I'm trying to get Mobx's autorun to work correctly.
My use case is I have one model that I like to serialize (or dehydrate) when it is changed and add that information to another model's data. This brings me rudimentary time travel of model states. Both are observables.
Edit: Idea in model separation is that one is app's data model and other should be completely separate library that I could use from the app. I need to track changes in the app regularly, but show UI for the state tool on the same page.
Now, autorun seems to make its own inferences of what I'm actually tracking. When I moved the model instance inside observing model's instantiation, autorun wasn't called anymore when changes happened. When model instance was created on the module top level, it worked as I expected. This was when I only changed one property of observing model (the one that gets changed by every autorun call). When I tried changing two things at once in the observing model, autorun was now called for these changes also, leading to a unending cycle (which Mobx caught).
I'd like to know how to express what I'm tracking with autorun function be more explicit, or wether there are other ways to keep track of model changes and update other model when anything happens.
Edit with code example.
This is what I did (greatly simplified):
class DataModel {
#observable one_state = null;
}
class StateStore {
#observable states = [];
}
let data = new DataModel();
let store = new StateStore();
autorun(() => {
store.states.push(data.one_state);
console.log("new data", toJSON(store.states));
});
data.one_state = "change 1";
data.one_state = "change 2";
And this creates circular dependency because autorun gets called for both original data model change and the resulting store change, whilst I'm only interested in tracking changes to the former.
Edit with working result:
class DataModel {
#observable one_state = null;
}
class StateStore {
#observable states = asFlat([]);
}
let data = new DataModel();
let store = new StateStore();
autorun(() => {
store.states.push(data.one_state);
});
data.one_state = "change 1";
data.one_state = "change 2";
As per #mweststrate answer, using asFlat with store's states variable and removing the logging from autorun broke the problem cycle.
It is a bit tough to answer this question without any real code. Could you share some code? But note that MobX works best if you make a small mind shift: instead of imperatively saying "if X happens Y should be changed" it is better to say "Y can be derived from X". If you think along those lines, MobX will really start to shine.
So instead of having two observable models, I think one of them should be a derivation of the other (by using computed indeed). Does that make sense? Otherwise, feel free to elaborate on your question a bit more :)
Edit:
Ok thanks for the code. You should remove the log statement to avoid it from looping; Currently you log the states model, so each time it changes, the autorun will run, adding the first item (again!), changing the stateModel etc...
Secondly I'm not sure whether the states list should be observable, but at least its contents should not be observable (since it is a snapshot and the data per state should not change). To express that, you can use the asFlat modifier, which indicats that the states collection should only be shallowly observable: #observable states = asFlat([]).
Does that answer your question?
Related
My Dev setup:
Qt version : Qt 5.15.0
OS: Embedded Linux
I have a list of information.
Assume I have a structure called MyStruct
My model class is having a member variable of QList of above structure, to hold data for my view. Whenever I am opening the view, I am updating the QList (Note: There may or may not be a change). Updating here is something like assigning a new QList to existing one. before assignment, I am calling beginResetModel and after assignment I am calling endResetModel,
void MyModelClass::SomeInsertMethod(const QList<MyStruct>& aNewData)
{
beginResetModel();
m_lstData = aNewData;
endResetModel();
}
One thing I believe can be improved, is putting a check, if the new data is different than the existing data and then doing the above. Something like this:
void MyModelClass::SomeInsertMethod(const QList<MyStruct>& aNewData)
{
if (m_lstData != aNewData)
{
beginResetModel();
m_lstData = aNewData;
endResetModel();
}
}
Apart from that, is there any possibilities of getting a performance issue for calling beginResetModel/endResetModel? I m seeing a very small delay in the view coming up in my application.
I checked the documentation of QAbstractItemModel for above methods. Didn't get anything specific to the performance issue.
The other way, which this can be done, is by individually comparing the elements of the lists and triggering a dataChanged signal with appropriate model index and roles. But I feel, this will unnecessarily introduce some additional loops and comparisons, which again may cause some other performance issue. Correct me if I am wrong.
Is there any advantage of using dataChanged over beginResetModel/EndResetModel?
Please let me know your views on the above.
I'm working on a list item that can have an image at the start of the view, and when edit mode is selected, a switch replaces the image and it becomes draggable. My initial thought was to use a data class for the image, since the developer would need to pass both the Painter object as well as a contentDescription for accessibility reasons, like this:
#Composable
fun ListItem(
startIcon: StartIcon? = null,
title: String
) {
Row(...) {
startIcon?.let {
Image(painter = startIcon.painter, contentDescription = startIcon.description)
}
...
