import string from datetime import datetime import random import string from datetime import datetime from typing import List import jinja2 from SpiffWorkflow import Task as SpiffTask, WorkflowException, NavItem from SpiffWorkflow.bpmn.PythonScriptEngine import Box from SpiffWorkflow.bpmn.specs.EndEvent import EndEvent from SpiffWorkflow.bpmn.specs.ManualTask import ManualTask from SpiffWorkflow.bpmn.specs.MultiInstanceTask import MultiInstanceTask from SpiffWorkflow.bpmn.specs.ScriptTask import ScriptTask from SpiffWorkflow.bpmn.specs.StartEvent import StartEvent from SpiffWorkflow.bpmn.specs.UserTask import UserTask from SpiffWorkflow.dmn.specs.BusinessRuleTask import BusinessRuleTask from SpiffWorkflow.specs import CancelTask, StartTask, MultiChoice from SpiffWorkflow.util.deep_merge import DeepMerge from SpiffWorkflow.util.metrics import timeit from jinja2 import Template from crc import db, app from crc.api.common import ApiError from crc.models.api_models import Task, MultiInstanceType, WorkflowApi from crc.models.data_store import DataStoreModel from crc.models.file import LookupDataModel, FileModel, File, FileSchema from crc.models.study import StudyModel from crc.models.task_event import TaskEventModel from crc.models.user import UserModel, UserModelSchema from crc.models.workflow import WorkflowModel, WorkflowStatus, WorkflowSpecModel from crc.services.data_store_service import DataStoreBase from crc.services.document_service import DocumentService from crc.services.file_service import FileService from crc.services.lookup_service import LookupService from crc.services.study_service import StudyService from crc.services.user_service import UserService from crc.services.workflow_processor import WorkflowProcessor class WorkflowService(object): TASK_ACTION_COMPLETE = "COMPLETE" TASK_ACTION_TOKEN_RESET = "TOKEN_RESET" TASK_ACTION_HARD_RESET = "HARD_RESET" TASK_ACTION_SOFT_RESET = "SOFT_RESET" TASK_ACTION_ASSIGNMENT = "ASSIGNMENT" # Whenever the lane changes between tasks we assign the task to specifc user. TASK_STATE_LOCKED = "LOCKED" # When the task belongs to a different user. """Provides tools for processing workflows and tasks. This should at some point, be the only way to work with Workflows, and the workflow Processor should be hidden behind this service. This will help maintain a structure that avoids circular dependencies. But for now, this contains tools for converting spiff-workflow models into our own API models with additional information and capabilities and handles the testing of a workflow specification by completing it with random selections, attempting to mimic a front end as much as possible. """ from crc.services.user_service import UserService @staticmethod def make_test_workflow(spec_id, validate_study_id=None): try: user = UserService.current_user() except ApiError as e: user = None if not user: user = db.session.query(UserModel).filter_by(uid="test").first() if not user: db.session.add(UserModel(uid="test")) db.session.commit() user = db.session.query(UserModel).filter_by(uid="test").first() if validate_study_id: study = db.session.query(StudyModel).filter_by(id=validate_study_id).first() else: study = db.session.query(StudyModel).filter_by(user_uid=user.uid).first() if not study: db.session.add(StudyModel(user_uid=user.uid, title="test")) db.session.commit() study = db.session.query(StudyModel).filter_by(user_uid=user.uid).first() workflow_model = WorkflowModel(status=WorkflowStatus.not_started, workflow_spec_id=spec_id, last_updated=datetime.utcnow(), study=study) return workflow_model @staticmethod def delete_test_data(workflow: WorkflowModel): db.session.delete(workflow) # Also, delete any test study or user models that may have been created. for study in db.session.query(StudyModel).filter(StudyModel.user_uid == "test"): StudyService.delete_study(study.id) user = db.session.query(UserModel).filter_by(uid="test").first() if user: db.session.delete(user) db.session.commit() @staticmethod def do_waiting(): records = db.session.query(WorkflowModel).filter(WorkflowModel.status==WorkflowStatus.waiting).all() for workflow_model in records: try: app.logger.info('Processing workflow %s' % workflow_model.id) processor = WorkflowProcessor(workflow_model) processor.bpmn_workflow.refresh_waiting_tasks() processor.bpmn_workflow.do_engine_steps() processor.save() except Exception as e: app.logger.error(f"Error running waiting task for workflow #%i (%s) for study #%i. %s" % (workflow_model.id, workflow_model.workflow_spec.name, workflow_model.study_id, str(e))) @staticmethod @timeit def test_spec(spec_id, validate_study_id=None, test_until=None, required_only=False): """Runs a spec through it's paces to see if it results in any errors. Not fool-proof, but a good sanity check. Returns the final data output form the last task if successful. test_until required_only can be set to true, in which case this will run the spec, only completing the required fields, rather than everything. """ workflow_model = WorkflowService.make_test_workflow(spec_id, validate_study_id) try: processor = WorkflowProcessor(workflow_model, validate_only=True) count = 0 while not processor.bpmn_workflow.is_completed(): processor.bpmn_workflow.get_deep_nav_list() # Assure no errors with navigation. exit_task = processor.bpmn_workflow.do_engine_steps(exit_at=test_until) if (exit_task != None): raise ApiError.from_task("validation_break", f"The validation has been exited early on task '{exit_task.task_spec.name}' and was parented by ", exit_task.parent) tasks = processor.bpmn_workflow.get_tasks(SpiffTask.READY) for task in tasks: if task.task_spec.lane is not None and task.task_spec.lane not in task.data: raise ApiError.from_task("invalid_role", f"This task is in a lane called '{task.task_spec.lane}', The " f" current task data must have information mapping this role to " f" a unique user id.", task) task_api = WorkflowService.spiff_task_to_api_task( task, add_docs_and_forms=True) # Assure we try to process the documentation, and raise those errors. # make sure forms have a form key if hasattr(task_api, 'form') and task_api.form is not None and task_api.form.key == '': raise ApiError(code='missing_form_key', message='Forms must include a Form Key.', task_id=task.id, task_name=task.get_name()) WorkflowService.populate_form_with_random_data(task, task_api, required_only) processor.complete_task(task) if test_until == task.task_spec.name: raise ApiError.from_task("validation_break", f"The validation has been exited early on task '{task.task_spec.name}' and was parented by ", task.parent) count += 1 if count >= 100: raise ApiError.from_task(code='validation_loop', message=f'There appears to be an infinite loop in the validation. Task is {task.task_spec.description}', task=task) WorkflowService._process_documentation(processor.bpmn_workflow.last_task.parent.parent) except WorkflowException as we: raise ApiError.from_workflow_exception("workflow_validation_exception", str(we), we) finally: WorkflowService.delete_test_data(workflow_model) return processor.bpmn_workflow.last_task.data @staticmethod def populate_form_with_random_data(task, task_api, required_only): """populates a task with random data - useful for testing a spec.""" if not hasattr(task.task_spec, 'form'): return form_data = task.data # Just like with the front end, we start with what was already there, and modify it. hide_groups = [] for field in task_api.form.fields: # Assure we have a field type if field.type is None: raise ApiError(code='invalid_form_data', message = f'Type is missing for field "{field.id}". A field type must be provided.', task_id = task.id, task_name = task.get_name()) # Assure we have valid ids if not WorkflowService.check_field_id(field.id): raise ApiError(code='invalid_form_id', message=f'Invalid Field name: "{field.id}". A field ID must begin with a letter, ' f'and can only contain letters, numbers, and "_"', task_id = task.id, task_name = task.get_name()) # Assure field has valid properties WorkflowService.check_field_properties(field, task) # Process the label of the field if it is dynamic. if field.has_property(Task.FIELD_PROP_LABEL_EXPRESSION): result = WorkflowService.evaluate_property(Task.FIELD_PROP_LABEL_EXPRESSION, field, task) field.label = result # If a field is hidden and required, it must have a default value or value_expression if field.has_property(Task.FIELD_PROP_HIDE_EXPRESSION) and field.has_validation(Task.FIELD_CONSTRAINT_REQUIRED): if not field.has_property(Task.FIELD_PROP_VALUE_EXPRESSION) and \ (not (hasattr(field, 'default_value')) or field.default_value is None): raise ApiError(code='hidden and required field missing default', message=f'Field "{field.id}" is required but can be hidden. It must have either a default value or a value_expression', task_id='task.id', task_name=task.get_name()) # If the field is hidden and not required, it should not produce a value. if field.has_property(Task.FIELD_PROP_HIDE_EXPRESSION) and not