Avoid Mini-Frameworks
(laike9m.com)52 points by laike9m 3 hours ago
52 points by laike9m 3 hours ago
If you look above the unorthodox library/framework distinction, I think the criticism is about birthing new (inadvertently leaky) abstraction layers with new semantics to capture the specifics of the domain. Often with either esoteric words attached to supposedly novel patterns, and/or unconventional usage of existing terminology.
The promise is to simplify and unify things but as noted, such efforts often have the opposite effects.
"Teams are struggling with properly adopting FooTech - our FooBarTool wraps it in a beautiful package and magically solves everything with a single command and one config file"
"We should template all this yaml"
Yeah had so many discussion with senior developers in my life to argue for just keeping things simple, but my god they love abstractions. They are clearly always very smart and understand the code base well. Maybe it’s their intelligence wanting to be more utilised or maybe they are bored and trying to over engineer simple problems
> Wrapping libraries makes a lot of sense
The best assumption to start with is that adapters are bad by default, because they are an unnecessary layer to maintain (and potentially a point of failure and bottleneck depending on what they are and do). Then, make the argument for the adapter as a guilty until proven innocent case.
If you can make a solid case for it, fine. There are many solid cases for adapters, e.g. drivers for a database or hardware.
Never write an adapter that you can handle more scalably, flexibly, and almost as easily by calling something directly.
React is 100% framework. They even bring their own DSL. It's absurd to call React library.
Library is something that can be pulled off project and replaced with something else. There's no non-trivial project where replacing React with anything would be possible. Every React web app is built around React.
I think absurd is a bit strong. It'd be absurd to call something like rails a library.
I think you can probably see that distinction already, but to spell it out React is described as a library precisely because it does just one thing - the view - and leaves it to you to figure out the entirety of the rest of the stack / structure of your app.
Framework, at least to me, but I also believe commonly, means something that lets you build a full application end to end using it.
You can't do that with React unless your app is just something that lives in the browser either in-memory or with some localstorage backing or something. If that's your app, then probably I'd agree React is your framework per se, but that's hardly ever the case.
By the way, back to my original point, I still do think these things are impossible to define and in lots of ways these terms don't matter - if it's a framework for you, it's a framework - but I just had to defend my position since you described it as absurd :-)
100% this. To this day the official website still describe itself as a library, and I'm convinced it's completely for marketing reasons, since 'framework' feels heavy and bloated, like Angular or Java Spring, while 'library' feels fast and lightweight, putting you in control.
Framework can be more or less modular, Angular or Ember choose to be 'battery included', while React choose to be more modular, which is simply choosing the other end of the spectrum on the convenience-versus-flexibility tradeoff.
React ostensibly only care about rendering, but in a way that force you to structure your whole data flow and routing according to its rules (lifecycle events or the 'rules of hooks', avoiding mutating data structures); No matter what they say on the official website, that's 100% framework territory.
Lodash or Moment.js, those are actual bona fide libraries, and nobody ever asked whether to use Vue, Angular or Moment.js, or what version of moment-js-router they should use.
I don't agree. What I said about React providing structure and (lifecyle) hooks was true from the first version.
The later stuff adds other ways of doing the same thing but a library it remains.
That's as self described by the React team, and I think the consensus more broadly.
> A library is something you call. > A framework is some kind of application scaffolding that normally calls you.
I think I broadly agree with this. In essence, libraries don't impose an application-level life cycle. Frameworks, generally, do.
The rest of the article, I don't know….
The most successful, long-running app I maintain has a mini-framework that allowed me to assemble what I need piecemeal rather than relying on any off-the-shelf framework that would have been obsoleted several times over in the seventeen-year lifespan of this code.
I guess about one in three things I do in it require me to dip into my framework code to look at it, but this is mostly to remember how it works! About one in five things have required me to make a small progressive change.
Twice in its lifetime a core component has been swapped out (mailer and database library).
And twice in its lifetime, because it is beginning to converge on a general web framework, I have considered porting the code out of it and into a general framework, which might make it easier to hand over. One day, I suspect, something will break compatibility in a way that makes that the sensible route, but the code works, fast, has a pretty obvious set of abstractions, and there are implicit examples of everything it can do in everything it already does.
Almost all articles like this start out with "here is a thing I claim is a generalised problem that I am sure you should not do", that is a well-meaning but false generalisation, and is then caveated to the point where no new point is being made.
Underneath they are always: don't write bad code. If you do, learn from it.
If I'd followed the advice of this article when I started this project, I would by now have rewritten the entire thing more than once, for little gain.
Much more concise and much more sensible: consider whether your additional levels of abstraction have value.
But do mini frameworks have value? Sure they do, especially if there is setup and teardown that every function needs to do.
> Is an `EventEmitter` a framework or a library?
