Although everyone has its own programming style, I personally don't think you should start with the game logic of the game without getting any visual output.
I like to start setting up the core components in order to draw a screen, implement all input/visual/audio APIs in an easy to use manner, setting up a basic gamestate system, entity manager incl. messaging etc. etc. I think this is actually a good programming behaviour, start by making a planning and design of what you actually need for your game.

Absolutely. A game, even if you make it without using an 'engine' that's pre-made, has components in it that can be called an 'engine'. The term 'engine' is used very generically.

I'm in no way saying start making levels before your game can display the levels, nor am I recommending to just 'code it as you go'.

Yes, design and plan out the architecture of your game, which includes the structure that makes the game run. My point is, don't try to make that structure so generic that it'll fit any game. Make the structure specific to the one game you are currently focusing on. If you make a game, you make an engine that is custom-made and specific for that game, whether you intend to or not, whether you know about engines or not. Later, you can remove those 'engine' parts, make them more flexible, and use them for another game. This becomes your 'engine'.
But if you try to make a engine that is flexible from the get-go (without having lots of experience), without having previously made multiple games, and without having a specific game in mind, your engine becomes useless because it lacks a game to keep the engine's features and goals in-check.

I'm not arguing that your game shouldn't be planned out, and engineered skillfully, but rather that you should skillfully engineer the game and engine together, so the requirements of the game actually steer the development of the engine, lest you make an engine that serves no purpose and fits no game without heavy modification.

I got the same advice when I started: make a game not an engine. But after finishing a couple of games I felt like I realy couldn't re-use much of my code.
Mostly because I thought things weren't done efficiently or weren't easy to use. At that point I started doing things the way I described above, using carefull design and starting from the ground up and now I can just use this framework I made and start coding the game almost right away.

But the key there is "after finishing a couple of games". You already understand how games work, before you were able to make re-usable components that work with games. If your very first programming project is making a game engine, you won't know how to make the engine because you never made a game.

So to summarize my advice: spend some time making a good design, if you write a very good reusable piece of code it'll safe time in the end by avoiding having to recode everything again.

Definitely. This applies to all software programming. Reusable pieces of code should be created for everything from text editors, to websites, to games. Reusable pieces of code on their own don't become engines. 'Make games' doesn't mean, 'Code poorly' or 'Don't plan', it means (when starting out) 'focus on your current project (the game), not on your next few projects'. It means, don't write your code for imaginary projects down the line, but write it for your current project today. Definitely write it reusable, clean, and stable the first time - you'll be able to adapt it and use it when you get to those later projects, and overtime it'll become very flexible and even more stable. But trying to write it flexible the first time (when you're a beginner), results in not useful pieces of code even for your current project.

Writing code without having solid requirements in mind, results in code that doesn't really fit any purpose.

But don't try to implement all kinds of features you're not going to use in the current game, you can easily implement those features as soon as you write the game you'll need it in.

Yes, exactly! That's one of the primary points 'write games, not engines' is trying to get across. You phrased it alot better than I did.

I'm not at all trying to say, don't write clean code. All code should be clean, almost all the code should be reusable, self-contained, and stable. I'm trying to say to write functional code (whether it be an engine or a helper library or just a few functions), you must have a goal in mind to steer the requirements, or the code usually becomes too general to use without you having to write boilerplate over it to make it usable.

Almost all games have 'engines', and so 'write games, not engines' doesn't mean to make a game without making the parts that run the game - it means to make a game (including the parts that run the game), but don't try to make a super-flexible engine before you even have a game in mind - tailor the engine to the game, the engine is part of the game, so let the game drive the design of the engine.

I don't think a beginner is really considering all kinds of low-level stuff when he talks about a game engine. And telling him to just code the game without thinking of the engine, might let them think they should rush into programming the game logic and skipping a good game design. But I tend to agree to most points of the last posts here and in the end it might actually be a good thing if you let the beginning programmers find out their code is not re-usable, so they can learn from their mistakes :)

As someone that has coded a framework 3 times now (I wouldn't call it an Engine, but more of "reusable classes" for my games), I would advise making SMALL games and then designing an engine/Framework around that.

After my 4th or 5th game, I scraped my design, and I'm currently reworking my framework to be more "engine" like, and I'm actually happier this way. But I would have no idea what to look for if i just started by attempting to build an engine. I'm not sure how much of this is related to a 2D engine since the concepts are easier, but there is still resource management and data structures. Start small. But always think about "How could I reuse this if I wanted to make another game".

If you know nothing, then you need to start from somewhere and build off from there. It is the only way to learn and gain an understanding of a particular subject.

