]> git.madduck.net Git - etc/taskwarrior.git/blobdiff - tasklib/task.py

madduck's git repository

Every one of the projects in this repository is available at the canonical URL git://git.madduck.net/madduck/pub/<projectpath> — see each project's metadata for the exact URL.

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.

SSH access, as well as push access can be individually arranged.

If you use my repositories frequently, consider adding the following snippet to ~/.gitconfig and using the third clone URL listed for each project:

[url "git://git.madduck.net/madduck/"]
  insteadOf = madduck:

Task: Enumarate only writable fields as modified
[etc/taskwarrior.git] / tasklib / task.py
index 5e2b71b66a48d4b87e6ad8c27ccbb932bb76ace4..c9fafa667f2376103cdc774f8189c27e23ce74d6 100644 (file)
+from __future__ import print_function
+import copy
+import datetime
 import json
+import logging
 import os
+import six
 import subprocess
-import tempfile
-import uuid
-
 
+DATE_FORMAT = '%Y%m%dT%H%M%SZ'
+REPR_OUTPUT_SIZE = 10
 PENDING = 'pending'
+COMPLETED = 'completed'
+
+VERSION_2_1_0 = six.u('2.1.0')
+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')
+
+logger = logging.getLogger(__name__)
 
 
 class TaskWarriorException(Exception):
     pass
 
 
-class Task(object):
+class TaskResource(object):
+    read_only_fields = []
 
-    def __init__(self, warrior, data={}):
-        self.warrior = warrior
+    def _load_data(self, data):
         self._data = data
+        # We need to use a copy for original data, so that changes
+        # are not propagated
+        self._original_data = data.copy()
 
     def __getitem__(self, key):
-        return self._data.get(key)
+        # This is a workaround to make TaskResource non-iterable
+        # over simple index-based iteration
+        try:
+            int(key)
+            raise StopIteration
+        except ValueError:
+            pass
+
+        return self._deserialize(key, self._data.get(key))
+
+    def __setitem__(self, key, value):
+        if key in self.read_only_fields:
+            raise RuntimeError('Field \'%s\' is read-only' % key)
+        self._data[key] = self._serialize(key, value)
+
+    def _deserialize(self, key, value):
+        hydrate_func = getattr(self, 'deserialize_{0}'.format(key),
+                               lambda x: x)
+        return hydrate_func(value)
+
+    def _serialize(self, key, value):
+        dehydrate_func = getattr(self, 'serialize_{0}'.format(key),
+                                 lambda x: x)
+        return dehydrate_func(value)
+
+    def __str__(self):
+        s = six.text_type(self.__unicode__())
+        if not six.PY3:
+            s = s.encode('utf-8')
+        return s
+
+    def __repr__(self):
+        return str(self)
+
+
+class TaskAnnotation(TaskResource):
+    read_only_fields = ['entry', 'description']
+
+    def __init__(self, task, data={}):
+        self.task = task
+        self._load_data(data)
+
+    def deserialize_entry(self, data):
+        return datetime.datetime.strptime(data, DATE_FORMAT) if data else None
 
-    def __setitem__(self, key, val):
-        self._data[key] = val
+    def serialize_entry(self, date):
+        return date.strftime(DATE_FORMAT) if date else ''
+
+    def remove(self):
+        self.task.remove_annotation(self)
 
     def __unicode__(self):
-        return self._data.get('description')
+        return self['description']
+
+    __repr__ = __unicode__
+
+
+class Task(TaskResource):
+    read_only_fields = ['id', 'entry', 'urgency', 'uuid']
+
+    class DoesNotExist(Exception):
+        pass
+
+    class CompletedTask(Exception):
+        """
+        Raised when the operation cannot be performed on the completed task.
+        """
+        pass
 