field.has_validation(Task.FIELD_CONSTRAINT_REQUIRED): if WorkflowService.evaluate_property(Task.FIELD_PROP_HIDE_EXPRESSION, field, task): continue # A task should only have default_value **or** value expression, not both. if field.has_property(Task.FIELD_PROP_VALUE_EXPRESSION) and (hasattr(field, 'default_value') and field.default_value): raise ApiError(code='default value and value_expression', message='This task has both a default_value and value_expression. Please fix this to only have one or the other.') # If we have a default_value or value_expression, try to set the default if field.has_property(Task.FIELD_PROP_VALUE_EXPRESSION) or (hasattr(field, 'default_value') and field.default_value): form_data[field.id] = WorkflowService.get_default_value(field, task) if not field.has_property(Task.FIELD_PROP_REPEAT): continue # If we are only populating required fields, and this isn't required. stop here. if required_only: if (not field.has_validation(Task.FIELD_CONSTRAINT_REQUIRED) or field.get_validation(Task.FIELD_CONSTRAINT_REQUIRED).lower().strip() != "true"): continue # Don't include any fields that aren't specifically marked as required. if field.has_property(Task.FIELD_PROP_REQUIRED_EXPRESSION): result = WorkflowService.evaluate_property(Task.FIELD_PROP_REQUIRED_EXPRESSION, field, task) if not result and required_only: continue # Don't complete fields that are not required. # If it is read only, stop here. if field.has_property("read_only") and field.get_property(Task.FIELD_PROP_READ_ONLY).lower().strip() == "true": continue # Don't mess about with read only fields. if field.has_property(Task.FIELD_PROP_REPEAT): group = field.get_property(Task.FIELD_PROP_REPEAT) if field.has_property(Task.FIELD_PROP_REPEAT_HIDE_EXPRESSION): result = WorkflowService.evaluate_property(Task.FIELD_PROP_REPEAT_HIDE_EXPRESSION, field, task) if not result: hide_groups.append(group) if group not in form_data and group not in hide_groups: form_data[group] = [{},{},{}] if group in form_data and group not in hide_groups: for i in range(3): form_data[group][i][field.id] = WorkflowService.get_random_data_for_field(field, task) else: form_data[field.id] = WorkflowService.get_random_data_for_field(field, task) if task.data is None: task.data = {} task.data.update(form_data) @staticmethod def check_field_id(id): """Assures that field names are valid Python and Javascript names.""" if not id[0].isalpha(): return False for char in id[1:len(id)]: if char.isalnum() or char == '_' or char == '.': pass else: return False return True @staticmethod def check_field_properties(field, task): """Assures that all properties are valid on a given workflow.""" field_prop_names = list(map(lambda fp: fp.id, field.properties)) valid_names = Task.valid_property_names() for name in field_prop_names: if name not in valid_names: raise ApiError.from_task("invalid_field_property", f'The field {field.id} contains an unsupported ' f'property: {name}', task=task) @staticmethod def post_process_form(task): """Looks through the fields in a submitted form, acting on any properties.""" if not hasattr(task.task_spec, 'form'): return for field in task.task_spec.form.fields: data = task.data if field.has_property(Task.FIELD_PROP_REPEAT): repeat_array = task.data[field.get_property(Task.FIELD_PROP_REPEAT)] for repeat_data in repeat_array: WorkflowService.__post_process_field(task, field, repeat_data) else: WorkflowService.__post_process_field(task, field, data) @staticmethod def __post_process_field(task, field, data): if field.has_property(Task.FIELD_PROP_DOC_CODE) and field.id in data: # This is generally handled by the front end, but it is possible that the file was uploaded BEFORE # the doc_code was correctly set, so this is a stop gap measure to assure we still hit it correctly. file_id = data[field.id]["id"] doc_code = task.workflow.script_engine.eval(field.get_property(Task.FIELD_PROP_DOC_CODE), data) file = db.session.query(FileModel).filter(FileModel.id == file_id).first() if(file): file.irb_doc_code = doc_code db.session.commit() else: # We have a problem, the file doesn't exist, and was removed, but it is still referenced in the data # At least attempt to clear out the data. data = {} if field.has_property(Task.FIELD_PROP_FILE_DATA) and \ field.get_property(Task.FIELD_PROP_FILE_DATA) in data and \ field.id in data: file_id = data[field.get_property(Task.FIELD_PROP_FILE_DATA)]["id"] data_args = (field.id, data[field.id]['label']) DataStoreBase().set_data_common(task.id, None, None, None, None, None, file_id, *data_args) @staticmethod