@focuson/fetcher
v1.44.3
Published
Used when we want to fetch data from APIs in order to display it. This provides a declarative and composable approach to state management
Downloads
53
Readme
State fetch
A common pattern in GUIs is 'showing the data revealed by an API'. For example we might show the data in shopping site, or show account data about customers.
When doing this typically there is a navigation area which determines what is being looked at, and then multiple places that show information about what is being looked at. Each time the state is changed a decision needs to be made about whether to get more information from the backing API(s).
The core idea is to have all this done as part of the 'flux loop'. So the components communicating by setting parameters in the state such as 'the selected thing' and possible a 'tag' that explains how to load the data. Thus the components don't 'call actions' or anything
This is currently under heavy development: fetchers are being added, signatures being changed, in order to make the library 'more polished'
Key Concepts
State
The state is just a blob of Json. It can come from anywhere. It could easily be the state of the '@focuson/state' project, but can also be 'just a javascript object'.
In the state we record 'what we want to be loaded' and 'what we have loaded'. The job of this project is to make that ' declarative'. So we just 'change what we want to be loaded' and the correct things are loaded.
Fetchers
A Fetcher is responsible for describing what will be fetched from the back end. It doesn't actually fetch it, but the method applyFetchers(...) will go and do the fetching. This separation makes it really easy to test what is going to happen.
/** The fetcher is responsible for modifying state. It is effectively a PartialFunction */
export interface Fetcher<State, T> {
/** This works out if we need to load. Typically is a strategy */
shouldLoad: (newState: State | undefined) => boolean,
/** This provides the info that we need to load. The first two parameters can be passed to fetch, and the third is how we process the result. Note that it is hard to guarantee that the json is a T, so you might want to check it!*/
load: LoadFn<State, T>,
/** For debugging and testing. It's idiomatically normal to make this the variable name of your fetcher, or some other description that makes sense to you */
description: string
}
export interface LoadFn<State, T> {(newState: State | undefined): LoadInfo<State, T>}
export type LoadInfo<State, T> = [RequestInfo, RequestInit | undefined, MutateFn<State, T>]
export interface MutateFn<State, T> {(s: State | undefined): (status: number, json: T) => State}
A PartialFunction is a function that might not be callable. You need to call the 'shouldLoad' method first to see if the load function is safe or sensible to load.
- shouldLoad. Given a state will determine whether this fetcher should say 'load it'
- load. Returns a function that has three results. The first two can be passed as in to 'fetch' and the last one is 'how do I update the state with the results of the loaded data'
Fetchers
How to use fetchers
Typically if you are using react you have a 'start react place' where you define what is renderered. You include in this a fetcherTree which is effectively a collection of fetchers, and this method ochestrates things. So mostly you don't ' use' fetchers. You 'set them up to describe how to load stuff' and then have some selection state that you 'set' in your components. The infrastructure works out what to be loaded in the same way that the react infrastructure is given a description of what to display and works out what to actually change
How to test fetchers
Low level 'what would load tests'
For many fetchers we will have a test that says 'given this state the fetcher would load this', and other tests that say 'given this state the fetcher doesn't load. Often these tests are one line long.
Mid level 'what would load tests' on the fetcher graph
Again typically we have fetcher graph. We write tests that 'set up a state' and then ask the question 'what would be loaded'. These are our 'unit tests' which are cheap to write, quick to execute and read good. They allow us to check all the permutations we are interested in. These test things like 'do we load from the correct url', 'are we only loading things when we need to load them', 'is the thing we are loading being added to the correct place.'
Contract tests
I test to have a contract test for each 'url pattern' that I use that checks it actually loads what we want. (i.e. 'load this resource', 'list this resource').
Smoke tests
I'll have a very small number of smoke tests that run against 'the real world' which can either be an acceptance environment, or against the current production systems (that's a policy decision that the project makes. Which ever choice they make they typically regard the other choice with horror). In a payment application they might log in, move one euro from an account to another... but no permutations or unhappy paths. These are just checking that the connectivity and the most critical paths work. The permutations and contracts are validated by much lower level, faster, cheaper and better tests.
