My new approach to software prototyping and the stack that makes it possible
In this blog post I am going to illustrate how, during the development of a pet project, I have changed the way I approach software development and prototyping, going from "backend first" to "frontend first, backend eventually". I will also mention the tools that make it possible and even enjoyable for a non frontend developer.
At $work I have always spent most of my time in the backend and infrastructure side of things but sometimes, especially when working on side projects, I have to go "full stack" and take care of frontend tasks, even if I kind of suck at it. For this reason I am always inclined to start with some sort of backend code, starting with a data model, core business logic and a REST API to (eventually) build a frontend on top of it...which most of the times means I will lose interest/momentum and the project dies.
At some point something clicked
One day (probably a lazy Sunday) I had an idea: I want an incredibly stupid way to collect random thoughts, links or any kind of text blob, I want to own it, it must live on my VPS and I have to finish it today! This time thoughts decided to follow a different path:
how important what I am going to write here is? can I afford to lose data?
what if I don't need to store anything in a durable storage?
ok, aim for least amount of features and see later if I need more
The outcome is a service that will keep all data in volatile memory, losing it at each restart and this is perfectly fine for what I needed at that time! I think this small tool planted the seed for what, I think, will be be my preferred method for future projects.
What has been so different with development of this tool is that I have tried and (more or less succeeded) to address the problem at hand without much ceremony, and it served its purpose quite well. At some point I've started to add more features "just because" but, at the end of the day, the core remained the same and now I even want to remove these new features because they are not adding value but are, effectively, distracting.
Fast forward few months, I am again thinking about a new project. My girlfriend needed/wanted a tool to make it easy to share a watermarked version of her drawings, even better if the shared link required a password to access the image.
Compared to the previous project, this one is slightly more complex, especially the user facing part, but I am confident that if I address the most basic requirements first, I should be able to give her a rough beta in few hours. Here are the features I have decided to include in the "MVP":
user can upload a file
user can chose a watermark size
user can eventually chose a password to protect the image
once the user clicks "Upload", the system applies the watermark and
render a page with the list of uploaded/watermarked files, each with a link to it
So far so good, the tool is there, ugly but working. What if we open this tool to other people? WOW! We can make our own product! Lets do it!
Armed with great energy I have started working on the new version but this time I have started building a full featured backend even if, at this point, is not 100% clear how a user is going to interact with it, how valuable it could be and all the usual unknowns. So why I am insisting with the old approach when it showed, multiple times, it is quite ineffective? I guess it is a matter of habits (because I don't want to call it stupidity).
A new approach
I often fail to make progress on my projects when it comes the turn of the frontend code, after having written most of the backend, maybe this time I can try the other way around! Common sense suggests to create mocks of the interface using tools like Balsamiq, Moqups and so on but, for some reasons I am totally unable to use those kind of tools plus I would like to have a taste of how to UI feels in the hands a user...what should I do now?
Recently I have started experimenting with ClojureScript and Rum (a library that wraps React) and so far I feel quite productive with it, why should I not use these tools to write the mock? The plan now looks like this:
quickly build an ugly but functional UI
mock the app state/data to reflect what I need in the UI
understand what feels wrong, get feedback from other people
plan next iteration
repeat until happy
Essential to this approach is the quick feedback loop you can get by using ClojureScript + Rum (or any other React wrapper like Reagent), immutable persistent data structures and a REPL; I "suspect" that, if we exclude the REPL, a similar setup can be achieved with other stacks but I am quite happy with the tools I have chosen.
How has it been so far?
I am not a frontend developer and every time I try to create some sort of UI I am pretty sure I will fail, lose interest and put the project on hold but, this time, it has been quite different. Being able to iterate quickly and having a fast feedback loop, helped me a lot to proceed at "great speed" and not losing focus. Each time I wanted to add a new feature or change how a component had to work, the process has been pretty straightforward:
create a new branch
try out the new idea by small, even tiny iterations with live feedback
adjust the data model as needed
if happy merge to master otherwise scrap everything and forget
Managing the app state
One of the points that I would like to spend so time on is the app state. Even if keeping all the state in one big map is quite handy and flexible, at some point, it becomes harder and harder to maintain, especially in the case where there are many ( > 4 or 5) entity types with cross references and the size of the data gets bigger and bigger; it can still be done but the state management code starts to get bigger and harder to maintain compared to the UI code; for my case, the UI code had to know too much about the data model itself impacting the separation of concerns and this was driving away the focus on the main goal which is to explore the space of possibilities for the UI I want to build.
Introducing Datascript into my tool-set
With the growing pain of state management I have decided to invest a little bit of time to find out which solutions are already available in this space and fortunately I have stepped into Datascript, an in memory database developed by the same person who gave us Rum :) This database has all the features I need plus some more; I am sure that I am still missing some handy features but so far I have been able to:
build a consistent data store
which supports cross references between entities
which I can easily query (using Datalog)
with support to joins
Datalog itself is quite simple even if a bit "alien" until you know how to use it; instead of writing about it here I prefer to give a link to the best learning material I have found http://www.learndatalogtoday.org/.
Armed with Datascript (for the data store) and Datalog (for the queries) the UI code can now interact with the app state without knowing much about it! This is a great relief because now I can focus again on building a UI for my project and spend the right amount of time (< 5%) on the data store.
Another good point of this approach is that, when the backend will be ready, it will be a matter of translating what I will get out of its APIs to the UI's internal data store and everything will continue to work as expected.
The goal of this post was to write down my current approach to software development with the hope to start a discussion with other developers which may be interested to explore and talk about novel methodologies which have worked great for them. If by reading this blob of text you are now also curious about ClojureScript and its ecosystem, I am even more happy!