stated-js

Stated

Stated
Licensing and documentation
API docs
Testing
Intro
Motivation
Getting Started
- Installation
- Running
  - REPL Mode
  - Oneshot mode
stated-js lib
- API documentation
- Basic usage
  - Stated-js package.json exports
  - Node.js based projects
REPL Commands
- .open
- All commands
- .out
- .color
- .log
- .set
- REPL command arguments
  - --options
    - strict
  - --xf
  - --ctx
Language & Syntax
- Expressions and Variables
- Dollars-Moustache
- Dollars-Variables
- Temporary Expressions
- References
- Expression Scope
- Rerooting Expressions
- Tags
Generative Templates
Reactive Behavior
- DAG
- visualizing the plan with .svg command
Concurrency
- Serialized Mutations
- Atomic State Updates
- Multi Version Concurrency Control (MVCC) and $forked
YAML
Complex Data Processing
Functions
- JSONata built-in
- Stated functions
- Custom Functions
  - Simple Custom Function Example
  - More Complex Function Example
Error Handling
- The error object
- Error Reporting
- $errorReport function
TemplateProcessor Snapshots
Understanding Plans
Planning
- MetaInfo Graph
- DFS Tree Traversal

Licensing and documentation

Stated is a cisco-open, Apache 2 Licensed, Open Source project at https://github.com/cisco-open/stated, our github main page.

API docs

If you would like to see a table of contents and developer API docs, jump over to our API docs page which is generated in CI and published here. Most developers will need to interact with the TemplateProcessor. It's API Docs are here

Testing

Are the examples in this README correct/reliable? Every markdown codeblock in this readme is tested on every commit.

stated logo

Intro

Stated is an engine of reactive state (a State Daemon). State is expressed as JSON or YAML. Rules for evolving state are written via embedded JSONata expressions in classic shell variable syntax ${}.

cat examples/hello.json
{
  "to": "world",
  "msg": "${'hello ' & to}"
}

You can use stated as a 'one shot' template processor that computes all bound states once. If we run the template above through stated, we see the ${} expression has been evaluated:

stated example/hello.json
{
  "to": "world",
  "msg": "hello world"
}

Variables can be passed in:

cat example/helloVar.json
{
  "msg": "${'hello ' & $TO}"
}
stated example/helloVar.json --ctx.TO=world
{
   "msg": "hello world"
}

The more interesting use of Stated is an engine of reactive state. Load a template in the REPL, .set a value and watch the state evolve.

stated
> .init -f example/hello.json
{
   "to": "world",
   "msg": "${'hello ' & to}"
}
> .set /to "Joshua"
{
   "to": "Joshua",
   "msg": "hello Joshua"
}
> .set /to "David Lightman"
{
   "to": "David Lightman",
   "msg": "hello David Lightman"
}

Stated templates can contain expressions, reusable functions, and can even use JS timeouts and intervals. Let's see a template that increments a counter every 10 ms, forever, as long it is running in the Stated engine. We will use the --tail command to tail the count variable until it reaches 100, then automatically disconnect the tail.

cat example/infiniteCount.json
{
  "count": 0,
  "counter": "${ $setInterval(function(){$set('/count', count+1)}, 10) }",
}
stated
> .init -f example/infiniteCount.json --tail "/count until $=100"
Started tailing... Press Ctrl+C to stop.
100

Stated is written in JS, and runs in the browser and in Node. Stated's REPL extends the Node.js REPL and allows you to interact with running templates. Stated uses asynchronous event loop I/O and can be used to orchestrate complex workflows:

> .init -f example/homeworlds.json
{
  "lukePersonUrl": "${ $fetch('https://swapi.tech/api/people/?name=luke').json().**.url}",
  "lukePersonDetails": "${ $fetch(lukePersonUrl).json().result[0]}",
  "lukeHomeworldURL": "${ lukePersonDetails.**.homeworld }",
  "homeworldDetails": "${ $fetch(lukeHomeworldURL).json() }",
  "homeworldName": "${ homeworldDetails.**.name }"
}
> .out /homeworldName
"Tatooine"

Unlike an ordinary program, Stated templates can be kept "alive" indefinitely. A change to any of the independent fields causes change propagation throughout the DAG. Stated includes a node REPL, stated.ts, for testing Stated json templates, and a JS library for embedding stated in applications. A typical REPL session consists of loading a template with the init command, viewing the computed output with the .out command and then setting values with the .set command and observing the changed output.

falken$ stated
> .init -f "example/ex08.json"
{
  "a": "${c}",
  "b": "${d+1+e}",
  "c": "${b+1}",
  "d": "${e+1}",
  "e": 1
}
> .out
{
  "a": 5,
  "b": 4,
  "c": 5,
  "d": 2,
  "e": 1
}
> .set /e 42
{
  "a": 87,
  "b": 86,
  "c": 87,
  "d": 43,
  "e": 42
}

Stated templates are modular and can be imported from a URL:

> .init -f "example/ex18.json"
{
  "noradCommander": "${ norad.commanderDetails  }",
  "norad": "${ $import('https://raw.githubusercontent.com/geoffhendrey/jsonataplay/main/norad.json')}"
}
> .out
{
  "noradCommander": {
    "fullName": "Jack Beringer",
    "salutation": "General Jack Beringer",
    "systemsUnderCommand": 4
  },
  "norad": {
    "commanderDetails": {
      "fullName": "Jack Beringer",
      "salutation": "General Jack Beringer",
      "systemsUnderCommand": 4
    },
    "organization": "NORAD",
    "location": "Cheyenne Mountain Complex, Colorado",
    "commander": {
      "firstName": "Jack",
      "lastName": "Beringer",
      "rank": "General"
    },
    "purpose": "Provide aerospace warning, air sovereignty, and defense for North America",
    "systems": [
      "Ballistic Missile Early Warning System (BMEWS)",
      "North Warning System (NWS)",
      "Space-Based Infrared System (SBIRS)",
      "Cheyenne Mountain Complex"
    ]
  }
}

Motivation

Consider this ordinary program:

let a=1;
let b=a;
a=42;
console.log(b); //prints out 1

In an ordinary sequential program the value of b is not affected by changes to the value of a at any point after the value of b has been assigned. But there are many situations where we do NOT want a sequential program execution Instead, we actually want b to change when a changes. Broadly, these cases fall under the rubric of "reactive" or "state driven" applications. When we try to build reactive applications upon a sequential execution model we are forced to code the data flow graph ourselves and things become very complex quickly. How could we make b change any time a changes in a sequential world? Perhaps naively like this?

let a=1;
let b=a;
function setA(val){
    a=val;
    b=a;
}

...or perhaps more generally like this:


let data = {
  a: 1,
  b: 1 
};

let handler = {
  set: function(target, property, value) {
    if (property === 'a') {
      target.b = value; // When 'a' changes, also change 'b'
    }
    target[property] = value;
    return true; // The set operation was successful
  }
};

let proxy = new Proxy(data, handler);

proxy.a = 2; // Setting a new value for 'a'
console.log(proxy.a); // Outputs: 2
console.log(proxy.b); // Outputs: 2

Every "coding" approach requires us to understand and implement code for propagating data dependencies. Stated solves for this by natively parsing and understanding dependencies.

{
  "a": 1,
  "b$": "a"
}

b$ is now declared to be continuously dependent upon a and reactive to any changes in a. This greatly simplifies the development of systems that are naturally reactive or dependency heavy, such as:

dynamic/continuous reactive UI state
lambda-like computations
workflows
configuration file

Getting Started

Installation

To install the stated-js package, you can use yarn or npm. Open your terminal and run one of the following commands:

Using Yarn:

yarn global add stated-js

Using Npm:

npm install -g stated-js

Running

REPL Mode

To use the Stated REPL (Read-Eval-Print Loop) it is recommended to have at least version 19.2 or higher of node.js. The Stated REPL is built on Node REPL. You can start the REPL by running the stated command in your terminal:

stated

The REPL will launch, allowing you to interact with the stated-js library. For example you can enter this command in the REPL:

> .init -f "example/ex01.json"

Oneshot mode

in oneshot mode, stated.ts simply computes the output template and exits. This is useful when you do not intend to change any if the fields after the initial output render

falken$ stated example/ex01.json
{
"to": "world",
"msg": "hello world"
}

stated-js lib

API documentation

API documentation is generated in CI and published to the project github page at https://cisco-open.github.io/stated/

Basic usage

Much more detailed progammer usage can be found by perusing https://cisco-open.github.io/stated/classes/TemplateProcessor.default.html To use the stated-js library in your own projects, you can require it as a dependency. Here's an example of how you can import it into your JavaScript file:

import TemplateProcessor from 'stated-js'

async function tryIt() {
    const tp = new TemplateProcessor({
        "a": "aaa",
        "b": "${a}"
    });
    await tp.initialize();
    await tp.setData("/a", 42);
    console.log(JSON.stringify(tp.output));
    // console will print...
    //    {
    //        "a": 42,
    //        "b": 42
    //    }     
}
tryIt().catch(err => console.error(err));

Stated-js package.json exports

The distribution of stated-js exports two different webpacks, one for ES Module (ie "import") which is in ./dist/bundle.mjs, and one for Common JS (ie "require") which is in ./dist/bundle-common-js.cjs. Using import or require in your JS code from a project with a package.json will automatically choose the correct dist target. Both the ES and Common JS exports are designed for use in the browser. Here is an excerpt of package.json showing exactly what is exported.

