All patches and comments are welcome. Please squash your changes to logical
commits before using git-format-patch and git-send-email to
patches@git.madduck.net.
If you'd read over the Git project's submission guidelines and adhered to them,
I'd be especially grateful.
1 from __future__ import print_function
10 DATE_FORMAT = '%Y%m%dT%H%M%SZ'
13 COMPLETED = 'completed'
15 VERSION_2_1_0 = six.u('2.1.0')
16 VERSION_2_2_0 = six.u('2.2.0')
17 VERSION_2_3_0 = six.u('2.3.0')
18 VERSION_2_4_0 = six.u('2.4.0')
20 logger = logging.getLogger(__name__)
23 class TaskWarriorException(Exception):
27 class SerializingObject(object):
29 Common ancestor for TaskResource & TaskFilter, since they both
30 need to serialize arguments.
33 def _deserialize(self, key, value):
34 hydrate_func = getattr(self, 'deserialize_{0}'.format(key),
35 lambda x: x if x != '' else None)
36 return hydrate_func(value)
38 def _serialize(self, key, value):
39 dehydrate_func = getattr(self, 'serialize_{0}'.format(key),
40 lambda x: x if x is not None else '')
41 return dehydrate_func(value)
43 def timestamp_serializer(self, date):
46 return date.strftime(DATE_FORMAT)
48 def timestamp_deserializer(self, date_str):
51 return datetime.datetime.strptime(date_str, DATE_FORMAT)
53 def serialize_entry(self, value):
54 return self.timestamp_serializer(value)
56 def deserialize_entry(self, value):
57 return self.timestamp_deserializer(value)
59 def serialize_modified(self, value):
60 return self.timestamp_serializer(value)
62 def deserialize_modified(self, value):
63 return self.timestamp_deserializer(value)
65 def serialize_due(self, value):
66 return self.timestamp_serializer(value)
68 def deserialize_due(self, value):
69 return self.timestamp_deserializer(value)
71 def serialize_scheduled(self, value):
72 return self.timestamp_serializer(value)
74 def deserialize_scheduled(self, value):
75 return self.timestamp_deserializer(value)
77 def serialize_until(self, value):
78 return self.timestamp_serializer(value)
80 def deserialize_until(self, value):
81 return self.timestamp_deserializer(value)
83 def serialize_wait(self, value):
84 return self.timestamp_serializer(value)
86 def deserialize_wait(self, value):
87 return self.timestamp_deserializer(value)
89 def deserialize_annotations(self, data):
90 return [TaskAnnotation(self, d) for d in data] if data else []
92 def serialize_tags(self, tags):
93 return ','.join(tags) if tags else ''
95 def deserialize_tags(self, tags):
96 if isinstance(tags, basestring):
97 return tags.split(',') if tags else []
100 def serialize_depends(self, cur_dependencies):
101 # Return the list of uuids
102 return ','.join(task['uuid'] for task in cur_dependencies)
104 def deserialize_depends(self, raw_uuids):
105 raw_uuids = raw_uuids or '' # Convert None to empty string
106 uuids = raw_uuids.split(',')
107 return set(self.warrior.tasks.get(uuid=uuid) for uuid in uuids if uuid)
110 class TaskResource(SerializingObject):
111 read_only_fields = []
113 def _load_data(self, data):
114 self._data = dict((key, self._deserialize(key, value))
115 for key, value in data.items())
116 # We need to use a copy for original data, so that changes
117 # are not propagated.
118 self._original_data = copy.deepcopy(self._data)
120 def __getitem__(self, key):
121 # This is a workaround to make TaskResource non-iterable
122 # over simple index-based iteration
129 return self._data.get(key) or self._deserialize(key, None)
131 def __setitem__(self, key, value):
132 if key in self.read_only_fields:
133 raise RuntimeError('Field \'%s\' is read-only' % key)
134 self._data[key] = value
137 s = six.text_type(self.__unicode__())
139 s = s.encode('utf-8')
146 class TaskAnnotation(TaskResource):
147 read_only_fields = ['entry', 'description']
149 def __init__(self, task, data={}):
151 self._load_data(data)
154 self.task.remove_annotation(self)
156 def __unicode__(self):
157 return self['description']
159 __repr__ = __unicode__
162 class Task(TaskResource):
163 read_only_fields = ['id', 'entry', 'urgency', 'uuid', 'modified']
165 class DoesNotExist(Exception):
168 class CompletedTask(Exception):
