Comment by jeswin | Hacker News Reader

Comment by jeswin | Hacker News Reader

Parent Full thread

I looked at this thread a while back and closed the window, hoping that those early comments don't influence the younger practitioners who are on HN.

As someone nearing 40 (mainly doing consulting work around boring business apps for a long time), my view is that Martin Folwer's biggest contribution is indeed the documentation, and defining the vocabulary to enable discussion around frequently encountered (but perhaps mundane) problems in Enterprise Application development. He isn't trying to break new ground, but I'm glad his writings exist. To pick a very random example, here's CQRS defined in an approachable manner[1]. It's a great link to send to a team of 50, out of which quite a few will be fresh out of college. And I am pretty sure Martin Folwer has seen plenty of real world code while he was working at Thoughtworks. The ideas in his writings aren't conceived in some vaccuum.

Some of his early writing isn't perhaps as relevant today. For instance, he talks about the flaws of Anemic Domain Model here [2]. Back in the day, many of us tried to enrich the domain model with particular emphasis on reusability. But those things are less relevant in today's heterogeous environments where logic lives across in multiple services (many outside your team's control). So just reading through his site will surface a lot of outdated ideas; but that's helped a lot of companies along the way when we didn't have to aim for web-scale tps. Actually, it might still work for 99% of apps even today.

HN tilts towards the experiences of Startups and young founders, but one good lesson to learn early is that we should be more open minded.

1: https://martinfowler.com/bliki/CQRS.html 2: https://martinfowler.com/bliki/AnemicDomainModel.html

I don't understand how anemic domain models are irrelevant now especially in a time where a lot of software practitioners seem to observe domain driven design. In the age of microservices, it is very important to determine clear context boundaries. "Logic lives across multiple services" sounds more like a design smell to me. In this case, changing even very simple business requirements could mean changes in multiple parts of the system, which could mean multiple redeployments of many different services. Also implies more complex integration (even e2e) testing.

I'm definitely guilty of using and promoting anemic models. I find that teams have a hard time deciding whether logic is service or domain or whatever and gravitate towards putting everything in one place if domain objects are allowed to be at all smart. On the other hand, people do well with the rule that logic belongs in procedural layers with clear names and some sort of maximum size/complexity.

I also find that these discussions matter less in microservices because the size of each service is so small that you usually don't need more than two layers: one for business logic and one for persistence.

I very much agree that microservices create their own challenges at service interfaces, but since these boundaries are usually expressed as http operations instead of object oriented function calls, I look for different solutions, especially various forms of system-wide introspection and testing.

First, anemic models work well in service contexts. You can't send a business rule from a web page to a web service as part of a data entity. And so such models have a valid place in the world.

Micro services are still services. As such we always validate data crossing a trust boundary - whether monolith or micro service. The rules to do that validation cannot travel with the data.

So guilt because anemic strikes me as... odd. Modern connected apps aren't written as single-tier Smalltalk apps.

You wouldn't send a business rule from a web page, you'd put it in the domain model on the server.

The web page has a UI model, not a domain model.

That's the point. And yet Fowler objects because it violates encapsulation and information hiding, because it requires a service layer and makes a model less expressive.

You can easily serialise a well fleshed out (i.e. non-anemic) object. Does the view layer in this particular example need to know the encapsulated business rule? In my experience, not usually. Otherwise, there's most likely a problem with responsibilities. And when that data comes back from the UI, it's also relatively easy to reconstitute it back to well fleshed out objects.

This I know. The point isn't how one might do it, or whether it can be done¹. The point is that Fowler has issues with anemic objects, even though there are patterns in distrubuted computing best solved by anemic objects.

¹ Note that reconstitution does not do away with anemic objects - reconstitution does not implement logic in a purely object-oriented way, i.e. the logic does not travel, but the data does.