"exports": {
    ".": {
      "import": "./dist/bundle.mjs",
      "require": "./dist/bundle-common-js.cjs"
    },
    "./dist/src/TemplateProcessor.js": "./dist/src/TemplateProcessor.js",
    "./dist/src/CliCore.js": "./dist/src/CliCore.js",
    "./dist/src/StatedREPL.js": "./dist/src/StatedREPL.js"
  }

Node.js based projects

If you are building for a Node.js environment you should not use the webpacked exports for import or require. Instead, use the 'raw' TemplateProcessor.js, CliCore.js, and StatedREPL.js. Write your import statement like this for Node.js :

import TemplateProcessor from 'stated-js/dist/src/TemplateProcessor.js';

When you run Node.js, you must pass --experimental-vm-modules flag to the Node.js runtime. This is due to the fact that stated-js is written using ES Module syntax.

REPL Commands

REPL commands can be used once you start the REPL by running stated from your prompt. All examples in this doc assume your working directory is the stated git repo's root folder.

.open

Stated provides a set of REPL commands to interact with the system. The .open command is the easiest way to get started opening an example template:

All commands

Command	Description	flags & args	Example
`.open`	Interactive command to open a template (defaults to examples folder)		`.open`
`.cd`	Change directory (then use .open command)		`.cd ..`
`.init`	Initialize the template from a JSON file.	• `-f <path>` • `--tags=<taglist>` •`--options=<json>` • `--xf=<path>` • `--importPath=<path>` • `--tail "<tailargs>" --ctx.<dotted-path>=val`	`.init -f "example/hello.json" --tags=FOO,BAR --xf=~/falken/myEnv.json --options={"strict":{"refs":true}} --importPath=~/falken/mytemplates --tail "/ until msg='hello world'" --ctx.TO=world`
`.set`	Set data to a JSON pointer path.	`<path> <data>`	`.set /to "jsonata"`
`.from`	Show the dependents of a given JSON pointer.	`<path>`	`.from /a`
`.to`	Show the dependencies of a given JSON pointer.	`<path>`	`.to /b`
`.in`	Show the input template.	`None`	`.in`
`.out`	Show the current state of the template.	`[<jsonPtr>]`	`.out` `.out /data/accounts`
`.state`	Show the current state of the template metadata.	`None`	`.state`
`.plan`	Show the execution plan for rendering the template.	`None`	`.plan`
`.note`	Show a separator line with a comment in the REPL output.	`<comment>`	`.note "Example 8"`
`.log`	Set the logging level	`[silent, error, warn, info, verbose, debug]`	`.log silent`
`.color`	Enable Colors	`[on,off]`	`.color on`
`.tail`	Tail part of the document for changes	`<jsonPointer> (until <jsonata_expr>)?`	`.tail /` `.tail "/ until foo='bar'"`
`.svg`	Serve an SVG diagram of the DAG	`--port <portnumber>` (defaults to 4242)	`.svg --port 3000`
`.restore`	Restore from a snapshot	• `-f <path>` • `--tags=<taglist>` • `--xf=<path>` • `--importPath=<path>` • `--tail "<tailargs>"`	`.restore -f "example/restoreSnapshot.json" --tail "/count until $=10"`

.out

The stated repl lets you experiment with templates. The simplest thing to do in the REPL is load a json file. The REPL parses the input, builds an execution plan, and executes the result. To see the result you have to use the .out

> .init -f "example/ex09.json"
{
  "a": [
    0,
    1,
    "${ $[0] + $[1] }"
  ]
}
> .out
{
  "a": [
    0,
    1,
    1
  ]
}

.color

You can set terminal colors for enhanced readability from a terminal

> .note color does not appear in this markdown file
"============================================================="
> .color on
> .init -f "example/ex01.json"
{
  "a": 42,
  "b": "${a}",
  "c": "${'the answer is: '& b}"
}

.log

the .log command can set the log level to [silent, error, warn, info, verbose, debug]. Enabling high log levels like debug can help you track down problems with expressions.

> .log debug
{
  "log level": "debug"
}

 .init -f "example/errors.json"
arguments: {"_":[],"f":"example/errors.json","filepath":"example/errors.json","tags":[],"oneshot":false,"options":{}}
verbose: initializing...
debug: tags: {}
verbose: evaluating template...
error: Error evaluating expression at /b
error: The right side of the "+" operator must evaluate to a number {"code":"T2002","position":3,"stack":"Error\n    at evaluateNumericExpression (/Users/ghendrey/proj/jsonataexperiments/node_modules/jsonata/jsonata.js:4122:25)\n    at evaluateBinary (/Users/ghendrey/proj/jsonataexperiments/node_modules/jsonata/jsonata.js:3900:30)\n    at async evaluate (/Users/ghendrey/proj/jsonataexperiments/node_modules/jsonata/jsonata.js:3490:26)\n    at async Object.evaluate (/Users/ghendrey/proj/jsonataexperiments/node_modules/jsonata/jsonata.js:5558:26)\n    at async TemplateProcessor._evaluateExprNode (file:///Users/ghendrey/proj/jsonataexperiments/src/TemplateProcessor.ts:637:25)\n    at async TemplateProcessor._evaluateExpression (file:///Users/ghendrey/proj/jsonataexperiments/src/TemplateProcessor.ts:556:28)\n    at async TemplateProcessor.evaluateJsonPointersInOrder (file:///Users/ghendrey/proj/jsonataexperiments/src/TemplateProcessor.ts:515:31)\n    at async TemplateProcessor.evaluateDependencies (file:///Users/ghendrey/proj/jsonataexperiments/src/TemplateProcessor.ts:358:16)\n    at async TemplateProcessor.evaluate (file:///Users/ghendrey/proj/jsonataexperiments/src/TemplateProcessor.ts:123:9)\n    at async TemplateProcessor.initialize (file:///Users/ghendrey/proj/jsonataexperiments/src/TemplateProcessor.ts:113:9)","token":"+","value":" is not a string"}
debug: Expression: a + ' is not a string'
debug: Target: {
  "a": 42,
  "b": "${a + ' is not a string'}"
}
debug: Result: null
verbose: _evaluateExpression at /b completed in 13 ms.
verbose: evaluation complete in 13 ms...
verbose: initialization complete...
{
  "a": 42,
  "b": "${a + ' is not a string'}"
}

.set

The stated REPL also allows you to dynamically set values in your templates, further aiding in debugging and development. In the example below .set /a/0 100 sets a[0] to 100. The syntax of /a/0 is RFC-6901 JSON Pointer.

> .init -f "example/ex09.json"
{
  "a": [
    0,
    1,
    "${ $[0] + $[1] }"
  ]
}
> .set /a/0 100
{
  "a": [
    100,
    1,
    101
  ]
}

REPL command arguments

--options

The cli and REPL both support --options which can be provided as arguments to other commands Options can also be used in oneshot mode. Note the use of backslashes to escape quotes in the JSON on the CLI

falken$ stated --options={\"strict\":{\"refs\":true}} example/strictref.json
error: /z does not exist, referenced from /c (strict.refs option enabled)
{
  a: 42,
  b: 42,
  c: {
    error: {
      name: 'strict.refs',
      message: '/z does not exist, referenced from /c (strict.refs option enabled)'
    }
  }
}

strict

The strict option currently supports the refs property. Setting {"strict":{"refs":true}} will cause templates to throw an Exception when references in the template cannot be resolved. Reference checking is only performed against the template itself; it is not performed agains variables that are injected into the execution context by the TemplateProcessor library.

> .log silent
{
  "log level": "silent"
}
> .init -f "example/strictref.json" --options={"strict":{"refs":true}}
{
  "a": 42,
  "b": "${a}",
  "c": "${z}"
}
> .out
{
  "a": 42,
  "b": 42,
  "c": {
    "error": {
      "name": "strict.refs",
      "message": "/z does not exist, referenced from /c (strict.refs option enabled)"
    }
  }
}

--xf

The --xf argument can be used to provide a context file. Context is used to provide JSONata Bindings

> .note here is a regular json file
"============================================================="
> .init -f "example/env.json" 
{
  "env": {
    "name": "Dr. Stephen Falken",
    "addr": "Goose Island, OR, USA"
  }
}
> .note let's use it as context to a stated template
"============================================================="
> .init -f "example/useEnv.json" --xf=example/env.json
{
  "name": "${$env.name}",
  "address": "${$env.addr}"
}
> .out
{
  "name": "Dr. Stephen Falken",
  "address": "Goose Island, OR, USA"
}

--ctx

The --ctx argument can be used to inject context variables into the template. Variables with a sinle $ like $foo refer to the JSONata Context. You can inject variables into the context using --ctx.<dot-path>=val.

> .init -f example/ctx.json --ctx.name david --ctx.games.choice1=chess --ctx.games.choice2 "global thermonuclear war"
{
  "msg": "${'Hello, ' & $name & ', how about a nice game of ' & $games.choice1}",
  "games": "${$games}"
}
> .out
{
  "msg": "Hello, david, how about a nice game of chess",
  "games": {
    "choice1": "chess",
    "choice2": "global thermonuclear war"
  }
}

Language & Syntax

Expressions and Variables

What makes a Stated template different from an ordinary JSON file? JSONata Expressions of course! Stated analyzes the Abstract Syntax Tree of every JSONata expression in the file, and learns what references are made by each expression to other fields of the document. The references of an expression become the dependencies of the field, which are used to build a DAG. The DAG allows Stated to know what expressions to compute if any fields of the document change. Fields of the document are changed either via the REPL .set function, or by calling the equivalent library function. Many classes of reactive application need to maintain state, and need to propagate state changes through the dependees of a particular field (a dependee of foo is a field that depends on foo). Stated can be used as state store for reactive applications.

Dollars-Moustache

returning to our example/hello.json, the msg field is a simple example of a dollars-moustache. Stated allows JSONata expressions to be embedded in a JSON document using ${<...JSONata here...>} syntax. The ${} tells stated that a field such as msg is not an ordinary string field, but rather a JSONata expression that has to be evaluated in order to compute the value of the field.

falken$ cat example/hello.json
{
"to": "world",
"msg": "${'hello ' & to}"
}

Dollars-Variables

There is also a more concise alternative to ${}. The field can simply be named with a trailing dollars sign.