-    def regenerate_uuid(self):
-        self['uuid'] = str(uuid.uuid4())
+    class DeletedTask(Exception):
+        """
+        Raised when the operation cannot be performed on the deleted task.
+        """
+        pass
+
+    class NotSaved(Exception):
+        """
+        Raised when the operation cannot be performed on the task, because
+        it has not been saved to TaskWarrior yet.
+        """
+        pass
+
+    def __init__(self, warrior, data={}, **kwargs):
+        self.warrior = warrior
+
+        # We serialize the data in kwargs so that users of the library
+        # do not have to pass different data formats via __setitem__ and
+        # __init__ methods, that would be confusing
+
+        # Rather unfortunate syntax due to python2.6 comaptiblity
+        self._load_data(dict((key, self._serialize(key, value))
+                        for (key, value) in six.iteritems(kwargs)))
+
+        # We keep data for backwards compatibility
+        # TODO: Should we keep this using unserialized access to _data dict?
+        self._data.update(data)
+
+    def __unicode__(self):
+        return self['description']
+
+    def __eq__(self, other):
+        if self['uuid'] and other['uuid']:
+            # For saved Tasks, just define equality by equality of uuids
+            return self['uuid'] == other['uuid']
+        else:
+            # If the tasks are not saved, compare the actual instances
+            return id(self) == id(other)
+
+
+    def __hash__(self):
+        if self['uuid']:
+            # For saved Tasks, just define equality by equality of uuids
+            return self['uuid'].__hash__()
+        else:
+            # If the tasks are not saved, return hash of instance id
+            return id(self).__hash__()
+
+    @property
+    def _modified_fields(self):
+        writable_fields = set(self._data.keys()) - set(self.read_only_fields)
+        for key in writable_fields:
+            if self._data.get(key) != self._original_data.get(key):
+                yield key
+
+    @property
+    def completed(self):
+        return self['status'] == six.text_type('completed')
+
+    @property
+    def deleted(self):
+        return self['status'] == six.text_type('deleted')
+
+    @property
+    def waiting(self):
+        return self['status'] == six.text_type('waiting')
+
+    @property
+    def pending(self):
+        return self['status'] == six.text_type('pending')
+
+    @property
+    def saved(self):
+        return self['uuid'] is not None or self['id'] is not None
+
+    def serialize_due(self, date):
+        if not date:
+            return None
+        return date.strftime(DATE_FORMAT)
+
+    def deserialize_due(self, date_str):
+        if not date_str:
+            return None
+        return datetime.datetime.strptime(date_str, DATE_FORMAT)
+
+    def serialize_depends(self, cur_dependencies):
+        # Check that all the tasks are saved
+        for task in cur_dependencies:
+            if not task.saved:
+                raise Task.NotSaved('Task \'%s\' needs to be saved before '
+                                    'it can be set as dependency.' % task)
+
+        # Return the list of uuids
+        return ','.join(task['uuid'] for task in cur_dependencies)
+
+    def deserialize_depends(self, raw_uuids):
+        raw_uuids = raw_uuids or ''  # Convert None to empty string
+        uuids = raw_uuids.split(',')
+        return set(self.warrior.tasks.get(uuid=uuid) for uuid in uuids if uuid)
+
+    def format_depends(self):
+        # We need to generate added and removed dependencies list,
+        # since Taskwarrior does not accept redefining dependencies.
+
+        # This cannot be part of serialize_depends, since we need
+        # to keep a list of all depedencies in the _data dictionary,
+        # not just currently added/removed ones
+
+        old_dependencies_raw = self._original_data.get('depends','')
+        old_dependencies = self.deserialize_depends(old_dependencies_raw)
+
+        added = self['depends'] - old_dependencies
+        removed = old_dependencies - self['depends']
+
+        # Removed dependencies need to be prefixed with '-'
+        return ','.join(
+                [t['uuid'] for t in added] +
+                ['-' + t['uuid'] for t in removed]
+            )
+
+    def deserialize_annotations(self, data):
+        return [TaskAnnotation(self, d) for d in data] if data else []
+
+    def deserialize_tags(self, tags):
+        if isinstance(tags, basestring):
+            return tags.split(',') if tags else []
+        return tags
+
+    def serialize_tags(self, tags):
+        return ','.join(tags) if tags else ''
 
     def delete(self):
-        self.warrior.delete_task(self['uuid'])
+        if not self.saved:
+            raise Task.NotSaved("Task needs to be saved before it can be deleted")
+
+        # Refresh the status, and raise exception if the task is deleted
+        self.refresh(only_fields=['status'])
+
+        if self.deleted:
+            raise Task.DeletedTask("Task was already deleted")
+
+        self.warrior.execute_command([self['uuid'], 'delete'], config_override={
+            'confirmation': 'no',
+        })
+
+        # Refresh the status again, so that we have updated info stored
+        self.refresh(only_fields=['status'])
+
 
     def done(self):
-        self.warrior.complete_task(self['uuid'])
+        if not self.saved:
+            raise Task.NotSaved("Task needs to be saved before it can be completed")
 