170 Raised when the operation cannot be performed on the completed task.
174 class DeletedTask(Exception):
176 Raised when the operation cannot be performed on the deleted task.
180 class NotSaved(Exception):
182 Raised when the operation cannot be performed on the task, because
183 it has not been saved to TaskWarrior yet.
187 def __init__(self, warrior, **kwargs):
188 self.warrior = warrior
190 # Check that user is not able to set read-only value in __init__
191 for key in kwargs.keys():
192 if key in self.read_only_fields:
193 raise RuntimeError('Field \'%s\' is read-only' % key)
195 # We serialize the data in kwargs so that users of the library
196 # do not have to pass different data formats via __setitem__ and
197 # __init__ methods, that would be confusing
199 # Rather unfortunate syntax due to python2.6 comaptiblity
200 self._load_data(dict((key, self._serialize(key, value))
201 for (key, value) in six.iteritems(kwargs)))
203 def __unicode__(self):
204 return self['description']
206 def __eq__(self, other):
207 if self['uuid'] and other['uuid']:
208 # For saved Tasks, just define equality by equality of uuids
209 return self['uuid'] == other['uuid']
211 # If the tasks are not saved, compare the actual instances
212 return id(self) == id(other)
217 # For saved Tasks, just define equality by equality of uuids
218 return self['uuid'].__hash__()
220 # If the tasks are not saved, return hash of instance id
221 return id(self).__hash__()
224 def _modified_fields(self):
225 writable_fields = set(self._data.keys()) - set(self.read_only_fields)
226 for key in writable_fields:
227 if self._data.get(key) != self._original_data.get(key):
231 def _is_modified(self):
232 return bool(list(self._modified_fields))
236 return self['status'] == six.text_type('completed')
240 return self['status'] == six.text_type('deleted')
244 return self['status'] == six.text_type('waiting')
248 return self['status'] == six.text_type('pending')
252 return self['uuid'] is not None or self['id'] is not None
254 def serialize_depends(self, cur_dependencies):
255 # Check that all the tasks are saved
256 for task in cur_dependencies:
258 raise Task.NotSaved('Task \'%s\' needs to be saved before '
259 'it can be set as dependency.' % task)
261 return super(Task, self).serialize_depends(cur_dependencies)
263 def format_depends(self):
264 # We need to generate added and removed dependencies list,
265 # since Taskwarrior does not accept redefining dependencies.
267 # This cannot be part of serialize_depends, since we need
268 # to keep a list of all depedencies in the _data dictionary,
269 # not just currently added/removed ones
271 old_dependencies = self._original_data.get('depends', set())
273 added = self['depends'] - old_dependencies
274 removed = old_dependencies - self['depends']
276 # Removed dependencies need to be prefixed with '-'
277 return 'depends:' + ','.join(
278 [t['uuid'] for t in added] +
279 ['-' + t['uuid'] for t in removed]
282 def format_description(self):
283 # Task version older than 2.4.0 ignores first word of the
284 # task description if description: prefix is used
285 if self.warrior.version < VERSION_2_4_0:
286 return self._data['description']
288 return "description:'{0}'".format(self._data['description'] or '')
292 raise Task.NotSaved("Task needs to be saved before it can be deleted")
294 # Refresh the status, and raise exception if the task is deleted
298 raise Task.DeletedTask("Task was already deleted")
300 self.warrior.execute_command([self['uuid'], 'delete'])
302 # Refresh the status again, so that we have updated info stored
308 raise Task.NotSaved("Task needs to be saved before it can be completed")
310 # Refresh, and raise exception if task is already completed/deleted
314 raise Task.CompletedTask("Cannot complete a completed task")
316 raise Task.DeletedTask("Deleted task cannot be completed")
318 self.warrior.execute_command([self['uuid'], 'done'])
320 # Refresh the status again, so that we have updated info stored
324 if self.saved and not self._is_modified:
327 args = [self['uuid'], 'modify'] if self.saved else ['add']
328 args.extend(self._get_modified_fields_as_args())
329 output = self.warrior.execute_command(args)
331 # Parse out the new ID, if the task is being added for the first time
333 id_lines = [l for l in output if l.startswith('Created task ')]
335 # Complain loudly if it seems that more tasks were created
337 if len(id_lines) != 1 or len(id_lines[0].split(' ')) != 3:
338 raise TaskWarriorException("Unexpected output when creating "
339 "task: %s" % '\n'.join(id_lines))
341 # Circumvent the ID storage, since ID is considered read-only
342 self._data['id'] = int(id_lines[0].split(' ')[2].rstrip('.'))
