In a previous post, I discussed a project I'm working on to level up the complexity present in my portfolio. After a few days of deliberation and prototyping I have decided to implement this application in Rust, rendering to an HTML5 canvas via WebAssembly.
I got a prototype working with some clickable buttons using a previous experiment as an example, but it was brittle. Every individual region had its own separate detection and handling logic - a mess to tweak or modify. I decided I wanted a better abstraction over the canvas before going any further, for blood pressure purposes.
Fast forward another two days, and I've essentially outlined a UI framework. I've decoupled the rendering from the game logic to the point where it could be separated out into its own external standalone library. It provides a set of traits you can implement for your game's types which fully handle creating the canvas element, mounting your app to it, and handling clicks. All you need to do is plug in the drawable widgets and it organizes your layout for you.
How does this not always happen? Or, rather, does this always happen? Does any sufficiently large project end up with its own highly decoupled bits of logic like this? Does the fact that I'm writing one from scratch mean I should be using a "real" framework instead? For this project I don't care, it's specifically an educational experience so I don't mind the extra labor, but would this sort of thing be a flag for you in a "real" project? How much of this would you let yourself implement from scratch before exploring the ecosystem? The extra control is nice, but it's time consuming, and this problem can be crowd-sourced.
Additionally, if you've found yourself with a DIY framework in the course of writing an application, what drives you to decide whether or not to actually separate and generalize it for public consumption? I don't know if the world needs another homebrew canvas UI framework, but if I've gone and written one anyway, maybe it's worthwhile to throw it out there. That said, it's also a whole other set of time-consuming tasks unrelated to this project.
Photo by Chris Abney on Unsplash
Top comments (23)
Oh Ben, oh why oh why oh why ;)
Why You Shouldn't Use A Web Framework
David Wickes ・ Jul 26 '18 ・ 4 min read
So my first question is: is it really a framework? Or is it a library? And how can you tell?
A rule of thumb I picked up from somewhere is that "you call libraries, frameworks call you". Which is pithy and easy to remember, and sounds true so :shrug:.
I'm so hazy on Rust right now that I can't even tell whether this is calling my code or I'm calling it!
Only if it's done right! But decoupling the logic doesn't mean that you've built a framework - you can architect your code around, say, MVC, without it being a 'framework'. So much of what you've said is really just the proper architecting of a project - keeping concerns separated.
You can write a nice UI library that describes and solves your problem domain - and I'm a big fan of building a language 'up' to meet the problem; writing a DSL (domain specific language) that I can use to solve my problem in.
I guess I think it's a framework when
Haha, right, right, don't say the "F" word around David :)
Good point - "framework" is likely not the correct word to apply to this solution. It sits around where React sits, but for a canvas - view layer only, but handles the view entirely.
It's a bit of both, though - I call it when I implement these traits (interfaces) for my types, but then it takes over and manages the entry point - it has some setup code that runs before everything else, and then manages the whole lifecycle of the running application, calling my code when needed to re-render. That's what makes it feel like a framework to me. I refactored away my own app's entry point, and now this separate apparatus handles those details.
I'm working with HTML, CSS, and JS and feel the same way. I write blocks of code that I like and would be easy to reuse in other sites so I save it as a VS Code snippet or a separate file or just add it to a growing list of reusable code of mine.
I asked myself "should I just jump into a framework/library? But I enjoy having the control of my code. I also have wondered if every developer has this problem or how many of them just jump into a framework/library.
That's it for me too - I want to build everything up myself from complete scratch, but I still want to make it easy on myself to use. This seems to be the natural conclusion.
Absolutely. I know people talk about using Frameworks/Libraries to speed up production, and perhaps it's just my inexperience, however, I'm almost "afraid" of using them 😂
My rule of thumb is to look at frameworks for the second go at a problem, after I've explored it myself. I can't choose a framework to solve a problem for me if I don't fully understand the shape of the problem I'm solving, but once I do, they really can save time if you pick the right one.
Lots of "if"s, though.
That's a good rule of thumb, I like it! Lol it is a lot of "if"s but I'm sure you, like me, would rather be self-sufficient and really learn along the way.
Wow, incredible response - thank you.
This is a great point that I hadn't thought of. It's true, I don't think any of the options I'm currently aware of would have been drop-in replacements for what I'm putting together.
