How would you define high quality code?

What would be the main attributes of good code?

Did you find this post useful? Show some love!
DISCUSSION (61)

There's a really great chapter about this in "Game Programming Patterns" by Robert Nystrom:

For me, good design means that when I make a change, it’s as if the entire program was crafted in anticipation of it. I can solve a task with just a few choice function calls that slot in perfectly, leaving not the slightest ripple on the placid surface of the code.

That sounds pretty, but it’s not exactly actionable. “Just write your code so that changes don’t disturb its placid surface.” Right.

Let me break that down a bit. The first key piece is that architecture is about change. Someone has to be modifying the codebase. If no one is touching the code — whether because it’s perfect and complete or so wretched no one will sully their text editor with it — its design is irrelevant. The measure of a design is how easily it accommodates changes. With no changes, it’s a runner who never leaves the starting line.

Be sure to read the whole thing!

I like this a lot. "Accommodation to changes" accounts for readability, organization, testing, and a lot of other things bundled in.

Coding only for the problem at hand right now: Bad.
Coding for every possible future need: Bad.
Coding to solve current problems while being accommodating towards future modifications: Good.

In his dry-rb at Deccan RubyConf last year Tim Riley said that well-written applications should be "change positive", i.e. conducive to future change. I made this expression part of my regular programming vocabulary and promote it every chance I get, like in this comment ;-)

Coding for every possible future need: Bad.

I have seen this scenario a few times and it obviously delays the whole project because requirements keep changing and so does the code.

A few points taken from Robert Nystrom are:

  • Key goal of software architecture: minimize the amount of knowledge you need to have in-cranium before you can make progress.
  • Get problem -> Learn Code -> Code Solution -> Clean up
  • Change to one piece of code doesn’t necessitate a change to another. We obviously need to change something, but the less coupling we have, the less that change ripples throughout the rest of the game.
  • Good architecture makes a huge difference in productivity. It’s hard to overstate how profound an effect it can have.
  • Good architecture takes real effort and discipline. Every time you make a change or implement a feature, you have to work hard to integrate it gracefully into the rest of the program. You have to take great care to both organize the code well and keep it organized throughout the thousands of little changes that make up a development cycle.
  • You have to think about which parts of the program should be decoupled and introduce abstractions at those points. 
  • Writing well-architected code takes careful thought, and that translates to time. Maintaining a good architecture over the life of a project takes a lot of effort. You have to treat your codebase like a good camper does their campsite: always try to leave it a little better than you found it.
  • Design requires a lot of experimentation and exploration. Especially early on, it’s common to write code that you know you’ll throw away.
  • If there is any method that eases these constraints, it’s simplicity. In my code today, I try very hard to write the cleanest, most direct solution to the problem. The kind of code where after you read it, you understand exactly what it does and can’t imagine any other possible solution.

This. There are few things that I feel like are as satisfying as being like "Oh crap, I need to make sure it can handle ABC" or "I need to change it for this new thing" and realizing you wrote it such that sure, it does need to change somewhere, but it has that capability for change without things getting messy. The opposite of having to add more and more logic to handle cases.

High quality code is the one that has a low wtf/hour ratio

This is the one and only true metric for good code quality. It never fails.

Exactly so, although I think that per minute would be a really big number hahaha.

An interesting point from Gridshore on this:

The most effective and certain method, however, is the code review. However, it is also very time consuming. It requires that one or more developers restrain from being (in the eyes of the customer) productive and give their attention to someone else’s code. Provided that the developer has a high standard of quality, it will definitely improve total quality of the code. Of course, there is the concept of pair programming, which could be seen as a form of continuous code reviewing. Highly effective, and can improve the skills of both developers.

These tools and processes can be very useful, but their overall efficiency depends on one thing: quality awareness within your development team. I’ll explain what I mean with “quality awareness”. It is the ability of developers to recognize code smells or failure to use design patterns at locations where they would typically be applied. This awareness it typically raises as developers gain more experience, but we’d typically like to help fate a little hand.

That is very simple: code that accomplishes the task is good.

Anything else, despite how concise or verbose, readable or wtfable, dry or wet, black or white it is, does not matter.

Tons of books written on a subject all defend the willingness of modern developers to write the code, that conforms all the good practices to have an excuse on why it’s not working.

Check Linux source code. It does not follow any single principle from these fancy books. And it is brilliant because of it serves, you know, the majority of devices in the universe.