def evaluate_property(property_name, field, task): expression = field.get_property(property_name) data = task.data if field.has_property(Task.FIELD_PROP_REPEAT): # Then you must evaluate the expression based on the data within the group only. group = field.get_property(Task.FIELD_PROP_REPEAT) if group in task.data: data = task.data[group][0] try: return task.workflow.script_engine.eval(expression, data) except Exception as e: message = f"The field {field.id} contains an invalid expression. {e}" raise ApiError.from_task(f'invalid_{property_name}', message, task=task) @staticmethod def has_lookup(field): """Returns true if this is a lookup field.""" """Note, this does not include enums based on task data, that is populated when the form is created, not as a lookup from a data table. """ has_ldap_lookup = field.has_property(Task.FIELD_PROP_LDAP_LOOKUP) has_file_lookup = field.has_property(Task.FIELD_PROP_SPREADSHEET_NAME) return has_ldap_lookup or has_file_lookup @staticmethod def get_default_value(field, task): has_lookup = WorkflowService.has_lookup(field) default = field.default_value # If there is a value expression, use that rather than the default value. if field.has_property(Task.FIELD_PROP_VALUE_EXPRESSION): result = WorkflowService.evaluate_property(Task.FIELD_PROP_VALUE_EXPRESSION, field, task) default = result # If no default exists, return None # Note: if default is False, we don't want to execute this code if default is None or (isinstance(default, str) and default.strip() == ''): if field.type == "enum" or field.type == "autocomplete": # Return empty arrays for multi-select enums, otherwise do a value of None. if field.has_property(Task.FIELD_PROP_ENUM_TYPE) and field.get_property(Task.FIELD_PROP_ENUM_TYPE) == "checkbox": return [] else: return {'value': None, 'label': None} else: return None if field.type == "enum" and not has_lookup: default_option = next((obj for obj in field.options if obj.id == default), None) if not default_option: raise ApiError.from_task("invalid_default", "You specified a default value that does not exist in " "the enum options ", task) return {'value': default_option.id, 'label': default_option.name} elif field.type == "autocomplete" or field.type == "enum": lookup_model = LookupService.get_lookup_model(task, field) if field.has_property(Task.FIELD_PROP_LDAP_LOOKUP): # All ldap records get the same person. return None # There is no default value for ldap. elif lookup_model: data = db.session.query(LookupDataModel).\ filter(LookupDataModel.lookup_file_model == lookup_model). \ filter(LookupDataModel.value == default).\ first() if not data: raise ApiError.from_task("invalid_default", "You specified a default value that does not exist in " "the enum options ", task) return {"value": data.value, "label": data.label, "data": data.data} else: raise ApiError.from_task("unknown_lookup_option", "The settings for this auto complete field " "are incorrect: %s " % field.id, task) elif field.type == "long": return int(default) elif field.type == 'boolean': default = str(default).lower() if default == 'true' or default == 't': return True return False else: return default @staticmethod def get_random_data_for_field(field, task): """Randomly populates the field, mainly concerned with getting enums correct, as the rest are pretty easy.""" has_lookup = WorkflowService.has_lookup(field) if field.type == "enum" and not has_lookup: # If it's a normal enum field with no lookup, # return a random option. if len(field.options) > 0: random_choice = random.choice(field.options) if isinstance(random_choice, dict): return {'value': random_choice['id'], 'label': random_choice['name'], 'data': random_choice['data']} else: # fixme: why it is sometimes an EnumFormFieldOption, and other times not? return {'value': random_choice.id, 'label': random_choice.name} else: raise ApiError.from_task("invalid_enum", "You specified an enumeration field (%s)," " with no options" % field.id, task) elif field.type == "autocomplete" or field.type == "enum": # If it has a lookup, get the lookup model from the spreadsheet or task data, then return a random option # from the lookup model lookup_model = LookupService.get_lookup_model(task, field) if field.has_property(Task.FIELD_PROP_LDAP_LOOKUP): # All ldap records get the same person. return WorkflowService._random_ldap_record() elif lookup_model: data = db.session.query(LookupDataModel).filter( LookupDataModel.lookup_file_model == lookup_model).limit(10).all() options = [{"value": d.value, "label": d.label, "data": d.data} for d in data] if len(options) > 0: return random.choice(options) else: raise ApiError.from_task("invalid enum", "You specified an enumeration field (%s)," " with no options" % field.id, task) else: raise ApiError.from_task("unknown_lookup_option", "The settings for this auto complete field " "are incorrect: %s " % field.id, task) elif field.type == "long": return random.randint(1, 1000) elif field.type == 'boolean': return random.choice([True, False]) elif field.type == 'file': doc_code = field.id if field.has_property('doc_code'): doc_code = WorkflowService.evaluate_property('doc_code', field, task) file_model = FileModel(name="test.png", irb_doc_code = field.id) doc_dict = DocumentService.get_dictionary() file = File.from_models(file_model, None, doc_dict) return FileSchema().dump(file) elif field.type == 'files': return random.randrange(1, 100) else: return WorkflowService._random_string() @staticmethod def _random_ldap_record(): return { "label": "dhf8r", "value": "Dan Funk", "data": { "uid": "dhf8r", "display_name": "Dan Funk", "given_name": "Dan", "email_address": "dhf8r@virginia.edu", "department": "Department of Psychocosmographictology", "affiliation": "Roustabout", "sponsor_type": "Staff"} } @staticmethod def _random_string(string_length=10): """Generate a random string of fixed length """ letters = string.ascii_lowercase return ''.join(random.choice(letters) for i in range(string_length)) @staticmethod def processor_to_workflow_api(processor: WorkflowProcessor, next_task=None): """Returns an API model representing the state of the current workflow, if requested, and possible, next_task is set to the current_task.""" navigation = processor.bpmn_workflow.get_deep_nav_list() WorkflowService.update_navigation(navigation, processor) spec = db.session.query(WorkflowSpecModel).filter_by(id=processor.workflow_spec_id).first() is_review = FileService.is_workflow_review(processor.workflow_spec_id) workflow_api = WorkflowApi( id=processor.get_workflow_id(), status=processor.get_status(), next_task=None, navigation=navigation, workflow_spec_id=processor.workflow_spec_id, spec_version=processor.get_version_string(), is_latest_spec=processor.is_latest_spec, total_tasks=len(navigation), completed_tasks=processor.workflow_model.completed_tasks, last_updated=processor.workflow_model.last_updated, is_review=is_review, title=spec.display_name, study_id=processor.workflow_model.study_id or None ) if not next_task: # The Next Task can be requested to be a certain task, useful for parallel tasks. # This may or may not work, sometimes there is no next task to complete. next_task = processor.next_task() if next_task: previous_form_data = WorkflowService.get_previously_submitted_data(processor.workflow_model.id, next_task) DeepMerge.merge(next_task.data, previous_form_data) workflow_api.next_task = WorkflowService.spiff_task_to_api_task(next_task, add_docs_and_forms=True) # Update the state of the task to locked if the current user does not own the task. user_uids = WorkflowService.get_users_assigned_to_task(processor, next_task) if not UserService.in_list(user_uids, allow_admin_impersonate=True): workflow_api.next_task.state = WorkflowService.TASK_STATE_LOCKED return workflow_api @staticmethod def update_navigation(navigation: List[NavItem], processor: WorkflowProcessor): # Recursive function to walk down through children, and clean up descriptions, and statuses for nav_item in navigation: spiff_task = processor.bpmn_workflow.get_task(nav_item.task_id) if spiff_task: # Use existing logic to set the description, and alter the state based on permissions. api_task = WorkflowService.spiff_task_to_api_task(spiff_task, add_docs_and_forms=False) nav_item.description = api_task.title user_uids = WorkflowService.get_users_assigned_to_task(processor, spiff_task) if (isinstance(spiff_task.task_spec, UserTask) or isinstance(spiff_task.task_spec, ManualTask)) \ and not UserService.in_list(user_uids, allow_admin_impersonate=True): nav_item.state = WorkflowService.TASK_STATE_LOCKED else: # Strip off the first word in the description, to meet guidlines for BPMN. if nav_item.description: if nav_item.description is not None and ' ' in nav_item.description: nav_item.description = nav_item.description.partition(' ')[2] # Recurse here WorkflowService.update_navigation(nav_item.children, processor) @staticmethod def get_previously_submitted_data(workflow_id, spiff_task): """ If the user has