Also, when returning error messages, attempt to include the task data for the task that caused the error.
Also, when attempting to delete any file, respond with an API error explaining the issue, and log the details.
I noticed the validation sometimes looks ahead for files, so looking at all the tasks now, not just the ready tasks for the lookup field.
Ran into an issue with validation where a workflow model was required, so I create one and delete it. Another refactor for another day.
Another speed improvement - data in the FileDataModel is deferred, and not queried until it is specifically used, as the new data structures need to use this model frequently.
Added a File class, that we wrap around the FileModel so the api endpoints don't change, but File no longer holds refences to versions or dates of the file_data model, we
figure this out based on a clean database structure.
The ApprovalFile is directly related to the file_data_model - so no chance that a reviewer would review the incorrect version of a file.py
Noticed that our FileType enum called "bpmn" "bpmm", hope this doesn't screw someone up.
Workflows are directly related to the data_models that create the workflow spec it needs. So the files should always be there. There are no more hashes, and thus no more hash errors where it can't find the files to rebuild the workflow.py
Not much to report here, other than I broke every single test in the system at one point. So I'm super concerned about this, and will be testing it a lot before creating the pull request.
From an API point of view you can do the following (and only the following)
/files?workflow_spec_id=x
* You can find all files associated with a workflow_spec_id, and add a file with a workflow_spec_id
/files?workflow_id=x
* You can find all files associated with a workflow_id, and add a file that is directly associated with the workflow
/files?workflow_id=x&form_field_key=y
* You can find all files associated with a form element on a running workflow, and add a new file.
Note: you can add multiple files to the same form_field_key, IF they have different file names. If the same name, the original file is archived,
and the new file takes its place.
The study endpoints always return a list of the file metadata associated with the study. Removed /studies-files, but there is an
endpoint called
/studies/all - that returns all the studies in the system, and does include their files.
On a deeper level:
The File model no longer contains:
- study_id,
- task_id,
- form_field_key
Instead, if the file is associated with workflow - then that is the one way it is connected to the study, and we use this relationship to find files for a study.
A file is never associated with a task_id, as these change when the workflow is reloaded.
The form_field_key must match the irb_doc_code, so when requesting files for a form field, we just look up the irb_doc_code.
Using the LDAP service for checking user details in development mode - even if you are using the back door.
Added a new Flask fucntion load-example-rrt-data that loads the rrt workflow, and not the CRC wrokflows.
Modified the "load-example-data" in the tests to use some test data, rather than loading up all the workflows[
in CRC each time, with a parameter to load crc data if that is required - which is enabled for just a handful of tests.
(Tests run in 1/4 the time now)
PB_ENABLED can be set to false in the configuration (either in a file called instance/config.py, or as an environment variable)
Added a check in the base_test, to assure that we are always running tests with the test configuration, and bail out otherwise. Setting TESTING=true as an environment variable will get this, but so well the correct ordering of imports. Just be dead certain the first file every test file imports is base_test.py.
Aaron was right, and we call the Protocol Builder in all kinds of awful places. But we don't do this now. So Carlos, you should have the ability to reuse a lot of the logic in the study_service now.
I dropped the poorly named "study-update" endpoint completely. We weren't using it. POST and PUT to Study still work just fine for doing exactly that.
All the tests now run and pass with the Protocol builder disabled. Tests that specifically check PB behavior turn it back on for the test, or mock it out.
updaing the user 'sso' endpoint to provide additional information for debugging.
Pulling information from ldap to stay super consistent on where we get our information.
Refactored calls into a new lookup_service to keep things tidy.
New keys for all enum/auto-complete fields:
PROP_OPTIONS_FILE = "spreadsheet.name"
PROP_OPTIONS_VALUE_COLUMN = "spreadsheet.value.column"
PROP_OPTIONS_LABEL_COL = "spreadsheet.label.column"
PROP_LDAP_LOOKUP = "ldap.lookup"
FIELD_TYPE_AUTO_COMPLETE = "autocomplete"
No Previous Task, No Last Task, No Task List. Just the current task, and the Navigation.