I was about to post a comment saying pretty much the same :-)

Most code is a means to an end, not an end it itself. Maybe it's because I spent a lot of my career working at/consulting startups, but future proof code doesn't matter much if it took so long to write that the company has no future (been there). As the old saying goes, "it's null until you ship it".

Now of course every software developer who considers themselves a craftsperson will instinctively object to this statement. But to me the real craft in programming is finding the best possible solution to often competing constraints/interests, and business needs and time allocations are just as much part of this as runtime complexities and memory consumption.

I better stop now, otherwise I might end up adding to the "[t]ons of books" you mentioned ;-)

That's not necessarily true in all cases.

You write code as language for a computer. When you speak to another person, whether it's functional or an expression, you can say it in multiple ways -- all accomplishing the same task, in bad, good, great, and best ways. "Accomplishes the task" is ambiguous, as is the task's definition. For example, a task to display an RSS feed on a page is different for a junior dev vs an intermediate, or even another junior dev. It's very well different for different types of people, roles, cultures, etc.

Code quality measurement is like checking someone's health. If your life purpose was defined with 100% accuracy and our biotech was 100% accurate could one accurately determine a person's health exactly. Because we're far from that point, we make estimations based on multiple dimensions like weight, height, sugar levels, etc; similarly, we have test coverage, penetration/smoke testing, analytics, user group sessions, etc.

“Accomplishes the task” is ambiguous [...]

How is that? Assuming we have a task definition, that is both complete and consistent, accomplishing it means everything is done as requested and nothing is done besides that.

There is no need to overcomplicate things with all that marketing stuff. All that test coverage, analytics, sessions, younameit are useless once we have a clean task definition. They all are used to camouflage the inability to state a problem in the first place.

Donald Knuth wrote exactly zero tests for TeX and the code is still better than 99% of what’s being produced today. With all the tests, sessions and other crap.

But it does matter. If you ever had to do any sort of code maintenance or expanding of existing code this is obvious. Code that is "clean"/"good" is easier to work with. "Does the code work?" is very important, but it doesn't even begin to be good code.

These books you mention, I would love to hear your arguments against their actual point instead of dismissing them by mentioning Linux or any other project, since that's not really the point.

If you ever had to do any sort of code maintenance [...]

Yeah, I had, sorta. In more than 10 different languages. In almost 30 years.

There is no single merit of how easy is the code to read, or how clean is it. What looks clean for me might be not as clean for you and vice versa.

That said, trying to apply an automated analysis to the code and claiming it’s a good assessment for readability is silly. That’s why I never use linters and static analysis tools of any kind. They do more harm than good.

Definition:

  • More theoretical ideal; perfection is not attainable
  • "Good" implies partially subjective
  • Not binary.; there are many states in-between
  • Not absolute; varies over time

Attributes:

  • Well designed without being over-engineered
  • State of the art, but not obscure or overtly clever
  • Documented, but no superfluous comments
  • Reviewed and refactored within time-constraints
  • Optimal in time and space while avoiding premature optimization
  • Meets business/technical requirements and common sense
  • Leverages tools of the trade (testing, static analysis, etc.) if/when appropriate
  • Maintainable and extendable in accord to its purpose

It's like art vs obscenity; I know it when I see it.

  • Reviewed and refactored within time-constraints

So, what would be the optimal time? Or is it somehow subjective depending on the circumstances?

It's subjective and variable depending on the circumstances:

  • Skill and experience of the author (I tend to gloss over PRs by people that do consistently good work)
  • Difficulty/importance of changes
  • ROI for additional time spent etc.

There's an implicit assumption that you can't spend "forever" on a piece of code; at some point it's got to be "good enough" so you can move on. It's also subject to diminishing returns.

Personally, the more that code reads like prose, the higher quality it is. High quality code tells you the story about the business solution that it is implementing.

My favorite code is that which reveals to me something about the business that I did not previously know. Plus, like businesses, it’s written in a way to accommodate the future only when the future arrives.

To me, high quality code means less lines of code, but has more power. That power might be a useful abstraction, or it's just a useful algorithm to solve repeatable problems.
And if you want more quality of your code, make it easy to test, the easier to test a code block, the more quality that code block has.
And if you want even more quality of your code, make it fast. If your language couldn't make it fast enough, change the language.

"Make it fast" means you should use better algorithms or data structures to solve the problem. Because as you know, to solve one problem, there're many ways to do it. So it's also called optimization of the code to use better libraries, for example.

Maintainable and efficient.

