267 points | by stanulilic3 days ago
Similar in spirit to Lehman’s Law of Continuing Change[0], the idea is that the less complexity a system has, the easier it is to change.
Rather than plan for the future with extensible code, plan for the future with straightforward code.
E.g. only abstract when the situation requires it, encourage simple duplication, use monoliths up front, scale vertically before horizontally, etc.
I’ve built many 0-1 systems, and this is the common thread among all of them.
[0] https://en.m.wikipedia.org/wiki/Lehman%27s_laws_of_software_...
If you start with code that's easy to delete, it's often possible to alter your data representation or otherwise transform the problem in a way that simply eliminates the edge cases. With the result being a design that is simpler by virtue of being more robust.
If you start with code that's hard to delete, usually by the time you discover your edge and corner cases it's already too late and you're stuck solving the problem by adding epicycles.
The most common examples are empty collections: either disallowing them even though it would be possible to handle them, or making a strange choice like using vacuous falsity, i.e.
all [] == False
Now, every layer above has to special-case these as well, even if they would be a completely normal case otherwise.
I find it more understandable, it’s just that DEs need to write their own compositors.
Example: screenshot. X11: "please tell me the pixels in the root window". Wayland: "please tell me the extension number of the portals extension so I can open a portal to pipewire so I can get the pipewire connection string so I can connect to the pipewire server so I can ..."
Example: get window position on screen.
Example: set window title.
X11 is a protocol about managing windows on a screen. Wayland is a protocol about sending pixel buffers to an unspecified destination. All the screen-related stuff, which is integral to X11, is hard to do in Wayland with a pile of optional extensions and external protocol servers which do not interact.
X11 is also more standardized, de facto, because there are fewer server implementations, which is not just an accident but is by design.
base mechanisms that interpret the raw problem correctly (e.g. pixels on a screen, mouse position) -> some policy that is in some ways mechanism with slightly more abstraction (e.g. drawing shapes) -> some more policy-mechanism abstraction (e.g. windows) ...
until you get to your desired layer of abstraction to work at. This goes along with modularity, composition, code reuse. X11 itself has many design flaws, but Wayland's design is untenable.
I think X11 has had that for a very long time. In the late 2000s when Beryl was still separate from Compiz, it was almost trivial to target things like dropdowns by a descriptive name and give them different effects than regular windows. Mine had an accordion effect while windows would burn up.
A rule I like to follow:
- first time: write it
- second time: copy it
- third time: maybe refactor it
Is this "the same" thing? If so - extract and reference. Or is it "a different" thing which is superficially similar? Then don't.
Knowing when two things are one thing or one thing is two things is most of our job, right?
I think a better way to look at it than “third time - consider refactor” is to follow this article and ask “will this ever need to be extended?”. If the answer is yes, then you should duplicate it.
This way you won’t get a flying dog in your OOP hellscape but you also won’t have to change your holiday service 9 million places when your shitty government decides to remove one of them (thanks Denmark). Between the two, I would personally prefer working on the one where I have to do the 9 million changes, but I would obviously prefer neither.
Yes, but often we don't know the domain enough but "this feature must be available yesterday". So add tests, copy, release. And when you have to either do it again or have to update this code and its original you should know more and be able to refactor and give good names to everything.
Example: So much early code is boilerplate CRUD, that it's tempting to abstract it. 9 times out of 10, you'll create a quasi-ORM that starts inheriting business logic and quickly grows omni-functions.
Eventually you may actually need this layer, assuming you're system miraculously scales to needing multiple services, datastores, and regions.
However this doesn't just apply to the obvious, and you may find omni-logic that made a feature more simple once and is currently blocking N new features.
Code is cheap, especially today. Complexity necessarily constrains, for better or worse.
If I need to introduce the same feature in multiple places in roughly the same way, that's a decent indication code wants to be the same and wants to change together. That's something to consider extracting.
Fixing the same bug in several places is a similar, but weaker indication. It's weaker, because a bug might also occur from using a framework or a library wrong and you do that in several places. Fixing the same business logic error in several places could mean to centralize some things.
