cr-connect-workflow/crc/models/study.py

166 lines
6.2 KiB
Python
Raw Normal View History

import marshmallow
from marshmallow import INCLUDE, fields
from marshmallow_enum import EnumField
from sqlalchemy import func
from crc import db, ma
from crc.api.common import ApiErrorSchema
from crc.models.file import FileModel, SimpleFileSchema, FileSchema
from crc.models.protocol_builder import ProtocolBuilderStatus, ProtocolBuilderStudy
from crc.models.workflow import WorkflowSpecCategoryModel, WorkflowState, WorkflowStatus, WorkflowSpecModel, \
WorkflowModel
class StudyModel(db.Model):
__tablename__ = 'study'
id = db.Column(db.Integer, primary_key=True)
title = db.Column(db.String)
last_updated = db.Column(db.DateTime(timezone=True), default=func.now())
protocol_builder_status = db.Column(db.Enum(ProtocolBuilderStatus))
primary_investigator_id = db.Column(db.String, nullable=True)
sponsor = db.Column(db.String, nullable=True)
hsr_number = db.Column(db.String, nullable=True)
ind_number = db.Column(db.String, nullable=True)
user_uid = db.Column(db.String, db.ForeignKey('user.uid'), nullable=False)
investigator_uids = db.Column(db.ARRAY(db.String), nullable=True)
requirements = db.Column(db.ARRAY(db.Integer), nullable=True)
on_hold = db.Column(db.Boolean, default=False)
def update_from_protocol_builder(self, pbs: ProtocolBuilderStudy):
self.hsr_number = pbs.HSRNUMBER
self.title = pbs.TITLE
self.user_uid = pbs.NETBADGEID
self.last_updated = pbs.DATE_MODIFIED
2020-07-01 15:01:36 +00:00
self.protocol_builder_status = ProtocolBuilderStatus.ACTIVE
if pbs.HSRNUMBER:
self.protocol_builder_status = ProtocolBuilderStatus.OPEN
if self.on_hold:
self.protocol_builder_status = ProtocolBuilderStatus.HOLD
class WorkflowMetadata(object):
def __init__(self, id, name, display_name, description, spec_version, category_id, state: WorkflowState, status: WorkflowStatus,
2020-04-09 18:25:14 +00:00
total_tasks, completed_tasks, display_order):
self.id = id
self.name = name
self.display_name = display_name
self.description = description
self.spec_version = spec_version
self.category_id = category_id
self.state = state
2020-03-27 18:55:53 +00:00
self.status = status
self.total_tasks = total_tasks
self.completed_tasks = completed_tasks
2020-04-09 18:25:14 +00:00
self.display_order = display_order
2020-03-27 18:55:53 +00:00
@classmethod
def from_workflow(cls, workflow: WorkflowModel):
instance = cls(
id=workflow.id,
name=workflow.workflow_spec.name,
display_name=workflow.workflow_spec.display_name,
description=workflow.workflow_spec.description,
spec_version=workflow.spec_version(),
category_id=workflow.workflow_spec.category_id,
state=WorkflowState.optional,
status=workflow.status,
total_tasks=workflow.total_tasks,
2020-04-09 18:25:14 +00:00
completed_tasks=workflow.completed_tasks,
display_order=workflow.workflow_spec.display_order
)
return instance
2020-03-27 18:55:53 +00:00
class WorkflowMetadataSchema(ma.Schema):
state = EnumField(WorkflowState)
status = EnumField(WorkflowStatus)
class Meta:
model = WorkflowMetadata
additional = ["id", "name", "display_name", "description",
2020-04-09 18:25:14 +00:00
"total_tasks", "completed_tasks", "display_order"]
unknown = INCLUDE
class Category(object):
def __init__(self, model: WorkflowSpecCategoryModel):
self.id = model.id
self.name = model.name
self.display_name = model.display_name
self.display_order = model.display_order
class CategorySchema(ma.Schema):
workflows = fields.List(fields.Nested(WorkflowMetadataSchema), dump_only=True)
class Meta:
model = Category
additional = ["id", "name", "display_name", "display_order"]
unknown = INCLUDE
class Study(object):
def __init__(self, title, last_updated, primary_investigator_id, user_uid,
id=None,
protocol_builder_status=None,
sponsor="", hsr_number="", ind_number="", categories=[],
files=[], approvals=[], **argsv):
self.id = id
self.user_uid = user_uid
self.title = title
self.last_updated = last_updated
self.protocol_builder_status = protocol_builder_status
self.primary_investigator_id = primary_investigator_id
self.sponsor = sponsor
self.hsr_number = hsr_number
self.ind_number = ind_number
self.categories = categories
self.approvals = approvals
self.warnings = []
self.files = files
@classmethod
def from_model(cls, study_model: StudyModel):
id = study_model.id # Just read some value, in case the dict expired, otherwise dict may be empty.
args = dict((k, v) for k, v in study_model.__dict__.items() if not k.startswith('_'))
instance = cls(**args)
return instance
def update_model(self, study_model: StudyModel):
for k,v in self.__dict__.items():
if not k.startswith('_'):
study_model.__dict__[k] = v
def model_args(self):
"""Arguments that can be passed into the Study Model to update it."""
self_dict = self.__dict__.copy()
del self_dict["categories"]
del self_dict["warnings"]
return self_dict
class StudySchema(ma.Schema):
id = fields.Integer(required=False, allow_none=True)
categories = fields.List(fields.Nested(CategorySchema), dump_only=True)
warnings = fields.List(fields.Nested(ApiErrorSchema), dump_only=True)
protocol_builder_status = EnumField(ProtocolBuilderStatus)
hsr_number = fields.String(allow_none=True)
sponsor = fields.String(allow_none=True)
ind_number = fields.String(allow_none=True)
files = fields.List(fields.Nested(FileSchema), dump_only=True)
approvals = fields.List(fields.Nested('ApprovalSchema'), dump_only=True)
class Meta:
model = Study
additional = ["id", "title", "last_updated", "primary_investigator_id", "user_uid",
"sponsor", "ind_number", "approvals", "files"]
unknown = INCLUDE
@marshmallow.post_load
def make_study(self, data, **kwargs):
"""Can load the basic study data for updates to the database, but categories are write only"""
2020-04-09 18:25:14 +00:00
return Study(**data)
2020-05-20 21:10:22 +00:00