Contents

• Best practicies
  • React Native
  • Typescript
  • Backend Python
  • Aidbox
• BS video library (meetup records)
• Learning materials
• SNOMED and Snowstorm
• React Native tips
• kubectl/k8s cheat sheet
• Tutorials
  • Setting up GitLab runner on AWS EC2

Best practicies

React Native

Debugging network requests in react-native

// To see all the requests in the chrome Dev tools in the network tab.
XMLHttpRequest = GLOBAL.originalXMLHttpRequest ?
    GLOBAL.originalXMLHttpRequest :
    GLOBAL.XMLHttpRequest;

Don't forget to remove it before pushing!

Open link in a running simulator

xcrun simctl openurl booted link-goes-here

Jest testing

There's a module.js, for example:

export const calculate = (a, b) => a + b;

How to mock it?

Mock a module with jest.mock

Common approach is to use jest.mock to automatically set all exports of a module to the Mock Function.

import * as module from "./module";

jest.mock("./module.js");

test("calls module.calculate", () => {
  module.calculate(1, 2);
  expect(module.calculate).toHaveBeenCalledWith(1, 2);
});

Mock a function with jest.spyOn + mockImplementation

You can mock a function then restore the original implementation:

import * as module from "./module";

test("calls math.calculate", () => {
  const calculateMock = jest.spyOn(module, "calculate");

  calculateMock.mockImplementation(() => 42);

  expect(module.calculate(1, 2)).toEqual(42);

  calculateMock.mockRestore();

  expect(calculateMock).toHaveBeenCalledWith(1, 2);
});

xcrun: error: SDK "iphoneos" cannot be located issue

sudo xcode-select --switch /Applications/Xcode.app

Fonts

Good tutorial how to add custom fonts: https://medium.com/@mehran.khan/ultimate-guide-to-use-custom-fonts-in-react-native-77fcdf859cf4

Typescript

Use type predicate for functions like filter to alter output type

const sleepMeasurements: (number | undefined)[]
const filteredMeasurements = sleepMeasurements.filter((v) => v !== undefined)) // will be (number | undefined)[]
// now use type predicate
const filteredMeasurements = sleepMeasurements.filter((v): v is number => v !== undefined)) // will be (number)[]

You can read more here User-Defined Type Guards

Determine array element's type

type Some = ['A', 'B', 'C']
type AnyOfSome = Some[number] // type AnyOfSome = 'A' | 'B' | 'C'

Backend Python

Always use _id instead of id in search params

Aidbox accepts both id and _id search params. But according to spec https://www.hl7.org/fhir/search.html the only valid param is _id. This is causing problems with access policies and may cause potential issues in future. So it'd be better to always use "_id" instead of "id" search params on frontend as well as backend part of an app.

Always user "token" type for SearchParameter instead of "string"

Mistakenly we use "string" as SearchParameter.type for some attributes with type "text", but it leads to serious issues, because FHIR search works as full-text search with substring partial matching. Example: SearchParameter User.email must have "token" type.

See FHIR Search specification about how different SearchParameter types work.

Try to avoid subscription in favor of the custom operation

If you need to mutate the original resource before/after save - use the custom operation to do it. It has some benefits:

• Version id won't be increased twice
• Easier to test
• You can make transactional queries: all or nothing

Don't use the display in references

Using display in references is a bad idea because it is denormalization and the original title can be changed so that you will need to actualize it. Actualization is a very difficult process and has some drawbacks:

• It changes version ID --- our UI usually can't handle version mismatch
• It takes time - you need to go through the database and find all references
• It can't work with other subscriptions (because actualization is a best practice for subscriptions)

If you don't want to have a headache with these problems, consider using _include every time to get the actual value.

User resource role in the system

User resources must contain only credentials and links to appropriate Patient/Practitioner.

Practitioners can be used for admin roles and other roles (see FHIR spec for details). Using references to User should be forbidden because makes a potential vulnerability (because of _include/_assoc that is not forbidden by default). Instead of this, always make references to appropriate Patient/Practitioner.

Access policy rules

They always should have additionalProperties: false for search params to forbid passing _assoc/_include/_revinclude.

Avoid using access_policy argument in the custom operation

Using access_policy argument in the customer operation makes it difficult to maintain policies for different roles

Generate future datetime ranges with DST/STD time shifting

In many countries there is a Daylight saving time shifting. While generating future time ranges it is necessary to consider such shiftings.

In Israel, DST->STD in 2020 is occuring on October 25. For example, there is a task to generate a datetime range from October 20 for 7 days with the 1 day step.