}
}
After reading comments on reddit about a medium article on the topic (see here) , most of the commenters seem to agree that using data classes for this purpose will create performance issues due to recomposition initialising new data classes willy nilly, and recommended using state instead.
How would i go about creating state for these two cases? It seems obvious that an edit mode would be a state, but I'm not sure if it overcomplicated for whether or not an image should be shown. Additionally, how would this look? What is the best way to manage this state while ensuring the view itself is stateless? I'm pretty new to Compose, and given how important it is i'd like to ensure i do not start out learning bad habits and antipatterns
Thanks!
Within the mobx-react documentation there are variations in how stores are created. For example, on the React Context page:
In the first code sample, the store is instantiated with useLocalStore:
const store = useLocalStore(createStore)
In the second code sample, the stores are initiated by directly "newing" the stores":
counterStore: new CounterStore(),
themeStore: new ThemeStore(),
By inference, the first is a "local" store (and thus needs useLocalStore), and the second is a "global" store and thus doesn't. However, it is not clear why this is, and what the subsequent differnce in behaviour is.
Why is useLocalStore not needed in second example, and what difference does this make to the behavior of the stores and mobx within React?
Thanks for any input
OK, I found the answer. useLocalStore turns a javascript literal into a store with observable properties. This is not needed if a store is created from a class object with observable attributes.
Thanks to #freddyc for the answer
useLocalStore has been deprecated in favor of useLocalObservable, see here.
Links from the question are no longer valid, mobx-react docs are now here:
https://mobx.js.org/react-integration.html
useLocalObservable(initializer, annotations)
is just a shorthand for:
useState(() => observable(initializer(), annotations, {autoBind: true}))[0]
useLocalStore is almost the same, but it was deprecated because it wasn't as practical.
Why is this needed?
useState() is the main way to store a variable in React.
observable({...}) is the main way to create observable objects in Mobx.
And when using Mobx+React you'll often need the useState + observable combo, so it is nice to have a shorthand.
When is it not needed?
If you are using a class to create your store then you don't need the observable wrapper and you can simply do:
const myStore = useState(() => new MyStore())[0];
If you are using React class components you can save the store on a class field.
class MyComponent
{
myStore = new MyStore();
render() {...}
}
Bonus tip:
Why do we use useState for memorizing values instead of useMemo?
useMemo is meant to be used more like a cache for performance optimization, rather than for storing app state.
class FilterCriteria {
#observable filter = new Map();
}
let criteria = new FilterCriteria ();
// setting up a reaction when something in the filter changes
// (property added, removed, or changed)
reaction(()=>criteria.filter, data => console.log(data.toJSON()));
criteria.filter.set('name', 'John'); // setting a new property.
I would expect the above code to print out { 'name': 'John' }, but it seems that the reaction is not running.
I suspect that I set up the reaction in the wrong way. I want to react whenever a new key is added, an existing key is removed or a key value is changed. I don't know the keys or values at compile time.
How am I supposed to do that?
UPDATE
I changed my code to
class FilterCriteria {
#observable filter = new Map();
#computed get json(){ return this.filter.toJSON(); }
}
...
reaction(()=>criteria.json, data => console.log(data));
and now it seems to work properly. The reaction sideffect is executed whenever I add, remove or change a value in the Map.
So the question is why the reaction did execute in the second but not in the first example?
UPDATE 2
I changed my code again for a second time. I reverted to almost the first version but this time instead of reacting on criteria.filter and logging data.toJSON(), i react on criteria.filter.toJSON() and I log data (toJSON is moved from the sideffect to the value being watched). This time the reaction runs normally.
class FilterCriteria {
#observable filter = new Map();
}
reaction(()=>criteria.filter.toJSON(), data => console.log(data));
Again, I don't understand why. If criteria.filter is not an observable in itself then how does the watched expression is reevaluated when something inside criteria.filter is changed?
UPDATE 4 (hope the final one) SOLUTION
According to MobX documentation, mobx reacts to any existing observable property that is read during the execution of a tracked function.
reaction side-effect executes when the observable property changes. In my example, when reacting to criteria.filter , the observable property that is read here is filter, but the filter itself never changes. It is the same map always. It is the properties of filter that change. So the reaction is never run for criteria.filter.
But when I react on criteria.filter.toJSON() or mobx.toJS(criteria.filter), the reaction is executed correctly.
So why is that? criteria.filter doesn't change, and toJSON is not an observable property. It is a function. same for mobx.toJS. It seems no properties are read here. But this is not correct. As the documentation states (but not so emphatically), the properties of criteria.filter are indeed read when toJSON or mobx.toJS is executed, because both functions create a deep clone of the map (thus iterating over every property).