If you do not know anything, then you should not even begin building an engine. It is a good challenge if you know enough.

Learning the basics is like riding a bike with training wheels. Learning GUI and graphics is like riding a bike with the training wheels off.

I've recently started a game project, myself being the only developer and working in collaboration with 3 artists. I settled on the Unity Engine for my 2D game because of how much is already done for you as a game programmer can focus on gameplay and not the boring stuff like creating collision libraries and object factories and object managers.. etc the list can go on forever. Also, the asset pipeline in Unity is extremely robust and can pretty much take anything you throw at it as far as 2D/3D assets. Give it a try if you have the time - well worth it.

Your engine is only good for what you test it for.

The big part of writing a reusable and good code is being able to test it in a real-world scenario. In case of a game engine, I would discuss that it is near impossible to write a good game engine without writing a game. Even then, this makes this engine only tested for this specific type of game [1].

Why it is so? Because a use-case is the main factor that drives any software. A good game code works for game-related use case, and not for an idea of a game-engine. If we test your engine by executing some demo of 3D terrain and a skybox, then our engine is good for making a demo for 3D terrain and a skybox. If we try to build a game on top of that, we will most likely find that our engine needs to change. Because our use-case have changed.

Do we really need an "engine"?

Why do we even want to write game engines?

I think we do not like the idea of our code being used for one game only. Why rewrite, when we can reuse the same thing again and again? Let's separate our engine-specific and game-specific code and make engine reusable this way.

Well, the problem is that games are quite different. And we like different games. Even in 2D side scrollers, we want to make new wildly insane mechanics different for every game and maybe even the level. The bad things start happening when the game-specific stuff starts to creep into our engine. Now, instead of being optimized for a single game, our engine needs to be optimized for both use cases. And sometimes it is not feasible, and we may end up using separate engine versions for separate games. We should have just made different games without wasting time for one engine.

The "good" code.

Let's just accept the real world, and embrace it. In the real world, no one ever creates perfect code (except me, but that is a joke [2]). Things break in the real world. Sometimes we need to rewrite big portions of the code. And we learn, often the hard way, that writing a good code requires a knowledge of what a bad code is/was.

And sometimes a good enough code for one game may be a bad code for another. And vice-versa.

It depends on what it is used for.

Group code by the Use Case.

The big thing driving a good software design nowadays is called decoupling and separation of concerns [3].

The Big Idea is to create the tools, not frameworks or engines. A tool is self-sufficient. It should not know about anything else but itself, and require the minimum of third-party components.

Inject a bit of pragmatism here. It is OK to copy-paste a helper function from our "engine" into the tool to make it more self-sufficient. Even better, optimize this helper function to work better for our tool. Our tool can have its own "vector" type, or use plain arrays best suited for its job, instead of using engine's "vector" type. It may not be DRY [4], but it might be OK.

A game is a test for the tools, as well as the reason the tools exist. It should remain that way.

The question of deciding what "tools" to write should be dictated by the requirements of the game, not a vague idea of what possible games there might be.

How to write a tool?

Let's write a game, and let the tools be the solutions to the problems we encounter while writing a game.

For example, write an input handling the way it is best suited for our game, and make it more and more tool-like and decoupled as we complete it. This way we can keep an eye on suitability for the game and the performance of our tool.

In the end we can move this "tool" into a separate library. We may use a software like CMake [5] to avoid thinking too much about compilation and linking. We may even write a small demo application that uses nothing but this tool.

We can write a small demo applications to demonstrate that our solutions work properly without the dependencies of the full game.

This is good for several reasons:

• If we ever want to evaluate our tool/solution for inclusion into another game, it is easy to see if they are suitable, because a use case is clearly demonstrated in a small amount of code that is required for a demo.
• This motivates to avoid adding dependencies to our libraries, because we will need to add dependencies to our demo applications.
• We see more immediate results while developing a game, because we have the working demo applications sooner than the finished game.
• In case something breaks or needs improving, we can use the demo as testing grounds instead of launching now-big game with all the other clutter.

And even if we do not succeed to completely move our tool into a separate library, our code is still better and more decoupled, because it is more "tool-like".

[1] Write Games, Not Engines, scientificninja's blog.

[2] “Nobody actually creates perfect code the first time around, except me. But there's only one of me.” [Tech Talk: Linus Torvalds on git, Google, 15 August 2007].

[3] Managing Decoupling, By Niklas Frykholm, GameDev

[4] DRY - Don't Repeat Yourself

[5] Cross Platform Test Driven Development Environment Using CMake, By AllEightUp, GameDev

(should I expand this into a full article? there is much more to say on this subject)