Maybe "dumb is robust" or "straightforward is robust" capture the sentiment better?
As a biomedical engineer who primarily writes software, it’s fun to consider analogies with evolution.
Copy/pasting and tweaking boilerplate is like protein-coding DNA that was copied and mutated in our evolutionary history.
Dealing with messy edge cases at a higher level is like alternative splicing of mRNA.
Used within a team setting, what is simple is entirely subjective to that set of experiences.
Example: Redis is dead simple, but it's also an additional service. Depending on the team, the problem, and the scale, it might be best to use your existing RDBMS. A different set of circumstances may make Redis the best choice.
Note: I love "dumb is robust," as it ties simplicity and straightforwardness together, but I'm concerned it may carry an unnecessarily negative connotation to both the problems and the team.
Simple isn't necessarily dumb.
You can keep twisting this question until you realize that without the behemoths of complexity that are modern operating systems (let alone CPUs), we wouldn't be able to afford the privilege to write "simple" code. And that no code is ever "simple", and if it is it just means that you're sitting on an adequate abstraction layer.
So we're back at square one. Abstraction is how you simplify things. Programming languages themselves are abstractions. Everything in this discipline is an abstraction over binary logic. If you end up with a mess of spaghetti, you simply chose the wrong abstractions, which led to counter-productive usage patterns.
My goal as someone who writes library code is to produce a framework that's simple to use for the end user (another developer). That means I'm hiding TONS of complexity within the walls of the infrastructure. But the result is simple-looking code.
Think about DI in C#, it's all done via reflection. Is that simple? It depends on who you ask, is it the user or the library maintainer who needs to parametrize an untyped generic with 5 different type arguments?
Obviously, when all one does is write business logic, these considerations fall short. There's no point in writing elegant, modular, simple code if there's no one downstream to use it. Might as well just focus on ease of readability and maintainability at that point, while you wait for the project to become legacy and die. But that's just one particular case where you're essentially an end user from the perspective of everyone who wrote the code you're depending on.
Why is your “AirBnb for dogs” startup with zero users worrying about C100K? Did AWS convince you to pay for serverless shit because they have your interests in mind, or to extract money from you?
Write code that is easy to delete, not easy to extend (2016) - https://news.ycombinator.com/item?id=24989351 - Nov 2020 (30 comments)
Write code that is easy to delete, not easy to extend (2016) - https://news.ycombinator.com/item?id=23914486 - July 2020 (109 comments)
Write code that is easy to delete, not easy to extend - https://news.ycombinator.com/item?id=18761739 - Dec 2018 (2 comments)
Write code that is easy to delete, not easy to extend - https://news.ycombinator.com/item?id=11093733 - Feb 2016 (133 comments)
____
I've come to believe the opposite, promoting it as "Design for Deletion."
I used to think I could make a wonderful work of art which everyone will appreciate for the ages, crafted so that every contingency is planned for, every need met... But nobody predicts future needs that well. Someday whatever I make is going to be That Stupid Thing to somebody, and they're going to be justified demolishing the whole mess, no matter how proud I may feel about it now.
So instead, put effort into making it easy to remove. This often ends up reducing coupling, but--crucially--it's not the same as some enthusiastic young developer trying to decouple all the things through a meta-configurable framework. Sometimes a tight coupling is better when it's easier to reason about. [...]
You might, but there's also going to be other people that will happily go ahead and create abstractions and logic that will form the very core of a project and entrench themselves to such a degree that they're impossible to get rid of.
For example, you might stumble upon CommonExcelFileParser, CommonExcelFileParserUtilities, HasExcelParseStatus, ProductImportExcelParser, ProductImportExcelParserView, ProductImportExcelParserResultHandler and who knows what else, the kind of stuff that ends up being foundational for the code around it, much like how if you start a front end project in React or Angular, migrating to anything else would be a Sisyphean task.