{
  "to": "world",
  "msg$": "'hello ' & to"
}

However the foo$ style can only be used when the expression is being assigned to a field. It won't work for array elements like this, where there is no field name. For array elements the ${} must be used:

[1, 2, "${$[0]+$[1]}"]

Temporary Expressions

The ! symbol is used mark fields as temporary. The ! can be used both as a prefix to the expression, and as a suffix to a key. Temporary fields are removed from the output. Notice how /b and /c! are removed from the output. Also notice that when an expression like ${`c!`.c1} refers to c! that backtics must be used.

> .init -f "example/tempVars.json"
{
   "a": 42,
   "b": "!${a}",
   "c!": {
      "c1": "../${a + 1}"
   },
   "d": "${`c!`.c1}"
}
> .out
{
   "a": 42,
   "d": 43
}

References

The most important thing stated does is analyze your embedded jsonata programs to understand their references. This means that as the user you don't have to tell stated what an expression depends on, and consequently you don't have to instruct Stated on how it should react to changes. It knows. In the example below, a JSONata block is used to produce defcon$. It defines local variables like $tmp which are pure JSONata constructs. The JSONata program also references fields of the input like MAX_DEFCON and threatLevel. States understands that if threatLevel changes, defcon$ must be recalculated. As shown below after viewing the output with the .out commnand, we mutate the threatLevel field which results in defcon$ changing from 3 to 5.

> .init -f "example/ex20.json"
{
  "defcon$": "($tmp:=$floor(intelLevel * threatLevel);$tmp:= $tmp<=MAX_DEFCON?$tmp:MAX_DEFCON;$tmp>=MIN_DEFCON?$tmp:MIN_DEFCON)",
  "MAX_DEFCON": 5,
  "MIN_DEFCON": 1,
  "intelLevel": 1.45,
  "threatLevel": 2.2
}
> .out
{
  "defcon$": 3,
  "MAX_DEFCON": 5,
  "MIN_DEFCON": 1,
  "intelLevel": 1.45,
  "threatLevel": 2.2
}
> .set /threatLevel 3.5
{
  "defcon$": 5,
  "MAX_DEFCON": 5,
  "MIN_DEFCON": 1,
  "intelLevel": 1.45,
  "threatLevel": 3.5
}

Expression Scope

Individual JSONata programs are embedded in JSON files between ${..}. What is the input to the JSONata program? The input, by default, is the object or array that the expression resides in. For instance in the example above, you can see that the JSONata $ variable refers to the array itself. Therefore, expressions like $[0] refer to the first element of the array.

Rerooting Expressions

In Stated templates, one way to declare a JSONata expression is by surrounding it by "dollars moustaches". E.g ${...some expression...}. JSONata expressions always have a context. The $ variable always points to the current context. The $$ variable always points to the input (root context) for an expression. In a Stated template, the root context for an expression is the object in which the expression is contained. For Example:

> .init -f "example/context.json"
{
   "a": {
      "b": "${[c,' and ',$.c,' and ',$$.c,' are the same thing. $ (current context) is /a, the object in which this expression resides']~>$join}",
      "c": "hello"
   }
}
> .out
{
   "a": {
      "b": "hello and hello and hello are the same thing. $ (current context) is /a, the object in which this expression resides",
      "c": "hello"
   }
}

Now we will show how we can change the context of an expression using 'rerooting.' Rerooting allows the expression's root context to be pointed anywhere in the json document. In the example below, consider greeting & ', ' & player1'. We want player1 to refer to the content at json pointer /player1 (the field named 'player1' at the root of the document). But our expression greeting & ', ' & player1 is located deep in the document at /dialog/partI. So how can we cause the root of the document to be the context for the JSONata expression greeting & ', ' & player1? You can reroot an expression in a different part of the document using relative rooting ../${<expr>} syntax or you can root an at the absolute doc root with /${<expr>}. The example below shows how expressions located below the root object, can explicitly set their context using the rooting syntax. Both absolute rooting, /${...} and relative rooting ../${...} are shown.

> .init -f "example/ex04.json"
{
  "greeting": "Hello",
  "player1": "Joshua",
  "player2": "Professor Falken",
  "dialog": {
    "partI": [
      "../../${greeting & ', ' &  player1}",
      "../../${greeting & ', ' &  player2}"
     ],
    "partII": {
      "msg3": "/${player1 & ', would you like to play a game?'}",
      "msg4": "/${'Certainly, '& player2 & '. How about a nice game of chess?'}"
    }
  }
}
> .out
{
  "greeting": "Hello",
  "player1": "Joshua",
  "player2": "Professor Falken",
  "dialog": {
    "partI": [
      "Hello, Joshua",
      "Hello, Professor Falken"
    ],
    "partII": {
      "msg3": "Joshua, would you like to play a game?",
      "msg4": "Certainly, Professor Falken. How about a nice game of chess?"
    }
  }
}

An advanced rerooting operator is the // absolute root operator. The / rooting operator, that we showed above, will never allow the expression to 'escape' outside of the template it was defined in. But what if we intend for a template to be imported into another template and we expect there to be a variable defined in the other template that we should use? This is where the // absolute root operator can be used. The // operator will set the expression context to the absolute root of whatever the final document is after all imports have been performed.

> .init -f "example/absoluteRoot.json"
{
   "to": "!${'Professor Falken'}",
   "greeting": "//${'Hello, ' & to}"
}
> .out
{
   "greeting": "Hello, Professor Falken"
}
> .init -f "example/importsAbsoluteRoot.json"
{
   "to": "Joshua",
   "message": "${$import('example/absoluteRoot.json')}"
}
> .out
{
   "to": "Joshua",
   "message": {
      "greeting": "Hello, Joshua"
   }
}

Generative Templates

Templates can contain generative expressions that cause their content to change over time. For instance the $setInterval function behaves exactly as it does in Javascript. Below, it causes the incr function to be called forever, every 10 ms.

> .init -f "example/tail.json"
{
  "incr": "${function(){$set('/counter',counter+1)}}",
  "counter": 0,
  "upCount": "${  $setInterval(incr, 10)  }"
}

The tail command can be used to watch changes on a particular field. Below we use --tail "/ until counter=5" to tail the document root until the value of counter reaches 5, at which point --tail stops producing output, though the incr function is still being called every 10 ms. field:

> .init -f "example/tail.json" --tail "/ until counter=5"
Started tailing... Press Ctrl+C to stop.
{
"incr": "{function:}",
"counter": 5,
"upCount": "--interval/timeout--"
}

Here was have a more complex example, again using $setInterval to increment a counter that drives rotation and scrolling of arrays and strings.

> .init -f "example/tailgraphs.yaml"
{
   "i": 0,
   "mod": "${i%10}",
   "ms": 30,
   "chars": [
      "この世界は広いですね。いろいろな人がいます。",
      "日本語はとても美しい言語です。詩を書くには最適。",
      "明日もきっといい日になる。希望を持って生きよう。",
      "プログラミングは楽しいですね。論理的思考が必要。",
      "本を読むことは、心を豊かにする。知識は無限だ。",
      "花が咲いて、山は緑が多い。自然は素晴らしい。",
      "春夏秋冬、日本の四季ははっきりしている。",
      "友達と一緒に遊ぶ時が最も楽しい時間だ。",
      "美しい海を見ていると、心が穏やかになる。",
      "映画を見るのは、私の趣味の一つです。"
   ],
   "wave": [
      "................==................................................................==................................................",
      ".............o......o..........................................................o......o.............................................",
      "..........o............o....................................................o............o..........................................",
      "........o................o................................................o................o........................................",
      "......o....................o............................................o....................o......................................",
      ".....o......................o..........................................o......................o.....................................",
      "...o..........................o......................................o..........................o...................................",
      "..o............................o....................................o............................o..................................",
      ".o...............................o.................................o...............................o................................",
      "..................................o..............................o..................................o..............................o",
      "...................................o............................o....................................o............................o.",
      "....................................o..........................o......................................o..........................o..",
      "......................................o......................o..........................................o......................o....",
      ".......................................o....................o............................................o....................o.....",
      ".........................................o................o................................................o................o.......",
      "...........................................o............o....................................................o............o.........",
      "..............................................o......o..........................................................o......o............",
      ".................................................==................................................................==..............."
   ],
   "chart": [
      "║█████████████████████████████",
      "║███████████████████████████████",
      "║██████████████████████",
      "║████████████",
      "║█████████",
      "║████",
      "║███████",
      "║██████████",
      "║██████████████████████",
      "║████████████████████████"
   ],
   "rot": "${  function($arr){$arr.($substring($, $$.i%$length($)) & $substring($, 0, $$.i%$length($)))}     }",
   "incr": "${ function(){ $set('/i', i+1) } }",
   "upCount": "${  $setInterval(incr, ms)  }",
   "scroll": "${  function($arr){ $arr[[$$.mod..9]] ~> $append($arr[[0..$$.mod]])}     }",
   "stars": "${[1..$$.i%20].($=19?'🚀':'⭐')~>$join}",
   "wavy": "${ (i;rot(wave)) }",
   "rotate": "${ (i;rot(chars)) }",
   "scrolled": "${ (i;scroll(chars))}",
   "moveBars": "${ (i;scroll(chart))}"
}

Setting --tail / instructs the REPL to tail the root of the document "in place" so the screen does not scroll and you can observe live changes to the json document.