Predefined fetchers
Fetcher when undefined
export function fetcherWhenUndefined<State, Child>(optional: Optional<State, Child>,
reqFn: ReqFn<State>,
description?: string): Fetcher<State, Child> { /* */
}
export interface ReqFn<State> {
(newState: State | undefined): [RequestInfo, RequestInit | undefined] | undefined
}
This is a simple fetcher that uses the parameter 'lens' to focus in on a child and if that child is not present will load it. This is a great fetcher to use when you just 'need some data at the start' such as 'reference data'. The parameter 'optional' points at the item that is to be loaded. The RegFn returns the url and any other 'stuff' (such as method or headers) needed to load the data. As every the description is for debugging and tests.
Example
interface State {
initialData?: Data
}
const loadInitialData = fetcherWhenUndefined<State, Data>(identityOptics<State>().focusQuery('initialData'), s => ["http://the/url", {}])
Often applications I work with have a 'sitemap' end point that details how to find things. To use it
const loadSiteMapF = fetcherWhenUndefined<State, SiteMap>(
stateToSiteMapL,
(s) => ["someSitemapUrl", undefined], "loadSiteMapF");
fetchAndMutate
This is a 'composed fetcher'. You give it a fetcher and a function, and it first does whatever the raw fetcher would do, and if the fetcher goes to get data, it is mutated by the function.
Examples of where you might want to use this
- You want to load something but it is in the wrong format, so the mutate function puts it in the right state
- You want to load a thing that can be navigated (a sitemap, a list of products...) and then want to set up the selection state so that the first thing is the selected thing
- Every time you load a new user you clear down the state. This is great with 'login' logic.
loadSelectedFetcher
export const loadSelectedFetcher = <State, Holder, T>(tagFn: (s: State) => (string | undefined)[],
holderPrism: DirtyPrism<Holder, [string [], T]>,
target: Optional<State, Holder>,
reqFn: ReqFn<State>,
description?: string): Fetcher<State, T>
{}
Imagine we have a shopping application. We have things like 'the selected department - books/washing machines/...', and 'the selected product'. In this world we want to have a selection state that details these things, and a place in the state that is our selected item (the target). This fetcher is responsible for loading the target when the selection state varies.
Generics
State
is the type of the full state. This fetcher know nothing about it, other than 'there is one'Holder
When we load the thing, we need to store things about it (such as the current value of the tags). This is stored in a Holder. There is a 'default holder' calledHolder<T>
but you can use your own to make the names nicerT
The type that will be fetched.
Parameters
holderPrism: DirtyPrism<Holder, [string [], T]>
. DirtyPrisms 'Look scary' but this is actually just a constructor/destructor. Given aHolder
it know how to rip it apart into the list of tags and a T, and given a list of tags and a T it knows how to make the holder.tagFn: (s: State) => (string | undefined)[]
given a State, this rips out the bits of data that we care about. If they change then theloadSelectedFetcher
will fetch something. In the example we are discussing this would return['selectedDepartmentId', 'selectedProductId'
. Note that it returns the current values of these, not their namestarget: Optional<State, Holder>
where to put the loaded targetreqFn: ReqFn<State>
how to load the targetdescription
for tests and debugging
Example
const loadApiF: Fetcher<State, Api> = loadSelectedFetcher<State, ApiData, Api>
((s: State) => [s.selState?.selectedEntity, s.selState?.selectedApi],
stateL.focusQuery('apiData'),
apiDataHolder('H/ApiData'),
s => [`http//someDomain/api/${s.selState.selectedEntity}/api/${s.selState.selectedApi}`],
"loadApiF")
export interface State {
selState: SelectionState
apiData?: ApiData,
}
export interface SelectionState {
selectedEntity?: string,
selectedThing?: string,
selectedApi?: string
}
export interface ApiData {
tags: string[]
api: Api, //the item being loaded
source?: string[], //other data we might want to load in the future that is about the api
status?: string,
desc?: string
}
function apiDataHolder(description: string): DirtyPrism<ApiData, [string[], Api]> {
return dirtyPrism(ad => [ad.tags, ad.api], ([tags, api]) => ({tags: tags, api: api}))
}
To use this you just need to set the selection state, and the infrastructure does the rest. If the
tags (selectedEntity
& selectedApi
) change then in the main state the apiData
will be reloaded. Note the pattern
than dependant data about the api (metadata that will be loaded by other fetchers if needed)
is part of the ApiData so that it is always cleared if a new api is loaded.
