dmn-eval-js is a Javascript rule engine to execute decision tables according to the DMN standard. This implementation is based on FEEL by EdgeVerve. It is tailored to evaluation of simple expression language (S-FEEL), plus some cherry-picked parts of FEEL.
dmn-eval-js allows to load and execute DMN decision tables from XML. DRGs are supported. Evaluation of decision tables is currently limited to those of hit policy FIRST (F), UNIQUE (U), RULE ORDER (R), and COLLECT (C) without aggregation.
Use dmn-eval-js to embed a DMN engine in your Javascript application (browser or NodeJS). If you are looking for a DMN engine as a standalone server instead, you may find dmn-server more suitable.
var { decisionTable, dateTime } = require('@hbtgmbh/dmn-eval-js');
dmn-eval-js uses loglevel for logging. By default, warning and error messages are shown only. For additional log output, you need to configure a logger namend 'dmn-eval-js' as follows:
...
var loglevel = require('loglevel');
...
var logger = loglevel.getLogger('dmn-eval-js');
logger.setLevel('info');
With loglevel 'info', you will see a few short log statements per decision call. With log level 'debug', you will additionally see what input was passed to each decision call (including nested decisions in DRGs), this might be verbose depending on what input values you pass for decision evaluation. Use loglevel 'error' to limit log statements to exceptional cases only.
dmn-eval-js parses DMN from XML content. It is up to you to obtain the XML content, e.g. from file system or service call. Parsing is asynchronous using a Promise, while evaluation (execution) of a decision is synchronous.
const { decisionTable } = require('dmn-eval-js');
const xmlContent = ... // wherever it may come from
decisionTable.parseDmnXml(xmlContent)
.then((decisions) => {
// DMN was successfully parsed
const context = {
// your input for decision execution goes in here
};
try {
const data = decisionTable.evaluateDecision('decide-approval', decisions, context);
// data is the output of the decision execution
// it is an array for hit policy COLLECT and RULE ORDER, and an object else
// it is undefined if no rule matched
... // do something with the data
} catch (err) {
// failed to evaluate rule, maybe the context is missing some data?
console.log(err)
};
})
.catch(err => {
// failed to parse DMN XML: either invalid XML or valid XML but invalid DMN
console.log(err)
});
Note than parsing a DMN decision table is a rather expensive operation. It should not be done for each evaluation of a decision table (the example above contains both parsing and evaluation just for the sake of completeness).
The result of the evaluateDecision(...)
call is
undefined
if no rule matched- an array of objects if the hit policy of the decision table is COLLECT or RULE ORDER (one array item for each matching rule)
- an object if the hit policy of the decision table is FIRST or UNIQUE and a rule matched
The object for a matching rule contains the evaluated output value(s) of the rule. The structure is defined by the output names. Qualified names with a dot (.) inside lead to nested objects. See the following example:
An object for a matching rule of the above table would look like this:
{
plainOutputProperty: '...',
output: {
property: '...',
nested: {
property: '...',
},
}
}
Input expressions are commonly (qualified) names, like so:
- customerAge
- customer.age
In the "context" object which is the input for decision execution, the corresponding properties / nested objects must exist. Example:
const context = {
customer: {
age: 18;
}
};
Input expressions are however not restricted to qualified names. You can use any expression according to S-FEEL, and additionally even function invocations, too, like so:
- employee.salary * 12
- convertToUSD(employee.salary)
In the case of functions, you need to define these in the given context object:
const context = {
employee: {
salary: 100000;
},
convertToUSD: (valueInEUR) => [
// your conversion logic here
};
};
Function implementation should be free of side-effects. Date and time instances that are arguments to functions are moment-js instances.
dmn-eval-js supports the following built-in functions from DMN 1.1:
- string functions:
starts with, ends with, contains, upper case, lower case
- boolean functions:
not
- list functions:
list contains, count, min, max, sum, mean, and, or, append, concatenate, insert before, remove, reverse, index of, union, distinct values, flatten
As input entries, simple unary tests according to the DMN specification are supported, with some additions:
- an endpoint can also be arithmetic expression
- a simple value can also be function invocation
- a simple literal can also be a null literal
- a date time literal can also be "date and time"
- brackets in arithmetic expressions are supported
- additional name symbols are not supported
Examples (the list is not complete though):
Input entry | matches if the input expression evaluates to... |
---|---|
42 | the numeric value 42 |
< 42 | a value less than 42 |
[41 .. 50] | a value between 41 and 50 (inclusive) |
10, 20 | either 10 or 20 |
<10, >20 | a value either less than 10 or greater than 20 |
"A" | the string "A" |
"A", "B" | the string "A" or "B" |
true | the boolean value true |
- | any value, even undefined |
any value, even undefined (sams as -) | |
null | the value null or undefined |
not(null) | any value other than null or undefined |
property | the same value as the property (must be given in the context) |
object.property | the same value as the property of the object |
f(a) | the same value as the function evaluated with the property (function and property must be given in the context) |
limit - 10 | the same value as the limit minus 10 |
limit * 2 | the same value as the limit times 2 |
[limit.upper, limit.lower] | a value between the value of two given properties of object limit |
date("2017-05-01") | the date value Mai 1st, 2017 (date is a built-in function) |
date(property) | the date which is defined by the value of the given property, the time if cropped to 00:00:00 |
date and time(property) | the date and time which is defined by the value of the given property (date and time is a built-in function) |
duration(d) | the duration specified by d, an ISO 8601 duration string like P3D for three days (duration is built-in either) |
duration(d) * 2 | twice the duration |
duration(begin, end) | the duration between the specified begin and end date |
date(begin) + duration(d) | the date that results by adding the given duration to the given date |
< date(begin) + duration(d) | any date before the date that results by adding the given duration to the given date |
Most combinations of the syntax elements above are valid, too. For example the following is a valid input entry (although it probably does not make any sense):
not(f(a + 1), [ date(b) + duration(c.d) .. g(d) ])
Sometimes, one whishes to use the value of an input expression as a parameter to a function in an input entry, for example to test that a given input string contains a certain substring, where each substring to test for constitutes a different rule. This could be used for example to derive the project name from the prefix of an issue ID, like so:
The starts with(string, substring)
function allows to test if a string starts with a given prefix,
but with S-FEEL it is not possible to use the value of an input expression as input variable to the function.