> .init -f "example/tailgraphs.yaml" --tail / Alt text for the GIF

Reactive Behavior

Stated is naturally reactive. In the example below, story will evaluate when the promises for partI and partII have both resolved, simply because story has references to partI and partII, each of which respectively is triggered by the resolution of the two fetch functions they each depend on.

> .init -f "example/ex21.json"
{
  "call": "${function($url){$fetch($url) ~> handleRes}}",
  "falcon": "${ call('https://swapi.tech/api/starships/?name=Millennium').result[0].properties.name}",
  "han": "${ call('https://swapi.tech/api/people/?name=han').result[0].properties.name}",
  "handleRes": "${ function($res){$res.ok? $res.json():$res.status?{'status': $res.status}:$res} }",
  "luke": "${ call('https://swapi.tech/api/people/?name=luke').result[0].properties.name}",
  "partI": "${ [han, 'piloted the', falcon] ~> $join(' ')}",
  "partII": "${ [luke, 'piloted the', xwing] ~> $join(' ')}",
  "story": "${ [partI, 'then', partII]~>$join(' ')}",
  "xwing": "${ call('https://swapi.tech/api/starships/?name=x-wing').result[0].properties.name}"
}
> .plan
[
  "/handleRes",
  "/call",
  "/falcon",
  "/han",
  "/luke",
  "/partI",
  "/xwing",
  "/partII",
  "/story"
]
> .out
{
  "story": "Han Solo piloted the Millennium Falcon then Luke Skywalker piloted the X-wing",
  "handleRes": "{function:}",
  "call": "{function:}",
  "partI": "Han Solo piloted the Millennium Falcon",
  "luke": "Luke Skywalker",
  "xwing": "X-wing",
  "partII": "Luke Skywalker piloted the X-wing",
  "han": "Han Solo",
  "falcon": "Millennium Falcon"
}

SVG command

The .svg command serves an SVG diagram of the DAG

> .init -f "example/ex21.json"
{
   "story": "${ [partI, 'then', partII]~>$join(' ')}",
   "handleRes": "${ function($res){$res.ok? $res.json():$res.status?{'status': $res.status}:$res} }",
   "call": "${function($url){$fetch($url) ~> handleRes}}",
   "partI": "${ [han, 'piloted the', falcon] ~> $join(' ')}",
   "luke": "${ call('https://swapi.tech/api/people/?name=luke').result[0].properties.name}",
   "xwing": "${ call('https://swapi.tech/api/starships/?name=x-wing').result[0].properties.name}",
   "partII": "${ [luke, 'piloted the', xwing] ~> $join(' ')}",
   "han": "${ call('https://swapi.tech/api/people/?name=han').result[0].properties.name}",
   "falcon": "${ call('https://swapi.tech/api/starships/?name=Millennium').result[0].properties.name}"
}
> .svg --port=4042
Server is running on port 4042
"http://localhost:4042"

Access the URL from your web browser to view the SVG diagram. starwars svg

YAML

Input can be provided in YAML. YAML is convenient because JSONata prorgrams are often multi-line, and json does not support text blocks with line returns in a way that is readable. For instance if we compare ex12.json and ex12.yaml, which is more readable?

falken$ cat ex12.json
{
  "url": "https://raw.githubusercontent.com/geoffhendrey/jsonataplay/main/games.json",
  "selectedGame": "${game.selected}",
  "respHandler": "${ function($res){$res.ok? $res.json():{'error': $res.status}} }",
  "game": "${ $fetch(url) ~> respHandler ~> |$|{'player':'dlightman'}| }"
}

In YAML the respHandler function can be written as a text block, whereas in JSON it must appear on a single line.

falken$ cat ex12.yaml

url: "https://raw.githubusercontent.com/geoffhendrey/jsonataplay/main/games.json"
selectedGame: "${game$.selected}"
respHandler$: |
  function($res){
    $res.ok? $res.json():{'error': $res.status}
  }
game$: "$fetch(url) ~> respHandler$ ~> |$|{'player':'dlightman'}|"

However, once a YAML file is parsed with the JavaScript runtime it becomes a JavaScript object. Hence, in the example below a YAML is the input file, but the REPL displays the resulting Javascript object using JSON syntax. As we can see below, loading the yaml file still results in the function being deisplayed as it's parsed in-memory JS representation.

> .init -f "example/ex12.yaml"
{
  "url": "https://raw.githubusercontent.com/geoffhendrey/jsonataplay/main/games.json",
  "selectedGame": "${game$.selected}",
  "respHandler$": "function($res){\n  $res.ok? $res.json():{'error': $res.status}\n}\n",
  "game$": "$fetch(url) ~> respHandler$ ~> |$|{'player':'dlightman'}|"
}

Setting Values in the stated REPL

> .init -f "example/ex09.json"
{
  "a": [
    0,
    1,
    "${ $[0] + $[1] }"
  ]
}
> .set /a/0 100
{
  "a": [
    100,
    1,
    101
  ]
}

Expression Scope

Rerooting Expressions

> .init -f "example/context.json"
{
   "a": {
      "b": "${[c,' and ',$.c,' and ',$$.c,' are the same thing. $ (current context) is /a, the object in which this expression resides']~>$join}",
      "c": "hello"
   }
}
> .out
{
   "a": {
      "b": "hello and hello and hello are the same thing. $ (current context) is /a, the object in which this expression resides",
      "c": "hello"
   }
}

> .init -f "example/ex04.json"
{
  "greeting": "Hello",
  "player1": "Joshua",
  "player2": "Professor Falken",
  "dialog": {
    "partI": [
      "../../${greeting & ', ' &  player1}",
      "../../${greeting & ', ' &  player2}"
     ],
    "partII": {
      "msg3": "/${player1 & ', would you like to play a game?'}",
      "msg4": "/${'Certainly, '& player2 & '. How about a nice game of chess?'}"
    }
  }
}
> .out
{
  "greeting": "Hello",
  "player1": "Joshua",
  "player2": "Professor Falken",
  "dialog": {
    "partI": [
      "Hello, Joshua",
      "Hello, Professor Falken"
    ],
    "partII": {
      "msg3": "Joshua, would you like to play a game?",
      "msg4": "Certainly, Professor Falken. How about a nice game of chess?"
    }
  }
}

> .init -f "example/absoluteRoot.json"
{
   "to": "!${'Professor Falken'}",
   "greeting": "//${'Hello, ' & to}"
}
> .out
{
   "greeting": "Hello, Professor Falken"
}
> .init -f "example/importsAbsoluteRoot.json"
{
   "to": "Joshua",
   "message": "${$import('example/absoluteRoot.json')}"
}
> .out
{
   "to": "Joshua",
   "message": {
      "greeting": "Hello, Joshua"
   }
}

DAG

Templates can grow complex, and embedded expressions have dependencies on both literal fields and other calculated expressions. stated is at its core a data flow engine. Stated analyzes the abstract syntax tree (AST) of JSONata expressions and builds a Directed Acyclic Graph (DAG). Stated ensures that when fields in your JSON change, that the changes flow through the DAG in an optimal order that avoids redundant expression calculation.

stated helps you track and debug transitive dependencies in your templates. You can use the from and to commands to track the flow of data. Their output is an ordered list of JSON Pointers, showing you the order in which changes propagate.

> .init -f "example/ex01.json"
{
"a": 42,
"b": "${a}",
"c": "${'the answer is: '& b}"
}
> .out
{
"a": 42,
"b": 42,
"c": "the answer is: 42"
}
> .from /a
[
"/a",
"/b",
"/c"
]
> .to /b
[
"/a",
"/b"
]
> .to /c
[
"/a",
"/b",
"/c"
]

The .plan command shows you the execution plan for evaluating the entire template as a whole, which is what happens when you run the out command. The execution plan always ensures the optimal data flow so that no expression is evaluated twice.

> .init -f "example/ex08.json"
{
  "a": "${c}",
  "b": "${d+1+e}",
  "c": "${b+1}",
  "d": "${e+1}",
  "e": 1
}
> .plan
[
  "/d",
  "/b",
  "/c",
  "/a"
]
> .out
{
  "a": 5,
  "b": 4,
  "c": 5,
  "d": 2,
  "e": 1
}

visualizing the plan with .svg command

> .init -f "example/ex21.json"
{
  "call": "${function($url){$fetch($url) ~> handleRes}}",
  "falcon": "${ call('https://swapi.tech/api/starships/?name=Millennium').result[0].properties.name}",
  "han": "${ call('https://swapi.tech/api/people/?name=han').result[0].properties.name}",
  "handleRes": "${ function($res){$res.ok? $res.json():$res.status?{'status': $res.status}:$res} }",
  "luke": "${ call('https://swapi.tech/api/people/?name=luke').result[0].properties.name}",
  "partI": "${ [han, 'piloted the', falcon] ~> $join(' ')}",
  "partII": "${ [luke, 'piloted the', xwing] ~> $join(' ')}",
  "story": "${ [partI, 'then', partII]~>$join(' ')}",
  "xwing": "${ call('https://swapi.tech/api/starships/?name=x-wing').result[0].properties.name}"
}
> .svg --port=4042
Server is running on port 4042
"http://localhost:4042"

Access the URL from your web browser to view the SVG diagram. starwars svg

Concurrency

Stated executes expressions in your template according to a plan. The ordering of the plan is critical to ensuring correctness. As we have seen in the section on reactive behavior, a plan is an array of JSON pointers, each json pointer pointing to an expression in the template that must be evaluated. The DAG section of this document discusses plans like this:

[
  "/handleRes",
  "/call",
  "/falcon",
  "/han",
  "/luke",
  "/partI",
  "/xwing",
  "/partII",
  "/story"
]

Now we are going to discuss what happens if we introduce concurrent behavior. Concurrency can happen if an external actor, like your code using the stated library, mutates a field of the template concurrently. For example by creating an array of Promises that each change a field, and calling Promises.all. Concurrency can also happen inside your template itself by using $timeout and $interval which schedule functions on the event loop. Stated cannot allow two plans to concurrently access the same set of variables in the template, as this will cause behavior that is inconsistent with the expected behavior of a plan. Therefore, this section discusses how Stated provides correct and performant behavior when concurrent mutations trigger concurrent plan execution.