ifEEqualsFetcher
We often only want data fetched if we are displaying it. Examples of this include 'status data', 'the selected item', 'the user history'. For this kind of scenario we have the 'ifEqualsFetcher'.
export function ifEEqualsFetcher<State>(condition: (s: State) => boolean, fetcher: Fetcher<State, any>, description?: string): Fetcher<State, any> {}
The fetcher is only invoked if the condition is true. Often the condition will be something like 'is the selected radio button equal to someValue'.
radioButtonFetcher
@Deprecated
We will probably delete this as the ifEqualsFetcher seems to do this job better and simple
function radioButtonFetcher<State>(
desiredRadioButton: (s: State) => (string | undefined),//The desired tag.
actualRadioButton: Optional<State, string>, // this is the tag that names the currently actually selected radio button
whichFetcher: (tag: string) => Fetcher<State, any>,
description?: string
): Fetcher<State, T>
Often our gui wants to load different aspects of a thing. For example when looking at a user we might want to go get
their profile, their order history, their favourite list. In this scenario only one is being shown at once so we have
a radioButtonFetcher
that works out what to show based on the desiredTag.
The react code that displays the component would be using the 'actual radio button' to select which view to show
generics
State
Themain state
. The fetcher knows nothing about this, other than that it exists
Parameters
desiredRadioButton: (s: State) => (string | undefined)
A function that given a state tells the fetcher what radio button the user wantsactualRadioButton: Optional<State, string>
This is the radio button that is 'currently loaded'. It is compared with the desiredRadioButtonwhichFetcher: (radioButton: string) => Fetcher<State, any>
Given a radio buttondescription?: string
Example
const loadApiChild = radioButtonFetcher<State>(
s => s.selState?.selectedRadioButton,
identityOptics<State>().focusQuery('radioButton'),
fromTaggedFetcher({'desc': loadApiDescF, 'src': loadApiSrc, 'status': loadApiStatus}),
"loadApiChild")
export interface State {
apiData?: ApiData,
selState: SelectionState,
radioButton?: 'desc' | 'src' | 'status'
}
export interface SelectionState {
selectedRadioButton?: string,
}
In this example when the code sets the s.selState?.selectedRadioButton
this is compared with the s.radioButton
and
if it varies the appropriate
fromTaggedFetcher({'desc': loadApiDescF, 'src': loadApiSrc, 'status': loadApiStatus}),
is fetched.
FetcherTree
interface FetcherTree<State> {
fetcher: Fetcher<State, any>,
children: FetcherChildLink<State, any>[]
}
interface FetcherChildLink<State, Child> {
lens: Lens<State, Child>,
child: FetcherTree<Child>
}
One fetcher on its own is usually not very useful. Typically we have to get all sorts of things from the back end. To
handle this we put fetchers into a fetcherTree. This is just a collection of fetchers with the nice feature that it
controls the order the fetchers are checked. Things higher up the tree are checked (and therefore loaded) before things
lower down the tree. The lens in the FetcherChildLink
is just a way to compose and reuse fetchers: usually it is the '
identity lens' as often all fetchers are based of 'State'.