completed this task previously, find the form data for the last submission.""" query = db.session.query(TaskEventModel) \ .filter_by(workflow_id=workflow_id) \ .filter_by(task_name=spiff_task.task_spec.name) \ .filter_by(action=WorkflowService.TASK_ACTION_COMPLETE) if hasattr(spiff_task, 'internal_data') and 'runtimes' in spiff_task.internal_data: query = query.filter_by(mi_index=spiff_task.internal_data['runtimes']) latest_event = query.order_by(TaskEventModel.date.desc()).first() if latest_event: if latest_event.form_data is not None: return latest_event.form_data else: missing_form_error = ( f'We have lost data for workflow {workflow_id}, ' f'task {spiff_task.task_spec.name}, it is not in the task event model, ' f'and it should be.' ) app.logger.error("missing_form_data", missing_form_error, exc_info=True) return {} else: return {} @staticmethod def spiff_task_to_api_task(spiff_task, add_docs_and_forms=False): task_type = spiff_task.task_spec.__class__.__name__ task_types = [UserTask, ManualTask, BusinessRuleTask, CancelTask, ScriptTask, StartTask, EndEvent, StartEvent] for t in task_types: if isinstance(spiff_task.task_spec, t): task_type = t.__name__ break else: task_type = "NoneTask" info = spiff_task.task_info() if info["is_looping"]: mi_type = MultiInstanceType.looping elif info["is_sequential_mi"]: mi_type = MultiInstanceType.sequential elif info["is_parallel_mi"]: mi_type = MultiInstanceType.parallel else: mi_type = MultiInstanceType.none props = {} if hasattr(spiff_task.task_spec, 'extensions'): for key, val in spiff_task.task_spec.extensions.items(): props[key] = val if hasattr(spiff_task.task_spec, 'lane'): lane = spiff_task.task_spec.lane else: lane = None task = Task(spiff_task.id, spiff_task.task_spec.name, spiff_task.task_spec.description, task_type, spiff_task.get_state_name(), lane, None, "", {}, mi_type, info["mi_count"], info["mi_index"], process_name=spiff_task.task_spec._wf_spec.description, properties=props ) # Only process the form and documentation if requested. # The task should be in a completed or a ready state, and should # not be a previously completed MI Task. if add_docs_and_forms: task.data = spiff_task.data if hasattr(spiff_task.task_spec, "form"): task.form = spiff_task.task_spec.form for i, field in enumerate(task.form.fields): task.form.fields[i] = WorkflowService.process_options(spiff_task, field) # If there is a default value, set it. if field.id not in task.data and WorkflowService.get_default_value(field, spiff_task) is not None: task.data[field.id] = WorkflowService.get_default_value(field, spiff_task) task.documentation = WorkflowService._process_documentation(spiff_task) # All ready tasks should have a valid name, and this can be computed for # some tasks, particularly multi-instance tasks that all have the same spec # but need different labels. if spiff_task.state == SpiffTask.READY: task.properties = WorkflowService._process_properties(spiff_task, props) # Replace the title with the display name if it is set in the task properties, # otherwise strip off the first word of the task, as that should be following # a BPMN standard, and should not be included in the display. if task.properties and "display_name" in task.properties: try: task.title = spiff_task.workflow.script_engine.evaluate_expression(spiff_task, task.properties[Task.PROP_EXTENSIONS_TITLE]) except Exception as e: # if the task is ready, we should raise an error, but if it is in the future or the past, we may not # have the information we need to properly set the title, so don't error out, and just use what is # provided. if spiff_task.state == spiff_task.READY: raise ApiError.from_task(code="task_title_error", message="Could not set task title on task %s with '%s' property because %s" % (spiff_task.task_spec.name, Task.PROP_EXTENSIONS_TITLE, str(e)), task=spiff_task) # Otherwise, just use the curreent title. elif task.title and ' ' in task.title: task.title = task.title.partition(' ')[2] return task @staticmethod def _process_properties(spiff_task, props): """Runs all the property values through the Jinja2 processor to inject data.""" for k, v in props.items(): try: template = Template(v) props[k] = template.render(**spiff_task.data) except jinja2.exceptions.TemplateError as ue: app.logger.error(f'Failed to process task property {str(ue)}', exc_info=True) return props @staticmethod def _process_documentation(spiff_task): """Runs the given documentation string through the Jinja2 processor to inject data create loops, etc... - If