Skip to content
Tools for developing Elm programs! 🔧
Branch: master
Clone or download
Latest commit 0601d75 Jun 18, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
example Bump tar from 2.2.1 to 2.2.2 in /example May 25, 2019
src removed unused functions, refactored a few things Mar 6, 2019
.gitignore dev-deps for example Dec 4, 2018
elm.json dump to 0.1.0 Dec 14, 2018
license some initial boilerplate Oct 31, 2018
readme.md Update readme.md Jun 18, 2019

readme.md

Elm DevTools

Try out my example

The module is still being developed and will probably go through major changes.

If you want to try it, you can clone the repo and run these commands from the /examples folder:

# install dependencies
npm install

# start example with linux
npm start

# start example with windows
npm run start:win

Goals

The overarching goal is to close the loop between writing Elm code and interacting with it. This concretely means:

  1. Code changes should be reflected immediately in a running instance of its application.
  2. Code changes should only break application-state if necessary, and if enabled, only partially.
  3. Application-states should be effortless to inspect, navigate, persist and share.
  4. Guidance from the compiler should be right in front of you when you make mistakes.
  5. The boilerplate necessary to enable the first four goals should be minimal, and if needed be, incremental.

I strongly believe that the optimal design and implementation of these goals will transform how we build our applications. Going through the loop of compiling code, reloading browser windows and getting the application into the right state costs seconds, but those seconds are spent at such a frequency that writing interactive applications is incredibly time-consuming. I hope it doesn't have to be.

Design

These devtools are based on the Elm architecture. Since programs are constructed using a few simple functions, they can be embedded and controlled, which also sets some limits for what these devtools can do.

Automatic Message Replay

Any previous state of an application can be recomputed using the initial state and a list of messages folded by an update-function. These state-transitions happen independently of their environment - no commands are sent during replay, only the state changes. This is very useful to:

  • Stop time to look at what happened.
  • Go to a previous point in time to try something else.
  • Reload without losing the state of your application.
  • Reload with code changes but only partial (or no) loss of application state.

Sessions

These devtools run in sessions. A session is essentially made up of:

  1. devtools settings and state.
  2. all messages that ever updated the debugged applications state.

Sessions keep track of what your doing during development. They persist through browser-reloads, code-changes, can be downloaded, sent, opened by collaborators, and submitted as bug-reports.

Support for Code-Changes

To reliably support code-changes in sessions, it is essential that interactions are recorded rather than states. Interactions are modeled in Elm applications with a single type, usually called Msg. Let's do an example:

type Msg = Increment | Decrement

update msg count = case msg of
    Increment -> count + 1
    Decrement -> count - 1

state = List.foldl update 0 [ Increment, Decrement, Increment ]

If you run this example, the value of state will be 1.

Refactor this into an application, and devtools would deal with changes to this code in different ways:

  • If Msg was given another constructor, called Reset which updates Reset -> 0, the value of state is still 1.
  • If Msg had Decrement removed (or changed), one of three strategies (of your choice) could be used:
    1. Restart the application, as Decrement has no meaning in the program anymore. The value of state is now 0, and we've accomplished nothing. This is how most web-app development works.
    2. Filter Decrement out of the list and update accordingly. The value of state is now 2. This works better or worse depending on the coupling between the updates your messages perform. If you remove LogIn and expect ViewAccount to work, you will be dissapointed.
    3. Take messages until you reach the first Decrement and update using those. The value of state is now 1. This is really great, as it captures the remaining valid sequence of updates the application could do. If LogIn goes away, you'll never try to ViewAccount.
  • If update is changed, any sequence of messages will still be (generally) valid. Increment might do a + 2 instead, but messages still capture your interactions, so you can tweak update until it works as intended. You can tweak view and subscriptions in the same way.

By recording interactions rather than state, Elm applications can be written while running, and you will only ever have partial to full state-loss when you modify the definition of messages, which is the very definition of making those interactions impossible or different.

You can’t perform that action at this time.