Module to evaluate JSON-structured expression trees
npm install fig-tree-evaluator!Logo
FigTree Evaluator is a module to evaluate JSON-structured expression trees.
A typical use case would be for evaluating configuration files, where you need to store dynamic values or arbitrary logic without allowing users to inject executable code (perhaps in a .json file, say). Examples could include:
- a form-builder app might need to allow complex conditional logic for form element visibility based on previous responses, or for validation beyond what is available in standard validation libraries.
- configure a decision tree to implement branching logic. (See implementation in 20_match.test.ts)
- extend JSON Forms with more complex logic and dynamic lookups: https://github.com/CarlosNZ/jsonforms-with-figtree-demo
A range of built-in operators are available, from simple logic, arithmetic and string manipulation, to data fetching from local sources or remote APIs. Plus, you can extend functionality with your own custom operators
The demo is powered by fig-tree-editor-react, a React component for editing FigTree expressions.
Fig-tree evaluates expressions structured in a JSON/Javascript object expression tree. A single "node" of the tree consists of an Operator, with associated parameters (or child nodes), each of which can itself be another Operator node -- i.e. a recursive tree structure of arbitrary depth and complexity.
A wide range of operators are available, but custom functions/operators can be added to your implementation if you wish to extend the base functionality.
For example:
``js`
{
operator: "+", // "Addition" operator
values: [1, 2, 3]
}
// -> 6
Or, with a deeper structure that results in the same final output:
`js`
{
operator: '+',
values: [
{
operator: '?', // conditional
condition: {
operator: '=', // equality
values: [
{
operator: 'objectProperties', // extracted from passed-in object
property: 'responses.Q1',
},
'correct',
],
},
valueIfTrue: 1,
valueIfFalse: 0,
},
{
operator: 'GET', // API lookup
url: 'https://my.server.com/api/get-count',
},
3,
],
}
// -> 6
Which would be represented diagrammatically with the following expression tree:
A playground for building and testing expressions is available here
npm install fig-tree-evaluator\yarn add fig-tree-evaluator
or\
`js
import { FigTreeEvaluator } from 'fig-tree-evaluator'
// New evaluator instance
const fig = new FigTreeEvaluator([ options ]) // See available options below
// Evaluate expressions
fig.evaluate(expression, [options]) // Options over-ride initial options for this evaluation
.then((result) => { // "evaluate" is async method
// Do something with result
})
// Or within async function:
const result = await fig.evaluate(expression, [options])
`
FigTreeEvaluator is written in Typescript, and the following types are available to import from the package:
- FigTreeOptions: options object, as per options belowOperator
- : string literal canonical Operator names (AND, OR, EQUAL, etc.)EvaluatorNode
- : Evaluator inputEvaluatorOutput
-
The options parameter is an object with the following available properties (all optional):
- data -- a single object containing any objects in your application that may wish to be inspected using the objectProperties operator. (See playground for examples). If these objects are regularly changing, you'll probably want to pass them into each separate evaluation rather than with the initial constructor.functions
- -- a single object containing any custom functions available for use by custom functions/operators.fragments
- -- commonly-used expressions (with optional parameters) that can be re-used in any other expression. See FragmentshttpClient
- -- pass your http client in here in order to use the HTTP-based operators (GET, POST, GraphQL) (uses browser's native fetch by default). graphQLConnection
- -- a GraphQL connection object, if using the graphQL operator. See operator details below.sqlConnection
- -- if you wish to make calls to an SQL database using the SQL operator, pass a connection to the database here. See operator details below.baseEndpoint
- -- If specified, any partial urls specified in the http-based operators (GET, POST) will be relative to to this base. Useful if you expect most http requests to be to the same server.headers
- -- A general http headers object that will be passed to all http-based operators (GET, POST, GraphQL). Useful for authentication headers, for example. Each operator and instance can have its own headers, though, so see specific operator reference for details.returnErrorAsString
- -- by default the evaluator will throw errors with invalid evaluation expressions (with helpful error messages indicating the node which threw the error and what the problem was). But if you have returnErrorAsString: true set, the evaluator will never throw, but instead return error messages as a valid string output. (See also the fallback parameter below). See Error handling section for more detail.allowJSONStringInput
- -- the evaluator is expecting the input expression to be a javascript object. However, it will also accept JSON strings if this option is set to true. We have to perform additional logic on every evaluation input to determine if a string is a JSON expression or a standard string, so this is skipped by default for performance reasons. However, if you want to send (for example) user input directly to the evaluator without running it through your own JSON.parse(), then enable this option.skipRuntimeTypeCheck
- -- we perform comprehensive type checking at runtime to ensure that each operator only performs its operation on valid inputs. If type checking fails, we throw an error detailing the explicit problem. However, if skipRuntimeTypeCheck is set to true, then all inputs are passed to the operator regardless, and any errors will come from whatever standard javascript errors might be encountered (e.g. trying to pass a primitive value when an array is expected => .map is not a function)caseInsensitive
- -- this only affects the equal/notEqual operators (see there for more detail). nullEqualsUndefined
- -- this only affects the equal/notEqual operators (see there for more detail). evaluateFullObject
- -- by default, FigTree expects the root of an input expression to be an Operator Node, and if not, will return the input unmodified. However, you may have cases where the evaluation expressions are deep within a larger structure (such as a JSON schema, for example). In this case, you can set evaluateFullObject to true and the evaluator will find any operator nodes within the structure and evaluate them within the object tree.excludeOperators
- -- an array of operator names (or aliases) to prohibit from being used in expressions. You may wish to restrict (for example) database access via FigTree configurations, in which case these exclusions can be defined when instantiating the FigTree instance (or updated on the fly).useCache
- -- caches the results from certain operators to avoid repeated network requests with the same input values. By default, this is set to true, and it can be overridden for specific nodes. See Memoization/Caching section for more detailmaxCacheSize
- -- the maximum number of results that will be held in the aforementioned cache (default: 50)maxCacheTime
- -- the maximum time (in seconds) that a result will be cached since last access (default: 1800 (30 minutes))noShorthand
- -- there is a shorthand syntax available for writing expressions. Internally, this is pre-processed into the standard expression form before evaluation. If you have no use for this and you'd rather all expressions were written with full verbosity, set noShorthand: true to save a small amount in performance by skipping internal pre-processing.
