Watchers
Basic Example
Computed properties allow us to declaratively compute derived values. However, there are cases where we need to perform "side effects" in reaction to state changes - for example, mutating the DOM, or changing another piece of state based on the result of an async operation.
With Composition API, we can use the watch
function to trigger a callback whenever a piece of reactive state changes:
<script setup>
import { ref, watch } from 'kdu'
const question = ref('')
const answer = ref('Questions usually contain a question mark. ;-)')
// watch works directly on a ref
watch(question, async (newQuestion, oldQuestion) => {
if (newQuestion.indexOf('?') > -1) {
answer.value = 'Thinking...'
try {
const res = await fetch('https://yesno.wtf/api')
answer.value = (await res.json()).answer
} catch (error) {
answer.value = 'Error! Could not reach the API. ' + error
}
}
})
</script>
<template>
<p>
Ask a yes/no question:
<input k-model="question" />
</p>
<p>{{ answer }}</p>
</template>
Watch Source Types
watch
's first argument can be different types of reactive "sources": it can be a ref (including computed refs), a reactive object, a getter function, or an array of multiple sources:
const x = ref(0)
const y = ref(0)
// single ref
watch(x, (newX) => {
console.log(`x is ${newX}`)
})
// getter
watch(
() => x.value + y.value,
(sum) => {
console.log(`sum of x + y is: ${sum}`)
}
)
// array of multiple sources
watch([x, () => y.value], ([newX, newY]) => {
console.log(`x is ${newX} and y is ${newY}`)
})
Do note that you can't watch a property of a reactive object like this:
const obj = reactive({ count: 0 })
// this won't work because we are passing a number to watch()
watch(obj.count, (count) => {
console.log(`count is: ${count}`)
})
Instead, use a getter:
// instead, use a getter:
watch(
() => obj.count,
(count) => {
console.log(`count is: ${count}`)
}
)
Deep Watchers
When you call watch()
directly on a reactive object, it will implicitly create a deep watcher - the callback will be triggered on all nested mutations:
const obj = reactive({ count: 0 })
watch(obj, (newValue, oldValue) => {
// fires on nested property mutations
// Note: `newValue` will be equal to `oldValue` here
// because they both point to the same object!
})
obj.count++
This should be differentiated with a getter that returns a reactive object - in the latter case, the callback will only fire if the getter returns a different object:
watch(
() => state.someObject,
() => {
// fires only when state.someObject is replaced
}
)
You can, however, force the second case into a deep watcher by explicitly using the deep
option:
watch(
() => state.someObject,
(newValue, oldValue) => {
// Note: `newValue` will be equal to `oldValue` here
// *unless* state.someObject has been replaced
},
{ deep: true }
)
Use with Caution
Deep watch requires traversing all nested properties in the watched object, and can be expensive when used on large data structures. Use it only when necessary and beware of the performance implications.
watchEffect()
watch()
is lazy: the callback won't be called until the watched source has changed. But in some cases we may want the same callback logic to be run eagerly - for example, we may want to fetch some initial data, and then re-fetch the data whenever relevant state changes. We may find ourselves doing this:
const url = ref('https://...')
const data = ref(null)
async function fetchData() {
const response = await fetch(url.value)
data.value = await response.json()
}
// fetch immediately
fetchData()
// ...then watch for url change
watch(url, fetchData)
This can be simplified with watchEffect()
. watchEffect()
allows us to perform a side effect immediately while automatically tracking the effect's reactive dependencies. The above example can be rewritten as:
watchEffect(async () => {
const response = await fetch(url.value)
data.value = await response.json()
})
Here, the callback will run immediately. During its execution, it will also automatically track url.value
as a dependency (similar to computed properties). Whenever url.value
changes, the callback will be run again.
You can check out this example with watchEffect
and reactive data-fetching in action.
TIP
watchEffect
only tracks dependencies during its synchronous execution. When using it with an async callback, only properties accessed before the first await
tick will be tracked.
watch
vs. watchEffect
watch
and watchEffect
both allow us to reactively perform side effects. Their main difference is the way they track their reactive dependencies:
watch
only tracks the explicitly watched source. It won't track anything accessed inside the callback. In addition, the callback only triggers when the source has actually changed.watch
separates dependency tracking from the side effect, giving us more precise control over when the callback should fire.watchEffect
, on the other hand, combines dependency tracking and side effect into one phase. It automatically tracks every reactive property accessed during its synchronous execution. This is more convenient and typically results in terser code, but makes its reactive dependencies less explicit.
Callback Flush Timing
When you mutate reactive state, it may trigger both Kdu component updates and watcher callbacks created by you.
By default, user-created watcher callbacks are called before Kdu component updates. This means if you attempt to access the DOM inside a watcher callback, the DOM will be in the state before Kdu has applied any updates.
If you want to access the DOM in a watcher callback after Kdu has updated it, you need to specify the flush: 'post'
option:
watch(source, callback, {
flush: 'post'
})
watchEffect(callback, {
flush: 'post'
})
Post-flush watchEffect()
also has a convenience alias, watchPostEffect()
:
import { watchPostEffect } from 'kdu'
watchPostEffect(() => {
/* executed after Kdu updates */
})
Stopping a Watcher
Watchers declared synchronously inside setup()
or <script setup>
are bound to the owner component instance, and will be automatically stopped when the owner component is unmounted. In most cases, you don't need to worry about stopping the watcher yourself.
The key here is that the watcher must be created synchronously: if the watcher is created in an async callback, it won't be bound to the owner component and must be stopped manually to avoid memory leaks. Here's an example:
<script setup>
import { watchEffect } from 'kdu'
// this one will be automatically stopped
watchEffect(() => {})
// ...this one will not!
setTimeout(() => {
watchEffect(() => {})
}, 100)
</script>
To manually stop a watcher, use the returned handle function. This works for both watch
and watchEffect
:
const unwatch = watchEffect(() => {})
// ...later, when no longer needed
unwatch()
Note that there should be very few cases where you need to create watchers asynchronously, and synchronous creation should be preferred whenever possible. If you need to wait for some async data, you can make your watch logic conditional instead:
// data to be loaded asynchronously
const data = ref(null)
watchEffect(() => {
if (data.value) {
// do something when data is loaded
}
})