I like the elegant simplicity of Redux, but it's so low-level and definitely feels like people are still figuring the "right" way to do a lot of things.
Heh. Can definitely vouch for that. I recently put together a links repo cataloging most of the Redux-related ecosystem, and was amazed at how many overlapping or similar libraries there were. 12 different promise middlewares, a couple dozen different side effects plugins, and I don't know _how_ many utilities to generate action creators, reducers, and constants.
Thanks for putting that together! I'd discovered it a while ago and have been checking out all the utility libraries, trying to find ones that work the way I want. The funny thing is that so far none do, so I'm now working on one of my own...
I may change my tune six months from now, but part of the problem I see is that new people getting into React are told to not learn redux to "keep things simple." While it's a little bit of a pain to set up, redux (+ redux-thunk) ultimately reduces the amount of complexity and magic considerably.
Of you can write stateless functional components and have async and not use them, but the disadvantages of components without lifecycle methods appear much greater if you're just using vanilla React.
I don't know much about Redux but these tricks seem like they should be less trick and more re-worked flow.
For the first one, if a burger is meant to be consumed differently than fries (serial versus parallel) they why are they using the same mechanism with the burger having a hacked on predicate , service type, dispatch, etc? Why wouldn't the burger just have a workflow that makes sense for how its invoked? For instance if the burger is tied to the UI and a backend request that must happen one at a time then the UI would disallow or provide a way of queueing which I guess this sorta supports but it seems forced and doesn't look like a normal flow.
The second part, ignoring async responses. I don't understand this part. Why can't you just replace your state and drop the existing handlers so nothing async from the previous state occurs? Does redux just not support this out of the box or is the author doing something very edge-case-y?
When I don't want to handle an async response anymore I remove the handle's association with the event it was going to handle. It works like this in native DOM, jQuery, even my own bias data point: msngr. But I'm curious why wouldn't redux be able to handle the same case?
There are many reasons why the UI should not care about how the server requests are ordered.
> When I don't want to handle an async response anymore I remove the handle's association with the event it was going to handle. It works like this in native DOM, jQuery, even my own bias data point: msngr. But I'm curious why wouldn't redux be able to handle the same case?
In our case we use the same connection but it switches "contexts" and we want to ignore everything from a previous context. In simpler situations you could make sure the your connection is destroyed before the UI is destroyed, yes.
> There are many reasons why the UI should not care about how the server requests are ordered.
Sure. It was just an example. My main point was the change here felt like a hack and they forced a mechanism that should be processed serially into one that is meant to handle parallel.
> In our case we use the same connection but it switches "contexts" and we want to ignore everything from a previous context. In simpler situations you could make sure the your connection is destroyed before the UI is destroyed, yes.
So wouldn't it be best to sever ties that will cause issues and reconnect them versus anything else? That's typically the pattern you use when you want to replace event handlers. The way in the article just wastes a lot of CPU cycles to support an awkward pattern I've never seen someone use before.
Just my 2 cents. But like I said I've never used Redux so maybe these patterns are normal here.
I just started with redux-saga which uses ES6 generators to handle app control flow.
After first impressions, I'd say it's worth a look if you're running into anything near as complicated as talked about here. It's very easy to reason about and test. gaearon - the creator of redux - seems to approve of it as well.
I just wrote our API layer with redux-saga too, I started with thunks but I didn't like dispatching thunks and actions, also it was hard to reason about flow and side effects:
If I update my search form, the results need to be updated, but the action is "update form", not "update form and update search results", the search result update is a side effect more than something that should be actually fired from the update form.
Saga is making side-effects (workflows) possible by subscribing to events.
So in this case I have a saga waiting for 'SEARCH_FORM_UPDATE', then it waits for 300ms, then starts the ajax request flow: a request start action, then a request complete action (or request error).
The code reads almost like synchronous code once you understand yield.
It also allows you to listen to actions that you do not control: we wanted to do something whenever a specific redux-router action happened. You can't do this with thunks, we would have needed a middleware to do this. With sagas you can simply hook up a flow to the action.
The isEatingBurger (isLoading) flag seems to be idiomatic in Redux and shows up in tutorials. In addition to preventing overlapping requests, it's useful for spinners, disabling submit buttons, etc. It's cool that you rolled the isLoading pattern into a service.
If you really want to pass around the promise, it's doable. To save in your component state, return the promise from your thunk so it gets returned by dispatch (thom_nic just posted this also). To save in your store, dispatch a 'save' action from the thunk with the promise as an argument. But I agree that it's probably a bad design.