I know one guy who thinks that Fowler is wrong in very large number of points. He works as an architect in big company now (FAANG) and he did some great projects before, so I certainly trust him. For example he thinks that anemic model is the right way to do and putting methods into data objects is wrong. Unfortunately he does not want to write a book yet, so I can't reveal anything, but I just want to underline that a good engineer should not blindly trust anything, even if it seems to be written by a well received author. I, myself, read his Refactoring book and it was great.

Unfortunately I'm not qualified enough to judge those things myself, I just learned that I must doubt about anything, even if it seems to come from big names.

> I just want to underline that a good engineer should not blindly trust anything, even if it seems to be written by a well received author.

Doesn't this also apply to the opinions of your friend who works at a FAANG? How do you judge between theirs and Fowler's view points?

Your point on vocab is so spot on. Simply identifying something as a pattern and naming it is tremendously valuable. Doing it in a way that's packageable and shareable with others is not easy at all.

Agree, but people tend to stop at Fowler and live it as gospel. There is a profound richness in the different ways to think and talk about computers, engineering, and information science, far more wonderous than any one person can catalog. Read SICP, Knuth, Code Complete, Codd. Hell, just YouTube around. If your vocabulary is prescribed by literally one guy in a fedora, you gotta get out there. Might I suggest Joe Armstrong [1] :)

[1] https://youtu.be/lKXe3HUG2l4

> Agree, but people tend to stop at Fowler and live it as gospel.

I know people do that, and it sucks. But I never got that from Fowler himself.

There are other authors (not naming names) who are pretty dogmatic often. But Fowler seems to genuinely try to capture patterns and techniques in an open-minded way.

It's the abuse of his material that is the problem.

Yeah he's no Zed Shaw. But actually I think Fowler is... well he skips a lot of steps and makes a lot of assertions. Here's an excerpt from the intro to Refactoring:

> The performance of software usually depends on just a few parts of the code, and changes anywhere else don't make an appreciable difference.

> But "mostly" isn't "alwaysly." Sometimes a refactoring will have a significant performance implication. Even then, I usually go ahead and do it, because it's much easier to tune the performance of well-factored code. If I introduce a significant performance issue during refactoring, I spend time on performance tuning afterwards. It may be that this leads to reversing some of the refactoring I did earlier--but most of the time, due to the refactoring, I can apply a more effective performance-tuning enhancement instead. I end up with code that's both clearer and faster.

> So my overall advice on performance with refactoring is: Most of the time you should ignore it. If your refactoring introduces performance slow-downs, finish refactoring first and do performance tuning afterwards.

One way to read this is basically what he writes: clear code is more important than efficient code, and clear code can help you make code more efficient if that's important.

Another way--indeed the way I read it--is "performance isn't the most important thing, and at least some of the time following my advice, performance will suffer, even if you undo a lot of the work I'm recommending you do." And I can respect that, but what he writes tries to have it both ways. At least when DHH makes this argument, he basically just says "yeah, buy more machines, they're cheap, engineers aren't." This "incidental" absence of drawbacks is just one of the things that makes all of Fowler's writing feel like marketing.

Another is all the hand waving assertions. There's a lot of it even in those three short paragraphs:

- How did he measure "much easier to tune the performance of well-factored code"? Honestly how would you even measure that?

- How did he measure "more effective performance-tuning enhancement", and also what does "effective" mean?

- What does "well-factored" mean?

- What does "clearer" mean when he says "I end up with code that's both clearer and faster"

Further, he reinvents and renames things. His (short) article about "Application Boundary" [1] is actually Conway's Law [2], which was nearly 40 years old at writing. And OK, no one knows everything. But that piece is written very prescriptively; here's an excerpt:

> We can draw application boundaries in hundred arbitrarily different ways. But it's our nature to group things together and organize groups of people around these groups. There's little science in how this works, and in many ways these boundaries are drawn primarily by human inter-relationships and politics rather than technical and functional considerations. To think about this more clearly I think we have to recognize this uncomfortable fact.