As mentioned above, options can be provided as part of the constructor as part of each separate evaluation. You can also change the options permanently for a given evaluator instance with:
fig.updateOptions(options)
You can also retrieve the current options state at any time with:
fig.getOptions()
It's also possible to run one-off evaluations by importing the evaluation method directly rather than using the constructor:
`js
import { evaluateExpression } from 'fig-tree-evaluator'
evaluateExpression(expression, [options]).then((result) => {
// Do something with result
})
`
Each operator has a selections of input properties associated with it, some required, some optional. For example, the conditional operator requires inputs equivalent to the javascript ternary operator, and are expressed as follows:
`js
{
operator: "conditional", // or "?"
condition:
valueIfTrue:
valueIfFalse:
}
`
However, it is also possible to provide the operator properties (or "operands") as a single children array, in which case the specific properties are interpreted positionally.
For example, the following two representations are equivalent conditional operator nodes:
`js
{
operator: "?", // conditional (alias)
condition: 1 + 1 ==== 2,
valueIfTrue: "True output",
valueIfFalse: "False output"
}
// same as:
{
operator: "?",
children: [ 1 + 2 === 2, "True output", "False output" ]
}
`
Most of the time named properties would be preferable; however there are occasional cases where the "children" array might be easier to deal with, or to build up from child nodes.
In each operator node, as well as the operator-specific properties, the following three optional properties can be provided:
- fallback: if the operation throws an error, the fallback value will be returned instead. The fallback property can be provided at any level of the expression tree and bubbled up from where errors are caught to parent nodes. Fallbacks are strongly recommended if there is any chance of an error (e.g. a network "GET" request that doesn't yet have its parameters defined). outputType
See Error handling
- (or type): will convert the result of the current node to the specified outputType. Valid values are string, number, boolean (or bool), and array. You can experiment in the demo app to see the outcome of applying different outputType values to various results.useCache
- : Overrides the global useCache value (from options) for this node only. See Caching/Memoization below for more info.
Remember that all operator node properties can themselves be operator nodes, including the fallback and outputType properties.
e.g.
`js
// Dynamic outputType, which uses the fallback value due to missing property
// for the conditional '?' operator:
{
operator: '+',
values: [9, 10, 11],
outputType: {
operator: '?',
condition: {
operator: '=',
values: ['three', 'four'],
},
valueIfTrue: 'number',
fallback: 'string',
},
}
// => "30"
`
For maximal flexibility, all operator names are case-insensitive, and also come with a selection of "aliases" that can be used instead, based on context or preference (e.g. the conditional operator can also be aliased as ? or ifThen). See specific operator reference for all available aliases.
Similarly, some property names accept aliases -- see individual operators for these.
The full list of available operators and their associated properties:
* denotes "required" properties
Logical AND
Aliases: and, &, &&
#### Properties
- values*: (array) -- any number of elements; will be compared using Javascript && operator
e.g.
`js`
{
operator: '&',
values: [true, true, true],
}
// => true
children array: [...values]
----
Logical OR
Aliases: or, |, ||
#### Properties
- values*: (array) -- any number of elements; will be compared using Javascript || operator
e.g.
`js`
{
operator: 'or',
values: [
true,
{
operator: 'and',
values: [true, false],
},
true,
],
}
// => true
children array: [...values]
----
Equality
Aliases: =, eq, equal, equals
#### Properties
- values*: (array) -- any number of elements; will be compared for strict equality. This includes simple types as well as deep equality of objects and arrays.caseInsensitive
- : (boolean, default false) -- when comparing string values, if this property is true, the case of the strings will be ignored (e.g. "Monday" == "mONdAY")nullEqualsUndefined
- : (boolean, default false) -- there are times when it is convenient for null to be considered equal to undefined. If this is desired, set this property to true, otherwise all equality checks will be "strict" equality. If you find that you want this setting enabled globally, then you can set it in the overall evaluator options instead of having to add this additional property to every equality expression.
e.g.
`js`
{
operator: '=',
values: [3, 3, 'three'],
}
// => false
children array: [...values]
----
Non-equality
Aliases: !=, !, ne, notEqual
#### Properties
- values*: (array) -- any number of elements; will be compared for inequality. This includes simple types as well as deep comparison of objects and arrays.caseInsensitive
- : (boolean, default false) -- as abovenullEqualsUndefined
- : (boolean, default false) -- as above
e.g.
`js`
{
operator: '=',
values: [3, 3, 'three'],
}
// => true
children array: [...values]
----
Addition, concatenation, merging
Aliases: +, add, concat, join, merge
#### Properties
- values*: (array) -- any number of elements. Will be added (numbers), concatenated (strings, arrays) or merged (objects) according their type.type
- : ('string' | 'array') -- if specified, operator will treat the values as though they were this type. E.g. if string, it will concatenate the values, even if they're all numbers. The difference between this property and the common outputType property is that outputType converts the result, whereas this type property converts each element before the "PLUS" operation.
e.g.
`js
{
operator: '+',
values: [4, 5, 6],
}
// => 15
{
operator: '+',
values: ['this', ' and ', 'that'],
}
// => 'this and that'
{
operator: '+',
values: [{one: 1, two: 2}, {three: 3}],
}
// => {one: 1, two: 2, three: 3}
{
operator: '+',
values: [4, 5, 6],
type: 'string'
}
// => "456"
{
operator: '+',
values: [4, 5, 6],
type: 'array'
}
// => [4, 5, 6]
`
children array: [...values]
----
Subtraction
Aliases: -, subtract, minus, takeaway
#### Properties
- values*: (array) -- exactly 2 numerical elements; the second will be subtracted from the first. (If non-numerical elements are provided, the operator will return NaN)
e.g.
`js
{
operator: '-',
values: [10, 8],
}
// => 2
{
operator: 'minus',
values: [0, 3.5, 10], // additional elements after the first two ignored
}
// => -3.5
{
operator: '-',
values: [4, "three"],
}
// => NaN
`
children array: [originalValue, valueToSubtract] (same as values)
----
Multiplication
Aliases: *, x, multiply, times
#### Properties
- values*: (array) -- any number of numerical elements. Returns the product of all elements. (If non-numerical elements are provided, the operator will return NaN)
e.g.