It is best described as a pattern implementation (observer / pub-sub).
Example Node.js EventEmitter usage:
const EventEmitter = require('events');
emitter.on('data', handler);
emitter.emit('data', value);
-----------
You explicitly instantiate it. You explicitly register listeners. You explicitly emit events. It does nothing unless you call it. There's no lifecycle, no main loop, no required structure.
EventEmitter is not a framework because it does not define application structure. It doesn't own the program's control flow. It doesn't decide when your code runs (beyond callbacks you register). It does not enforce conventions or architecture.
People sometimes call it a framework incorrectly because of two sources of confusion:
1. Callback-based APIs feel like inversion of control. But this is partial IoC, not framework-level iOc.
2. It is often embedded inside frameworks. Examples: Express routes, React synthetic events, Electron internals.
A library. It doesn’t define how you structure your application in any way.
> A library is something you call.
> A framework is some kind of application scaffolding that normally calls you.
There is no real distinction between these two.
I never heard the term mini framework before but I like it. Applied within that context it makes sense. I was at Google 2011-2013 through an acquisition and witnessed some of what the author is describing. Actually something very specific comes to mind. There was a SQL library built on top of BigTable, Google's internal columnar data store, I think it was called Megastore [1]. The team implementing a new product decided to use it over BigTable directly because what it would mean for ease of development but it turned out as soon as they needed to do any data migration it was locking the entire database (or I guess the table? Aka BigTable). Anyway at the time this was a major issue and they had to revert to BigTable best practices as opposed to what this library was doing. Because essentially it was a library layer, not some new data store. Eventually Google built a different SQL based data store to replace the whole thing which you can see the open source version known as CockroachDB, which some ex-googlers invented.
Moral of the story, abstractions always fail the edge cases, especially at scale. But when the entire eng org adopts something, it works much better. Everyone has to be bought in. Which at Google scale is hard.
https://research.google/pubs/megastore-providing-scalable-hi...
As a rule of thumb, "magic" is a code smell. Libraries should be preferred over frameworks whenever possible.
A toolbelt of small utility-like composables are often easier to maintain and reason about. This results in added explicitness (i.e. less magic, fewer surprises).
Personal experience shows that the immediate efficiency gains of a framework often get diminished in the face of all the hacks people introduce later, just to work around the remaining 10% of cases that the framework did not anticipate or traded-off against.
Please note this is a comment based on personal experience and professional preference.
BOCTAOE.
I don‘t like dismissing technologies on the basis of being „magic“, since the magic could often just as well be called abstraction, and the line between them is often personal preference.
The abstracted-away logic in a Laravel application can either be called magic or abstraction, but so can the optimizations of a database query planner.
I think often you still need to know the underlying mechanism, but it is still useful to get the innards out of the way.
It's useful to get "glue" code out of the way while building, but to the point in the article it all becomes very difficult to debug and maintain once there are problems in the that layer.
Spring Boot and other similar frameworks come to mind; by forcing huge amounts of indirection you lose a lot of visibility of your call stack because the convenient "glue" code is now orchestrating everything at runtime, but that code isn't yours, and it isn't easily inspected or fixed.
In case "framework" is understood as something that calls my code and that forces me to write my code in a certain way, I totally agree.
And I think twice before I use a framework. Frameworks enforce a certain way of programming which you can never be sure to match the problems you will have to solve in the future. Libraries don't do this - at least not to the extent of a framework. Libraries are composable building blocks.
Nevertheless, there may be applications where frameworks are beneficial (e.g. GNU Radio).
I had to look up BOCTAOE (But Of Course There Are Obvious Exceptions)
"Good magic decomposes into sane primitives" highlights an essential distinction: not all magic is bad (but it's not always clear at first which kind of magic is in play).
I remember using Laravel a while back and it had quite some magic but it was done right and made a lot of things much easier
This is a curse of lots of software companies:
** mini-frameworks is a realization of the creator's mental model, but it's not everyone's mental model**
People being smart enough to make their own understanding work well - but not smart enough to see they are just pushing their way of doing things and not working on something “generally understood”. A
This was great. I enjoyed the alternative definition of "framework" as something that introduces new concepts (differing from the more common idea that it's code that calls YOUR code.)
This article reminded me of two classic pieces of writing.
The first is 20+ years old now: Joel Spolsky's law of leaky abstractions:
https://www.joelonsoftware.com/2002/11/11/the-law-of-leaky-a...
One of the reasons these mini-frameworks lead to problems is that they leak. They don't cover every case which means you inevitably have to fully understand what they are doing for you in order to debug them or work around their limitations. Your cognitive load goes up!
The second is Will Larson's "Migrations: the sole scalable fix to tech debt."
https://lethain.com/migrations/
The OP complains that they've never seen a technology migration at Google that finished. Will advocates for migrations to be driven by a team that both coaches other teams on the migration and then, crucially, finish the job themselves to make absolutely sure it gets done to 100% completion.