How can you say "there's little science in how this works" immediately after "it's in our nature to..."? And for what it's worth, there's plenty of work around this; look at any discussion of permission/capability systems, consensus systems, sandboxes, etc. There is literally science about it.

I have no doubt that Fowler is earnest and well-meaning, I appreciate his contributions to (what I reluctantly accept is) my field, and I'm happy that he's frequently many people's gateway to thinking more critically about software. But he's far from the final word on it, and too often he writes like he is.

And, at the risk of making an over-long post even longer, there are two problems with that. I've already been explicit about the first: people take it as gospel and don't look further. That's not great for our field.

But even worse, it ends up changing our culture from one of wonder, exploration, and experimentation to one of prescription and certitude. I have no idea what the best way to factor a program is; it's very hard for me (or anyone) to say why one expression of information or a computation is "better" than another, and that is an endless source of delight for me. God help me the day I figure it all out.

[1]: https://martinfowler.com/bliki/ApplicationBoundary.html

[2]: https://en.wikipedia.org/wiki/Conway%27s_law

Are you applying the principle of charity when reading Fowler? The quote about performance seem totally sensible and basically akin to Knuths "We should forget about small efficiencies, say about 97% of the time...". Focus on maintainability and only sacrifice clarity for performance when you know there is a performance problem. What part of this do you honestly disagree with?

You can pick any writing apart by being deliberately obtuse. It doesn't help the conversation. E.g rhetorically asking the meaning of "clear" and "well-factored" code, when basically all of his writing are about exactly that.

I'm sure some people are taking him too much as gospel and don't look further - since this happen to any sufficiently popular writer. Just look at all the people on HN which read PG's essays as gospel and think all problems of architecture and maintainability can be solved by using Lisp - you can't really blame PG for that either.

Yes you will have to apply you own experience and critical thinking and understanding of the problem domain. This is true for everything you read or hear.

> Are you applying the principle of charity...?

Well, when I read that passage, I tried to figure out what he might mean. And mostly I distilled it down to what I wrote: "clarity is more important than performance, and will usually lead to better performance". So even though he's lacking a lot of specifics, sure I think I get his meaning.

> "That's basically exactly what Knuth said"

No Knuth said "ignore small inefficiencies". Fowler said "clarity via refactoring is more important than performance, and will usually lead to better performance". Those are very different.

> Focus on maintainability and only sacrifice clarity for performance when you know there is a performance problem. What part of this do you honestly disagree with?

A good and well known example are entity systems in game engines. If you build such a system using OO principles--polymorphism, encapsulation, messaging--you will have an entity system that doesn't support very many entities on common hardware. It will also have very complex and hard to predict behavior as entities are created and destroyed, and as they all react to different messages, even if you break down and do it all synchronously.

Unless you have a lot of certainty that any game built with that engine will have a (far) below average entity need, you'll end up building the wrong thing, even following best practices, and no amount of refactoring will help you; it's a fundamental design limitation. In fact, you have to design a system that breaks these rules in order to get good performance, and I'm not talking about wonky code here and there; I'm talking about building something that (for example) lets you avoid vtables entirely.

So that's an example of why I disagree with Fowler on refactoring and performance. I think he's basically as wrong as you can be about it. It's a complex subject, and no, it doesn't just fall out of "clear" code, whatever that is.

> You can pick any writing apart by being deliberately obtuse. It doesn't help the conversation. E.g rhetorically asking the meaning of "clear" and "well-factored" code, when basically all of his writing are about exactly that.

I can understand how I probably came across as overly pedantic, but I don't know any other way to pin down all the vagueness. A lot of people find Haskell very clear and idiomatic C unreadable, or a lot of people can read fucked up SQL but can't make heads or tails out of ES6. If I'm gonna burn engineering time on making something "clearer", I'd like some idea of what that is.

