That being said, I also wanted to show that the pattern does not degrade too much when you don't use them. I understand that State Machine semantics can translate into very verbose code and I'd trade state for the delineation of business logic into actions, model and next-action-predicate without second thoughts.
As we talked before, I understand your reservations about my silly string concatenations, but I feel a lot better when I don't use templates and databinding, they tend to erode the intent of the patent that the View is a function of the Model. It you absolutely have to use a framework, I would prefer something like React or Cycle.js
In the end I just want to show that you can build great looking Web apps without a framework. As I mentioned before, the ngrx list app, took nearly 10 minutes to install on my Mac (npm install)
This is where you could serialize actions when needed, for instance the counter example I keep coming up with does not work when concurrent actions are expected (you'd be presenting stale values)
You can only execute one action at a time (on a given model) and all the others have to be queued. When you are dealing with a typical GUI and only the user can send actions, you block the even loop until the new state representation is rendered (or if you don't want to block, you queue it up in nap())
unlike what Mykola Bilokonsky (@mbilokonsky) was saying, to separate actions from model is not an anti-pattern at all (how would it be?), however, if you don't have the nap() function, you'd run concurrency issues. Again, this is standard TLA+ regardless of what @brucou is saying.
Thanks for your input. I understand that it's not hard coded relationship and it's up to the developer to keep the code strict to follow sam dataflow. This was also a way for me to learn vue.js. I'll give react and cycle a try for the same examples and see how they fare. In vue.js it's pretty easy to make the view be based off state exposing different actions which I liked.
maybe the command pattern
So, I am building a example using Redux-React-SAM. Redux is handling all Store mutations inside the Store (Model). I'd love to have your feedback https://github.com/gunar/sam-redux
Hello there, just have a small question that kind of prevents me from doing some toy applications and starting to grasp this SAM thing. When using functions to create the views (instead of templates or JSX, etc), how do define the event handlers? In the examples I have seen some stuff like
return '<button onclick=somefunction(somedata)>', but I don't see that going a long way: what if I don't have a reference (even within a namespace) in the global scope? Thanks.
There is no requirement whatsoever to do so, or even use HTML at all. This is just a pattern that simply states that the view is created as a function of the model, in a similar way to React.js. The same thing could be achieved in Objective-C/Swift, Java (Android SDK), ...
I know, but for me it is about HTML and JS. And I really prefer good old javascript over a yet another framework (actually I found about SAM because of a comment on a blog post about the gazillion tools you need for JS development). And I think that it is easier to grasp the essence of something instead of getting distracted with yet-another-cool-thing (that cycle.js for instance)
So no easy fix for my problem? :D
By the way @jdubray, what are your opinions about Ember? It offers some solutions to some of your problems. For instance, the models on the client don't have to match exactly what you are retrieving from the server. First you have a so-called "adapter" that can modify the data before it comes in or gets out, and then (and most importantly) your client models can have computed properties, e.g. if you have a Person model, and the age is coming from the server, you can define a property that computes on the fly "young" or "old" based on the age attribute.
@jdubray the problem I see is that, if I have a function myAction, and I want it to be bound to the onclick, I have to write (as a string) something like onclick="myAction". But that will only work if that function is reachable from the global scope, which is something you want to avoid at all costs when developing serious JS applications.
If I pack my whole application within an object that acts as a namespace, for instance MYAPP, and I expose my function there, I could write onclick="MYAPP.myAction", but still it would be better not to do so, and somehow bind the handler in another way.
@jdubray I've put your Rocket example in a Fiddle. I've added a new button "Dummy Action" to test the nap() function. As I imagined, as your
nap() doesn't know if it already has set a timer or not, it may set multiple timers i.e. running any other action during the countdown sets another timer. The end result is the same as these multiple timers don't interfere with one another, still, this could become a problem scaling. I don't like having to store a flag (mutate the state) just to indicate that a timer has been set. Do you know of any other solution?
That's the absolute beauty of V = f(M)
@gunar yes and no, because remember, unlike Redux / Sagas, SAM sequences precisely action-model-state. If you look at Redux thunks, they write a "script" regardless of the control state reached. That is fundamentally flawed. That is why I wanted to talk to Dan originally, because Redux is so close, but at the same time, it's leading people in the wrong direction. We are back to the usual action-oriented spaghetti.
Front-End Architecture is such a perfect use case for SAM, because precisely it is a state machine. That's why FEA failed for the past 40 years, because we all ignored the state machine beneath.
Because After A, B may not be allowed, you can't arbitrarily write it that way
Yes it's awesome not to have to worry about entering invalid states, but imagine a very complex app sending numerous actions, that'd trigger many timers. From a SAM perspective that's not a problem at all, but has performance costs on the browser. But I guess that's beyond the scope of this discussion.
Remember, you also don't have to follow the exact semantic of SAM if you know what you are doing. In other words, that's what we have been doing for decades, writing state machines as a series of actions. The problem is that we rarely know what we are doing and SAM provides just enough constrains to eliminate most problems without much overhead.
Algorithms tend to not require patterns like SAM, so I am not suggesting that we start using SAM everywhere all the time. I am saying however that SAM semantics (TLA+) work everywhere all the time, so when you see value over more "optimized code" you should go back to the basics of SAM.
Does that make sense?
Since I learned about TLA+, I have looked at ways to use the semantics as you code so you have less bug to analyze later...
I am lazy :-)
I'll never make that claim
I can't claim you can nest n-level, but 2-3 levels should be ok.
Would that work for you?
The way you describe it seems that I will use the actions from low levels in high level views. I wondering about composing views too. For example, I have a SAM components handling different things with different views and I need to compose then on a larger SAM component that orchestrate then.
@adonirammedeiros so you don't want your low level views/actions to know about higher order SAM cycles?