import datetime import glob import os import xml.etree.ElementTree as ElementTree from crc import app, db, session from crc.models.file import FileType, FileModel, FileDataModel, CONTENT_TYPES from crc.models.study import StudyModel from crc.models.user import UserModel from crc.models.workflow import WorkflowSpecModel, WorkflowSpecCategoryModel from crc.services.file_service import FileService from crc.services.workflow_processor import WorkflowProcessor from crc.models.protocol_builder import ProtocolBuilderStatus class ExampleDataLoader: @staticmethod def clean_db(): session.flush() # Clear out any transactions before deleting it all to avoid spurious errors. for table in reversed(db.metadata.sorted_tables): session.execute(table.delete()) session.flush() def load_all(self): self.load_reference_documents() categories = [ WorkflowSpecCategoryModel( id=0, name='irb_review', display_name='From PB', display_order=0 ), WorkflowSpecCategoryModel( id=1, name='core_info', display_name='Core Info', display_order=1 ), WorkflowSpecCategoryModel( id=2, name='approvals', display_name='Approvals', display_order=2 ), WorkflowSpecCategoryModel( id=3, name='data_security_plan', display_name='Data Security Plan', display_order=3 ), WorkflowSpecCategoryModel( id=4, name='finance', display_name='Finance', display_order=4 ), WorkflowSpecCategoryModel( id=5, name='notifications', display_name='Notifications', display_order=5 ), WorkflowSpecCategoryModel( id=6, name='status', display_name='Status', display_order=6 ), ] db.session.add_all(categories) db.session.commit() # Pass IRB Review self.create_spec(id="irb_api_personnel", name="irb_api_personnel", display_name="Personnel", description="TBD", category_id=0, display_order=0) self.create_spec(id="irb_api_details", name="irb_api_details", display_name="Protocol Builder Data", description="TBD", category_id=0, display_order=1) self.create_spec(id="documents_approvals", name="documents_approvals", display_name="Documents & Approvals", description="Status of all approvals and documents required from Protocol Builder", category_id=0, display_order=2) self.create_spec(id="ide_supplement", name="ide_supplement", display_name="IDE Supplement Info", description="Supplemental information for the IDE number entered in Protocol Builder", category_id=0, display_order=3) self.create_spec(id="ind_supplement", name="ind_supplement", display_name="IND Supplement Info", description="Supplement information for the Investigational New Drug(s) specified in Protocol Builder", category_id=0, display_order=4) # Core Info self.create_spec(id="protocol", name="protocol", display_name="Protocol", description="Upload the Study Protocol here.", category_id=1, display_order=0) self.create_spec(id="core_info", name="core_info", display_name="Core Info", description="TBD", category_id=1, display_order=1) # Approvals self.create_spec(id="ids_full_submission", name="ids_full_submission", display_name="Investigational Drug Service (IDS) Full Submission", description="TBD", category_id=2, display_order=0) self.create_spec(id="ids_waiver", name="ids_waiver", display_name="Investigational Drug Service (IDS) Waiver", description="TBD", category_id=2, display_order=1) self.create_spec(id="rsc_hire_submission", name="rsc_hire_submission", display_name="RSC/HIRE Submission", description="TBD", category_id=2, display_order=2) self.create_spec(id="rsc_hire_committee", name="rsc_hire_committee", display_name="RSC/HIRE Committee", description="TBD", category_id=2, display_order=3) # Data Security Plan self.create_spec(id="data_security_plan", name="data_security_plan", display_name="Data Security Plan", description="Create and generate Data Security Plan", category_id=3, display_order=0) # Finance self.create_spec(id="sponsor_funding_source", name="sponsor_funding_source", display_name="Sponsor Funding Source", description="TBD", category_id=4, display_order=0) self.create_spec(id="finance", name="finance", display_name="Finance Data", description="TBD", category_id=4, display_order=1) # Notifications self.create_spec(id="notifications", name="notifications", display_name="Notifications", description="TBD", category_id=5, display_order=0) # Status self.create_spec(id="enrollment_date", name="enrollment_date", display_name="Enrollment Date", description="Study enrollment date", category_id=6, display_order=0) self.create_spec(id="abandoned", name="abandoned", display_name="Abandoned", description="Place study into Abandoned status", category_id=6, display_order=1) # Top Level (Master Status) Workflow self.create_spec(id="top_level_workflow", name="top_level_workflow", display_name="Top Level Workflow", description="Determines the status of other workflows in a study", category_id=None, master_spec=True) def create_spec(self, id, name, display_name="", description="", filepath=None, master_spec=False, category_id=None, display_order=None): """Assumes that a directory exists in static/bpmn with the same name as the given id. further assumes that the [id].bpmn is the primary file for the workflow. returns an array of data models to be added to the database.""" global file file_service = FileService() spec = WorkflowSpecModel(id=id, name=name, display_name=display_name, description=description, is_master_spec=master_spec, category_id=category_id, display_order=display_order) db.session.add(spec) db.session.commit() if not filepath: filepath = os.path.join(app.root_path, 'static', 'bpmn', id, "*") files = glob.glob(filepath) for file_path in files: noise, file_extension = os.path.splitext(file_path) filename = os.path.basename(file_path) is_status = filename.lower() == 'status.bpmn' is_primary = filename.lower() == id + '.bpmn' try: file = open(file_path, 'rb') data = file.read() content_type = CONTENT_TYPES[file_extension[1:]] file_service.add_workflow_spec_file(workflow_spec=spec, name=filename, content_type=content_type, binary_data=data, primary=is_primary, is_status=is_status) except IsADirectoryError as de: # Ignore sub directories pass finally: if file: file.close() return spec def load_reference_documents(self): file_path = os.path.join(app.root_path, 'static', 'reference', 'irb_documents.xlsx') file = open(file_path, "rb") FileService.add_reference_file(FileService.IRB_PRO_CATEGORIES_FILE, binary_data=file.read(), content_type=CONTENT_TYPES['xls']) file.close()