Serialized Mutations

A template has a single set of variables. What you see is what you get. The fields of the JSON are the variables of the program. Using $timeout and $interval causes concurrent behavior. In an ordinary program, concurrent modifications to shared state cause unexpected behavior. Not so in stated. Much like a database that implements SERIALIZABLE isolation level, stated queues and serializes mutations to the template. Lets see what happens when we map over an array of 10 integers, using a $timeout to generate concurrent mutations for each integer:

$ cat example/concurrentTimeouts.yaml
start: |
  ${
      [1..10].(
                $setTimeout(
                    function(){
                      $set('/step1', $string($))
                    }, $random()*100)
              )
  }
step1: NA
step2: ${step1 & ':' & step1}
step3: ${step2 & ':' & step2}
step4: ${$set('/final/-', step3)}
final: []

the $random timeout causes the /step1 plan to execute out of order with regard to the iputs [1..10]. However, due to Stated plan serialization, each input propagates through the execution plan while other inputs are locked out. This guarantees that every value in the final array is correct. For instance we never see "3:3:9:9" which would indicate that the input 3 had been allowed to begin its plan execution before `9' had completed its plan. As expected, the outputs are correct, but in a random order.

 "final": [
   "NA:NA:NA:NA",
   "5:5:5:5",
   "4:4:4:4",
   "6:6:6:6",
   "10:10:10:10",
   "2:2:2:2",
   "9:9:9:9",
   "7:7:7:7",
   "8:8:8:8",
   "3:3:3:3",
   "1:1:1:1"
]

Atomic State Updates

Stated Workflows provides an atomic primitive that prevents lost-updates with concurrent mutations of arrays. It does this by using a special JSON pointer defined by RFC 6902 JSON Patch for appending to an array. The example above uses this technique to with $set('final/-', step2). Although serialized mutations actually prevents concurrent updates, the $forked command allows for plans to truly run in parallel. Atomic array patching ensures that none of the concurrent plans will erase data from any other plan when they share an array.

Multi Version Concurrency Control (MVCC) and $forked

When plans are concise data computations, serialization can be a good strategy. However, what if the steps of the plan involves talking to external rest services? Serializing these plans will dramatically reduce the throughput of plan execution. This is where the $forked and $joined functions come into play. In the example below, we map over the names array and each element ($) is $forked on the /name. When a plan is forked, a copy is made in memory of the state. This is called Multi Version Concurrency Control (MVCC) and it allows each forked execution plan to operate on an isolated, complete, state. There is no serialization of plans. Each of the 10 names has a plan that is executing concurrently.

names: ${['luke', 'han', 'leia', 'chewbacca', 'darth', 'ben', 'c-3po', 'yoda'].($forked('/name',$))}
name: obi
personDetails: ${ $fetch('https://swapi.tech/api/people/?name='&name).json().result[0]}
homeworldURL: ${ personDetails.properties.homeworld}
homeworldDetails: ${ $fetch(homeworldURL).json().result }
homeworldName: ${ $joined('/homeworlds/-', homeworldDetails.properties.name)}
homeworlds: []

> .init -f example/homeworlds-forked.yaml --tail "/homeworlds until $count($)=9"
Started tailing... Press Ctrl+C to stop.
[
"Corellia",
"unknown",
"Tatooine",
"Alderaan",
"Tatooine",
"Tatooine",
"Tund",
"Stewjon",
"Kashyyyk"
]

Let's add some timing so we can understand tha effect of $forked on performance for I/O heavy plans like this, with multiple HTTP fetches.

$ cat example/homeworlds-forked-timed.yaml
startTime$: $millis()
data$: |
  (
    startTime$;
    ['luke', 'han', 'leia', 'chewbacca', 'darth', 'ben', 'c-3po', 'yoda'].($forked('/name',$))
  )
name: obi
personDetails: ${ $fetch('https://swapi.tech/api/people/?name='&name).json().result[0]}
homeworldURL: ${ personDetails.properties.homeworld}
homeworldDetails: ${ $fetch(homeworldURL).json().result }
homeworldName: ${ $joined('/homeworlds/-', homeworldDetails.properties.name)}
homeworlds: []
totalTime$: $string($millis()-startTime$) & ' ms'

Note that we make sure to include a trivial dependency on startTime$ from the data$ expression. This ensures that startTime$ is calculated first. We will run the .plan command to verify this.

> .init -f example/homeworlds-forked.yaml --tail "/homeworlds until $count($)=9"
Started tailing... Press Ctrl+C to stop.
[
   "unknown",
   "Tatooine",
   "Tatooine",
   "Kashyyyk",
   "Tatooine",
   "Alderaan",
   "Corellia",
   "Tund",
   "Stewjon"
]
> .plan
[
  "/startTime$",
  "/data$",
  "/personDetails",
  "/homeworldURL",
  "/homeworldDetails",
  "/homeworldName",
  "/totalTime$"
]
> .out /homeworlds
[
  "Tatooine",
  "Tatooine",
  "Alderaan",
  "unknown",
  "Tatooine",
  "Kashyyyk",
  "Tund",
  "Corellia",
  "Stewjon"
]
> .out /totalTime$
"781 ms"

Let's contrast the performance with the default SERIALIZED isolation level.

$ cat  example/homeworlds-serialized-timed.yaml
startTime$: $millis()
data$: |
  (
    startTime$;
    ['luke', 'han', 'leia', 'chewbacca', 'darth', 'ben', 'c-3po', 'yoda'].(
      $setTimeout(function(){$set('/name',$)})
    )
  )
name: obi
personDetails: ${ $fetch('https://swapi.tech/api/people/?name='&name).json().result[0]}
homeworldURL: ${ personDetails.properties.homeworld}
homeworldDetails: ${ $fetch(homeworldURL).json().result }
homeworldName: ${ $set('/homeworlds/-', homeworldDetails.properties.name)}
homeworlds: []
totalTime$: (homeworlds;$string($millis()-startTime$) & ' ms')

> .init -f example/homeworlds-serialized-timed.yaml --tail "/homeworlds until $count($)=9"
[
  "Stewjon",
  "Corellia",
  "Alderaan",
  "unknown",
  "Tund",
  "Tatooine",
  "Tatooine",
  "Kashyyyk",
  "Tatooine"
]
> .out /homeworlds
[
  "Tatooine",
  "Corellia",
  "Alderaan",
  "Kashyyyk",
  "Tatooine",
  "Tund",
  "Tatooine",
  "unknown",
  "unknown"
]
> .out /totalTime$
"6290 ms"

The $forked variation runs in 781 ms, while the SERIALIZED version ran in 6290 ms. The forked version is 8x faster.

YAML

falken$ cat ex12.json
{
  "url": "https://raw.githubusercontent.com/geoffhendrey/jsonataplay/main/games.json",
  "selectedGame": "${game.selected}",
  "respHandler": "${ function($res){$res.ok? $res.json():{'error': $res.status}} }",
  "game": "${ $fetch(url) ~> respHandler ~> |$|{'player':'dlightman'}| }"
}

In YAML the respHandler function can be written as a text block, whereas in JSON it must appear on a single line.

falken$ cat ex12.yaml

url: "https://raw.githubusercontent.com/geoffhendrey/jsonataplay/main/games.json"
selectedGame: "${game$.selected}"
respHandler$: |
  function($res){
    $res.ok? $res.json():{'error': $res.status}
  }
game$: "$fetch(url) ~> respHandler$ ~> |$|{'player':'dlightman'}|"

> .init -f "example/ex12.yaml"
{
  "url": "https://raw.githubusercontent.com/geoffhendrey/jsonataplay/main/games.json",
  "selectedGame": "${game$.selected}",
  "respHandler$": "function($res){\n  $res.ok? $res.json():{'error': $res.status}\n}\n",
  "game$": "$fetch(url) ~> respHandler$ ~> |$|{'player':'dlightman'}|"
}

Complex Data Processing

The example below uses JSONata $zip function to combine related data.

> .init -f "example/ex03.json"
{
  "data": {
    "fn": [
      "john",
      "jane"
    ],
    "ln": [
      "doe",
      "smith"
    ]
  },
  "names": "${ $zip(data.fn, data.ln) }"
}
> .out
{
  "data": {
    "fn": [
      "john",
      "jane"
    ],
    "ln": [
      "doe",
      "smith"
    ]
  },
  "names": [
    [
      "john",
      "doe"
    ],
    [
      "jane",
      "smith"
    ]
  ]
}

The example below uses the $sum function to compute a costs of each product, and then again uses $sum to sum over the individual product costs to get the totalCost. In a round-about fashion each individual product pulls in its cost from the costs array.

> .init -f "example/ex10.json"
{
  "totalCost": "${$sum(costs)}",
  "costs": "${products.$sum(quantity * price)}",
  "products": [
    {
      "name": "Apple",
      "quantity": 5,
      "price": 0.5,
      "cost": "/${costs[0]}"
    },
    {
      "name": "Orange",
      "quantity": 10,
      "price": 0.75,
      "cost": "/${costs[1]}"
    },
    {
      "name": "Banana",
      "quantity": 8,
      "price": 0.25,
      "cost": "/${costs[2]}"
    }
  ]
}
> .plan
[
  "/costs",
  "/products/2/cost",
  "/products/1/cost",
  "/products/0/cost",
  "/totalCost"
]
> .out
{
  "totalCost": 12,
  "costs": [
    2.5,
    7.5,
    2
  ],
  "products": [
    {
      "name": "Apple",
      "quantity": 5,
      "price": 0.5,
      "cost": 2.5
    },
    {
      "name": "Orange",
      "quantity": 10,
      "price": 0.75,
      "cost": 7.5
    },
    {
      "name": "Banana",
      "quantity": 8,
      "price": 0.25,
      "cost": 2
    }
  ]
}

Here is a different approach in which cost of each product is computed locally then rolled up to the totalCost. Note the difference in the execution plan between the example above and this example.