-    def save(self, delete_first=True):
-        if self['uuid'] and delete_first:
-            self.delete()
-        if not self['uuid'] or delete_first:
-            self.regenerate_uuid()
-        self.warrior.import_tasks([self._data])
+        # Refresh, and raise exception if task is already completed/deleted
+        self.refresh(only_fields=['status'])
 
-    __repr__ = __unicode__
+        if self.completed:
+            raise Task.CompletedTask("Cannot complete a completed task")
+        elif self.deleted:
+            raise Task.DeletedTask("Deleted task cannot be completed")
 
+        self.warrior.execute_command([self['uuid'], 'done'])
 
-class TaskWarrior(object):
-    DEFAULT_FILTERS = {
-        'status': 'pending',
-    }
+        # Refresh the status again, so that we have updated info stored
+        self.refresh(only_fields=['status'])
+
+    def save(self):
+        args = [self['uuid'], 'modify'] if self.saved else ['add']
+        args.extend(self._get_modified_fields_as_args())
+        output = self.warrior.execute_command(args)
+
+        # Parse out the new ID, if the task is being added for the first time
+        if not self.saved:
+            id_lines = [l for l in output if l.startswith('Created task ')]
+
+            # Complain loudly if it seems that more tasks were created
+            # Should not happen
+            if len(id_lines) != 1 or len(id_lines[0].split(' ')) != 3:
+                raise TaskWarriorException("Unexpected output when creating "
+                                           "task: %s" % '\n'.join(id_lines))
+
+            # Circumvent the ID storage, since ID is considered read-only
+            self._data['id'] = int(id_lines[0].split(' ')[2].rstrip('.'))
+
+        self.refresh()
+
+    def add_annotation(self, annotation):
+        if not self.saved:
+            raise Task.NotSaved("Task needs to be saved to add annotation")
+
+        args = [self['uuid'], 'annotate', annotation]
+        self.warrior.execute_command(args)
+        self.refresh(only_fields=['annotations'])
+
+    def remove_annotation(self, annotation):
+        if not self.saved:
+            raise Task.NotSaved("Task needs to be saved to add annotation")
+
+        if isinstance(annotation, TaskAnnotation):
+            annotation = annotation['description']
+        args = [self['uuid'], 'denotate', annotation]
+        self.warrior.execute_command(args)
+        self.refresh(only_fields=['annotations'])
+
+    def _get_modified_fields_as_args(self):
+        args = []
+
+        def add_field(field):
+            # Task version older than 2.4.0 ignores first word of the
+            # task description if description: prefix is used
+            if self.warrior.version < VERSION_2_4_0 and field == 'description':
+                args.append(self._data[field])
+            elif field == 'depends':
+                args.append('{0}:{1}'.format(field, self.format_depends()))
+            else:
+                # Use empty string to substitute for None value
+                args.append('{0}:{1}'.format(field, self._data[field] or ''))
+
+        # If we're modifying saved task, simply pass on all modified fields
+        if self.saved:
+            for field in self._modified_fields:
+                add_field(field)
+        # For new tasks, pass all fields that make sense
+        else:
+            for field in self._data.keys():
+                if field in self.read_only_fields:
+                    continue
+                add_field(field)
+
+        return args
+
+    def refresh(self, only_fields=[]):
+        # Raise error when trying to refresh a task that has not been saved
+        if not self.saved:
+            raise Task.NotSaved("Task needs to be saved to be refreshed")
+
+        # We need to use ID as backup for uuid here for the refreshes
+        # of newly saved tasks. Any other place in the code is fine
+        # with using UUID only.
+        args = [self['uuid'] or self['id'], 'export']
+        new_data = json.loads(self.warrior.execute_command(args)[0])
+        if only_fields:
+            to_update = dict(
+                [(k, new_data.get(k)) for k in only_fields])
+            self._data.update(to_update)
+            self._original_data.update(to_update)
+        else:
+            self._data = new_data
+            # We need to create a clone for original_data though
+            # Shallow copy is alright, since data dict uses only
+            # primitive data types
+            self._original_data = new_data.copy()
+
+
+class TaskFilter(object):
+    """
+    A set of parameters to filter the task list with.