Basically, if the input expression is issueId
, and the input entry is starts with(issueId, "CAM")
, then the input entry
will be evaluated to true
if the value of issueId
is for example CAM-42
, but a rule with this input entry
will still not match since true
does not equal the value of the input expression, which is CAM-42
.
dmn-eval-js follows a pragmatic approach to allow for input variable in input entries.
The convention is: if an input expression is a qualified name (in DMN sense), and
an input entry contains a function which takes the same qualified name as one of its parameters,
and the function evaluates to true
, then the rule will match with respect to this
input entry (it may still not match because of non-matching other input entries).
The decision table shown above can therefore be used to derive the project name from the issue prefix.
A simple expression according to the DMN specification is supported as output entry, with the same additions as mentioned for input entries. Since output entries are expressions, not comparisons, values like the following are not allowed:
- < 1
- [1 .. 2]
- not("A")
- empty values (this includes the dash -)
Since version 1.2.0, dmn-eval-js allows function and properties that are referenced by input expressions, input entries, and output entries, to be undefined or missing from the input context. The names of undefined functions and properties are logged with log level 'info'. Undefined values are handled as follows:
Evaluation of input expressions, input entries, and output entries
Input expressions, input entries, and output entries evaluate to undefined if they contain a function or a property which is not found in the input context or is contained there with undefined value. Undefinedness cannot be compared or checked for equality: for undefined values 'a' and 'b', the expression
a = b
evaluated to undefined, not to true.
There is one exception though: there is a built-in function 'defined' which returns true if the given argument is neither null nor undefined, and false if the given argument is null or undefined. If the given argument is undefined, this is not logged to console, since in this case the undefinedness is expected and nothing to inform about.
Matching of rules
If an input expression evaluates to undefined, any rule whose corresponding input entry is not empty (neither the empty string literal nor the dash -) does not match, regardless of to which value (or undefined) the input entry evaluates.
If an input entry evaluates to undefined, the containing rule does not match, regardless of to which value the corresponding input expression evaluates.
Decision result
If an output entry of a matching rule evaluates to undefined, the variable defined by the output name is set to undefined, if the hit policy is UNIQUE or FIRST. If the hit policy is COLLECT or RULE ORDER, the undefined output entry value is not added to the result list.
Custom functions
If you cannot rule out undefined values, your custom functions should check their arguments for undefined values, and return undefined in turn if one or more of the arguments are themselves undefined.
For input expressions, input values, or output values of the following type:
date(property)
date and time(property)
the value of 'property' can be created in the following way:
// for dates only: date string in the format YYYY-MM-DD
const context = {
property: '2018-03-01';
}
// for date and time: ISO8601 date/time string
const context = {
property: '2018-03-01T14:30:00+01:00';
}
// Javascript date - a) from numerical year, month, ... values
const context = {
property: new Date(2018, 2, 1, 0, 0, 0); // note that this will be implicitly in the local time zone!
}
// Javascript date - b) from string with explicit time zone
const context = {
property: new Date('2018-03-01T00:00:00+01:00');
}
// moment-js value
const context = {
property: moment.parseZone('2018-03-01T00:00:00+01:00')
}
Syntactically, any Date or moment-js value is fine, regardless of how you created it.
If your custom functions take date or date and time values, these are of an internal type which is compatible with the moment-js API. Make sure not to modify them by calling moment-js operations with side effects. If needed, clone the arguments first.
If your custom functions return date or date and time values, these must be of the same internal type. The functions required to create them is exported by dmn-eval-js:
const { decisionTable, dateTime } = require('dmn-eval-js');
// creates a date in the internal date format,
// d may be a Javascript Date, a moment-js date,
// or an instance of the internal date format
dateTime.date(d);
// creates a date and time in the internal date format,
// d may be a Javascript Date, a moment-js date,
// or an instance of the internal date and time format
dateTime['date and time'](d);
# install dependencies
npm install
# run test cases
npm test
# watch for changes in source and grammar
gulp watch
# generate parser from grammar
gulp generate
# lint source files
npm run lint
# lint-fix source files
npm run lintfix
For comprehensive set of documentation on DMN, you can refer to :