{"rowid": 257, "title": "The (Switch)-Case for State Machines in User Interfaces", "contents": "You\u2019re tasked with creating a login form. Email, password, submit button, done.\n\u201cThis will be easy,\u201d you think to yourself.\nLogin form by Selecto\nYou\u2019ve made similar forms many times in the past; it\u2019s essentially muscle memory at this point. You\u2019re working closely with a designer, who gives you a beautiful, detailed mockup of a login form. Sure, you\u2019ll have to translate the pixels to meaningful, responsive CSS values, but that\u2019s the least of your problems.\nAs you\u2019re writing up the HTML structure and CSS layout and styles for this form, you realize that you don\u2019t know what the successful \u201clogged in\u201d page looks like. You remind the designer, who readily gives it to you. But then you start thinking more and more about how the login form is supposed to work.\n\nWhat if login fails? Where do those errors show up?\nShould we show errors differently if the user forgot to enter their email, or password, or both?\nOr should the submit button be disabled?\nShould we validate the email field?\nWhen should we show validation errors \u2013 as they\u2019re typing their email, or when they move to the password field, or when they click submit? (Note: many, many login forms are guilty of this.)\nWhen should the errors disappear?\nWhat do we show during the login process? Some loading spinner?\nWhat if loading takes too long, or a server error occurs?\n\nMany more questions come up, and you (and your designer) are understandably frustrated. The lack of upfront specification opens the door to scope creep, which readily finds itself at home in all the unexplored edge cases.\nModeling Behavior\nDescribing all the possible user flows and business logic of an application can become tricky. Ironically, user stories might not tell the whole story \u2013 they often leave out potential edge-cases or small yet important bits of information.\nHowever, one important (and very old) mathematical model of computation can be used for describing the behavior and all possible states of a user interface: the finite state machine.\nThe general idea, as it applies to user interfaces, is that all of our applications can be described (at some level of abstraction) as being in one, and only one, of a finite number of states at any given time. For example, we can describe our login form above in these states:\n\nstart - not submitted yet\nloading - submitted and logging in\nsuccess - successfully logged in\nerror - login failed\n\nAdditionally, we can describe an application as accepting a finite number of events \u2013 that is, all the possible events that can be \u201csent\u201d to the application, either from the user or some other external entity:\n\nSUBMIT - pressing the submit button\nRESOLVE - the server responds, indicating that login is successful\nREJECT - the server responds, indicating that login failed\n\nThen, we can combine these states and events to describe the transitions between them. That is, when the application is in one state, an an event occurs, we can specify what the next state should be:\n\nFrom the start state, when the SUBMIT event occurs, the app should be in the loading state.\nFrom the loading state, when the RESOLVE event occurs, login succeeded and the app should be in the success state.\nIf login fails from the loading state (i.e., when the REJECT event occurs), the app should be in the error state.\nFrom the error state, the user should be able to retry login: when the SUBMIT event occurs here, the app should go to the loading state.\nOtherwise, if any other event occurs, don\u2019t do anything and stay in the same state.\n\nThat\u2019s a pretty thorough description, similar to a user story! It\u2019s also a bit more symbolic than a user story (e.g., \u201cwhen the SUBMIT event occurs\u201d instead of \u201cwhen the user presses the submit button\u201d), and that\u2019s for a reason. By representing states, events, and transitions symbolically, we can visualize what this state machine looks like:\n\nEvery state is represented by a box, and every event is connected to a transition arrow that connects two states. This makes it intuitive to follow the flow and understand what the next state should be given the current state and an event.\nFrom Visuals to Code\nDrawing a state machine doesn\u2019t require any special software; in fact, using paper and pencil (in case anything changes!) does the job quite nicely. However, one common problem is handoff: it doesn\u2019t matter how detailed a user story or how well-designed a visualization is, it eventually has to be coded in order for it to become part of a real application.