+    """
+
+    def __init__(self, filter_params=[]):
+        self.filter_params = filter_params
+
+    def add_filter(self, filter_str):
+        self.filter_params.append(filter_str)
+
+    def add_filter_param(self, key, value):
+        key = key.replace('__', '.')
+
+        # Replace the value with empty string, since that is the
+        # convention in TW for empty values
+        value = value if value is not None else ''
+
+        # If we are filtering by uuid:, do not use uuid keyword
+        # due to TW-1452 bug
+        if key == 'uuid':
+            self.filter_params.insert(0, value)
+        else:
+            self.filter_params.append('{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.filter_params = list(self.filter_params)
+        return c
+
+
+class TaskQuerySet(object):
+    """
+    Represents a lazy lookup for a task objects.
+    """
+
+    def __init__(self, warrior=None, filter_obj=None):
+        self.warrior = warrior
+        self._result_cache = None
+        self.filter_obj = filter_obj or TaskFilter()
+
+    def __deepcopy__(self, memo):
+        """
+        Deep copy of a QuerySet doesn't populate the cache
+        """
+        obj = self.__class__()
+        for k, v in self.__dict__.items():
+            if k in ('_iter', '_result_cache'):
+                obj.__dict__[k] = None
+            else:
+                obj.__dict__[k] = copy.deepcopy(v, memo)
+        return obj
+
+    def __repr__(self):
+        data = list(self[:REPR_OUTPUT_SIZE + 1])
+        if len(data) > REPR_OUTPUT_SIZE:
+            data[-1] = "...(remaining elements truncated)..."
+        return repr(data)
+
+    def __len__(self):
+        if self._result_cache is None:
+            self._result_cache = list(self)
+        return len(self._result_cache)
+
+    def __iter__(self):
+        if self._result_cache is None:
+            self._result_cache = self._execute()
+        return iter(self._result_cache)
+
+    def __getitem__(self, k):
+        if self._result_cache is None:
+            self._result_cache = list(self)
+        return self._result_cache.__getitem__(k)
+
+    def __bool__(self):
+        if self._result_cache is not None:
+            return bool(self._result_cache)
+        try:
+            next(iter(self))
+        except StopIteration:
+            return False
+        return True
+
+    def __nonzero__(self):
+        return type(self).__bool__(self)
+
+    def _clone(self, klass=None, **kwargs):
+        if klass is None:
+            klass = self.__class__
+        filter_obj = self.filter_obj.clone()
+        c = klass(warrior=self.warrior, filter_obj=filter_obj)
+        c.__dict__.update(kwargs)
+        return c
+
+    def _execute(self):
+        """
+        Fetch the tasks which match the current filters.
+        """
+        return self.warrior.filter_tasks(self.filter_obj)
+
+    def all(self):
+        """
+        Returns a new TaskQuerySet that is a copy of the current one.
+        """
+        return self._clone()
+
+    def pending(self):
+        return self.filter(status=PENDING)
+
+    def completed(self):
+        return self.filter(status=COMPLETED)
+
+    def filter(self, *args, **kwargs):
+        """
+        Returns a new TaskQuerySet with the given filters added.
+        """
+        clone = self._clone()
+        for f in args:
+            clone.filter_obj.add_filter(f)
+        for key, value in kwargs.items():
+            clone.filter_obj.add_filter_param(key, value)
+        return clone
+
+    def get(self, **kwargs):
+        """
+        Performs the query and returns a single object matching the given
+        keyword arguments.
+        """
+        clone = self.filter(**kwargs)
+        num = len(clone)
+        if num == 1:
+            return clone._result_cache[0]
+        if not num:
+            raise Task.DoesNotExist(
+                'Task matching query does not exist. '
+                'Lookup parameters were {0}'.format(kwargs))
+        raise ValueError(
+            'get() returned more than one Task -- it returned {0}! '
+            'Lookup parameters were {1}'.format(num, kwargs))
+
+
+class TaskWarrior(object):
     def __init__(self, data_location='~/.task', create=True):
-        if not os.path.exists(data_location):
+        data_location = os.path.expanduser(data_location)
+        if create and not os.path.exists(data_location):
             os.makedirs(data_location)
         self.config = {
             'data.location': os.path.expanduser(data_location),
         }
+        self.tasks = TaskQuerySet(self)
+        self.version = self._get_version()
 