> .init -f "example/ex11.json"
{
  "totalCost": "${ $sum(products.cost) }",
  "products": [
    {
      "name": "Apple",
      "quantity": 5,
      "price": 0.5,
      "cost": "${ quantity*price }"
    },
    {
      "name": "Orange",
      "quantity": 10,
      "price": 0.75,
      "cost": "${ quantity*price }"
    },
    {
      "name": "Banana",
      "quantity": 8,
      "price": 0.25,
      "cost": "${ quantity*price }"
    }
  ]
}
> .plan
[
  "/products/0/cost",
  "/products/1/cost",
  "/products/2/cost",
  "/totalCost"
]
> .out
{
  "totalCost": 12,
  "products": [
    {
      "name": "Apple",
      "quantity": 5,
      "price": 0.5,
      "cost": 2.5
    },
    {
      "name": "Orange",
      "quantity": 10,
      "price": 0.75,
      "cost": 7.5
    },
    {
      "name": "Banana",
      "quantity": 8,
      "price": 0.25,
      "cost": 2
    }
  ]
}

Functions

JSONata built-in

Stated functions behave exactly as functions described in JSONata Functions. All JSONata functions are available in Stated.

Stated functions

Stated provides many functions not provided out of the box by JSONata.

$timeout

$interval

$fetch

JSONata provides the standard JS fetch function. You can use it exactly as you would in JS, except that Stated wraps fetch so that any exceptions are caught and converted to error objects. This is because JSONata does not have excpetions as a language construct.

This example fetches JSON over the network and uses the JSONata transform operator to set the player field on the fetched JSON. Here is the JSON file that it downloads and operates on. You can see why DAG and evaluation order matter. selectedGame does not exist until the game field has been populated by fetch.

> .init -f "example/ex12.json"
{
  "url": "https://raw.githubusercontent.com/geoffhendrey/jsonataplay/main/games.json",
  "selectedGame": "${game.selected}",
  "respHandler": "${ function($res){$res.ok? $res.json():{'error': $res.status}} }",
  "game": "${ $fetch(url) ~> respHandler ~> |$|{'player':'dlightman'}| }"
}
> .out
{
  "url": "https://raw.githubusercontent.com/geoffhendrey/jsonataplay/main/games.json",
  "selectedGame": "Global Thermonuclear War",
  "respHandler": "{function:}",
  "game": {
    "titles": [
      "chess",
      "checkers",
      "backgammon",
      "poker",
      "Theaterwide Biotoxic and Chemical Warfare",
      "Global Thermonuclear War"
    ],
    "selected": "Global Thermonuclear War",
    "player": "dlightman"
  }
}

$import

The sequence below explains by example that the $import function which is used to fetch and initialize remote templates (or local literal templates) into the current template

> .note "let's take a simple template..."
"============================================================="
> .init -f "example/ex17.json"
{
  "commanderDetails": {
    "fullName": "../${commander.firstName & ' ' & commander.lastName}",
    "salutation": "../${$join([commander.rank, commanderDetails.fullName], ' ')}",
    "systemsUnderCommand": "../${$count(systems)}"
  },
  "organization": "NORAD",
  "location": "Cheyenne Mountain Complex, Colorado",
  "commander": {
    "firstName": "Jack",
    "lastName": "Beringer",
    "rank": "General"
  },
  "purpose": "Provide aerospace warning, air sovereignty, and defense for North America",
  "systems": [
    "Ballistic Missile Early Warning System (BMEWS)",
    "North Warning System (NWS)",
    "Space-Based Infrared System (SBIRS)",
    "Cheyenne Mountain Complex"
  ]
}
> .note "now let's see what it produced"
"============================================================="
> .out
{
  "commanderDetails": {
    "fullName": "Jack Beringer",
    "salutation": "General Jack Beringer",
    "systemsUnderCommand": 4
  },
  "organization": "NORAD",
  "location": "Cheyenne Mountain Complex, Colorado",
  "commander": {
    "firstName": "Jack",
    "lastName": "Beringer",
    "rank": "General"
  },
  "purpose": "Provide aerospace warning, air sovereignty, and defense for North America",
  "systems": [
    "Ballistic Missile Early Warning System (BMEWS)",
    "North Warning System (NWS)",
    "Space-Based Infrared System (SBIRS)",
    "Cheyenne Mountain Complex"
  ]
}
> .note "what happens if we put it on the web and fetch it?"
"============================================================="
> .init -f "example/ex19.json"
{
  "noradCommander": "${ norad.commanderDetails  }",
  "norad": "${ $fetch('https://raw.githubusercontent.com/geoffhendrey/jsonataplay/main/norad.json') ~> handleRes }",
  "handleRes": "${ function($res){$res.ok? $res.json():{'error': $res.status}} }"
}
>  .note "If we look at the output, it's just the template."
"============================================================="
> .out
{
  "noradCommander": {
    "fullName": "../${commander.firstName & ' ' & commander.lastName}",
    "salutation": "../${$join([commander.rank, commanderDetails.fullName], ' ')}",
    "systemsUnderCommand": "../${$count(systems)}"
  },
  "norad": {
    "commanderDetails": {
      "fullName": "../${commander.firstName & ' ' & commander.lastName}",
      "salutation": "../${$join([commander.rank, commanderDetails.fullName], ' ')}",
      "systemsUnderCommand": "../${$count(systems)}"
    },
    "organization": "NORAD",
    "location": "Cheyenne Mountain Complex, Colorado",
    "commander": {
      "firstName": "Jack",
      "lastName": "Beringer",
      "rank": "General"
    },
    "purpose": "Provide aerospace warning, air sovereignty, and defense for North America",
    "systems": [
      "Ballistic Missile Early Warning System (BMEWS)",
      "North Warning System (NWS)",
      "Space-Based Infrared System (SBIRS)",
      "Cheyenne Mountain Complex"
    ]
  },
  "handleRes": "{function:}"
}
> .note "Now let's use the import function on the template"
"============================================================="
> .init -f example/ex16.json
{
   "noradCommander": "${ norad.commanderDetails  }",
   "norad": "${ $fetch('https://raw.githubusercontent.com/geoffhendrey/jsonataplay/main/norad.json') ~> handleRes ~> $import}",
   "handleRes": "${ function($res){$res.ok? $res.json():{'error': $res.status}} }"
}
> .out
{
   "noradCommander": {
      "fullName": "Jack Beringer",
      "salutation": "General Jack Beringer",
      "systemsUnderCommand": 4
   },
   "norad": {
      "commanderDetails": {
         "fullName": "Jack Beringer",
         "salutation": "General Jack Beringer",
         "systemsUnderCommand": 4
      },
      "organization": "NORAD",
      "location": "Cheyenne Mountain Complex, Colorado",
      "commander": {
         "firstName": "Jack",
         "lastName": "Beringer",
         "rank": "General"
      },
      "purpose": "Provide aerospace warning, air sovereignty, and defense for North America",
      "systems": [
         "Ballistic Missile Early Warning System (BMEWS)",
         "North Warning System (NWS)",
         "Space-Based Infrared System (SBIRS)",
         "Cheyenne Mountain Complex"
      ]
   },
   "handleRes": "{function:}"
}
> .note "You can see above that 'import' makes it behave as a template, not raw JSON."
"============================================================="
> .note "We don't have to fetch ourselves to use import, it will do it for us."
"============================================================="
> .init -f "example/ex18.json"
{
  "noradCommander": "${ norad.commanderDetails  }",
  "norad": "${ $import('https://raw.githubusercontent.com/geoffhendrey/jsonataplay/main/norad.json')}"
}
> .out
{
  "noradCommander": {
    "fullName": "Jack Beringer",
    "salutation": "General Jack Beringer",
    "systemsUnderCommand": 4
  },
  "norad": {
    "commanderDetails": {
      "fullName": "Jack Beringer",
      "salutation": "General Jack Beringer",
      "systemsUnderCommand": 4
    },
    "organization": "NORAD",
    "location": "Cheyenne Mountain Complex, Colorado",
    "commander": {
      "firstName": "Jack",
      "lastName": "Beringer",
      "rank": "General"
    },
    "purpose": "Provide aerospace warning, air sovereignty, and defense for North America",
    "systems": [
      "Ballistic Missile Early Warning System (BMEWS)",
      "North Warning System (NWS)",
      "Space-Based Infrared System (SBIRS)",
      "Cheyenne Mountain Complex"
    ]
  }
}

Importing bits of other templates

Suppose you just want to import a function defined in another template. The $import function understands URL fragments with JSON Pointers. In this example we use a URL with a fragment to import just a function from a remote template. Note that the URL ends in #/resourceMapperFn

> .init -f "example/resourceMapperB.json"
{
  "input": {
    "foo": 42,
    "bar": "something",
    "zap": "zing"
  },
  "resourceMapperAFn": "${$import('https://raw.githubusercontent.com/cisco-open/stated/main/example/resourceMapperA.json#/resourceMapperFn')}",
  "resourceMapperBFn": "${ function($in){$in.foo < 30 and $in.zap='zing'?[{'type':'B', 'id':$in.foo, 'bar':$in.bar, 'zap':$in.zing}]:[]}  }",
  "BEntities": "${ (resourceMapperBFn(input))}",
  "entities": "${ BEntities?BEntities:resourceMapperAFn(input)}"
}
> .out
{
  "input": {
    "foo": 42,
    "bar": "something",
    "zap": "zing"
  },
  "resourceMapperAFn": "{function:}",
  "resourceMapperBFn": "{function:}",
  "BEntities": [],
  "entities": [
    {
      "Type": "A",
      "id": 42,
      "bar": "something"
    }
  ]
}

