Monday, July 01, 2013

Architecture Rituals and Patterns: On the Importance of the Middle Way Once and Only Once Balances Yagni - Part 2

...he heard a master teaching a girl how to play a string instrument. The master told her that if the string was too loose, it would not sound, but if the string was too tight, it would break: the string had to be at just the right tension to produce music and harmony. At that moment, he understood the middle way...

In the previous part of this series, I told you about YAGNI and OAOO and said that OAOO and YAGNI are like Yin and Yang. They balance each other. Without OAOO, YAGNI would slowly strangle us. With OAOO, you always have a system that you can improve (a phrase by Ron Jeffries).

I also "tied the cat," built systems as I saw others build them, was at the Shu level of ShuHaRi when programming enterprise systems, put my DAOs, put my Service, and put my Controller.

All the "well-architected" systems I knew did this, so I did it too. If a well-architected system looks like this in the end, it must be because it looked like this from the beginning... right?

Well, no. Things don't look the same when you're building them as when they're finished. If, for example, you try to build an arch without extra supports (which you will later remove), you could wrongly conclude that if you can't build it "without using supports," you're doing it wrong... That's not the case.

I discovered I was wrong almost by chance, I had already read that OAOO and YAGNI are like Yin and Yang. They balance, but I hadn't understood it beyond the sound of the words in the phrase... When I understood it, I wrote that article about epicycles.

I learned it when I came into contact with PHP programmers...

Speak your truth quietly and clearly; listen to others, even the dull and the ignorant; they too have their story. Desiderata

Those PHP programmers didn't know what YAGNI was... they knew nothing about the principle of building the simplest thing that could work (KISS)... they had never visited c2, and some of them had difficulty reading English...

And yet...

Yet precisely because of that, they followed YAGNI and KISS to a fascinating extreme, they did absolutely nothing that wasn't essential to finish on time the work assigned to them. Yes, their code was almost unbearable to the sight of someone like me at that time...

But what I needed months to build, they did in weeks; what I needed weeks for, they did in days; and what I needed days for, they did in hours...

On the other hand, their intense (and intuitive?) adherence to YAGNI and KISS also had its price, the technical debt they accumulated was brutal, and while they could put functionality in the hands of the user in record time, they also had to redo everything almost from scratch if they were asked for a change, there simply was no reuse...

There was no balance... But for me, it was a shock to look at the other extreme, and the other extreme looked back at me... I could spend hours refactoring code to eliminate redundancy to the maximum, whether the system was ready for the delivery date was a secondary factor... I was all OAOO... they were all YAGNI... I like to think that we learned a lot from each other..

Yagni means that we should only implement what we need now, not what we foresee that we are going to need. This strategy (when it works) eliminates some costs because it does not allow them to come into existence (they were things that we actually never came to need), and it delays the costs of those things that we will eventually need to a point in a future date. However, a commonly used counter-argument is that it is better to implement generic things now than in the future...

If we implement everything incrementally by invoking YAGNI, the system can become increasingly difficult to modify, especially if every time we implement something new, we do it always in the simplest possible way

It is good to do the simplest thing that could work, however, we must remember to reapply that rule to keep the whole system as simple as possible. This means that we will refactor the code so that nothing violates OAOO.

This refactoring, by centralizing any functionality into a single class or method, leaves the system well positioned for the next change. Anything you need to do afterward, you can be almost sure that you can do it better by improving or replacing a specific and isolated point of your code.

Now let's see an example:

  • I have to make a registration screen, which includes

    a <select> element that displays the countries of the world, so that when a person registers on my site, I can present them with content relevant to their country (or countries) of origin. As I'm going to apply YAGNI, I won't do more than this:

  • Controller:
public List<Country> getCountriesList() { logger.info("Getting countries!"); List<Country> countries = getSimpleJdbcTemplate().query( "select id, name from country", new CountryMapper()); return countries; }

Following YAGNI, I don't have to do more. If this ends up being the only screen in my entire system that displays countries, I won't have wasted my time on complexity without benefits.

But then, it turns out that in another part of the system, on the screen to register orders, when I have to put in the address where I want the product I'm purchasing to be delivered, and among the information to indicate in the address, I have to include the country

It is then, and only then, -when I already have 2 conceptually separate places, and for which it is not convenient to use the same controller- that I add a service:

  • Countries Service:
public List<Country> getListOfCountries() { logger.info("Getting countries!"); List<Country> countries = getSimpleJdbcTemplate().query( "select id, name from country", new CountryMapper()); return countries; }
  • Controller: User Registration
public List<Country> getProductsToList() { return getCountryServices().getListOfCountries(); }
  • Controller: Order Address Registration
public List<Country> getProductsToList() { return getCountryServices().getListOfCountries(); }

This is why when a system is finished, it gives the impression that "everything is built with layers from the beginning." But that's a mistake, a product of seeing the system as something static, as something that "was born" with the form it had at a moment in time, and not as something organic, whose structure changes as it adapts to the needs of the user who acquired it.

The architecture of the project is not something that is predefined before understanding the system. The architecture is an emergent structure, derived from the user stories that are gradually built as the system is shaped. If the system does not include among its stories the need to operate on multiple different databases (Oracle, SQL Server, etc.), then the DAO layer should never have to exist. If it doesn't add value, it's superfluous.

In the case of the previous example, we can go even further. Nowadays, most applications are RIAs built on HTML5 and JavaScript, where the logic in Java (or in C#, or in Groovy, or in Scala...) merely acts as a layer of data exchange and security between the persistence layer (typically a database) and the business logic and presentation layer (which are usually located entirely in JavaScript)

So why do we insist on using POJOs that don't add value? Why doesn't our code look like this:

public List<Map<String, ?>> getCountriesList() { logger.info("Getting countries!"); List<Map<String, ?>> countries = getSimpleJdbcTemplate().queryForList("select id, name from country"); return countries; }

Why complicate our existence with one (or several) POJO (or DTO) if our system is so simple (at the Java level) that the methods exposed by our controllers only do CRUD and it is the JavaScript code that really brings the data to life? What's the point of working with POJOs, if our domain model is anemic?

Approaches like OData, sweep away in one fell swoop all this paraphernalia of making controllers to slowly and step by step build a query API that could be automatically derived from the data model. But OData is just a query mechanism, which comes to eliminate the need to keep "tying the cat"... that is, to make a custom layer of controllers each time.

Why don't Oracle, SQLServer, MySQL, PostgreSQL, etc. directly offer an OData API (or something similar) instead of forcing us to reinvent it each time?

It seems that our human traditions make us live in cycles within cycles:

  1. We invented Cobol, to expose our data through procedures.
  2. We invented relational databases (whose implementation we never completed) to allow access to data through query expressions (SQL).
  3. We then invented stored procedures, to procedurally encapsulate our query expressions.
  4. We then invented object-relational mappers, to achieve compatibility between multiple databases and be able to query them with a unified query language.
  5. We then invented SOAP, to procedurally encapsulate our objects.
  6. We discovered that in parallel, someone had invented REST, which allows us to query resources using query expressions represented by a URL.

And so... again and again, procedures, expressions, procedures, expressions, procedural approach, declarative approach... and every time we reinvent the wheel, we feel like we are breaking new ground... when in reality, we are just giving the wheel of fashion another turn.

Will we ever find the balance? (Or perhaps we have already found it... and just need to realize it...)

 Originally published in spanish in Javamexico

 

No comments:

Requirements Analysis: Negative Space

A while ago, I was part of a team working on a crucial project. We were confident, relying heavily on our detailed plans and clear-cut requi...