It is a portfolio project, but I'm concerned about this anyway. I suppose worst case scenario is that I release and don't have time to maintain it and nobody uses it, which is more or less how many people currently use it...
True enough. Time spent solving new problems is rarely if ever wasted.
I like your classifications - Mechanic-style work does seem up my alley interest-wise.
If it's a portfolio piece, then putting it out on your GitHub as OSS is a default place folks expect to find your portfolio pieces.
If nobody uses it or contributes then it is like 99.9999% of GitHub repos and an even larger proportion of portfolio repos.
If people do use or contribute then it is a better portfolio piece.
Hah, fair enough :)
Also let me know if you do post to GitHub. I've been reading up on Rust lately and writing a PR would be nice practice.
Absolutely! Still quite underbaked for now but I will keep you posted.
I believe it is the natural tendency for our projects to end up with their own framework. I mean, our whole job is to automate and save time for others. So it is only natural that we see repeated patterns in our code and then also try to do that for ourselves as well. In the small we call it DRY and in the large, I suppose, frameworks.
The real danger in frameworks though is that they are coupled to everything. Later, something new comes up which the framework doesn't handle. Then like any piece of coupled code, changing the framework can cause rippling breakage throughout the code base. It's why on a deployed project we don't simply upgrade to the latest version of external frameworks when they come out -- to reap any new benefit usually requires first fixing breaking changes. So the upgrade has to be evaluated and planned out. And sometimes the choice may be not to upgrade because there is too much work (based on how dependent you are on the framework).
Don't get me wrong. There is definitely a positive value proposition to be had in centrally handling common plumbing and providing extensibility points for adding new features / capabilities. However, I found once I decided to formalize it into an (ostensibly reusable) abstraction, that's when the trouble began. You end up running into case after case where you either: a) make an opinionated choice that limits the consumer's options b) parameterize a choice which makes your framework more complex to use.
Frankly any app frameworks I made even when I tried to abstract them for reuse, I always felt they were lacking in some way. And in any new project I would start over from scratch and try to "do it right this time". Nowadays I do functionally the same thing, but this time without the pretense of making it reusable across projects. Instead, I reuse the relevant capabilities (sections of code or libraries) in different mixes across apps, as opposed to plugging into a grand abstraction. I believe this might be characterized as what Dan North calls the ginger cake pattern.
Interesting - that makes a lot of sense. This is good to keep in mind - I think I'm going to continue with this present tack, but within the context solely of creating this working application first before I start thinking about what this proto-framework might need to provide in a more general setting.
It's like Neil said, "It's very hard for a general-purpose framework to happen to provide the best solution for your specific problem domain."
Thanks for your perspective! Looking forward to watching this talk.
Oh absolutely, pursue that spark of interest. I will say that the ones I made were valuable learning experiences. Often in the "what not to do" kinda way, but still valuable. :)
I enjoy your thoughtful questions.
few months back I also tried to develop a MVC framework which written in PHP . Although this framework never upto the mark but I got to know more about the internal working of MVC based web frameworks . Increasing projects complexity is always help , definetly recommend to other to try this kind of projects .
you can find its code here github.com/thecodestuff/leggo
Best of luck for your
I have a Clojure(script) project that's kind of heading towards a framework. Current API is custom, websocket based, and considering whether to switch to GraphQL separating the two more. But it's overkill if I just use it myself. I didn't want to use an additional framework to clojurescript, which worked, but it might have been less work if I did use one.
What's the point of the framework? I think the point is to have agreed-upon way of building things. Whether the framework is 3rd party or not is not very relevant, unless you're thinking about hiring new people. It's gonna take less time for new people to get familiar with how to build your thing if you're using a 3rd party framework, but really only if the framework is darn popular. If the framework is not popular, there's no reason to use it because new people will have to get familiar with it as well.
It seems like what you're doing is completely right, separating code into smaller independent modules. If an independent framework / engine comes out of it, there's nothing wrong with it. You may choose to publish it or to keep it private, and that's totally up to you - if you're not willing to market or support it, might as well keep it private.
When do you follow your advice and when do you draw the line and say: we should not have done this ourselves?
I read the part about learning and doing it anyway to answer that question, but buy vs build isn't always that simple. How and when can justify investing money in building a framework?
Amazing! Thank you do much for your reply. I definitely was mixing up ending up with a framework vs setting out to build one.
I'm sure this always happens, let's arrange the madness, oh shoot it's another framework.