To use effects managers or not to use effects managers. The Elm guide basically says "don't" and then proceeds to not provide much documentation about how to write an effects manager if you wanted to. I guess one could take that as a hint.
TL;DR You don't need to write effects managers but you will end up writing more (and more explicit) code if you don't. If you replace commands with the list of out messages pattern discussed elsewhere on this list — an internal update function doesn't return a command but rather a list of out messages to be processed by the next level up (where processing can include simply forwarding) — then you can, I believe, do everything that a command-only effects manager can do. At the top level of your app, you recognize the out messages destined for the pseudo-effects-manager, collect them up, and deliver them to an update function for that pseudo-effects-manager. When the pseudo-effects-manager has responses, it delivers them via out messages that the top level can relay down to the appropriate component. This does tend to result in tagger functions getting stored in the pseudo-effects-manager portion of the model. Effects manager store functions in their state and we would expect the same for pseudo-effects-managers. This does result in more plumbing code than effects managers. That's because effects managers are taking advantage of special routing support in the language and runtime and this approach requires building it all oneself. That said, one of the benefits often cited for functional programming and the justification for things like threading context information through view functions instead of relying on mutable globally visible state is that it is both more clear and more explicit. So, if magic bothers you and you prefer to be explicit, this approach should be preferable to the effects manager approach. All that said, what about subscriptions? Here we have more of a problem. The subscriptions function cannot mutate the model so it can't be built by flowing up an internal subscription equivalent and handing those to our pseudo-effects-manager to decide what to do because the pseudo-effects-manager can't change state either. A real effects manager, existing outside of the model, can and generally does update in response to the subscriptions. Effects managers enable one to have state outside of one's model that can mutate even when one's model cannot. So, what could we do? If we are willing to update pseudo-subscriptions on every update — and as I understand it, that's what the Elm runtime does with real subscriptions — then we could certainly do this at the end of our top level update functions. In other words, subscriptions are feasible but maybe a bit less pretty in how the logic flows. Mark On Wed, Jun 1, 2016 at 12:45 AM James Wilson <[email protected]> wrote: > Thanks for your help; I'll have a ponder and probably end up taking an > effect manager route again (I started this way and then ended up with some > weird hybrid where my cache was in the main elm app and an effect manager > did the "lower level" api stuff; but I'm not so happy with it) > > As another effect manager code point if you're intrigued how they work > (and my comments are in any way useful, which they may not be); here's an > effect manager I made for sending and receiving things from a backend I'm > wiring my app up to (so it uses custom Cmds and Subs): > > > https://github.com/jsdw/tl-asset-browser/blob/3e9e8527f65ae0a0a2d5d00d01779bdfdf03701d/src/elm/Api.elm > > Not sure whether it's any help (may just be easier looking at the elm-lang > effect managers!) but just in case :) > > > On Wednesday, 1 June 2016 08:31:11 UTC+1, Peter Damoc wrote: > > I haven't used an effect manager for this because I haven put in the time > needed to learn how to create effect managers. :) > > If what I've shown here can be accomplished with an effect manager then > that's the way it should be done. :) > > > > > > > On Wed, Jun 1, 2016 at 10:21 AM, James Wilson <[email protected]> wrote: > > Thanks, that looks like basically exactly what I'd have guessed :) It's > super useful seeing an actual code sample with these ideas in. > > One thing I wonder now is; why not use an effect manager for this? It > basically seems to fit the exact same space (allows you to create a custom > Req like thing that can be mapped and batched and passed up the component > hierarchy - except that it's just a Cmd instead and plays nice with other > Cmds; allows you to maintain and update state (the cache) as you go; allows > you to "convert" Reqs to tasks to be run - just Cmds again now). In fact, > effect managers don't really seem to help you do anything other than what's > described here (plus a subscription side if you want it). Are there any > cons to using an effect manager here that you have in mind? > > On Tuesday, 31 May 2016 20:43:41 UTC+1, Peter Damoc wrote: > > The updating of the cache sounds to me like this: > > 1. if we have the info in cache, just supply the info without a HTTP GET > 2. if we don't have the info in cache, return a different Msg that > encapsulates the msg that requested the original information and the info > required for the cache update. > > Here is a quick update of the code I've previously posted to include this > caching mechanism. > > https://gist.github.com/pdamoc/d492ab58023926cd4d4950f12e5e170d > > > > > On Tue, May 31, 2016 at 10:05 PM, James Wilson <[email protected]> wrote: > > The key part that's not coded in the gist is the use of a cache/global > state object, however I think I see what you're getting at - pass back up > the chain a Req object, say, and at the top we can turn it into a Cmd > using, say, some