarrior
+ if not os.path.exists(self.taskrc_location):
+ self.taskrc_location = '/'
+
+ self.version = self._get_version()
self.config = {
- 'data.location': os.path.expanduser(data_location),
'confirmation': 'no',
'dependency.confirmation': 'no', # See TW-1483 or taskrc man page
'recurrence.confirmation': 'no', # Necessary for modifying R tasks
+ # 2.4.3 onwards supports 0 as infite bulk, otherwise set just
+ # arbitrary big number which is likely to be large enough
+ 'bulk': 0 if self.version >= VERSION_2_4_3 else 100000,
}
+
+ # Set data.location override if passed via kwarg
+ if data_location is not None:
+ data_location = os.path.expanduser(data_location)
+ if create and not os.path.exists(data_location):
+ os.makedirs(data_location)
+ self.config['data.location'] = data_location
+
self.tasks = TaskQuerySet(self)
- self.version = self._get_version()
def _get_command_args(self, args, config_override={}):
- command_args = ['task', 'rc:/']
+ command_args = ['task', 'rc:{0}'.format(self.taskrc_location)]
config = self.config.copy()
config.update(config_override)
for item in config.items():
command_args.append('rc.{0}={1}'.format(*item))
- command_args.extend(map(str, args))
+ command_args.extend(map(six.text_type, args))
return command_args
def _get_version(self):
stdout, stderr = [x.decode('utf-8') for x in p.communicate()]
return stdout.strip('\n')
- def execute_command(self, args, config_override={}, allow_failure=True):
+ def get_config(self):
+ raw_output = self.execute_command(
+ ['show'],
+ config_override={'verbose': 'nothing'}
+ )
+
+ config = dict()
+ config_regex = re.compile(r'^(?P<key>[^\s]+)\s+(?P<value>[^\s].+$)')
+
+ for line in raw_output:
+ match = config_regex.match(line)
+ if match:
+ config[match.group('key')] = match.group('value').strip()
+
+ return config
+
+ def execute_command(self, args, config_override={}, allow_failure=True,
+ return_all=False):
command_args = self._get_command_args(
args, config_override=config_override)
logger.debug(' '.join(command_args))
else:
error_msg = stdout.strip()
raise TaskWarriorException(error_msg)
- return stdout.strip().split('\n')
+
+ # Return all whole triplet only if explicitly asked for
+ if not return_all:
+ return stdout.rstrip().split('\n')
+ else:
+ return (stdout.rstrip().split('\n'),
+ stderr.rstrip().split('\n'),
+ p.returncode)
def enforce_recurrence(self):
# Run arbitrary report command which will trigger generation
# of recurrent tasks.
- # TODO: Make a version dependant enforcement once
- # TW-1531 is handled
- self.execute_command(['next'], allow_failure=False)
+
+ # Only necessary for TW up to 2.4.1, fixed in 2.4.2.
+ if self.version < VERSION_2_4_2:
+ self.execute_command(['next'], allow_failure=False)
def filter_tasks(self, filter_obj):
self.enforce_recurrence()