Basically, premature abstraction / abstraction at the wrong boundaries coupled with the political power to force adoption of the bad abstraction. Yep. That’s an annoying attribute of working at a big corp. I prefer working in tiny teams and companies for this and many other reasons.
Mini-frameworks would be dangerous at a place like Google where smart people think they are smart. I think they're heaven sent for smart people who think they're stupid.
Distributed (multithreaded, concurrent, ...) systems are a counterexample that are highly vulnerable to snake oil. In normal software it makes sense to build up from a small set of intellectually coherent primitives. In those cases you inevitably end up with poor performance and usually reliability if you try that. Java started out with a snake oil approach to threading (synchronized!) and Doug Lea talked some sense into them and now we have java.util.concurrent which has a rich set of primitives which are pragmatic and meet the real requirements, not a vision of purity.
On the other hand, If it was a mini-framework to pound out numerous not-so-simple HTML form applications it could greatly enrich your life and your team's.
Making good abstractions is hard.
And it is very easy to start feel like you know you have a great abstraction when you don't. And unfortunately it's easy and fun to make abstractions, kinda like making babies. And has kind of similar weight to it too.
I just have to say.. stay safe out there.
I wrote this article to reflect a pattern I observed (and hate) while working in Google, but I'm sure this is not a company specific problem. Would be interested to hear other people's stories :)
I can be found here:
This is why I use Ruby on Rails. Shared constraints and boring conventions age better than clever mini-frameworks built around one team's mental model.
It's why I've always eschewed stuff like flask even when I felt like Django was going to be overkill. The problem is that, especially for web apps, everybody needs the same stuff. You need to handle cookies and forms and auth. You need to be able to inspect and manage your data. You need to be able to do tokens and password resets.
At the end of the day. You end up cobbling together a bespoke, worse version of Django anyway.
This. This. This. I currently work in a codebase where so much code has been abstracted away for “cleanliness” that it’s impossible to understand what code is actually running.
The worst is when three lines of completely standard code (immediately understandable to anybody inline) get „helpfully” lifted out into a utility function.
By the way, the reason behind all of this, like so many ills of our industry, is the completely broken promotion culture.
> Introduce new concepts that doesn't exist in the original stack
That is also true for "macro" frameworks.
> Wraps around the company/org-shared tech stack or framework
That is often also true for "macro" frameworks.
> Creators claim that the framework "magically" solves many problems, and push more people to use it
That is often also true for "macro" frameworks.
---
It is not clear from the reader's perspective what actually characterizes a "micro" framework. It's also not clear why the size is the issue here, when all complaints seems to be about design or quality.
Is googletest a micro or macro framework? Is google/zx a micro or a macro framework? Give us some clarifying examples. Actual things people can look for, not internal unknowable projects. There must be some exceptions too (silver bullet rules don't exist), mention them.
Also, rethink the title. Maybe "makeshift frameworks" is better terminology, as it more accurately reflects the problem that is described in the content.
If you have a very specific product with limited scope, a micro-framework would work just fine. My experience in the real world™ is as such: people start with micro-frameworks and keep bolting on stuff to the point where it would have been better if they started with a macro-framework in the first place. At least there is better compatibility between framework components and a clear upgrade process. I agree with the "makeshift framework" terminology by the way. One way or another, my experience is that products that start with micro-frameworks, over time turn into a "makeshift framework" over time regardless. If the scope is clear and limited from the start, micro-frameworks are great. If unsure, micro-framework is a no go (for me).
My experience in the real world is that the majority of people choose the largest "macro" framework available and go with that. It's what happens most often.
The "micro framework" phase happens when that "macro" framework fails to deliver something. It happens way less often than a team picking a big estabilished tool.
However, the sizes never mattered. That is likely what causes the confusion in the first place ("it's large so it must have lots of things I want", "it's small so it must be easy to understand").
The real red herring is focusing on the size (or LOC, or any vague metric) instead of other more relevant architectural properties.
> real and only difference between a library and a framework, is whether it introduces new concepts
This isn't what is normally understood in software engineering by those terms.
A library is something you call.
A framework is some kind of application scaffolding that normally calls you.
You can use more than one library. You normally only have one framework in-process.
I found the blog post a little hard to parse. Is it an argument against wrapping frameworks, or wrapping libraries?
I agree that wrapping frameworks is fraught with danger. I can't quite agree for wrapping libraries. Wrapping libraries makes a lot of sense if you're only using a tiny fraction of the library functionality, the breadth of the wrapper's API is much smaller than the library's API, wrapping it enables you to substitute it (whether for a smaller / faster / whatever dependency in the future, or for testing, etc.), and so on.