-    def _generate_command(self, command):
-        args = ['task', 'rc:/']
-        for item in self.config.items():
-            args.append('rc.{0}={1}'.format(*item))
-        args.append(command)
-        return ' '.join(args)
-
-    def _execute(self, command):
-        p = subprocess.Popen(self._generate_command(command), shell=True,
-                             stdout=subprocess.PIPE, stderr=subprocess.PIPE)
-        stdout, stderr = p.communicate()
+    def _get_command_args(self, args, config_override={}):
+        command_args = ['task', 'rc:/']
+        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))
+        return command_args
+
+    def _get_version(self):
+        p = subprocess.Popen(
+                ['task', '--version'],
+                stdout=subprocess.PIPE,
+                stderr=subprocess.PIPE)
+        stdout, stderr = [x.decode('utf-8') for x in p.communicate()]
+        return stdout.strip('\n')
+
+    def execute_command(self, args, config_override={}):
+        command_args = self._get_command_args(
+            args, config_override=config_override)
+        logger.debug(' '.join(command_args))
+        p = subprocess.Popen(command_args, stdout=subprocess.PIPE,
+                             stderr=subprocess.PIPE)
+        stdout, stderr = [x.decode('utf-8') for x in p.communicate()]
         if p.returncode:
-            raise TaskWarriorException(stderr.strip())
+            if stderr.strip():
+                error_msg = stderr.strip().splitlines()[-1]
+            else:
+                error_msg = stdout.strip()
+            raise TaskWarriorException(error_msg)
         return stdout.strip().split('\n')
 
-    def _format_filter_kwarg(self, kwarg):
-        key, val = kwarg[0], kwarg[1]
-        if key in ['tag', 'tags']:
-            key = 'tags.equal'
-        key = key.replace('__', '.')
-        return '{0}:{1}'.format(key, val)
-
-    def get_tasks(self, **filter_kwargs):
-        filters = self.DEFAULT_FILTERS
-        filters.update(filter_kwargs)
-        filter_commands = ' '.join(map(self._format_filter_kwarg,
-                                       filters.items()))
-        command = '{0} export'.format(filter_commands)
+    def filter_tasks(self, filter_obj):
+        args = ['export', '--'] + filter_obj.get_filter_params()
         tasks = []
-        for line in self._execute(command):
+        for line in self.execute_command(args):
             if line:
-                tasks.append(Task(self, json.loads(line.strip(','))))
+                data = line.strip(',')
+                try:
+                    tasks.append(Task(self, json.loads(data)))
+                except ValueError:
+                    raise TaskWarriorException('Invalid JSON: %s' % data)
         return tasks
 
-    def get_task(self, task_id):
-        command = '{0} export'.format(task_id)
-        return Task(self, json.loads(self._execute(command)[0]))
-
-    def add_task(self, description, project=None):
-        args = ['add', description]
-        if project is not None:
-            args.append('project:{0}'.format(project))
-        self._execute(' '.join(args))
-
-    def delete_task(self, task_id):
-        self._execute('{0} rc.confirmation:no delete'.format(task_id))
-
-    def complete_task(self, task_id):
-        self._execute('{0} done'.format(task_id))
+    def merge_with(self, path, push=False):
+        path = path.rstrip('/') + '/'
+        self.execute_command(['merge', path], config_override={
+            'merge.autopush': 'yes' if push else 'no',
+        })
 
-    def import_tasks(self, tasks):
-        fd, path = tempfile.mkstemp()
-        with open(path, 'w') as f:
-            f.write(json.dumps(tasks))
-        self._execute('import {0}'.format(path))
+    def undo(self):
+        self.execute_command(['undo'], config_override={
+            'confirmation': 'no',
+        })