The request sequence IDs are useful for many situations where you need to order or cancel async requests. If you have multiple in-flight requests for an autocomplete field, you only want to set isLoading=false when the last request completes, and you don't want the result of a later request to overwrite the result of an earlier request if the responses come back out-of-order.
My usual approach is to avoid middlewares for async-requests and to keep this stuff in a separate layer. For example, it can look like an ApiClient object with ES7 async-methods. Such methods would:
In addition to redux-saga (https://github.com/yelouafi/redux-saga) which others have mentioned, redux-loop (https://github.com/raisemarketplace/redux-loop) is another interesting "declarative" approach to sequencing operations that uses promises. I'm not using either (yet), but I wrote up my notes on the differences between them and the thunk middleware approach recently (https://blog.boldlisting.com/connecting-redux-to-your-api-ea...). One aspect of these discussions that I find is lost or at least very implicit is the notion of storing metadata about the status of async operations as well as their result is elided together, but there's reason to make mindful choices about whether or not to do that.
BTW, the "Ignoring Async Responses" section here is a useful pattern, even if you're not working on browser developer tools ;). I've been working on a web app and have a polling timeout that I cancel on logout. Not quite the same, but logout triggers a reset of all the state obtained while authenticated, but the timeout ids are also in the store and canceled before being cleared.
"In an asynchronous world, you have 3 actions indicating asynchronous work: start, done, and error. In our system, they all are of the same type (like ADD_BREAKPOINT) but the status field indicates the event type."
This is a great suggestion (I have used a very similar schema before, although in flux), and should be the standard suggestion in redux tutorials.
Not enforcing this can lead to a proliferation of inconsistently named constants, especially if a team is learning flux/redux and has not yet settled on a naming schema.
As a follow-up, one alternative method I have been exploring for solving the author's problems (an alternative to wait-service) is as follows:
* if the action creator is passed a UUID, it includes that UUID in all of the start|succeed|error events it dispatches.
* There is a request store that sits on top of the base store, which watches for events with UUIDs. If it finds one, it stores in its own map UUID => { state: start|succeed|done, IDs: [Array of IDs that came back from the AJAX request] }
* the request store exposes UUID => { state, array of resources } (i.e. it talks to the base store for you)
* Components that know exactly which resource they need (e.g. id=1) continue listening to the base store
* Components that want the results of an ajax call pass a new UUID every time their filter state changes, and thus can be certain that their UI matches the data they are displaying
One problem I have with standard action is that on failure I sometimes want to have more than just an Error as payload. Something like `payload: {error, seqId}`. You could add it to error itself, but I find it more consistent to put it into the payload directly.
Not enforcing this can lead to a proliferation of inconsistently named constants
Can you explain this further? Our team uses the ACTION, ACTION_SUCCESS, ACTION_FAILURE pattern, and we haven't run into any issues. I like it because it's a bit easier to parse the sequence of actions in the console at a glance, and async actions are more clearly distinguished from synchronous ones. The main downside is having a bit more boilerplate.
I think that's a good pattern, which is to say I think you're enforcing it correctly.
I think that as a follow up, you might consider whether stores should respond to multiple ACTION's. You may find that all of the actions are forms of "take this item with id: X and replace it with this updated item" or an array of such commands
The EAT_BURGER seems slightly odd, as i've never hit a situation where I would allow a second one to be fired while the first is still pending. I.E. upon EAT_BURGER.start i would disable the UI for the EAT_BURGER action to be fired. I guess that's an idealised world and you've hit something more complicated that requires it though, and the solution is good
I particularly like the async call tracking - very simple and neatly solves your issue!
re: an action caller needing to wait on the async action to complete: I thought that was a non-issue. If you use redux-thunk and your action creator returns a promise, the caller gets it. So while you're no longer triggering off of a redux action/store state change -- which may or may not be seen as a good thing -- you can get the promise and add a `.then()` which will fire after the async action completes.
Basically: component fires a 'delete' action, and then does page navigation when delete is successful. (Not saying this is a good idea to do it that way, just that you can do it.) Or did I miss his point?
My post says exactly this. Read the full thing. There are other cases where you do not have a reference to the promise. (think a function being called again at some indeterminate time)
Basically a UI optimization. Instead of firing multiple requests to the server a lot, we know that a specific UI event may happen multiple times in a short period, so we "throttle" it by only performing the request once at a time and only re-requesting it if it was requested again.
So far it's always been a UI optimization like that.
You could handle this at the connection layer, but I liked how it integrates with actions better (it gives a finer control over when actions are dispatched)
I tweeted my feelings on this last week:
https://twitter.com/tlrobinson/status/717258992989306880