[edw519]

What I’m about to say will be blindingly obvious to the Enterprise crowd...The rules must be considered as carefully as the entities. Enterprise developers have known this for years: that’s why you see rules engines, table-driven designs, and visual workflow editors in many Enterprise applications.

What a refreshing statement at a time when we're bombarded with so many annoying "SQL is Dead" posts.

Try designing an enterprise production/distribution system where:

  - set-ups must be minimized
  - backlog must be minimized
  - inventory must be minimized
  - trucks must be full
  - warehouse space is limited
  - deliveries must be on time, but not too early
  - sales people must have stock on hand
  - plant absorption must be maximized
  - bills must be paid on time
  - down time must be zero
  - working capital must be put to the best use
  - stockholders must be satisfied

Say what you want about enterprise programmers, but they get stuff built that handles all of these while academicians fuck around with linear algebra and OO castles for years.

After this, Monopoly sounds like a day off. Great post!

[dazzawazza]

Of course many of the concrete methods enterprise developers use come from these 'academicians' so don't be too quick to judge them.

We all stand on their shoulders... it's fine to give em a good kick every so often but remember who is holding us up.

[edw519]

I agree that we all stand on the shoulders of giants, just not the same ones you cite.

As a trained mathematician, I wish I could agree with you and cite all the wonderful applications of classical thinking that I've used to solve real enterprise problems, but alas, I can't. It was only after I abandoned the constant search for "elegant" solutions to real world problems that I was able to embrace good old fashioned "blocking and tackling" to solve the problem at hand.

I've lost track of how many times I was challenged to "use the simplex method" to attack multiple simulaneous mutually exclusive constraints when the most effective approach ended up being, "Put all the data into a relational database and keep spitting it out different ways until our people figure it out." The scientist in me wishes it wasn't true, but the pragmatist knows better.

Sorry to dump on the academic world so much, but I suppose I wouldn't if I could cite more success stories.

The Monopoly example just struck a nerve; a classic example of the difference between theory and practice.

[jerf]

Ah, you just have to do it differently. You can't approach an enterprise problem with "real math"; real math is usually pretty brittle in practice, in the sense that the way the axiomatic systems tend to get set up you can ram the slightest deviation from them through to prove anything, thus breaking the system wide open. Practical programming does need to be more robust than that. Even Haskell requires unsafePerformIO and while end users should not use it, some important libraries certainly do. Even algorithms work tends to be not very useful.

On the other hand, the way in which you think of mathematical entities can be very important and useful, particularly in thinking of things in terms of constraints and preconditions and the resulting guarantees the code can make. Again, not in a flat-out "code proof" sense, but as a way of approaching the code. This has allowed me to write some surprisingly capable code in a multi-team setting by carefully considering the minimal requirements I can impose on the other teams to get a desired result, then making sure I enforce those requirements carefully (since I am in a language where I can't do it with a type system), and then on those basic promises build another layer of the system that is the thing I am actually working on. If I came at this with a hacky hacky chop chop approach it would fail... just as the previous effort did, ahem.

Also, in programming as in math, a thing is what it does, profoundly. This can take a long time to come to grips with. A typical programming entity "is" quite a bit more than your typical math entity, though, for better and for worse.

Preconditions, postconditions, and constraints also make great testing fodder, by which I mean, they make great practical testing fodder. Writing code in practice is a lot easier if you have some assurance that those things are actually true.

I don't necessarily even sit down and sketch this all out, again, because being too rigid causes your rigid assumptions to break, but this stuff is always in the back of my head, and it will get you quite far once you master it....

... and I say that in general. edw519 I wouldn't care to second guess how you're doing things at this point. :)

[joubert]

The relational database model is actually a very good example of mathematical rigor, and comes out of the world of research.

[edw519]

Good point. Sorry I missed that. (Although I have to admit I don't remember that last time I actually took anything to 4th Normal Form. It just seems like too much hammer for most problems at hand.)

[barrkel]

May be completely besides the point, but my first OO instinct upon approaching the question is to model rules etc. as objects themselves. Only the most naive OO modeler would think that OO objects must be nouns within some physical analogue of the system - to my mind, anyhow.

[MartinCron]

I've been working on a board-game program lately and I've created a "referee" object that's responsible for enforcing rules, handling workflow, turn taking, etc. Works really well for me.

[dspeyer]

Why would you want to do that? They have no state. You're almost never going to do anything with them except invoke them from specific places. This sounds like complexity for its own sake.

[barrkel]

First up, there's nothing wrong with immutable objects; indeed, immutable objects have benefits that mutable objects don't. So not having mutable state is not a good argument against modeling them as objects in an object-oriented language.

Secondly, rules do have state, in so far as one rule is distinct from another. Perhaps a rule has preconditions (which could be represented as a binary tree and evaluated through recursion) that must be true before the rule is "active". Perhaps rules have effects, a list of state changes that they impose on the rest of the game world. You could model these as functions, or alternatively again as trees, evaluated through recursion. If the rule system needs to be flexible enough, they could be serializable and deserializable from a text or binary format.

If the game is a simple one, with a fixed set of rules that never change, there's little need for this. But if you look at a game like Monopoly, there's a lot of complexity that comes up in real-life games that isn't so simple to encapsulate in rule lists, such as making complex deals, haggling, borrowing money from other players, even receiving charitable donations from other players to prevent the game ending prematurely.

Move further into business rule systems and the need for flexibility increases again. I've architected systems which needed a generic rule-based system for data validation and workflow transitions; and these rules had themselves rules for when they applied, as older rules needed to be grandfathered in to older data, but expired for new data.

[ghurlman]

Not as refreshing, when you realize this article was written in 2006.

[edw519]

What's changed? Perhaps the technology a little, but as for the real world problems, not nearly as much. Give the frameworks for thinking a little more credit than that.

[ghurlman]

The thinking/mindshare of the NoSql crowd has changed, that's all - your other points are certainly valid then as they are now.