`js
{
operator: '*',
values: [5, 7],
}
// => 35
{
operator: 'x',
values: [2, 3.5, 10], // additional elements after the first two ignored
}
// => 70
{
operator: 'times',
values: [4, "three"],
}
// => NaN
`
children array: [...values]
----
Division
Aliases: /, divide, รท
#### Properties
- values: (array) -- exactly 2 numerical elements; the first will be divided by the second. (If non-numerical elements are provided, the operator will return NaN)dividend
- (or divide): (number) -- the number that will be divideddivisor
- (or by): (number) -- the number to divide dividend byoutput
- ('quotient' | 'remainder') -- by default, the operator returns a floating point value. However, if quotient is specified, it will return the integer part of the result; if remainder is specified, it will return the remainder after division (i.e. value1 % value2)
Note that the input values can be provided as either a values array or dividend/divisor properties. If both are provided, values takes precedence.
e.g.
`js
{
operator: '/',
values: [35, 7],
}
// => 5
{
operator: '/',
divide: 20,
by: 3,
output: 'quotient'
}
// => 6
{
operator: 'divide',
dividend: 20,
divisor: 3,
output: 'remainder'
}
// => 2
`
children array: [dividend, divisor] (same as values)
----
Greater than (or equal to)
Aliases: >, greaterThan, higher, larger
#### Properties
- values*: (array) -- exactly 2 values. Can be any type of value that can be compared with Javascript > operator.strict
- : (boolean, default false) -- if true, value 1 must be strictly greater than value 2 (i.e. >). Otherwise it will be compared with "greater than or equal to" (i.e. >=)
e.g.
`js
{
operator: '>',
values: [10, 8]
}
// => true
{
operator: '>',
values: ["alpha", "beta"]
}
// => false
{
operator: '>',
values: [4, 4],
strict: true
}
// => false
`
children array: [firstValue, secondValue] (same as values)
----
Less than (or equal to)
Aliases: <, lessThan, lower, smaller
#### Properties
- values*: (array) -- exactly 2 values. Can be any type of value that can be compared with Javascript < operator.strict
- : (boolean, default false) -- if true, value 1 must be strictly lower than value 2 (i.e. <). Otherwise it will be compared with "less than or equal to" (i.e. <=)
e.g.
`js
{
operator: '<',
values: [10, 8]
}
// => false
{
operator: '<',
values: ["alpha", "beta"]
}
// => true
{
operator: '<',
values: [4, 4],
strict: false
}
// => true
`
children array: [firstValue, secondValue] (same as values)
----
Count elements in array
Aliases: count, length
#### Properties
- values*: (array) -- any number of elements. Returns array.length
e.g.
`js`
{
operator: 'count',
values: [10, 8, "three", "four"]
}
// => 4
children array: [...values]
----
Return different values depending on a condition expression
Aliases: ?, conditional, ifThen
#### Properties
- condition*: (boolean) -- a boolean value (presumably the result of a child expression)valueIfTrue
- (or ifTrue)*: the value returned if condition is true
- (or ifFalse)*: the value returned if condition is false
e.g.
`js`
{
operator: '?',
condition: {
operator: '=',
values: [
{
operator: '+',
values: [5, 5, 10],
},
20,
],
},
ifTrue: 'YES',
ifFalse: 'NO',
}
// => YES
children array: [condition, valueIfTrue, valueIfFalse]
Note: For more complex branching logic, the "match" operator can be used (it matches more than just a boolean condition)
----
Compares an input string against a regular expression pattern
Aliases: regex, patternMatch, regexp, matchPattern
#### Properties
- testString (or string, value)*: (string) -- the string to be compared against the regex patternpattern
- (or regex, regexp, regExp, re)*: (string) a regex pattern to test testString against
Returns true (match found) or false (no match)
e.g.
`js`
{
operator: 'regex',
string: "home@myplace.com",
pattern: '^[A-Za-z0-9.]+@[A-Za-z0-9]+\\.[A-Za-z0-9.]+$' // Simple Email validation
}
// => true
children array: [testString, pattern]
----
Extracts values from data objects in your application
Aliases: objectProperties, dataProperties,data, getData, objProps, getProperty, getObjProp
#### Properties
- property (or path, propertyName)*: (string) -- the path to the required property in the objectadditionalData
- (or additionalObjects, additional, data): (object) -- any other objects whose properties can be referenced in property (see below)
Data objects are normally expected to be passed in to the evaluator as part of the options, not as part of the expression itself. This is because the source objects are expected to be values internal to your application, whereas the evaluator provides an externally configurable mechanism to extract (and process) application data. (However, it is possible to pass data objects directly as part of the expression using the additionalObjects property, so (in theory) data objects could be dynamically generated from other expressions.)
For example, consider a user object and an fig-tree evaluator instance:
`js
const user = {
firstName: 'Peter',
lastName: 'Parker',
alias: 'Spider-man',
friends: ['Ned', 'MJ', 'Peter 2', 'Peter 3'],
enemies: [
{ name: 'The Vulture', identity: 'Adrian Toomes' },
{ name: 'Green Goblin', identity: 'Norman Osborne' },
],
}
const fig = new FigTreeEvaluator()
const expression = getExpressionFromConfig()
fig.evaluate(expression, { data: { user } })
`
Here is the result of various values of expression:
`js
{
operator: 'objectProperties',
property: 'user.firstName',
}
// => "Peter"
{
operator: 'getData',
path: 'user.friends[1]',
}
// => "MJ"
{
operator: 'getProperty',
path: 'user.enemies.name',
}
// => ["The Vulture", "Green Goblin"]
`
Notice the last example pulls multiple values out of an array of objects, in this case the "name". This is essentially a shorthand for:
`js`
fig.evaluate(
{ operator: 'getProperty', path: 'user.enemies' },
{ data: { user } }
).map((e) => e.name)
The "objectProperties" operator uses object-property-extractor internally, so please see the documentation of that package for more information.
The "objectProperties" operator will throw an error if an invalid path is provided, so it is recommended to provide a fallback value for the expression:
`js`
{
operator: 'objectProperties',
property: 'user.middleName',
fallback: 'Not found!'
