"Steak before sizzle"

Get something done and working, even if performance sucks. Then do your performance logging and optimizations. Don't attempt to pre-optimize, because you may make a good situation worse.

That doesn't preclude selecting correct algorithms and such beforehand; ie - don't be stupid about it. But don't try to be clever before you know you need it, either.

Second thing I always think should be necessary for any software project:

Get your security design working first.

Don't try to patch in security after the fact; if this isn't at the top of your design and planning, it should be (assuming the project needs it or can be foreseen to need it). Too often I have witnessed the opposite - and it almost never works without major refactoring (or scrapping and starting over).

I like it!

> That doesn't preclude selecting correct algorithms and such beforehand; ie - don't be stupid about it. But don't try to be clever before you know you need it, either.

Exactly. "Avoid premature optimisation" doesn't mean "make it as slow as possible", it just means "use std::vector until it becomes an issue instead of hand-coding some arcane data structure in assembler".

Security, Durability, Availability.

From my perspective, that's the order of importance too, but any one without the other two and you're not going to gain or retain customers.

On your second point ... Im not sure. Sony is still in business. Equifax and co are still printing money. OPM is still around. Security and trust doesnt seem so binary in practice.

Tbe ahove applies to accessibility too, by the same reasoning.

Often if "performance sucks" then it isn't done and isn't working. Performance is a client requirement like any other.

Yet it is what I sometimes see when this performance discussions take place.

You know, sales is everything and I agree that "there are two types of software: software that people complain about and software that nobody talks about", but some of us take some pride on doing stuff with enough performance and not pure crap...

Myst was at the cutting edge of technology at the time. They had to write some weird custom extensions in order to jam color into HyperCard at all. (Myst was initially written in HyperCard!) One could argue that designing the game around static images that could be rendered offline was itself an optimization. It gave the illusion of an immersive 3D space within the very harsh limitations of early Macs.

Also, the couple of QuickTime videos used in the game were very carefully integrated into a surrounding static image to give the illusion of an entire live scene while only animating a small rectangle of it because that's all most computers could handle at the time.

Myst is a fantastic example of making very thoughtful performance decisions given the constraints at the time.

That they managed to pull it off for home computers in 1993 is nothing short of a performance focused mindset which led them to static images and QuickTime overlays. As soon as they had the ability, they released realMyst which was a realtime 3D version of the original, followed by realMyst Masterpiece in the last 5-10 years.

For that matter, Minecraft may be poorly optimized (it is) but without the effort spent on the chunking system that renders 16x16x(128/256) blocks as a single mesh, it wouldn't run at all. There's plenty of low hanging fruit (which may be more difficult to retrofit into the existing engine, or in the JVM itself) that other games have found and utilized, but without that initial performance optimization Minecraft wouldn't run at all.

Take as much pride as you want. Just remember to, you know, ship something from time to time.

Perfect example.

You're not making games, you're selling entertainment. Focus on the latter first, and optimize accordingly.

I see this time and time again - beautifully architected and performant platforms that completely bomb because noone actually wants it.

I agree with that. And yet, Minecraft is possibly the worst example:

- It's an exception to the general wisdom that games do have to worry about performance in order to deliver fun. Just because Minecraft became popular despite it's abysmal performance, doesn't mean games can get away with that in general.

- Notch never expected Minecraft to become that popular. In a sense, Minecraft was an accident.

- Minecraft's performance was so bad, a lot of people had to install a mod called OptiFine just to be able to play it on their hardware.

That being said, I agree that performance might actually be integral to your product. Quake 3, for example, probably would have bombed if everyone was playing sub-10fps.

Either way, the priority is "figure out who wants your product and why." That helps determine when and where to focus your optimization efforts.

That's not true in the way that I think you're using video games as an example. Video games must be fun, being fun requires being performant enough, and for graphically demanding games that requires lots of optimizing. But there are tons of games that are not graphically demanding.

Game studios were one of the first areas to care for UX.

During the 80's and early 90's many kids got into games by starting as group testers after school.

It wasn't always the same group of kids.

Writing casual games with AAA tech is needless overengeniring.

I don’t know about this. I generally end up building the things that have already been sold.

In other words—don’t go to extremes in any direction.

It sounds very much like this conpany took an extreme perf-first approach from the outset, and advocate a reactionary overcompensated opposite extreme in their postmortem.

Except that's not the argument in the comment you're reacting to. Here's what the comment actually said:

> Anyone with a reasonable amount of experience is often able to look ahead - well before any code is written - and see how performance could be better if road A is taken instead of road B.

In other words, some optimizations come with no significant cost. As a banal, perhaps exaggerated example, if you're writing code that needs to store data in a collection and you know that it's going to perform random access to that data on a regular basis, you're not likely to pick a linked list over some kind of array.

What @abraae is criticizing is the practice of asserting "best practices" that are likely to be taken out of context, widely propagated and misinterpreted.

Knuth's famous assertion "premature optimization is the root of all evil" is an excellent example. People love quoting it, but few seem to take into account the context, to the point that it has been largely forgotten. Here's the quote in its context:

Programmers waste enormous amounts of time thinking about, or worrying about, the speed of noncritical parts of their programs, and these attempts at efficiency actually have a strong negative impact when debugging and maintenance are considered. We should forget about small efficiencies, say about 97% of the time: premature optimization is the root of all evil. Yet we should not pass up our opportunities in that critical 3%.

To be clear, it's not Knuth's fault that people are taking his very rational and sensible advice and reducing it to "yeah, you should totally disregard performance when first writing your code". This is not about pointing fingers and assigning blame, it's about being responsible when offering advice. Knuth's advice was reasonable and nuanced, and it still got taken out of context and reduced to something different and arguably harmful. Now look at Gabriel's advice, which is arguably a lot easier to take out of its context and misinterpret, and imagine the consequences it could have.

That is what @abraae is reacting to.

EDIT: There's another thing that really, really bothers me about your reply:

> Rule #1 of any business: Sales cures all.

This, right here, is what I personally feel is wrong with software nowadays. It's how we end up with poorly written, bloated, bug-ridden, unstable, unreliable, insecure crap that we end up using anyway, because there's no better alternative.

See https://medium.com/@jaredpochtar/on-performance-and-software... and the full section on https://medium.com/@gabriel_20625/technical-lessons-from-bui...

> Most of the code we wrote was fast enough (<2ms) that making it faster wouldn’t be noticeable, so it would definitely be a waste of time to optimize. Making code that’s 0.1% of your runtime 100x faster only makes your latency <1% lower.

Specifically, when you look at that numbered list describing your code-writing practices, the numbers 1 and 3 stand out as hyperbolic. People will* take that stuff out of context and pass it around. Hell, people are already having a hard time staying reasonable in this very discussion thread.

I can't speak for anyone else, but don't take my comments personally. You guys made your own product and ran with it, which is more than I can say for myself, and I respect you for that. I also agree with what you're trying to say. However, I stand by my criticism of the hyperbole, because I've seen people who take that stuff too literally and proceed to write crap.

While premature optimization is definitely a problem, In my experience not handling it correctly before the app is in prod usually results in having to fix logical bugs , refactor for performance and add new features users are clamouring for at the same time.

We don't get lot of chances with users to say you will eventually fix these issues for them, you will simply loose them .

As all things there has to be a balance between both ends

It seems to be within the set of concepts and suggestions that cannot fully sink in and be truly appreciated until some degree of failure or experience relating to them has occurred.

Yeah, they read it. they get it. But they still don't get it, if you know what I mean.

But wow - this one sure does take on some laser etched permanence once it does get through.