a markdown file exists with the same name as the task id, it will use that file instead of the documentation. """ documentation = spiff_task.task_spec.documentation if hasattr(spiff_task.task_spec, "documentation") else "" try: doc_file_name = spiff_task.task_spec.name + ".md" data_model = FileService.get_workflow_file_data(spiff_task.workflow, doc_file_name) raw_doc = data_model.data.decode("utf-8") except ApiError: raw_doc = documentation if not raw_doc: return "" try: template = Template(raw_doc) return template.render(**spiff_task.data) except jinja2.exceptions.TemplateError as ue: raise ApiError.from_task(code="template_error", message="Error processing template for task %s: %s" % (spiff_task.task_spec.name, str(ue)), task=spiff_task) except TypeError as te: raise ApiError.from_task(code="template_error", message="Error processing template for task %s: %s" % (spiff_task.task_spec.name, str(te)), task=spiff_task) except Exception as e: app.logger.error(str(e), exc_info=True) @staticmethod def process_options(spiff_task, field): # If this is an auto-complete field, do not populate options, a lookup will happen later. if field.type == Task.FIELD_TYPE_AUTO_COMPLETE: pass elif field.has_property(Task.FIELD_PROP_SPREADSHEET_NAME): lookup_model = LookupService.get_lookup_model(spiff_task, field) data = db.session.query(LookupDataModel).filter(LookupDataModel.lookup_file_model == lookup_model).all() if not hasattr(field, 'options'): field.options = [] for d in data: field.options.append({"id": d.value, "name": d.label, "data": d.data}) elif field.has_property(Task.FIELD_PROP_DATA_NAME): field.options = WorkflowService.get_options_from_task_data(spiff_task, field) return field @staticmethod def get_options_from_task_data(spiff_task, field): if not (field.has_property(Task.FIELD_PROP_VALUE_COLUMN) or field.has_property(Task.FIELD_PROP_LABEL_COLUMN)): raise ApiError.from_task("invalid_enum", f"For enumerations based on task data, you must include 3 properties: " f"{Task.FIELD_PROP_DATA_NAME}, {Task.FIELD_PROP_VALUE_COLUMN}, " f"{Task.FIELD_PROP_LABEL_COLUMN}", task=spiff_task) prop = field.get_property(Task.FIELD_PROP_DATA_NAME) if prop not in spiff_task.data: raise ApiError.from_task("invalid_enum", f"For enumerations based on task data, task data must have " f"a property called {prop}", task=spiff_task) # Get the enum options from the task data data_model = spiff_task.data[prop] value_column = field.get_property(Task.FIELD_PROP_VALUE_COLUMN) label_column = field.get_property(Task.FIELD_PROP_LABEL_COLUMN) items = data_model.items() if isinstance(data_model, dict) else data_model options = [] for item in items: if value_column not in item: raise ApiError.from_task("invalid_enum", f"The value column '{value_column}' does not exist for item {item}", task=spiff_task) if label_column not in item: raise ApiError.from_task("invalid_enum", f"The label column '{label_column}' does not exist for item {item}", task=spiff_task) options.append(Box({"id": item[value_column], "name": item[label_column], "data": item})) return options @staticmethod def update_task_assignments(processor): """For every upcoming user task, log a task action that connects the assigned user(s) to that task. All existing assignment actions for this workflow are removed from the database, so that only the current valid actions are available. update_task_assignments should be called whenever progress is made on a workflow.""" db.session.query(TaskEventModel). \ filter(TaskEventModel.workflow_id == processor.workflow_model.id). \ filter(TaskEventModel.action == WorkflowService.TASK_ACTION_ASSIGNMENT).delete() db.session.commit() for task in processor.get_current_user_tasks(): user_ids = WorkflowService.get_users_assigned_to_task(processor, task) for user_id in user_ids: WorkflowService.log_task_action(user_id, processor, task, WorkflowService.TASK_ACTION_ASSIGNMENT) @staticmethod def get_users_assigned_to_task(processor, spiff_task) -> List[str]: if processor.workflow_model.study_id is None and processor.workflow_model.user_id is None: raise ApiError.from_task(code='invalid_workflow', message='A workflow must have either a study_id or a user_id.', task=spiff_task) # Standalone workflow - we only care about the current user elif processor.workflow_model.study_id is None and processor.workflow_model.user_id is not None: return [processor.workflow_model.user_id] # Workflow associated with a study - get all the users else: if not hasattr(spiff_task.task_spec, 'lane') or