}
// => "Not found!"
children array: [property]
Example using "data" passed in dynamically as part of expression:
`js`
{
operator: 'objectProperties',
property: 'user.name',
additionalData: {
operator: '?',
condition: { operator: '=', values: [{ operator: '+', values: [7, 8, 9] }, 25] },
valueIfTrue: { user: { name: 'Bilbo' } },
valueIfFalse: { user: { name: 'Frodo' } },
},
}
// => "Frodo"
----
Replace values in a string using simple parameter (positional or named properties) substitution
Aliases: stringSubstitution, substitute, stringSub, replace
#### Properties
- string*: (string) -- a parameterized (%1, %2, or {{named}}) string, where the parameters are to be replaced by dynamic values. E.g. "My name is %1 (age %2)", or "My name is {{name}} (age {{age}})"
- (or replacements, values)*: (array | object) -- the values to be substituted into string. Will be either an array or object depending on whether you're using positional replacements or named properties (see below).trimWhiteSpace
- (or trimWhitespace, trim): (boolean, default true) -- strips whitespace from the beginning or end of the substitution valuessubstitutionCharacter
- (or subCharacter, subChar): ("%" or "$") -- by default, when using positional replacement, it looks for the % token (i.e %1, %2, etc), but this can be changed to $ (i.e. $1, $2, $3, etc) by setting this property to $.numberMapping
- (or numMap, numberMap, pluralisation, pluralization, plurals): (object) -- when replacing with named properties and you have replacement values that are numbers, it's possible to map values or ranges to specific string outputs. This can be used to produce correct pluralisation, for example. See below for more details.
Substitution can be done using either positional replacement, or with named properties:
#### Positional replacement
The values in the substitutions array are replaced in the original string by matching their order to the numerical order of the parameters.
e.g.
`js
{
operator: 'stringSubstitution',
string: 'My name is %1 (age %2)',
substitutions: [
'Steve Rogers',
{
operator: '-',
values: [2023, 1918],
},
],
}
// => "My name is Steve Rogers (age 106)"
{
operator: 'replace',
// Using $1, $2 instead of %1, %2 this time:
string: '$1 is actually $2 $3',
substitutionCharacter: "$",
substitutions: [
// Using the 'user' object from above (OBJECT_PROPERTIES operator)
{
operator: 'objectProperties',
property: 'user.alias',
},
{
operator: 'objectProperties',
property: 'user.firstName',
},
{
operator: 'objectProperties',
property: 'user.lastName',
},
],
}
// => "Spider-man is actually Peter Parker"
// Parameters can be repeated:
{
operator: 'stringSubstitution',
string: 'A %1 says: "%2 %2 %2"',
substitutions: ['bird', 'Tweet!'],
}
// => 'A bird says: "Tweet! Tweet! Tweet!"'
// Replacement tokens can be escaped using the standard "\" escape character:
{
operator: 'stringSubstitution',
string: 'The price of $1 is \$5',
subChar: '$',
substitutions: [ 'a coffee', 'not used' ],
}
// => The price of a coffee is $5
`
children array: [string, ...substitutions] trimWhiteSpace
( and substitutionCharacter not available, since substitutions can be an arbitrary number of items)
e.g.
`js`
{
operator: 'replace',
children: ['I am %1 %2', 'Iron', 'Man'],
}
// => "I am Iron Man"
#### Named property replacement
Replacement tokens can be indicated in the main string with a named value, using {{ syntax, e.g. "Your name is {{firstName}} {{lastName}} and your best friend is {{friends[0]}}". Then the substitutions property should be an object with those property names:
`js`
{
operator: 'stringSubstitution',
string: 'Your name is {{firstName}} {{lastName}} and your best friend is {{friends[0]}}',
substitutions: {
firstName: 'Steve',
lastName: "Rogers"
friends: [ "Bucky Barnes", "Peggy Carter" ]
}
}
// => "Your name is Steve Rogers and your best friend is Bucky Barnes"
Note the use of as per objectProperties.
Additionally, the substitutions can actually be provided directly in the associated data object and the operator will search for them there if not found in the substitutions property. This could be achieved simply by nesting a getData node inside the substitutions property, but because this is a very common scenario (the values provided will normally be dynamic based on application state), this shorthand is provided as a convenience.
`js
// With "data" object:
{
info: { where: "Spain", what: "plain" }
}
// These two expressions are equivalent:
{
operator: 'stringSubstitution',
string: 'The rain in {{where}} falls mainly on the {{what}}',
substitutions: {
where: { operator: "getData", property: "info.where" },
what: { operator: "getData", property: "info.where" }
}
}
// or
{
operator: 'stringSubstitution',
string: 'The rain in {{info.where}} falls mainly on the {{info.what}}',
}
// => "The rain in Spain falls mainly on the plain"
`
#### Number mapping
If the replacement values are numbers, we can extend this functionality with a special numberMapping object, which allows for different replacements depending on the value, which is handy for pluralisation, for example.
The syntax for the numberMapping property is:`js`
{
propertyName1: {
1: "Output if value is {}",
2: "Output if value is 2",
">5": "Output if value is greater than 5",
"<0": "Output if value is less than 0"
"other": "Fallback output if none of the others match: {} count"
// {} is a replacement for the numerical value itself
},
propertyName2: { ...etc }
}
The number map can have as few or as many match options as desired -- if no match is found (or if no property at all), the number will be returned as-is.
e.g.
`js
{
operator: 'stringSubstitution',
string: 'Hi {{name}}, we have {{count}} attending this event.',
values: {
name: "Tatiana",
count: {operator: "getData", property: "numOfPeople" }
},
numberMap: {
count: {
0: "no one",
1: "just one person",
">10": "too many people",
"other": "{} people"
}
}
}
// Output with varying values for "numOfPeople" passed into evaluation
// "data" object:
// { numOfPeople: 5 }
// => "Hi Tatiana, we have 5 people attending this event."
// { numOfPeople: 0 }
// => "Hi Tatiana, we have no one attending this event."
// { numOfPeople: 100 }
// => "Hi Tatiana, we have too many people attending this event."
// { numOfPeople: 1 }
// => "Hi Tatiana, we have just one person attending this event.
`
Note: children array not available for named properties
----
Split strings into arrays
Aliases: split, arraySplit
#### Properties
- value (or string)*: (string) -- string to be splitdelimiter
- (or separator): (string) -- substring to split value on (Default: " " (space)) trimWhiteSpace
- (or trimWhitespace, trim): (boolean, default true) -- strips whitespace from the beginning or end of resulting substrings excludeTrailing
- (or removeTrailing, excludeTrailingDelimiter): (boolean, default true) -- if false, if the input string ends with the delimiter, the last member of the output array will be an empty string. this, that, another,
i.e. (delimiter ",") => ["this", "that", "another", ""]
The last two parameters (timeWhiteSpace and excludeTrailing) should rarely be needed to be changed from their default values.
e.g.