Maintainable means it is easy to understand, sensibly organized, it is painless to make modifications that should be expected (client wants to add another type of report), and as easy as possible to make changes that aren't (client wants a totally new feature)

Efficient means the code uses as few resources as possible to accomplish its job, so that it is as fast and scalable as possible.

I'm going to use the #likeimfive approach here.

If code is like cake, then good code is like cake that you can enjoy eating and that will not leave you with stomach aches later on.

Note that 'doing what it is supposed to' is not an attribute of good code any more than 'being edible' is an attribute of good cake. That's just a fundamental real-world requirement for when someone asks you to write code / bake a cake. It doesn't have anything to do with either of those being good.

Just to highlight.. so you mean cake is considered good based on the experience of eating it.

In the case of cake, yes. How that translates to code is that code is also experienced as being good by how you interact with it later, i.e. modifying it in any way (rather than just looking at it).

In my opinion "good" code is the one that is disposable. Is a codebase where you can identify with confidence the parts that need to be changed or removed in order to make a new feature or meet a new requirement.

You won't necessarily recognize it when you see it or when you write it for the first time, but you'll know when the times comes and you have to mantain it.

My personal definition is very subjective: A code is of high quality if

  • someone other than the author can read and understand it, without investing too much time and coginitive effort
  • it follows the conventions of your field (using familiar variable names, familiar abbreviations, familiar unit systems), whatever they might be.
  • and one thing (source file / class / package) solves one task.

I’ll chime in with some characteristics I try to impart to my code:

  • simple: keep the code as simple as possible. Avoid over-engineering.
  • readable: make the intent of the code as clear as possible.
  • orthogonal: give modules of code clear and independent responsibilities. Ideally modules that work together should be composable in any combination.
  • tested: code should have automated tests by default.
  • graceful with errors: modules should have well thought out considerations for edge cases and error conditions.
  1. Easily accessible - simple file structure and well organised
  2. Self-explanatory - meaningful names for files, variables, methods, etc.
  3. That doesn't mean commented code is unacceptable.
  4. Aesthetics - Well indented - tabs size, spaces, etc.
  5. With test cases.
  6. Reviewed by collaborators.
  7. Readable. Not over optimised.

Well, there is a lot of books, articles and talks for that matter, but all of them lead to two things: low coupling and tests.

If it's feature, it must be connected to codebase through interfaces and properly tested (mocks, benchmarks for critical parts etc).

If it's refactoring, tests must remain untouched.

Looks pretty simple, but it's not.

One way to measure good code is that you need the knowledge of your problem to solve it, but no more. For example, if you need to fix a bug regarding to who financial transactions are sent, you obviously need to know which beneficiary to use in each use case. You should not need to worry about algorithms calculating the amount sent or interaction with external systems. This allows the developer to focus on one problem and one problem only.
Quite often this is achieved by modular abstractions, e.g. hide the external system behind an interface and supply the transferred amount through classes with calculated getter properties.

High quality code is code written by someone who cares.

What does it mean to care?

  • Knowing the language beyond the syntax, i.e. writing concise ideomatic code

  • Naming things in a way which helps understanding the purpose of things

  • Grouping things together, which belong together

Or as Ward Cunningham once put it:

You know you are working with clean code when each routine you read
turns out to be pretty much what you expected. You can call it beautiful
code when the code also makes it look like the language was made for the problem.

I heared once, that instead of naming it »software engineering« it should be renamed to »software gardening«. This would connotate the constant effort which has to be put into a codebase in order to keep it beautiful and keep weeds at a minimum.

High quality code is

  • Maintainable: it must be readable, understandable, changeable, and easily removable
  • Effective: reaching the desired goal with the least mental/computational workload
  • Robust: fail only if neccessary, and then early and verbose
  • Secure: written with possible vulnerabilites in mind

The projects I've worked with that had the best codebase to them actually made it difficult to write bad code with it. Bad code typically comes when there is too much going on and it is not well abstracted, so developers have trouble understanding it and end up writing more bad code, making the problem even worse.

In contrast, a good codebase has very clearly-defined structures and well-architected patterns that guide you toward the solution intended by its creators.

For example, Laravel was created for a specific purpose, and every API it opens up to developers drives you right where it wants you. You really have to put in a lot of work to use Laravel in any way other than it was designed for, and the way it encourages you to develop in it feels very natural and clean. The underlying codebase of Laravel is what I consider to be the best codebase I've ever seen and worked with.