In simple applications we probably have a single fetcher tree. In complex ones we will often make subtrees and stich them together. This allows the tests for the subtrees to be independent of the rest of the world.
Example
const loadApiF: Fetcher<State, Api> = loadSelectedFetcher<State, SelectionState, ApiData, Api>(
stateL.focusOn('selState'), apiDataHolder('H/ApiData'))((sel: SelectionState) => [sel?.selectedEntity, sel?.selectedApi],
stateL.focusQuery('apiData'),
loadFromSiteMap((siteMap, sel) =>
sel.selectedEntity && sel.selectedApi && siteMap[sel.selectedEntity].api[sel.selectedApi]._links.self),
"loadApiF")
const loadThingF: Fetcher<State, Thing> = loadSelectedFetcher<State, SelectionState, Holder<Thing>, Thing>(
stateL.focusOn('selState'), holderIso<Thing>('H/Thing'))(s => [s.selectedEntity, s.selectedThing],
stateL.focusQuery('thing'),
loadFromSiteMap((siteMap, sel) => sel.selectedEntity && sel.selectedThing && siteMap[sel.selectedEntity].things[sel.selectedThing].href),
'LoadThingF')
let apiDataL: Optional<State, ApiData> = stateL.focusQuery('apiData')
const loadApiDescF: Fetcher<State, string> = fetcherWhenUndefined<State, string>(
apiDataL.focusQuery('desc'),
loadFromSiteMap((siteMap, sel) => sel.selectedEntity && sel.selectedApi && siteMap[sel.selectedEntity].api[sel.selectedApi]._links.description),
"loadApiDescF")
const loadApiStatus: Fetcher<State, string> = fetcherWhenUndefined<State, string>(
apiDataL.focusQuery('status'),
loadFromSiteMap((siteMap, sel) => sel.selectedEntity && sel.selectedApi && siteMap[sel.selectedEntity].api[sel.selectedApi]._links.status),
"loadApiStatus")
const loadApiSrc: Fetcher<State, string[]> = fetcherWhenUndefined<State, string[]>(
apiDataL.focusQuery("source"),
loadFromSiteMap((siteMap, sel) => sel.selectedEntity && sel.selectedApi && siteMap[sel.selectedEntity].api[sel.selectedApi]._links.source),
"loadApiSrc")
const loadApiChild = radioButtonFetcher<State, ApiData>(
s => s.selState?.selectedRadioButton,
identityOptics<State>().focusQuery('radioButton'),
fromTaggedFetcher({'desc': loadApiDescF, 'src': loadApiSrc, 'status': loadApiStatus}),
"loadApiChild")
const demoTree: FetcherTree<State> = fetcherTree(
loadSiteMapF,
child(loadThingF),
child(loadApiF,
child(loadApiChild)))
Testing
The key to testing is the wouldLoad
function
function wouldLoad<T>(ft: FetcherTree<T>, state: T | undefined, depth?: number): WouldLoad[]
This is fantastic in tests
it("when sitemap loaded and api not loaded but radio selected", () => {
expect(wouldLoad(demoTree, {
sitemap: exampleSiteMap,
selState: {selectedEntity: "be", selectedApi: "a1", selectedRadioButton: "src"}
})).toEqual([
{fetcher: loadSiteMapF, load: false},
{fetcher: loadThingF, load: false},
{fetcher: loadApiF, load: true, reqData: ["/be/a1/api", {}]},
{fetcher: loadApiChild, load: true, reqData: ["/be/a1/source", {}]}
]
)
})
Here we have our fetcherTree
called demoTree
. We give it a state and wouldLoad
tells us what it would load.
WouldLoad
allows us to worry about 'what should happen'. If we have tested the individual fetchers, we can now test
the orchestration of those fetchers assuming that they work. This allows us to write lots of permutations covering lots
of possibility cheaply and quickly without worry about test data management.
Note that when you use this recommended style of testing, you do need to test the fetchers as well