`js
{
operator: 'split',
children: ['Alpha, Beta, Gamma, Delta', ','],
}
// => ['Alpha', 'Beta', 'Gamma', 'Delta']
`
children array: [value, delimiter] trimWhiteSpace
( and excludeTrailing not available, since array can only support one optional parameter)
---
The following three operators (GET, POST, GraphQL) make http requests, so require an http client. If using fig-tree in the browser, it will use the native fetch() method by default, so no configuration is required. However, if using in node, or you wish to use a different http client (if your project is already using axios, say), you can specify it with the httpClient option.
The httpClient object is an abstraction around an http package in order to standardise the implementation for use in fig-tree. Two such "wrappers" are provided in the FigTree package, for:
- axios
- node-fetch (same API as browser fetch, but for node)
To specify one of these for use, just pass the client directly to the httpClient option, like so:
#### Axios
`js
import axios from 'axios'
import { FigTreeEvaluator, AxiosClient } from 'fig-tree-evaluator'
const fig = new FigTreeEvaluator({
httpClient: AxiosClient(axios),
...otherOptions
})
`
#### Node-Fetch
`js
import fetch from 'node-fetch'
import { FigTreeEvaluator, FetchClient } from 'fig-tree-evaluator'
const fig = new FigTreeEvaluator({
httpClient: FetchClient(fetch),
...otherOptions
})
`
If you wish to use a client other than these two, you must provide an function that takes the client as a parameter and returns an object of the following type structure:
`ts
interface HttpClient {
get: (req: Omit
post: (req: Omit
throwError: (err: unknown) => void
}
// where HttpRequest is an input of the following shape:
interface HttpRequest {
url: string
params?: { [key: string]: string }
data?: Record
headers?: Record
method?: 'get' | 'post'
}
`
And then in FigTree, for example:
`js
import { MyHttpWrapper } from './customWrappers'
import { someClient } from 'some-library'
const fig = new FigTreeEvaluator({
httpClient: MyHttpWrapper(someClient),
...otherOptions
})
`
See the implementation for axios and node-fetch in the repo for specific details.
Http GET request
Aliases: get, api
Note: if used in node, or you're using an http client other than fetch, you will need to explicitly provide an httpClient option. See details
#### Properties
- url (or endpoint)*: (string) -- url to be queriedparameters
- (or queryParams, queryParameters, urlQueries): (object) -- key-value pairs for any query parameters for the requestheaders
- : (object) -- any additional headers (such as authentication) required for the requestreturnProperty
- (or outputProperty): (string) -- an object path for which property to extract from the returned data. E.g. if the API returns {name: {first: "Bruce", last: "Banner"}, age: 35} and you specify returnProperty: "name.first, the operator will return "Bruce" (Uses the same logic as the objectProperties internally)
As mentioned in the options reference above, a baseEndpoint string and headers object can be provided in the constructor. These are applied to all subsequent requests to save having to specify them in every evaluation. (Additional/override headers can always be added to a specific evaluation, too.)
e.g.
`js
{
operator: 'GET',
url: 'https://restcountries.com/v3.1/name/zealand',
returnProperty: 'name.common',
outputType: 'string' // This extracts the string from the returned array value
}
// => "New Zealand"
{
operator: 'get',
endpoint: {
operator: '+',
values: ['https://restcountries.com/v3.1/name/', 'india'],
},
parameters: { fullText: true },
outputProperty: '[0].name.nativeName.hin',
}
// => { "official": "เคญเคพเคฐเคค เคเคฃเคฐเคพเคเฅเคฏ", "common": "เคญเคพเคฐเคค" }
`
children array: [urlObject, parameterKeys, ...values, returnProperty]
- urlObject: either a url string, or an object structured as {url:
e.g.
`js`
{
operator: 'get',
children: [
'https://restcountries.com/v3.1/name/cuba', // url
['fullText', 'fields'], // parameterKeys
'true', // parameter value 1
'name,capital,flag', // parameter value 2
'flag', // returnProperty
],
outputType: 'string',
}
// => "๐จ๐บ"
----
Http POST request
Aliases: post
The "POST" operator is basically structurally the same as GET.
Note: if used in node, or you're using an http client other than fetch, you will need to explicitly provide an httpClient option. See details
#### Properties
- url/endpoint*, headers, returnProperty/outputProperty -- same as "GET" operator.parameters
- (or bodyJson, data) -- passed to the Post request as body JSON rather than url query parameters (hence the different aliases)
e.g.
`js
{
operator: "post",
endpoint: "https://jsonplaceholder.typicode.com/posts",
parameters: {
title: "New Blog Post",
body: "Just a short note...",
userId: 2
},
returnProperty: "id"
}
// => 101
`
children array: [urlObject, parameterKeys, ...values, returnProperty] (same as "GET")
----
Http GraphQL request (using POST)
Aliases: graphQL, graphQl, graphql, gql
This operator is essentially a special case of the "POST" operator, but structured specifically for GraphQL requests.
Note: if used in node, or you're using an http client other than fetch, you will need to explicitly provide an httpClient option. See details
#### Properties
- query*: (string) -- the GraphQL query stringvariables
- : (object) -- key-value pairs for any variables used in the query
- (or endpoint): (string) -- url to be queried (Only required if querying a different url to that specified in the GraphQLConnection object in fig-tree options)headers
- : (object) -- any additional headers (such as authentication) required for the requestreturnNode
- (or returnProperty, outputProperty): (string) -- an object path for which property to extract from the returned data (same as "GET" and "POST").
As mentioned in the options reference above, a headers object can be provided in the constructor. These are applied to all subsequent requests to save having to specify them in every evaluation, although additional/override headers can always be added to a specific evaluation, too.
Often, GraphQL queries will be to a single endpoint and only the query/variables will differ. In that case, it is recommended to pass a GraphQL connection object into the FigTreeEvaluator constructor options.
The required connection object is:
`ts`
{
endpoint: string // url
headers?: { [key: string]: string } // key-value pairs
}
The following example expression uses the GraphQL connection (specified in constructor options): {endpoint: 'https://countries.trevorblades.com/'}jsquery getCountry($code: String!) {
{
operator: 'graphQL',
query:
countries(filter: {code: {eq: $code}}) {
name
emoji
}
},
variables: { code: 'NZ' },
returnNode: 'countries[0]',
}
// => { "name": "New Zealand", "emoji": "๐ณ๐ฟ" }
`
children array: [query, endpoint, variableKeys, ...variableValues, returnNode]
- query: the GraphQL query (string)endpoint
- : url string; to use the endpoint provided in the GraphQL connection options, pass empty string "" herevariableKeys
- : an array of strings representing the keys the GraphQL variables object...variableValues
- : one value for each key specified in variableKeys
- (optional): the return property, as per "GET" and "POST" operators
e.g.