Comparatively, Android's codebase is a bit of a mess. Don't get me wrong, I love Android and it is my favorite platform to develop for, but its codebase is messy and confusing, needlessly complex, often contradictory, and with very little direction given to the developer. It is definitely getting better now that it has matured the core framework, with Google starting to focus on better APIs for UI components and a more opinionated approach with Jetpack, but I still struggle to understand how most of the internal classes work, no matter how long I study their code.

Therefore, a good codebase is the base of good code, isn't it? :)

I think what complicates this question for me is the intent of the code and future usage which is hard to predict. As a business owner I want the best bang for my buck to get to market as quickly as possible. As a maintainer of software my viewpoint is going to be more on how I can transition something over time. Very rarely am I thinking of how elegantly a solution is solved than how does it fit in an overall architecture.

I would look at the following criteria:

  1. When I make a change, how confident am I that the change will do exactly what I expect it to?
  2. Also, how long do I need to look for the right place to make the change?
  3. Are simple things simple to implement?
  4. How long do I need to look at the codebase to figure out "the system", i.e. how everything is organised?

These all sort of tie together, something with no unit tests for example is likely to fail on 1, whereas something with extensive but brittle unit tests will fail on 3.

I think the quality of code has mostly to do with who needs to work with it.

Sometimes techniques that are hard to grasp can improve attributes of the software, like performance.

Some highly skilled people can write very performant code, but only other highly skilled people can work with this code.

Should we make the code more readable and easier to grasp but sacrifice performance so it can't be used by as much users as before?

I think 90% of the time the question is yes.

Write cleaner code and make it more maintainable, the bit of performance gains you could achieve aren't worth the trouble.

But sometimes it is worth it, sometimes you need to write high performance code so it works for the huge amount of people who only have slow devices to run your software.

Loosely coupled and extensible. How this is achieved is up for debate in most circles.

Clean Code by Uncle Bob is a great start. For me it comes down to small class sizes, avoiding breaking changes from modifying a portion of code. Don't use hard coded values, don't use hard to read variable names. Make it so that the next person to read your code can understand what it's doing, this includes your future self.

Write unit tests, and this is also part of keeping functions small.

Uncle Bob says:

The answer to clean code is craftsmanship.

There are two parts to learning craftsmanship: knowledge and work. You must gain the knowledge of principles, patterns, practices and heuristics that a craftsman knows, and you must also grind that knowledge into your eyes and gut by working hard and practicing.

Learning to write clean code is hard work. It requires more than just the knowledge of principles and patterns. You must sweat over it. You must practice it yourself, and watch yourself fail. You must watch others practice it and fail, You must see them stumble and retrace their steps.

That is good advice for a beginner!

Some books might define it as "high quality code is when you don't break your software while doing small changes", which is totally not true, if you designe your program to play music, it shouldn't play movies too. For me, high quality code is when your codes are optimized, easy to read and well organized.
If you're afraid of not having a quality code then don't worry, i myself had some very bad habits of writing everything in one file 3 years ago but i fixed it as soon as i realised how hard it was for other people to read my codes. I fixed it with habits, you start by trying to write a well organized file then repeat it with every project you make until it becomes a habit.

Ideally it should have all these properties:

  • Readable
  • Maintainable
  • Robust
  • Flexible
  • Low overall cyclomatic complexity
  • Modular with clearly defined dependencies
  • Automated testing of behavior and use cases, including alternate paths and exception paths

Don't you use manual testing at all?

Sure, but the quality pretty quickly suffers in that you have no regression tests unless you very clearly specify the manual test cases and test steps. These manual tests have to be followed very carefully at least before each new release and deployment.

When I feel very confident about some code I leave out tests but often discover stupid bugs like a boolean being the reverse of what is expected in a specific state. Depending on the difficulty of the deployment process and the critical level of the project, I ramp up the amount of automated tests.

Quality code is stable and easily maintained.

Professionally I do not even have an experience to use for reference, but as a student I would say this:
-> Quick run
-> Quick reading
-> easy to understand

High quality code is 1) not easily reproducible and 2) valuable.

note: this philosophy skews my coding conventions against the grain pretty hard.

One, that keeps running a decade later and is still easy to understand, although it was modified many times by multiple developers over the years.

  • Fewer lines of code in methods.
  • Code following SOLID Principles.
  • Code should be understandable easily when you look at it.

I'd say this is one of those questions that is probably better left unanswered. Does your software deliver value? If yes then quality doesn't matter.

Classic DEV Post from Apr 27

What is your personal Programming ethics?

...

Julia Torrejón
Live. Learn. Code
Join dev.to

Better understand your code.