In practice, that means that people end up building a whole platform and you basically have to stick with it, even though some of the choices made might cause bunches of problems in the future and, due to all of the coupling, refactoring is way harder than it would be in an under-abstracted codebase.
I'm not sure what to do then. People seem to like doing that more than applying KISS and YAGNI and making code easy to delete.
* Software has a tendency to become maximally complex. You either have an actually complex domain, or the developers will find a way to increase the complexity (..because otherwise, they're bored)
* Good software is modular and easy to remove. Consequently, good software will keep getting replaced until it's bad and cannot be removed anymore
Frameworks and libraries not really, for those you still have to adjust to whatever happens in the world but at much saner pace.
Biggest issue is when devs want to write “framework” when they work on business line application where they have frameworks that they are already using like Rails/Asp.Net etc.
This is correct, but from my experience of working in the same company for over a decade: You'll learn to foresee requirements. Especially the "we'll never need that" ones that become business critical after a few months/years...
However it would have all been so much easier if they'd realized their business domain called for "Foo Revisions" in the first place.
Unless you’re writing the Linux kernel you shouldn’t write it like the Linux kernel.
I recently did our first semi-automated removal of exposed graphql resolvers, metrics about how often a given resolver was already available so parsing that yielded the set of resolvers I *couldn't* delete. Graphql already has a deprecated annotation, but our service didn't handle that annotation in any special way. I added observability to flag if any deprecated functions have been called & then let that run for sufficiently long in prod, then you can safely delete externally exposed code.
I think external consumption as you frame it is a good example of this. It’s fair to give consumers a reasonable warning about the depreciation of a service, but if you can’t actually shut it off when you want to, then you’ve not designed your system to let things be easily deleted.
Which is fair. If that works for you, then so things that way. I suspect it may not work too well if you’re relying on tests and observations to tell you if things are breaking. Not that I have anything against tests, but it’s not exactly a great safe-guard if you have to let them tell you if you broke something in a long complicated chain. Not least because you’re extremely unlikely to have test-coverage which will actually protect you.
> To write code that’s easy to delete: repeat yourself to avoid creating dependencies, but don’t repeat yourself to manage them. Layer your code too: build simple-to-use APIs out of simpler-to-implement but clumsy-to-use parts. Split your code: isolate the hard-to-write and the likely-to-change parts from the rest of the code, and each other. Don’t hard code every choice, and maybe allow changing a few at runtime.
> The problem with code re-use is that it gets in the way of changing your mind later on.
This is simply incorrect, especially in the generality in which it is stated. If you change your mind and the code was copy-pasted to ten places, then you have to change ten places. On the other hand, if the code is in a function, then you only need to change it once. And if you do find that one of the ten invocations should not be changed, then you can still copy-paste - or make the function more general.
Like crossing a street without looking, copy-pasting is almost always a bad idea.
Of course, the answer is “don’t make bad abstractions”, but we all know how that one goes with a team and changing product reqs.
I would trade that for one piece of mediocre abstracted code any day.
Oh yeah and everything in the codebase is copy and pasted.
Ah yes, but what happens if you have to change 3 of the function invocations in one way, 5 in another, and the other two need to be completely rewritten because those aren't even using the same abstraction any more?
If it's all in one function, most developers will try to change that function to make all 10 cases work, when it should never have been one function in the first place.
It is much much easier to fix ten copy-paste places than to untangle a knot that should never have been tied, once it's holding pieces of your system together.
In a many cases I'd still rather have three or more versions of a function, many which may just be very thin shims to accommodate that scenario than 10 copy/pastes of variations. Or shim at the call site and keep one function if that suits.
Languages like Erlang which can have different versions of a function, selected by pattern matching (with optional guards) make this convenient:
Name(Pattern11,...,Pattern1N) [when GuardSeq1] ->
Body1;
...;
Name(PatternK1,...,PatternKN) [when GuardSeqK] ->
BodyK.You can't prevent future morons from doing moronic stuff in the future. They'll just find another moronic thing to do.
The author would probably argue that you should have moved that code to a module / function.