346 def add_annotation(self, annotation):
348 raise Task.NotSaved("Task needs to be saved to add annotation")
350 args = [self['uuid'], 'annotate', annotation]
351 self.warrior.execute_command(args)
354 def remove_annotation(self, annotation):
356 raise Task.NotSaved("Task needs to be saved to remove annotation")
358 if isinstance(annotation, TaskAnnotation):
359 annotation = annotation['description']
360 args = [self['uuid'], 'denotate', annotation]
361 self.warrior.execute_command(args)
364 def _get_modified_fields_as_args(self):
367 def add_field(field):
368 # Add the output of format_field method to args list (defaults to
370 serialized_value = self._serialize(field, self._data[field]) or ''
371 format_default = lambda: "{0}:'{1}'".format(field, serialized_value)
372 format_func = getattr(self, 'format_{0}'.format(field),
374 args.append(format_func())
376 # If we're modifying saved task, simply pass on all modified fields
378 for field in self._modified_fields:
380 # For new tasks, pass all fields that make sense
382 for field in self._data.keys():
383 if field in self.read_only_fields:
390 # Raise error when trying to refresh a task that has not been saved
392 raise Task.NotSaved("Task needs to be saved to be refreshed")
394 # We need to use ID as backup for uuid here for the refreshes
395 # of newly saved tasks. Any other place in the code is fine
396 # with using UUID only.
397 args = [self['uuid'] or self['id'], 'export']
398 new_data = json.loads(self.warrior.execute_command(args)[0])
399 self._load_data(new_data)
402 class TaskFilter(SerializingObject):
404 A set of parameters to filter the task list with.
407 def __init__(self, filter_params=[]):
408 self.filter_params = filter_params
410 def add_filter(self, filter_str):
411 self.filter_params.append(filter_str)
413 def add_filter_param(self, key, value):
414 key = key.replace('__', '.')
416 # Replace the value with empty string, since that is the
417 # convention in TW for empty values
418 attribute_key = key.split('.')[0]
419 value = self._serialize(attribute_key, value)
421 # If we are filtering by uuid:, do not use uuid keyword
424 self.filter_params.insert(0, value)
426 # Surround value with aphostrophes unless it's a empty string
427 value = "'%s'" % value if value else ''
429 # We enforce equality match by using 'is' (or 'none') modifier
430 # Without using this syntax, filter fails due to TW-1479
431 modifier = '.is' if value else '.none'
432 key = key + modifier if '.' not in key else key
434 self.filter_params.append("{0}:{1}".format(key, value))
436 def get_filter_params(self):
437 return [f for f in self.filter_params if f]
441 c.filter_params = list(self.filter_params)
445 class TaskQuerySet(object):
447 Represents a lazy lookup for a task objects.
450 def __init__(self, warrior=None, filter_obj=None):
451 self.warrior = warrior
452 self._result_cache = None
453 self.filter_obj = filter_obj or TaskFilter()
455 def __deepcopy__(self, memo):
457 Deep copy of a QuerySet doesn't populate the cache
459 obj = self.__class__()
460 for k, v in self.__dict__.items():
461 if k in ('_iter', '_result_cache'):
462 obj.__dict__[k] = None
464 obj.__dict__[k] = copy.deepcopy(v, memo)
468 data = list(self[:REPR_OUTPUT_SIZE + 1])
469 if len(data) > REPR_OUTPUT_SIZE:
470 data[-1] = "...(remaining elements truncated)..."