spiff_task.task_spec.lane is None: associated = StudyService.get_study_associates(processor.workflow_model.study.id) return [user['uid'] for user in associated if user['access']] if spiff_task.task_spec.lane not in spiff_task.data: return [] # No users are assignable to the task at this moment lane_users = spiff_task.data[spiff_task.task_spec.lane] if not isinstance(lane_users, list): lane_users = [lane_users] lane_uids = [] for user in lane_users: if isinstance(user, dict): if 'value' in user and user['value'] is not None: lane_uids.append(user['value']) else: raise ApiError.from_task(code="task_lane_user_error", message="Spiff Task %s lane user dict must have a key called 'value' with the user's uid in it." % spiff_task.task_spec.name, task=spiff_task) elif isinstance(user, str): lane_uids.append(user) else: raise ApiError.from_task(code="task_lane_user_error", message="Spiff Task %s lane user is not a string or dict" % spiff_task.task_spec.name, task=spiff_task) return lane_uids @staticmethod def log_task_action(user_uid, processor, spiff_task, action): task = WorkflowService.spiff_task_to_api_task(spiff_task) form_data = WorkflowService.extract_form_data(spiff_task.data, spiff_task) task_event = TaskEventModel( study_id=processor.workflow_model.study_id, user_uid=user_uid, workflow_id=processor.workflow_model.id, workflow_spec_id=processor.workflow_model.workflow_spec_id, spec_version=processor.get_version_string(), action=action, task_id=task.id, task_name=task.name, task_title=task.title, task_type=str(task.type), task_state=task.state, task_lane=task.lane, form_data=form_data, mi_type=task.multi_instance_type.value, # Some tasks have a repeat behavior. mi_count=task.multi_instance_count, # This is the number of times the task could repeat. mi_index=task.multi_instance_index, # And the index of the currently repeating task. process_name=task.process_name, # date=datetime.utcnow(), <=== For future reference, NEVER do this. Let the database set the time. ) db.session.add(task_event) db.session.commit() @staticmethod def extract_form_data(latest_data, task): """Extracts data from the latest_data that is directly related to the form that is being submitted.""" data = {} if hasattr(task.task_spec, 'form'): for field in task.task_spec.form.fields: if field.has_property(Task.FIELD_PROP_REPEAT): group = field.get_property(Task.FIELD_PROP_REPEAT) if group in latest_data: data[group] = latest_data[group] else: value = WorkflowService.get_dot_value(field.id, latest_data) if value is not None: WorkflowService.set_dot_value(field.id, value, data) return data @staticmethod def get_dot_value(path, source): ### Given a path in dot notation, uas as 'fruit.type' tries to find that value in ### the source, but looking deep in the dictionary. paths = path.split(".") # [a,b,c] s = source index = 0 for p in paths: index += 1 if isinstance(s, dict) and p in s: if index == len(paths): return s[p] else: s = s[p] if path in source: return source[path] return None @staticmethod def set_dot_value(path, value, target): ### Given a path in dot notation, such as "fruit.type", and a value "apple", will ### set the value in the target dictionary, as target["fruit"]["type"]="apple" destination = target paths = path.split(".") # [a,b,c] index = 0 for p in paths: index += 1 if p not in destination: if index == len(paths): destination[p] = value else: destination[p] = {} destination = destination[p] return target @staticmethod def process_workflows_for_cancels(study_id): workflows = db.session.query(WorkflowModel).filter_by(study_id=study_id).all() for workflow in workflows: if workflow.status == WorkflowStatus.user_input_required or workflow.status == WorkflowStatus.waiting: WorkflowProcessor.reset(workflow, clear_data=False) @staticmethod def get_workflow_from_spec(workflow_spec_id, user): workflow_model = WorkflowModel(status=WorkflowStatus.not_started, study=None, user_id=user.uid, workflow_spec_id=workflow_spec_id, last_updated=datetime.now()) db.session.add(workflow_model) db.session.commit() return workflow_model @staticmethod def get_standalone_workflow_specs(): specs = db.session.query(WorkflowSpecModel).filter_by(standalone=True).all() return specs @staticmethod def get_primary_workflow(workflow_spec_id): # Returns the FileModel of the primary workflow for a workflow_spec primary = None file = db.session.query(FileModel).filter(FileModel.workflow_spec_id==workflow_spec_id, FileModel.primary==True).first() if file: primary = file return primary