Setting up local imports with --importPath

You can import local files by specifying a folder where stated will look for the imported files

> .init -f "example/localImport.json" --importPath=./example
{
  "noradCommander": "${ norad.commanderDetails  }",
  "norad": "${ $import('ex17.json')}"
}
> .out
{
  "noradCommander": {
    "fullName": "Jack Beringer",
    "salutation": "General Jack Beringer",
    "systemsUnderCommand": 4
  },
  "norad": {
    "commanderDetails": {
      "fullName": "Jack Beringer",
      "salutation": "General Jack Beringer",
      "systemsUnderCommand": 4
    },
    "organization": "NORAD",
    "location": "Cheyenne Mountain Complex, Colorado",
    "commander": {
      "firstName": "Jack",
      "lastName": "Beringer",
      "rank": "General"
    },
    "purpose": "Provide aerospace warning, air sovereignty, and defense for North America",
    "systems": [
      "Ballistic Missile Early Warning System (BMEWS)",
      "North Warning System (NWS)",
      "Space-Based Infrared System (SBIRS)",
      "Cheyenne Mountain Complex"
    ]
  }
}

Import JS functions

Repl and cli support importing javascript functions. If provided file to --xf has a .js or .mjs extension, it will be loaded and all exported functions will be added to TemplateProcessor's execution context.

An example "src/test/test-export.js" exports 2 functions

const barFunc = (input) => `bar: ${input}`;

// explicitly define exported functions and their names
export const foo = () => "foo";
export const bar = barFunc;

Which can be used in the stated tempalate context

> .init -f example/importJS.json --xf=example/test-export.js
{
  "res": "${ $bar($foo()) }"
}
> .out
{
  "res": "bar: foo"
}

This can be combined with the --importPath option to import files relative to that path

> .init -f example/importJS.json --importPath=example --xf=test-export.js
{
  "res": "${ $bar($foo()) }"
}
> .out
{
  "res": "bar: foo"
}

$open

Allowing expressions to open local files is a security risk. For this reason the core TemplateProcessor does not support the $open function. However, the CLI/REPL which are for local usage allow the $open function. Additionally, programs that want to allow properly guarded $open operations may inject a $open function of their choosing into the TemplateProcessor contexet. $open accepts a relative path, and parses the JSON or YAML file on that path into an object.

> .note This shows two equivalent ways to open a json or yaml file using $open
"============================================================="
> .cd example
"Current directory changed to: /Users/falken/proj/jsonataexperiments/example"
> .init -f "importLocal.json"
{
   "a": "${'ex01.json'~>$open~>$import}",
   "b": "${$import($open('ex01.json'))}"
}
> .out
{
   "a": {
      "a": 42,
      "b": 42,
      "c": "the answer is: 42"
   },
   "b": {
      "a": 42,
      "b": 42,
      "c": "the answer is: 42"
   }
}
> .cd ..
"Current directory changed to: /Users/falken/proj/jsonataexperiments"

$set

You have already seen how the REPL .set command works. The REPL simply calls the internal $set function. The $set function can be called from your JSONata blocks or function. The $set function is used to push data into other parts of the template. The function signature is $set(jsonPointer, value). The set command returns an array of JSON Pointers that represent the transitive updates that resulted from calling set. In the example below $set('/systems/1', 'JOSHUA') is used to push the string "JOSHUA" onto the systems array.

> .init -f "example/ex13.json"
{
  "systems": [
    "WOPR"
  ],
  "onBoot": "${ $set('/systems/1', 'JOSHUA')}",
  "newSystem": "${systems[1]}"
}
> .out
{
  "newSystem": "JOSHUA",
  "onBoot": [
    "/systems/1",
    "/newSystem"
  ],
  "systems": [
    "WOPR",
    "JOSHUA"
  ]
}

$debounce

Debouncing is a technique used in software development, particularly in web development and event handling, to control or limit the frequency of a particular action or event when it occurs rapidly or frequently. It helps prevent excessive or redundant executions of the action, making applications more efficient and responsive.

In this example we rapidly change a counter and feed it into a debounced function that appends the count to an accumulator array, acc. The interval is terminated when count reaches 100, and due to debouncing, only a single value, 100, is written into acc. We use the --tail "/ until acc[0]=100" to tail the root document (/) until a jsonata expression condition is met. The condition we check is acc[0]=100 which is the expected debounced value.

> .init -f "example/debounce.yaml"
{
   "acc": [],
   "appendAcc": "${ function($v){$set('/acc/-', $v)} ~> $debounce(15)}",
   "counter": "${   function(){($set('/count', $$.count+1); $$.count)}    }",
   "count": 0,
   "rapidCaller": "${$setInterval(counter~>appendAcc, 10)}",
   "stop": "${ count=100?($clearInterval($$.rapidCaller);'done'):'not done'  }"
}

> .init -f "example/debounce.yaml" --tail "/ until acc[0]=100"
Started tailing... Press Ctrl+C to stop.
{
  "acc": [
    100
  ],
  "counter": "{function:}",
  "count": 100,
  "rapidCaller": "--interval/timeout--",
  "stop": "done"
}

$defer

While $debounce is used to create a debounced function, $defer can be a more concise approach when you simply want a "slowed down" version of a rapidly changing variable. $defer takes two arguments:

The json pointer to the field you wish to defer (or "slow down")
An optional number of milliseconds, T.

$defer will always produce one initial value, which comes from the deferred field. $defer will not produce a subsequent value until the deferred field has remained unchanges for T ms. For example, suppose you are collecting a query string from a user input. Each character entered mutates the query field, but we don't want to do anything with query unless the user pauses or stops typing characters. In the example below, an $setInterval call is used to simulate a user entering the characters of a sampleQuery into the query field at a rate of one every 25ms. Fast typer! The deferredQuery$ will begin with the inital value of query ("") and remain unchanged until the querystops receiving changes and reaches its final state of "Would you like to play a game? How about a nice game of chess?"

> .init -f "example/defer.yaml"
{
   "sampleQuery": "Would you like to play a game? How about a nice game of chess?",
   "query": "",
   "deferredQuery$": "$defer('/query', 100)",
   "counter$": "function(){  $set('/count', count+1)}",
   "count": 0,
   "appendQuery$": "$set('/query', sampleQuery~>$substring(0,count))",
   "rapidCaller$": "$setInterval(counter$, 25)",
   "stop$": "count=$length(sampleQuery)\n  ?($clearInterval(rapidCaller$);'done')\n  :'simulating typing'  \n"
}

To watch the data changing we can use the --tail until to tail the changing template output until it reaches the final state.

> .init -f example/defer.yaml --tail "/ until deferredQuery$ = 'Would you like to play a game? How about a nice game of chess?'"
Started tailing... Press Ctrl+C to stop.
{
  "sampleQuery": "Would you like to play a game? How about a nice game of chess?",
  "query": "Would you like to play a game? How about a nice game of chess?",
  "deferredQuery$": "Would you like to play a game? How about a nice game of chess?",
  "counter$": "{function:}",
  "count": 62,
  "appendQuery$": [
    "/query"
  ] ,
  "rapidCaller$": "--interval/timeout--",
  "stop$": "done"
}

Here is a screencapture showing the two commands above executed in the REPL. tailing deferred

$rateLimit

Rate limiting allows to ensure than no than one function call is executed withing some time. For exameple, we want to ensure that a function calling external APIs does not overload it.

>.init -f example/rateLimit.yaml
{
  "acc": [],
  "appendAcc": "${ function($v){$set('/acc/-', $v)} ~> $rateLimit(100)}",
  "counter": "${   function(){($set('/count', $$.count+1); $$.count)}    }",
  "count": 0,
  "rapidCaller": "${$setInterval(counter~>appendAcc, 10)}",
  "stop": "${ count=100?($clearInterval($$.rapidCaller);'done'):'not done'  }",
  "accCount": "${ $count(acc) }"
}

Below output demonstrates, that rateLimit function calls to to set acc to once in no less than 100ms, which will result in only 10 counts added o the acc array, the first one, the last one, and 10 in between.

>.init -f example/rateLimit.yaml --tail "/ until accCount=12"
Started tailing... Press Ctrl+C to stop.
{
  "acc": [
    1,
    10,
    19,
    28,
    38,
    48,
    57,
    66,
    75,
    85,
    94,
    100
  ],
  "counter": "{function:}",
  "count": 100,
  "rapidCaller": "--interval/timeout--",
  "stop": "done",
  "accCount": 12
}

Custom Functions

Stated let's you define and call functions. Custom functions are simply JSONata functions. Functions are not a feature of Stated, they are a feature of JSONata.

Simple Custom Function Example

> .init -f "example/ex05.json"
{
  "hello": "${ (function($to){'hello ' & $to & '. How about a nice game of ' & $$.game})}",
  "to": "David",
  "game": "chess",
  "greeting$": "hello(to)"
}
> .out
{
  "hello": "{function:}",
  "to": "David",
  "game": "chess",
  "greeting$": "hello David. How about a nice game of chess"
}

More Complex Function Example

Here is an elaborate example of functions. The fibonnaci function itself is pulled into the last element of x using the expression /${fibonacci}. The first element of the array contains $[2]($[1]). Can you see that it invokes the fibonacci function passing it the value 6? Hint: $[2] is the last element of the array which will pull in the fibonacci function and $[1] is the middle element of the array, holding the static value 6. So $[2]($[1]) expands to fibonacci(6). The value 6th fibonacci number is 8, which is what fibonacci(6) returns.