Now, in the beginning, the Map did not contain any property. So how is it that newly added properties are tracked, since they did not exist (to be read) when tracking begun? This is a map's feature. Maps provide observability for not yet existing properties too.
In MobX 5 you can track not existing properties of observable objects (not class instances) too, provided that they were instatiated with observable or observable.object. Class instances don't support this.
In mobx you have two options when you want to observe changes to something that is observable. reaction and observe. Reaction allows you to specify when you want some function to be called when a specific aspect of the observable changes. This could be changes to an array length, keys, properties, really anything. observe will trigger some function any time that the observable has changed.
I suspect the reason that your reaction hasn't been triggered is because of the first function. () => criteria.filter. This will not be triggered when a key is added/removed or a value changed. Instead, it will be triggered when filter actually changes. And since filter is really a reference to the Map, it will never change, even when the Map itself changes.
Here are some examples to illustrate my point:
If you want to trigger a reaction when a key has been added or removed, you may want your function to be:
() => criteria.filter.keys()
The result of this function will be different when a key has been added or removed. Similarly, if you want to trigger a reaction for when a value has been modified, something like this should work:
() => criteria.filter.values()
So some combination of those two should be what you need to listen to changes to keys/values. Alternatively, you could use observe, which will trigger on every change and require you to check what has changed to ensure that your specific conditions have been met to warrant calling a function (ie. key/value change)
UPDATE: Here is an example that illustrates the problem
#observable map = new Map();
Lets say that the value of map in memory is 5. So when you check map === map, it is equivalent to 5 === 5 and will evaluate to true.
Now, looking at the first code snippet you posted:
reaction(() => map, data => console.log(map.toJSON()));
Every time you add/remove a key or change a value, that first function will run. And the result will be 5, since that is what we said the value in memory is for this example. It will say: the old value is 5, and the new value is 5, so there is no change. Therefore, the reaction will not run the second function.
Now the second snippet:
reaction(() => map.toJSON(), data => console.log(data));
At first the result of the function will be: {} because the Map is empty. Now lets add a key:
map.set(1, 'some value');
Now, the result of the first function will be:
{"1": "some value"}
Clearly, this value is different than {}, so something has changed, and the second function of the reaction is called.
My Redux store is normalized, i.e. it's quite flat and each entity type has it's own slice.
Here is a simplified example of my Redux store:
drawings
1: {name:'D1', thumbnailId: 33}
2: {name:'D2', thumbnailId: 34}
thumbnails
33: {filePath: 'path to local file'}
34: {filePath: null (i.e. needs to be downloaded)}
The listview that shows the drawings whith respective thumbnail needs to be re-rendered when:
Changes in drawings-slice occurs, i.e. new, removed or updated drawings
Changes in any of the referenced thumbnails occurs, i.e. thumbnail 34 eventually gets downloaded (download is handled async by a Redux-Saga)
My current mapStateToProps is obviously flawed, as it does excessive selections from the Redux store. This happens before the actual view gets hold of its new props, so I cannot control this from shouldComponentUpdate.
There are several places in my app where I need a better solution for this.
Here is my flawed mapStateToProps (I'm using Immutable.js):
(state, ownProps) => {
const drawings = selectProjectDrawings(state, ownProps.projectId).map(drawing => {
const thumbnailFileGuid = drawing.get('thumbnailFileGuid');
if (!thumbnailFileGuid) return drawing;
const filePath = selectFile(state, thumbnailFileGuid).get(THUMBNAIL_FILE_PATH_FIELD);
return filePath ? drawing.set('_thumbnailFilePath', filePath) : drawing;
});
return {
drawings: drawings
};
}
Edit:
A really bad thing with my current solution is that I'm creating new drawing object by augmenting them with _thumbnailPath. This means that I cannot compare the object references in shouldComponentUpdate, as the reference is always altered.
Being able to compare object references is one of the main argument for not mutating the objects, but I have thrown away this opportunity with my flawed solution.
I would suggest placing this logic behind a selector. reselect is a selector memoization library which means that it will do a deep compare of the values for you and if the returned object and all the properties within stay the same, (including deep tree like structures) then it will return the original object.
React then will notice this is the same object instance (not a new one) and not re-render. It is OK if you recreate the drawing object within the selector, since you are not capturing this in any persistent state.
Side note: As your app grows, you will notice that all connected components will get mapStateToProps called even if it has nothing to do with it, therefore using a memoized selector really helps here. Otherwise your componentShouldUpdate gets complex real quick.