Superficially, they contradict themselves on the topic. When read slowly, they use copy-paste as a way to indicate what code should be abstracted, and what really is a pattern to follow.
yet, we keep bringing this stuff up like it's some sort of genius insight / silver bullet
the problems in engineering rarely stem from the lack of principles and have way more to do with mismanaged projects, arbitrary deadlines, shifting priorities, unreliable sources of data, misunderstood business logic and all those fancy acronyms, all the SCRUM and agile in the world will never make up for all that
For example, abstracting every piece of similar code to make it "DRY" because they don't understand that it's about concepts not code.
If:
1. You know where the 'creases' of orthogonality are. You've carved the turkey 1000 times and you never get it wrong anymore.
2. As a result, there is hardly any difference in complexity between code that is and isn't easy to extend.
Then write code that is easy to extend, not delete.
The question is whether your impression of the above is true. It won't be for most junior developers, and for many senior ones. If orthogonality isn't something you preoccupy yourself with, it probably won't be.
In my experience, the most telling heuristic is rewriting propensity. I'm talking about rewriting while writing, not about refactoring later. Unless something is obvious, you won't get the right design on the first write. You certainly won't get the correct extensible design. If you're instructed to write it just once, then by all means make it easy to delete.
If you think you are good enough to qualify you almost certainly don't qualify. If you do qualify then chances are you probably don't think you do.
f(x) = 6x^2 - 5x + 1
g(x,a,b) = ax + b
f(x,q()) = q(x,3,-1) q(x,2,-1)
f(x,g)
However, it's algebraically orthogonal. It has advantages in some cases, for instance if you later add code for a 6th-order polynomial and need to use its zeroes for something else.
We know that it could be better in some cases. Is it a good bet to predict that it will be better overall? The problem domain can fracture across a thousand orthogonal "creases" like this one. The relevant skill is in making the right bets.
Here's an example that's not orthogonal. Let's say we think the 6 coefficient might be more likely to change in the future:
g(x,a) = ax^2 - 5x - 1
f(x,q()) = q(x,6)
f(x,g)
I don't see why this would be useful.
If you want the code to do something different later, change, replace or extend it then... when you actually know what it needs to do
It is my understanding that we should try to build solutions to current problems, and be open to future use cases that could involve small additions in functionality.
It would be stupid to design an unmodifiable system just because some parts can be deleted and we are not sure what future needs are. Code should always be easy to extend, in my opinion.
But what is an unmodifiable system? If it's code in your control, it can be changed, right?
I hope I won't offend anyone pointing at it [1]. This is a somewhat popular tool to evaluate macroeconomic policies in the EU.
A combination of language choice (C# as a natural language Microsoft Excel bosses tend to request from their interns to use), the usual churn of academic undergrads and loads of other cultural failures are the reasons this monster exsts.
Someone should write a book how to make the worst ever codebase, and start with EUROMOD.
[1] https://github.com/ec-jrc/JRC-EUROMOD-software-source-code
I'm not sure how constructive that would be. I'm still hurting because the IT department decided the only way to deploy my Java app was through rsyncing to a running Tomcat installation, allowing class files from several deployments previous to resurface in memory causing some beautiful bugs.
Or the time they decided to buy a Hadoop cluster at a cost of EUR 100k which I told IT dept they wouldn't be able to connect to from the outside world because the network rules are carved in stone. They bought it, and guess what, network ops said no.
The ten foot high touch screen and the car emissions data stored in Excel files and the 80 million euros spent on a website or the time the partner research group refused to release the data we had funded so we couldn't run workshops or release the project (around EUR 2 million).
The waste.
You can delete while resync'ing but I guess the issue is not in resyncing itself, but rather in the disempowerment of individual contributors.
You could have argued to add --delete for your case, as well as requesting a shutdown before and a start after, but I guess explaining this to countless morons is too much to ask from a humble developer.
OTOH, this resyncing story probably means that you were allowed to choose the wrong development framework to start with. Because resyncing PHP is much more reasonable.
We are rewriting the codebase from scratch in Rust and Svelte.