top level global state as well as whatever other data we > need. This may lead to a request being made to the server or it may not. > > The other part of the puzzle is actually updating the cache when a request > is made. Req.toCmd for instance could return an updated GlobalState so that > it's able to cache "pending" states on values (so that we can avoid > duplicating requests). To update the cache when the response actually comes > in we could have toCmd return a Cmd.batch of 2 commands, one that will > fail/succeed and send a message to the component that initiated the Req, > and one that will send a message aimed at the top level cache itself. > > Thanks Peter, I'll definitely mull over this! > > On Tuesday, 31 May 2016 19:45:42 UTC+1, Peter Damoc wrote: > > ADT in Elm is one of its most powerful weapons. > > You could encapsulate your requests in a type and use this type at top > level to fulfill them. > > For example: instead of returning Cmd msg you return some Req msg that can > be turned into a Cmd msg at top level based on some context information. > > Here is a gist with a skeleton of how I view this implemented: > https://gist.github.com/pdamoc/a47090e69b75433efa60fe4f70e6a06a > > I've sent the base of the URL as a simple String in `Req.toCmd` but you > can imagine a more complex type holding all kind of information (e.g. > cache, auth, etc ) . > Also, I've kept the type of the Req simple (only saved the rest of the URL > based on the user and the request) but one could use it to store all the > info needed when you will turn the Req into a Cmd. > > > > > > > On Tue, May 31, 2016 at 7:29 PM, James Wilson <[email protected]> wrote: > > In Elm, each component basically has its own internal state (which is > actually all just a slice of one global model). In my app, I also want > global state that is independant of any components; for example a > clientside cache of various API responses (asset details - there could be > many thousands, user authentication status). > > I want any component to be able to call methods that make use of this > global state. For example, a method to obtain details for items in the > current view might first look at the global state to see if these items are > cached. If they arent, the call would provide a Cmd to be issued that gets > the items (and puts them in the cache), while simultaneously updating the > state to indicate that they are being loaded (so that the same request > again from another component doesnt trigger another call to the backend). > If they are cached, they can be easily returned from there. A first shot at > a signature might look something like: > > getItem : GlobalState -> ID -> Tag -> (GlobalState, Cmd msg) > > > > However we could partially apply functions that exist on some globalState > instantiation to hdie the initial state being passed in and end up with: > > state.items.getItem : ID -> Tag -> (GlobalState, Cmd msg) > > > > The downside of this approach is that I have to thread this state through > multiple calls that might make use of it, and thread it back up explicitly > through the update functions to get it back to the top. At the top we'd > then have something like (excuse any mistakes!): > > update msg model = case msg of > SubMsg m -> > let (newSubModel, subCmds, newGlobalState) = SubComponent.update m > model.subModel > in ({ model | state = newGlobalState, subModel = newSubModel}, Sub.map > SubMsg subCmds) > ... > > > An alternative approach is to hold this global state in an effect manager, > and so in the app you'd end up using the Cmd/Sub mechanism to ask for > things from the state and internally initiate API requests to update the > state as necessary. We'd end up with an API more like: > > getItem : ID -> Tag -> Cmd msg > > > or > > state.items.getItem : ID -> Tag -> Cmd msg > > > where the returned Cmd would either lead to an item being sent to the > component immediately via a cache (where Tag is a Msg type the component > knows about) or after it was obtained via some backend. This would make all > retrieving of state async but seems to simplify the interface (perhaps at > the cost of more complexity in implementing the effect manager). > > Which approach do people think is best for working with global state > (neither is an option if you have a better way!)? Do you get away with not > needing this kind of thing (and if so, how)? I'd love to hear back, > especially from those that have had experience building larger apps in Elm! > > -- > You received this message because you are subscribed to the Google Groups > "Elm Discuss" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > For more options, visit https://groups.google.com/d/optout. > > > > > -- > There is NO FATE, we are the creators. > blog: http://damoc.ro/ > > -- > You received this message because you are subscribed to the Google Groups > "Elm Discuss" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > For more options, visit https://groups.google.com/d/optout. > > > > > -- > There is NO FATE, we are the creators. > blog: http://damoc.ro/ > > -- > You received this message because you are subscribed to the Google Groups > "Elm Discuss" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > For more options, visit https://groups.google.com/d/optout. > > > > > -- > There is NO FATE, we are the creators. > blog: http://damoc.ro/ > > -- > You received this message because you are subscribed to the Google Groups > "Elm Discuss" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > For more options, visit https://groups.google.com/d/optout. > -- You received this message because you are subscribed to the Google Groups "Elm Discuss" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