\nWith the state machine model described above, the same visual description can be mapped directly to code. Traditionally, and as the title suggests, this is done using switch/case statements:\nfunction loginMachine(state, event) {\n    switch (state) {\n        case 'start':\n            if (event === 'SUBMIT') {\n                return 'loading';\n            }\n            break;\n        case 'loading':\n            if (event === 'RESOLVE') {\n                return 'success';\n            } else if (event === 'REJECT') {\n                return 'error';\n            }\n            break;\n        case 'success':\n            // Accept no further events\n            break;\n        case 'error':\n            if (event === 'SUBMIT') {\n                return 'loading';\n            }\n            break;\n        default:\n            // This should never occur\n            return undefined;\n    }\n}\n\nconsole.log(loginMachine('start', 'SUBMIT'));\n// => 'loading'\nThis is fine (I suppose) but personally, I find it much easier to use objects:\nconst loginMachine = {\n  initial: \"start\",\n  states: {\n    start: {\n      on: { SUBMIT: 'loading' }\n    },\n    loading: {\n      on: {\n        REJECT: 'error',\n        RESOLVE: 'success'\n      }\n    },\n    error: {\n      on: {\n        SUBMIT: 'loading'\n      }\n    },\n    success: {}\n  }\n};\n\nfunction transition(state, event) {\n  return machine\n    .states[state] // Look up the state\n    .on[event]     // Look up the next state based on the event\n    || state;      // If not found, return the current state\n}\n\nconsole.log(transition('start', 'SUBMIT'));\nAs you might have noticed, the loginMachine is a plain JS object, and can be written in JSON. This is important because it allows the machine to be visualized by a 3rd-party tool, as demonstrated here:\n\nA Common Language Between Designers and Developers\nAlthough finite state machines are a fundamental part of computer science, they have an amazing potential to bridge the application specification gap between designers and developers, as well as project managers, stakeholders, and more. By designing a state machine visually and with code, designers and developers alike can:\n\nidentify all possible states, and potentially missing states\ndescribe exactly what should happen when an event occurs on a given state, and prevent that event from having unintended side-effects in other states (ever click a submit button more than once?)\neliminate impossible states and identify states that are \u201cunreachable\u201d (have no entry transition) or \u201csunken\u201d (have no exit transition)\nadd features with full confidence of knowing what other states it might affect\nsimplify redundant states or complex user flows\ncreate test paths for almost every possible user flow, and easily identify edge cases\ncollaborate better by understanding the entire application model equally.\n\nNot a New Idea\nI\u2019m not the first to suggest that state machines can help bridge the gap between design and development.\n\nVince MingPu Shao wrote an article about designing UI states and communicating with developers effectively with finite state machines\nUser flow diagrams, which visually describe the paths that a user can take through an app to achieve certain goals, are essentially state machines. Numerous tools, from Sketch plugins to standalone apps, exist for creating them.\nIn 1999, Ian Horrocks wrote a book titled \u201cConstructing the User Interface with Statecharts\u201d, which takes state machines to the next level and describes the inherent difficulties (and solutions) with creating complex UIs. The ideas in the book are still relevant today.\nMore than a decade earlier, David Harel published \u201cStatecharts: A Visual Formalism for Complex Systems\u201d, in which the statechart - an extended hierarchical state machine model - is born.\n\nState machines and statecharts have been used for complex systems and user interfaces, both physical and digital, for decades, and are especially prevalent in other industries, such as game development and embedded electronic systems. Even NASA uses statecharts for the Curiosity Rover and more, citing many benefits:\n\nVisualized modeling\nPrecise diagrams\nAutomatic code generation\nComprehensive test coverage\nAccommodation of late-breaking requirements changes\n\nMoving Forward\nIt\u2019s time that we improve how we communicate between designers and developers, much less improve the way we develop UIs to deliver the best, bug-free, optimal user experience. There is so much more to state machines and statecharts than just being a different way of designing and coding. For more resources:\n\nThe World of Statecharts is a comprehensive guide by Erik Mogensen in using statecharts in your applications\nThe Statechart Community on Spectrum is always full of interesting ideas and questions related to state machines, statecharts, and software modeling\nI gave a talk at React Rally over a year ago about how state machines (finite automata) can improve the way we develop applications. The latest one is from Reactive Conf, where I demonstrate how statecharts can be used to automatically generate test cases.\nI have also been working on XState, which is a library for \u201cstate machines and statecharts for the modern web\u201d. You can create and visualize statecharts in JavaScript, and use them in any framework (and soon enough, multiple different languages).\n\nI\u2019m excited about the future of developing web and mobile applications with statecharts, especially with regard to faster design/development cycles, auto-generated testing, better error prevention, comprehensive analytics, and even the use of model-based reinforcement learning and artificial intelligence to greatly improve the user experience.", "year": "2018", "author": "David Khourshid", "author_slug": "davidkhourshid", "published": "2018-12-12T00:00:00+00:00", "url": "https://24ways.org/2018/state-machines-in-user-interfaces/", "topic": "code"}