This project aims at providing a set of tools to build complex and evolving business rules. It helps when:
- You decided to split the computation into smaller parts working together.
- You need to keep old versions of the rules working.
- You want to persist the rules leading to a result.
- You want to make your tests independent from your production data.
- You want to introspect the computation of a result.
The idea is very similar to function composition or Rack's middlewares. It is a layering abstraction where each layer works for the next layers in order for the whole to produce a single result.
An engine respond to #call
, taking a context
in argument. It produces a
result that is extracted from the context by the #result
method. Before
doing that, the engine apply its rules.
Each rule have two methods: #trigger?
and #apply
. #apply
runs only when
trigger?
is true. #apply
writes stuff to the context for other rules and
for the engine to produce the result. Rules can be initialized with a
configuration.
A typical usage for this kind of engine is to use it to compute the price of a service or a good. But, this is not limited to that use-case as an engine would be able to produce any kind of results.
Here is an example from the Elephant Carpaccio kata. The specs are:
Accept 3 inputs from the user:
- How many items
- Price per item
- 2-letter state code
Output the total price. Give a discount based on the total price, add state tax based on the state and the discounted price.
This is the smallest example and the best one to refer to. More examples are available:
If you compare this approach with a simple method like this one:
require "bigdecimal"
require "bigdecimal/util"
STATE_TAXES = {
"UT" => "0.0685".to_d,
"NV" => "0.0800".to_d,
"TX" => "0.0625".to_d,
"AL" => "0.0400".to_d,
"CA" => "0.0825".to_d,
}
DISCOUNT_RATES = [
[ 50_000_00, "0.15".to_d ],
[ 10_000_00, "0.10".to_d ],
[ 7_000_00, "0.07".to_d ],
[ 5_000_00, "0.05".to_d ],
[ 1_000_00, "0.03".to_d ],
[ 0, "0.00".to_d ],
]
def pricing(item_count, unit_price, state)
price = item_count * unit_price
discount_rate = DISCOUNT_RATES.find { |limit, _| limit <= price }.last
state_tax = STATE_TAXES.fetch(state)
price * (1 - discount_rate) * (1 + state_tax)
end
... then you'll find significant differences:
- Over-abstraction versus under-abstraction
- 1st example uses the layering abstraction provided by the gem
- 2nd example uses nothing, YAGNI
- Generalization vs specialization
- 1st approach is more generic and could handle changes
- 2nd approach is more specialized and would require a rewrite
- Complexity versus brievety
- 1st example is more verbose and thus is harder to grasp
- 2nd example is more concise and fits in the head
- Configuration versus constants
- 1st example relies on rule's configuration
- 2nd example relies on
STATE_TAXES
andDISCOUNT_RATES
- Observability vs black-box
- 1st example allows to provide more information (
state_tax
and so on) - 2nd example only gives a single result
- 1st example allows to provide more information (
- Data-independent versus hard-coded values
- 1st example considers as much logic as possible as data
- 2nd example mixes data and logic together (with hidden dependencies and assumptions)
- Temporal extensibility or versionning
- 1st example can compute the price using different rules
- Without discounts or with different discount rate per client
- With tax rates from on year or another
- 2nd example will have to introduce options, or even different methods
- 1st example can compute the price using different rules
- Testability
- 1st example could be tested at various levels without any mocks
- 2nd example have to mock hidden dependencies from the implementation
Overall, it is about finding the right level of abtsraction. This tiny framework helps you by providing you a little abstraction. Even if you're not using this gem directly, it can give you some ideas behind it.