Use the token endpoint to set the current task, even if it is a "READY" task in the api.
Previous Task can be set by identifying the prior task in the Navigation (I'm hoping)
Prefering camel case to snake case on all new apis. Maybe clean the rest up later.
running all extension/properties through the Jinja template processor so you can have custom display names using data, very helpful for building multi-instance displays.
Properties was returned as an array of key/value pairs, which is just mean. Switched this to a dictionary.
I noticed we were saving the workflow every time we loaded it up, rather than only when we were making changes to it. Refactored this to be a little more careful.
Centralized the saving of the workflow into one location in the processor, so we can make sure we update all the details about that workflow every time we save.
The workflow service has a method that will log any task action taken in a consistent way.
The stats models were removed from the API completely. Will wait for a use case for dealing with this later.
Added error checking such that attempting to submit data for a task that is not in the "READY" state throws an error message.
For some reason I'm getting lots of errors in the tests as they try to hit API endpoints they were not hitting before, so adding a number of mocks to some of the study service tests.
INCOMPLETE = 'Incomplete in Protocol Builder',
ACTIVE = 'Active / Ready to roll',
HOLD = 'On Hold',
OPEN = 'Open - this study is in progress',
ABANDONED = 'Abandoned, it got deleted in Protocol Builder'
Found a problem where the documentation for elements was being processed BEFORE data was loaded from a script. There still may be some issues here.
Ran into an issue with circular dependencies - handling it with a new workflow_service, and pulling computational logic out of the api_models - it was the right thing to do.
some ugly fixes in the file_service for improving panda output from spreadsheet processing that I need to revist.
now that the spiff-workflow handles multi-instance, we can't have random multi-instance tasks around.
Improved tests around study deletion.
Created a Study object (seperate from the StudyModel) that can cronstructed on request, and contains a different data structure than we store in the DB. This allows us to return underlying Categories and Workflows in a clean way.
Added a new status to workflows called "not_started", meaning we have not yet instantiated a processor or created a BPMN, they have no version yet and no stored data, just the possiblity of being started.
The Top Level Workflow or "Master" workflow is now a part of the sample data, and loaded at all times.
Removed the ability to "add a workflow to a study" and "remove a workflow from a study", a study contains all possible workflows by definition.
Example data no longer creates users or studies, it just creates the specs.
Fixing adding a study so all workflows are again added, will add status on those workflows based on output from the master bpmn diagram, which is coming shortly.
Added a validate_workflow_specification endpoint that allows you to check if the workflow will execute from beginning to end using random data.
Minor fixes to existing bpmns to allow them to pass.
All scripts must include a "do_task_validate_only" that restricts external calls and database modifications, but performs as much logic as possible.
The protocol builder service now returns real models, not dictionaries, forcing proper validation and fail-fast behavior.
Changed the name of the "status" spec, to "top_level_workflow" and removing any connection a workflow or study has with this specification. It is only unused to determine status in real time, and is not reused or tracked.
Modified the required documents script to return a dictionary and not an array, making it easier to speak to specific values in the BPMN and DMN.
Working on new ways to test the top_level_workflow in the context of updates, this is still a work in progress.
Making use of several modifications to the Spiff library that enables more complex expressions in DMN models. This is evident in the new DMN models for the top_level_workflow
Required Documents is becoming complicated, so making this it's own script task, removing it from study_info.py
The file_service is now very aware of this irb_documents file, so it will always need to exist. We seed this file
during setup, but it can be overwritten by the configurator.
Split the API specific models out from the workflow models to help me keep this straight.
Added tests to help me understand the errors thrown the and resolution path when a workflow specification changes in the midst of a running workflow.
Adding a new script that script tasks can use to add in data about the study.
Moving all the test workflow specifications out of the main load.
fixing a pile of tests so they can find workflow specs that are now moved into the test directory.
Dan Funk
Carlos Lopez
Aaron Louie
Aaron Louie