And my strong feeling is that it's very situational, contextual, and subjective. What's clear in pixman is confounding in Django; what's elegant in Java is weird as hell in Python. I'd love to read a book that explored these differences and gave me more insight into how to express myself more clearly across them. Refactoring is just not that book at all, and in fact it encourages lots of engineers to apply its concepts everywhere, which of course makes things less idiomatic, less context-aware, and thus, ironically, less clear.

> clarity is more important than performance,

But he does not say that! You are deliberately misrepresenting the meaning of the quote. He says he focus on clarity first and only optimize for performance if absolutely necessary. You can disagree with that, but it is a totally different thing.

If you have legitimate disagreements which his viewpoints you should be able to state your argument without needing to misrepresent what he says. Then we can have a conversation and maybe learn something.

I mean I quoted the entire 3 paragraphs up there, so of course I'm paraphrasing.

Fowler advocates strongly for OOP. Presumably he'd find an OO entity system clearer (if you're gonna quibble with this, imagine how much more productive this conversation would be if we had an understanding of what "clearer" meant, then recoil in horror as you realize thousands of developers are having this exact same unproductive discussion, all because one guy said he had the answers but didn't really give specifics, like Fermat's final theorem but for devs), but the fact remains that if you built a modern entity system this way you'd have to rewrite it to get even average performance out of it. No amount of refactoring will fix the base design.

So I feel like I have disagreed directly with his point, and I'm happy to discuss further. I do, weirdly, feel a little like you're not applying the principle of charity to me though. So let's both endeavor to be a little more understanding.

> I do, weirdly, feel a little like you're not applying the principle of charity to me though.

Fair enough, we can all be guilty of that. What part of your argument have I misunderstood or misrepresented?

I think I've pretty directly disagreed with:

> focus on clarity first and only optimize for performance if absolutely necessary

I think that in a lot of very important, very common use cases (take your pick: game engines, embedded programming, OS development, language runtimes, linear algebra libraries, etc. etc. etc.) this is extremely bad advice. It's... probably good advice for enterprise applications; but that's a pretty small subset of all software. I think that if you start with clarity here and do performance optimization after the fact you will build something that can't be used as intended--almost always (of course, not "alwaysly", see Python).

More broadly, I think this is a far too narrow way of looking at "clarity" vs. "performance". I have some idea of what I think clarity is, and I have some idea of what I think performance is, and they're both pretty nuanced and deep topics. I think they're two big bags of variables in an even larger bag of engineering variables, and I think Fowler does us all a huge disservice when he skips that entire discussion with what is your synopsis up top there. Maybe you're writing Python and you need a certain level of dynamic nature in order to reach your desired level of flexibility and clarity. That can be fine! The point here is that these are rich topics, and they deserve better than what I see as a pretty blithe dismissal.

Agreed. But that's a fault of content consumers not producers, no? I have read SCIP (in college), Code Complete (first book I got in my first job), Gang of Four (2nd book), Clean Code, etc.

Unfortunately we are in an age where blog posts and videos are a lot more accessible than dense books. That sucks, and means we all have to expend extra energy to find a mix of good and diverse ideas and thoughts.

> But that's a fault of content consumers not producers, no?

Yeah, mostly. But at least a little is that Fowler writes and presents pretty authoritatively. And he should, because that's what consultants do, but we should remember that he's a consultant for a very niche area of software ("Enterprise") whenever evaluating his advice.

But then, very often, he also make notes that there is no one good design and, when presenting an ideas, he also presents pros and cons of given solution.

You're completely off-topic. Nobody's mentioning CQRS or anemic models.

This is an article about architecture with zero references to other authors, studies or proof of any kind. It presents several graphs without any supporting evidence.

=> it's absolutely normal to request evidence for those claims. This is software engineering, not software literature, where we try to guess what the author meant and marvel in awe.

It's exactly the younger practitioners on HN which should ignore such articles and demand properly supported texts.

Too much of this industry is driven by fads and personalities. You're exactly correct, of course, but I don't think your point will be well received.

Reply on news.ycombinator.com