474 if self._result_cache is None:
475 self._result_cache = list(self)
476 return len(self._result_cache)
479 if self._result_cache is None:
480 self._result_cache = self._execute()
481 return iter(self._result_cache)
483 def __getitem__(self, k):
484 if self._result_cache is None:
485 self._result_cache = list(self)
486 return self._result_cache.__getitem__(k)
489 if self._result_cache is not None:
490 return bool(self._result_cache)
493 except StopIteration:
497 def __nonzero__(self):
498 return type(self).__bool__(self)
500 def _clone(self, klass=None, **kwargs):
502 klass = self.__class__
503 filter_obj = self.filter_obj.clone()
504 c = klass(warrior=self.warrior, filter_obj=filter_obj)
505 c.__dict__.update(kwargs)
510 Fetch the tasks which match the current filters.
512 return self.warrior.filter_tasks(self.filter_obj)
516 Returns a new TaskQuerySet that is a copy of the current one.
521 return self.filter(status=PENDING)
524 return self.filter(status=COMPLETED)
526 def filter(self, *args, **kwargs):
528 Returns a new TaskQuerySet with the given filters added.
530 clone = self._clone()
532 clone.filter_obj.add_filter(f)
533 for key, value in kwargs.items():
534 clone.filter_obj.add_filter_param(key, value)
537 def get(self, **kwargs):
539 Performs the query and returns a single object matching the given
542 clone = self.filter(**kwargs)
545 return clone._result_cache[0]
547 raise Task.DoesNotExist(
548 'Task matching query does not exist. '
549 'Lookup parameters were {0}'.format(kwargs))
551 'get() returned more than one Task -- it returned {0}! '
552 'Lookup parameters were {1}'.format(num, kwargs))
555 class TaskWarrior(object):
556 def __init__(self, data_location='~/.task', create=True):
557 data_location = os.path.expanduser(data_location)
558 if create and not os.path.exists(data_location):
559 os.makedirs(data_location)
561 'data.location': os.path.expanduser(data_location),
562 'confirmation': 'no',
563 'dependency.confirmation': 'no', # See TW-1483 or taskrc man page
565 self.tasks = TaskQuerySet(self)
566 self.version = self._get_version()
568 def _get_command_args(self, args, config_override={}):
569 command_args = ['task', 'rc:/']
570 config = self.config.copy()
571 config.update(config_override)
572 for item in config.items():
573 command_args.append('rc.{0}={1}'.format(*item))
574 command_args.extend(map(str, args))
577 def _get_version(self):
578 p = subprocess.Popen(
579 ['task', '--version'],
580 stdout=subprocess.PIPE,
581 stderr=subprocess.PIPE)
582 stdout, stderr = [x.decode('utf-8') for x in p.communicate()]
583 return stdout.strip('\n')
585 def execute_command(self, args, config_override={}):
586 command_args = self._get_command_args(
587 args, config_override=config_override)
588 logger.debug(' '.join(command_args))
589 p = subprocess.Popen(command_args, stdout=subprocess.PIPE,
590 stderr=subprocess.PIPE)
591 stdout, stderr = [x.decode('utf-8') for x in p.communicate()]
594 error_msg = stderr.strip().splitlines()[-1]
596 error_msg = stdout.strip()
597 raise TaskWarriorException(error_msg)
598 return stdout.strip().split('\n')
600 def filter_tasks(self, filter_obj):
601 args = ['export', '--'] + filter_obj.get_filter_params()
603 for line in self.execute_command(args):
605 data = line.strip(',')
607 filtered_task = Task(self)
608 filtered_task._load_data(json.loads(data))
609 tasks.append(filtered_task)
611 raise TaskWarriorException('Invalid JSON: %s' % data)
614 def merge_with(self, path, push=False):
615 path = path.rstrip('/') + '/'
616 self.execute_command(['merge', path], config_override={
617 'merge.autopush': 'yes' if push else 'no',
621 self.execute_command(['undo'])