`jsquery getCountry($code: String!) {
{
operator: 'GraphQL',
children: [
countries(filter: {code: {eq: $code}}) {
name
emoji
}
},`
"", // default endpoint
['code'], // variable keys
'NZ', // variable value
'countries.emoji', // return node
],
type: 'string',
}
// => "๐จ๐บ"
----
Query an SQL database
Aliases: sql, pgSql, postgres, pg, sqlLite, sqlite, mySql
#### Properties
- query*: (string) -- SQL query string, with parameterised replacements (i.e. $1, $2, etc)values
- (or replacements): (array / object) -- replacements for the query parameterssingle
- (or singleRecord): (boolean) -- by default, results are returned as an array of objects. However, if your query is expected to just return a single record, you can set single: true and just the record object will be returned (i.e. not in an array). Note that if the query does fetch multiple records, only the first will be returned.flatten
- (or flat): (boolean) -- Instead of returning an object, flatten: true will just return an array of values. e.g, instead of {name: "Tom", age: 49}, it will return ["Tom", 49]. This would usually be used in conjunction with the single property -- if not, it will return an array of flattened arrays.
#### Examples
The following additional examples query a default installation of the Northwind demo database.
`js
{
operator: 'sql',
query: "SELECT contact_name FROM customers where customer_id = 'FAMIA';",
single: true,
flatten: true
}
// => "Aria Cruz"
{
operator: 'sql',
query: SELECT product_name FROM public.products
WHERE category_id = $1 AND supplier_id != $2,
values: [1, 16],
flatten: true,
}
// => ["Chai","Chang","Guaranรก Fantรกstica","Cรดte de Blaye","Chartreuse verte",
// "Ipoh Coffee","Outback Lager","Rhรถnbrรคu Klosterbier","Lakkalikรถรถri"]
{
operator: 'sql',
query: 'SELECT COUNT(*) FROM employees',
flatten: true,
type: 'number',
}
// => 9
`
children array: [queryString, ...substitutions]
(single and flatten can not be part of children array)
#### Connecting to the database
In order to query the SQL database, fig-tree must be provided with a database connection object in its sqlConnection option. An SQLConnection is an abstraction around a specific database connection in order to standardise the implementation for use in fig-tree. Two such "wrappers" are provided in the FigTree package for the following database connections:
- PostgreSQL using node-postgres: SQLNodePostgres wrappersqlite
- SQLite using/sqlite3: SQLite wrapper
You will need to have the appropriate package installed separately, and they can be implemented as follows:
##### PostgreSQL
`js
import { Client } from 'pg' // node-postgres
import { FigTreeEvaluator, SQLNodePostgres } from 'fig-tree-evaluator'
const pgConfig = {
// database config, see node-postgres documentation
user: 'postgres',
host: 'localhost',
database: 'northwind',
port: 5432,
...etc
}
const pgConnect = new Client(pgConfig)
pgConnect.connect()
const fig = new FigTreeEvaluator({
sqlConnection: SQLNodePostgres(pgConnect),
...otherOptions
})
fig.evaluate({
operator: "SQL",
query: "SELECT contact_name FROM customers where customer_id = 'FAMIA';",
single: true,
flatten: true
})
.then((result) => console.log(result)) // => "Aria Cruz"
`
##### SQLite
`js
import sqlite3 from 'sqlite3'
import { open, Database } from 'sqlite'
import { FigTreeEvaluator, SQLite } from 'fig-tree-evaluator'
open({
filename: '/path/to/sqlite.db',
driver: sqlite3.Database
}).then((db) => {
const fig = new FigTreeEvaluator({
sqlConnection: SQLite(db),
...otherOptions
})
fig.evaluate(...) // Continue app operations
})
`
To create additional abstractions for other database connections, you need to provide a function that takes the library's connection object as a parameter and returns an object with a .query() method with the following type structure:
`ts
interface SQLConnection {
query: (input: QueryInput) => Promise
}
// where QueryInput is:
interface QueryInput {
query: string
values?: (string | number | boolean)[] | object
single?: boolean
flatten?: boolean
}
// and QueryOutput is any FigTree output`
You then implement in FigTree options the same way as the two described above.
Check out SQLNodePostgres and SQLite in the repo for specific details.
----
Return an object constructed by separate keys and values
Aliases: buildObject, build, object
The "buildObject" operator would primarily be used to construct an object input for another operator property (e.g. variables on "GraphQL") out of elements that are themselves evaluator expressions.
#### Properties
- properties (or values, keyValPairs, keyValuePairs)*: (array) -- array of either:`
- objects of the following shape:
ts`
{
key: string
value: any
}
- key/value pairs in sequence, e.g.
Each object or pair of elements provides one key-value pair in the output object
e.g.
`js
{
operator: 'buildObject',
properties: [
{ key: 'one', value: 1 },
{ key: 'two', value: 2 },
{
// Using "user" object from earlier
key: { operator: 'objectProperties', property: 'user.friends[0]' },
value: { operator: '+', values: [7, 8, 9] },
},
],
}
// => { one: 1, two: 2, Ned: 24 }
// OR, this is equivalent...
{
operator: 'buildObject',
properties: [
"one", 1, "two", 2,
{ operator: 'objectProperties', property: 'user.friends[0]' },
{ operator: '+', values: [7, 8, 9] }
]
}
// => { one: 1, two: 2, Ned: 24 }
`
children: [...properties] (same as properties array above)
----
Return different values depending on a matching expression
Aliases: match, switch
The "match" operator is equivalent to a "switch"/"case" in Javascript. It is similar to the "conditional" operator, but can handle matching to any number of values, not just true/false. This provides a way to construct elaborate decision trees.
#### Properties
- matchExpression (or matchValue)*: (string | number | boolean) -- a node that returns a value to be compared against possible cases.branches
- (or arms or cases): (object) -- an object whose keys are compared against the matchExpression. The value of the matching key is returned....branches
- -- as an alternative to the branches object, matching key/values can be placed at the root of the node (see example)
e.g.