This code will prevent to an error starting from October 25, because datetime.timedelta doesn't consider the DST->STD time shift:

while current_date < end_date:
  # datetime.timedelta doesn't consider time shifting
  current_date += datetime.timedelta(days=1)

To fix this use DST shift adding/subtraction:

next_date = current_date + datetime.timedelta(days=1)

# Normalize datetime to the local timezone considering DST/STD
next_date = next_date.tzinfo.normalize(next_date)

# Add/subtract DST (3600 seconds) /STD (0 seconds) time shift
next_date = next_date + (current.dst() - next_date.dst())

Do not forget to cover this logic with tests by checking that all time values (hours, minutes, etc) in your range are equal and the range parts quantity is correct

Create resources via dict not via args

Bad

organization = sdk.client.resource(
  "Organization",
  name="beda.software",
  active=False,
  someAttribute="attribute value"
)

Good

organization = sdk.client.resource(
  "Organization",
  **{"name": "beda.software", "active": False, "someAttribute": "attribute value"}
)

Use dasherize search parameters names and URL paths in operations

Bad

"Encounter.hasInvoice": {
  "name": "hasInvoice",
  #...
},

Good

"Encounter.has-invoice": {
  "name": "has-invoice",
  #...
},

Bad

@sdk.operation(["GET"], ["Resource", {"name": "resource_id"}, "$operation_with_resource"])

or

@sdk.operation(["GET"], ["Resource", {"name": "resourceId"}, "$operationWithResource"])

Good

@sdk.operation(["GET"], ["Resource", {"name": "resource-id"}, "$operation-with-resource"])

QA, Backend Tests

Mocking free functions

Use module path the function is looked up in (it might be different from the module that defines the function). Official unittest documentation has a detailed description

Aidbox

Aidbox Migrations

We can use Aidbox REST Console to run SQL queries.

Reference: Postgres JSON Functions and Operators

Ex. 1: Deletion of an aidbox resource attribute

Updated resource example (the filter attribute to be deleted):

id: d6fdcc6a-4513-45df-8f06-797a4727ce7f
resourceType: MedicalRule
sourceQuery:
  id: Source
  type: batch
  entry:
    - request:
        url: $snowstorm
        method: POST
      resource:
        ecl: '<< 372586001 |Hypotensive agent (substance)|'
        filter: '...'  // <- to be deleted
        resourceType: SnowstormRequest

Migration SQL query (for all resources of this type):

update medicalrule 
set resource = jsonb_set(
   resource, 
   '{sourceQuery,entry,0,resource}', 
   (resource #> '{sourceQuery,entry,0,resource}')- 'filter'
)

Aidbox Resources

How to add an object attribute with Any content to an Aidbox resource

"Condition.property": {
    "path": ["property"],
    "resource": {"resourceType": "Entity", "id": "Condition"},
    "extensionUrl": "http://beda.software/fhir-extensions/condition-property",
    "schema": {"type": "object"},
},

Exceptions and OperationOutcome

It is better to return Exception or HTTPError in form of OperationOutcome resource.

You need fhirpy or aidboxpy version 1.3.0 or higher to import OperationOutcome. You need aidbox-python-sdk to enable sdk operations return relevant HTTPError if OperationOutcome exception raised.

Usage example:

from app.sdk import sdk
from fhirpy.base.exceptions import OperationOutcome, IssueType, IssueSeverity

@sdk.operation(["POST"], ["Observation", "$create"], )
async def observation_custom_create(operation, request):
    try:
        observation_data = request["resource"]["observation_data"]
    except KeyError as e:
        error_text = f"Attribute: {str(e)} is required"
        raise OperationOutcome(reason=error_text, code=IssueType.required.value, severity=IssueSeverity.error.value)

Backups

Restore backup to local dev environment

docker-compose up -d devbox-db
cat backup.dump |  docker exec -i environment_devbox-db_1 pg_restore  -d devbox

Emacs

Crash with watching hidden files

https://github.com/facebook/create-react-app/pull/10706/files

curl  https://raw.githubusercontent.com/facebook/create-react-app/a3dd649530f4c0c305fc183842399c9d36ecf97f/packages/react-dev-utils/ignoredFiles.js > node_modules/react-dev-utils/ignoredFiles.js
curl  https://raw.githubusercontent.com/facebook/create-react-app/a3dd649530f4c0c305fc183842399c9d36ecf97f/packages/react-dev-utils/ignoredFiles.js > ./node_modules/react-scripts/node_modules/react-dev-utils/ignoredFiles.js