`js
// Simple decision tree
{
operator: 'match',
matchExpression: {
operator: 'objectProperties',
property: 'weather',
},
branches: {
sunny: {
operator: 'match',
match: {
operator: 'objectProperties',
property: 'humidity',
},
cases: { high: 'NO', normal: 'YES' },
},
cloudy: 'YES',
},
rainy: {
operator: 'match',
match: {
operator: 'objectProperties',
property: 'wind',
},
branches: { strong: 'NO', weak: 'YES' },
},
}
// With:
// data = { weather: "sunny", humidity: "high" } => "NO"
// data = { weather: "rainy", wind: "weak" } => "YES"
`
This expression could also be written as (with branch/case keys at the root level)"
`js`
{
operator: 'match',
matchExpression: {
operator: 'objectProperties',
property: 'weather',
},
sunny: {
operator: 'match',
match: {
operator: 'objectProperties',
property: 'humidity',
},
high: 'NO',
normal: 'YES',
},
cloudy: 'YES',
rainy: {
operator: 'match',
match: {
operator: 'objectProperties',
property: 'wind',
},
strong: 'NO',
weak: 'YES',
},
}
children array: [matchExpression, key1, value1, key2, value2, ...]
The pairs of key/values are constructed into the branches object, the same way the objects are built using children in the "buildObject" operator.
For an example of a complex decision tree implementation, which includes aliases, fallbacks and a range of operators, see the "match" test case file (20_match.test.ts).
----
Pass-thru (does nothing)
Aliases: passThru, _, pass, ignore, coerce, convert
This operator simply returns its input. Its purpose is to allow an additional type conversion (using outputType) before passing up to a parent node.
#### Properties
- value (or _, data)*: (any) -- the value that is returned
e.g.
`js`
{
operator: 'pass',
value: { operator: '+', values: [50, 0], type: 'string' },
outputType: 'array',
}
// => ["500"]
----
There is one final operator, CUSTOM_FUNCTIONS, which opens the door to extending FigTree's functionality. You can either call this operator directly, or make your functions "custom operators" in their own right.
But first you need to define your functions in FigTree's options, for example:
`js`
const fig = new FigTreeEvaluator({
functions: {
double: (x) => x * 2,
getCurrentYear: () => new Date().toLocaleString('en', { year: 'numeric' }),
changeCase: ({string, toCase}) => toCase === "upper" ?
string.toUpperCase() : string.toLowerCase()
average: (...numbers) => (numbers.reduce(
(a, n) => a + n, 0)) / numbers.length,
},
})
You can also define functions with extended metadata -- see Metadata for more on that.
Aliases: customFunctions, customFunction, functions, function, runFunction
#### Properties
- functionPath (or functionsPath, functionName, funcPath*): (string) -- name of the function in the options.functions objectargs
- (or arguments, variables): (array | any) -- input arguments for the function. If an array, will be passed in as multiple arguments.input
- : Input argument if function only takes a single argument. Note that, if your function takes a single array as its argument, you'll need to pass it in the input property -- the args property will spread the array into multiple arguments.
Here is the result of various expressions:
`js
{
operator: 'functions',
functionPath: 'double',
input: 50,
}
// => 100
{
operator: 'stringSubstitution',
string: "The average age of our students is %1"
replacements: [
{
operator: 'function',
functionPath: 'average',
args: [18, 35, 27, 55, 17, 28],
}
]
}
// => "The average age of our students is 30"
{
operator: '+',
values: [
{
operator: 'customFunctions',
functionPath: 'changeCase',
args: ['The current year is: ', 'upper'],
},
{
operator: 'customFunctions',
functionPath: 'getCurrentYear',
},
],
}
// => "THE CURRENT YEAR IS: 2023"
`
children array: [functionPath, ...args]
Your expressions can actually refer to functions directly in the operator property, effectively allowing you to create any number of custom operators.
Converting the above examples to this format:
`js
{
operator: 'double',
input: 50,
}
// => 100
{
operator: 'stringSubstitution',
string: "The average age of our students is %1"
replacements: [
{
operator: 'average',
args: [18, 35, 27, 55, 17, 28],
}
]
}
// => "The average age of our students is 30"
{
operator: '+',
values: [
{
// Notice the properties of a single input object can be
// passed as properties on the main Operator node
operator: 'changeCase',
string: 'The current year is: ',
toCase: 'upper'
},
{ operator: 'getCurrentYear' },
],
}
// => "THE CURRENT YEAR IS: 2023"
`
These custom operators even support the shorthand syntax -- they should behave just like standard operators.
If you have a node that is used more than once in a complex expression, it's possible to just evaluate the repeated node once, and refer to it throughout using an "alias" reference. This allows for a simpler expression (reduces code duplication) as well as a performance improvement, since the aliased node is only evaluated once, providing a simple memoization mechanism. (See also Caching/Memoization)
For example, if you have the expression:
`js`
{
operator: "?",
condition: {
operator: "!=",
values: [
{
operator: "GET",
children: [
"https://restcountries.com/v3.1/name/zealand",
[],
"name.common"
],
type: "string"
},
null
]
},
valueIfTrue: {
operator: "GET",
children: [
"https://restcountries.com/v3.1/name/zealand",
[],
"name.common"
],
type: "string"
},
valueIfFalse: "Not New Zealand"
}
The GET operation is used twice -- once to compare it for non-equality with null, and once again to return its value if true. This is particularly wasteful since it is a network request.
We can create an alias for this whole node, resulting in this equivalent expression:
`js`
{
$getCountry: {
operator: "GET",
children: [
"https://restcountries.com/v3.1/name/zealand",
[],
"name.common"
],
type: "string"
},
operator: "?",
condition: {
operator: "!=",
values: [
"$getCountry",
null
]
},
valueIfTrue: "$getCountry",
valueIfFalse: "Not New Zealand"
}
Alias nodes are defined as part of the expression object, using $ prefix to identify them as such, in this case $getCountry, which returns the result of the network request that was called twice in the previous expression.
Alias nodes are evaluated first, then the results are substituted in whenever they are referenced elsewhere.
Like all expression nodes, alias nodes can themselves contain complex expressions with their own alias nodes defined within. As long as the alias references are descendent nodes of the alias definition, they will be resolved.
(If an alias reference does not having a matching definition, the reference value will just be returned as a literal string, e.g. "$getCountry")
You may find that the expressions you are building for your configuration files often use very similar sub-expressions (but perhaps with only some input values that differ). For example, your expressions might be regularly looking up a "countries" database and fetching the capital city, such as:
`js`
{
operator: 'GET',
url: {
operator: 'stringSubstitution',
string: 'https://restcountries.com/v3.1/name/%1',
replacements: ['New Zealand'],
},
returnProperty: '[0].capital',
outputType: 'string',
}
// => "Wellington"
In this case, you can pre-define the main part of the expression as part of an expression "fragment" in the evaluator options.
The idea is that you can "hard-code" some common expressions into your app, so that external configuration expressions can be simpler when using these common elements.
The syntax is similar to Alias Nodes, in that any string values prefixed with $ will be treated as "parameters" that will be replaced during evaluation. In the above example, you would define the fragment when instantiating the evaluator like so:
`js`
const fig = new FigTreeEvaluator({
fragments: {
getCapital: {
operator: 'GET',
url: {
operator: 'stringSubstitution',
string: 'https://restcountries.com/v3.1/name/%1',
replacements: [ "$country" ],
},
returnProperty: '[0].capital',
outputType: 'string',
metadata: {
// Not required, but useful for external consumers -- see Metadata below
description: "Fetches the capital city of a country",
parameters: { $country: { type: 'string', required: true } },
}
},
},
})
Then any subsequent expressions can use this fragment by specifying a special "Fragment Node", which contains the fragment and (optionally) parameters fields:`js`
{
fragment: "getCapital",
parameters: {
$country: { operator: "getData", property: "path.to.country.name" }
}
}
Like the branches field in the "Match" operator, the properties of the parameters field can be specified at the root level as well -- it just depends on whichever is most appropriate for your use case. So the following is equivalent to the previous fragment node:`js`
{
fragment: "getCapital",
$country: { operator: "getData", property: "path.to.country.name" }
}
See 22_fragments.test.ts for more complex examples.
Unlike Alias Nodes, which are evaluated once and then the result re-used whenever the alias is referenced, Fragments are evaluated every time, as the input parameters may differ.
It's possible to express FigTree expressions in a more compact syntax, as follows:
- Fragment and Operator nodes can be represented by putting the name of the fragment or operator as a property name (prefixed by $), then putting the parameters in an object as the property value. For example: `
js`
{
operator: 'regex',
string: "home@myplace.com",
pattern: '^[A-Za-z0-9.]+@[A-Za-z0-9]+\\.[A-Za-z0-9.]+$'
}
can be written as:
js`
{
$regex:
{
string: 'home@myplace.com',
pattern: '^[A-Za-z0-9.]+@[A-Za-z0-9]+\\.[A-Za-z0-9.]+$'
},
}
- For operator nodes (not fragments, as they always require named parameters), parameters can be represented positionally (equivalent to the children property in normal operator nodes) in an array. The above example could also be expressed as: `
js`
{
$regex: [ 'home@myplace.com', '^[A-Za-z0-9.]+@[A-Za-z0-9]+\\.[A-Za-z0-9.]+$' ]
}
- Operator nodes with a single parameter can just be placed directly as the single property value. For example:
`js`
{
operator: "getData",
property: "user.firstName"
}
can become, simply:
js`
{ $getData: "user.firstName" }
For more examples, see 23_shorthand.test.ts, or have a play with the demo app app.
FigTree Evaluator has basic memoization functionality for certain nodes to speed up re-evaluation when the input parameters haven't changed. There is a single cache store per FigTree instance which persists for the lifetime of the instance. By default, it remembers the last 50 results with each result being valid for half and hour, but these values can be modified using the maxCacheSize and maxCacheTime options.
Currently, caching is only implemented for the following operators, since they perform requests to external resources, which are inherently slow:
- GET (useCache default: true)useCache
- POST ( default: false) useCache
- PG_SQL ( default: true)useCache
- GRAPH_QL ( default: true)useCache
- CUSTOM_FUNCTIONS ( default: false)
This is different to the memoization provided by Alias Nodes:
- Alias nodes are still evaluated once for every evaluation -- they're more for re-use within a complex expression.
- Cached nodes will persist between different evaluations as long as the input values are the same as a previously evaluated node.
Caching is enabled by default for most of the above operators, but this can be overridden by setting useCache: false in options, either globally or per expression. If you're querying a database or API that is likely to have a different result for the same request (i.e. data has changed), then you probably want to turn the cache off.
It's possible to manually set/get the cache store, which can be used (for example) to save the cache in local storage to persist it between page reloads or new FigTree instances.
Use the methods:
`js
fig.getCache() // returns key-value store
fig.setCache(cache) // where "cache" is the object retrieved by .getCache()
`
By default, FigTree will throw errors that can be caught and handled by your application. The Error object is a FigTreeError type, which extends the standard Error object (with name, message and stack fields) with the following interface:
`ts`
interface FigTreeError extends Error {
errorData?: Record
operator?: Operator
expression?: EvaluatorNode
prettyPrint: string // nicely formatted summary
}
There are two alternatives to throwing:
1. Fallback: If the fallback property is specified in the expression, this will be returned whenever an error occurs at that node (or below). See Fallback option.returnErrorAsString: true
2. Formatted string: By using the option, FigTree will return a nicely formatted string describing the error. This is the same string returned in the prettyPrint property of the FigTreeError. The formatted string will be structured like so:
`
Operator:
{
...errorData
} is specific data returned by the error thrown by an internal process (such as a network request using fetch).
For example, a 403 (forbidden) error thrown by the GET operator (with Axios HTTP client) would return a string like:`
Operator: GET - AxiosError
Request failed with status code 403
{
"status": 403,
"error": "Forbidden",
"url": "http://httpstat.us/403",
"response": {
"success": false,
"message": "jwt must be provided"
}
}
And, if thrown, the error object would contain:
`js`
{
name: "AxiosError",
message: "Request failed with status code 403"
operator: "GET",
errorData: {
status: 403,
error: 'Forbidden',
url: 'http://httpstat.us/403',
response: {
success: false,
message: "jwt must be provided"
}
}
expression: {
operator: 'API',
url: 'http://httpstat.us/403',
},
prettyPrint: '[...the above formatted string...]'
...otherProperties // all AxiosError and standard Error properties
}
It may seem that returning the error as a string results in the same output as catching the error and displaying error.prettyPrint`. There is one key difference, however: if the error occurs deeper in the expression tree, a thrown error will be bubbled up immediately, whereas if it's returned as a string (or a fallback, for that matter), it'll be treated as any other string