> .init -f "example/ex06.json"
{
  "x": [
    "${$[2]($[1])}",
    6,
    "/${fibonacci}"
  ],
  "fibonacci": "${ function($n){$n=1?1:$n=0?0:fibonacci($n-1)+fibonacci($n-2)}}"
}
> .out
{
  "x": [
    8,
    6,
    "{function:}"
  ],
  "fibonacci": "{function:}"
}

Let's take a more complex example where we generate MySQL instances:

> .init -f "example/mysql.json"
{
  "name": "mysql",
  "count": 1,
  "pn": 3306,
  "providerName": "aws",
  "tmp": {
    "host": "/${ [1..count].{'database_instance.host':'mysql-instance-' & $ & '.cluster-473653744458.us-west-2.rds.amazonaws.com'}}",
    "port": "/${ [1..count].{'database_instance.port:':$$.pn}}",
    "provider": "/${[1..count].{'cloud.provider': $$.providerName}}",
    "instanceId": "/${[1..count].{'cloud.database_instance.id':'db-mysql-instance-' & $formatBase($,16)}}",
    "instanceName": "/${[1..count].{'database_instance.name':'MySQL instance' & $}}",
    "clusterName": "/${[1..count].{'database_instance.cluster._name':'MySQL cluster' & $}}"
  },
  "instances": "${$zip(tmp.host, tmp.port, tmp.provider, tmp.instanceId, tmp.instanceName, tmp.clusterName)~>$map($merge)}"
}
> .out
{
  "name": "mysql",
  "count": 1,
  "pn": 3306,
  "providerName": "aws",
  "tmp": {
    "host": {
      "database_instance.host": "mysql-instance-1.cluster-473653744458.us-west-2.rds.amazonaws.com"
    },
    "port": {
      "database_instance.port:": 3306
    },
    "provider": {
      "cloud.provider": "aws"
    },
    "instanceId": {
      "cloud.database_instance.id": "db-mysql-instance-1"
    },
    "instanceName": {
      "database_instance.name": "MySQL instance1"
    },
    "clusterName": {
      "database_instance.cluster._name": "MySQL cluster1"
    }
  },
  "instances": {
    "database_instance.host": "mysql-instance-1.cluster-473653744458.us-west-2.rds.amazonaws.com",
    "database_instance.port:": 3306,
    "cloud.provider": "aws",
    "cloud.database_instance.id": "db-mysql-instance-1",
    "database_instance.name": "MySQL instance1",
    "database_instance.cluster._name": "MySQL cluster1"
  }
}
> .from /count
[
  "/count",
  "/tmp/clusterName",
  "/tmp/host",
  "/tmp/instanceId",
  "/tmp/instanceName",
  "/tmp/port",
  "/tmp/provider",
  "/instances"
]
> .set /count 3
{
  "name": "mysql",
  "count": 3,
  "pn": 3306,
  "providerName": "aws",
  "tmp": {
    "host": [
      {
        "database_instance.host": "mysql-instance-1.cluster-473653744458.us-west-2.rds.amazonaws.com"
      },
      {
        "database_instance.host": "mysql-instance-2.cluster-473653744458.us-west-2.rds.amazonaws.com"
      },
      {
        "database_instance.host": "mysql-instance-3.cluster-473653744458.us-west-2.rds.amazonaws.com"
      }
    ],
    "port": [
      {
        "database_instance.port:": 3306
      },
      {
        "database_instance.port:": 3306
      },
      {
        "database_instance.port:": 3306
      }
    ],
    "provider": [
      {
        "cloud.provider": "aws"
      },
      {
        "cloud.provider": "aws"
      },
      {
        "cloud.provider": "aws"
      }
    ],
    "instanceId": [
      {
        "cloud.database_instance.id": "db-mysql-instance-1"
      },
      {
        "cloud.database_instance.id": "db-mysql-instance-2"
      },
      {
        "cloud.database_instance.id": "db-mysql-instance-3"
      }
    ],
    "instanceName": [
      {
        "database_instance.name": "MySQL instance1"
      },
      {
        "database_instance.name": "MySQL instance2"
      },
      {
        "database_instance.name": "MySQL instance3"
      }
    ],
    "clusterName": [
      {
        "database_instance.cluster._name": "MySQL cluster1"
      },
      {
        "database_instance.cluster._name": "MySQL cluster2"
      },
      {
        "database_instance.cluster._name": "MySQL cluster3"
      }
    ]
  },
  "instances": [
    {
      "database_instance.host": "mysql-instance-1.cluster-473653744458.us-west-2.rds.amazonaws.com",
      "database_instance.port:": 3306,
      "cloud.provider": "aws",
      "cloud.database_instance.id": "db-mysql-instance-1",
      "database_instance.name": "MySQL instance1",
      "database_instance.cluster._name": "MySQL cluster1"
    },
    {
      "database_instance.host": "mysql-instance-2.cluster-473653744458.us-west-2.rds.amazonaws.com",
      "database_instance.port:": 3306,
      "cloud.provider": "aws",
      "cloud.database_instance.id": "db-mysql-instance-2",
      "database_instance.name": "MySQL instance2",
      "database_instance.cluster._name": "MySQL cluster2"
    },
    {
      "database_instance.host": "mysql-instance-3.cluster-473653744458.us-west-2.rds.amazonaws.com",
      "database_instance.port:": 3306,
      "cloud.provider": "aws",
      "cloud.database_instance.id": "db-mysql-instance-3",
      "database_instance.name": "MySQL instance3",
      "database_instance.cluster._name": "MySQL cluster3"
    }
  ]
}

Error Handling

The error object

if a JSONata expression evaluation throws an exception, the exception is converted to an error and placed into the template output.

> .log silent
{
  "log level": "silent"
}
> .init -f "example/errors.json"
{
  "a": 42,
  "b": "${a + ' is not a string'}"
}
> .out
{
  "a": 42,
  "b": {
    "error": {
      "name": "JSONata evaluation exception",
      "message": "The right side of the \"+\" operator must evaluate to a number"
    }
  }
}

Error Reporting

The .errors command will produce a report of all errors in the template

> .log silent
{
  "log level": "silent"
}
> .init -f "example/errors.json"
{
  "a": 42,
  "b": "${a + ' is not a string'}"
}
> .errors
{
  "/b": {
    "name": "JSONata evaluation exception",
    "message": "The right side of the \"+\" operator must evaluate to a number"
  }
}

$errorReport function

JSONata does not have a throw/catch syntax. However, JSONata has a $error function that you can use to throw an error. However, doing so will end the execution of the JSONata expression. If you wish to simply record the fact that an error occured without exiting the expression, use the $errorReport function which returns an error object but does not throw.

> .init -f example/errorReport.json
{
   "a": [
      0,
      1,
      2,
      "${$errorReport('oops', 'my_custom_error')}",
      4,
      5
   ],
   "b": "${($errorReport('e0');$errorReport('e1');$errorReport('e2'))}"
}
> .out
{
   "a": [
      0,
      1,
      2,
      {
         "error": {
            "message": "oops",
            "name": "my_custom_error"
         }
      },
      4,
      5
   ],
   "b": {
      "error": {
         "message": "e2"
      }
   }
}
> .errors
{
   "/a/3": {
      "error": {
         "message": "oops",
         "name": "my_custom_error"
      }
   },
   "/b": [
      {
         "error": {
            "message": "e0"
         }
      },
      {
         "error": {
            "message": "e1"
         }
      },
      {
         "error": {
            "message": "e2"
         }
      }
   ]
}

TemplateProcessor Snapshots

TemplateProcessor.snapshot allows you to capture template state at a point in time. The snapshot can be used to restore and continue template execution from that point. The snapshot is a JSON object that can be serialized and stored in a file.

To create a periodic snapshot of the template every time it changes

const tempalte = {
   "count": 0,
   "counter": "${ $setInterval(function(){$set('/count', count+1)}, 10) }",
   "stop": "${ count=10?($clearInterval($$.counter);'done'):'not done'  }"
};
const tp = new TemplateProcessor(template);
await tp.initialize();
await fs.writeFile("./myTemplateSnapshot.json", tp.snapshot());

To continue template execution from the snapshot, you can either programmatically restore the snapshot

or use .restore command in the REPL.

> .restore -f example/restoreSnapshot.json --tail "/count until $=10"
Started tailing... Press Ctrl+C to stop.
10

Understanding Plans

This information is to explain the planning algorithms to comitters. As a user you do not need to understand how Stated formulates plans. Before explaining how a plan is made, let's show the end-to-end flow of how a plan is used with an example:

A template is shown with a graph tree visualizing the JSON. The template contains 3 expressions. For any valid template there is at least one ordering of all expressions that produces a correct output. We call such as correct ordering a Plan. A Plan is an ordered Array whose elements are JSON Pointers to expressions. To illustrate the importance of a correct plan, what would happen if the first expression executed was ${e.i+100}? The result would be garbage/undefined because e.i has not been evaluated.
From the template, a plan is formulated (we will explain how this happens in the following section)
The Plan is executed sequentially producing the final output

Planning

Planning is the process of

Analyzing each expression to produce a MetaInfo Node recording facts about the Node such as its dependencies
Organizing MetaInfo into a tree structure mirroring the template JSON
Traversing the tree using a DFS algorithm to produce a single ordering of expression evaluation (a Plan)

MetaInfo Graph

A plan is built by traversing a metainfo graph that has these features.

labels - each node is labeled with its JSON POinter
ordinary nodes - A node such as /a/d is an ordinary node with a value 100
expression nodes - A node such as /b/e contains a JSONata expression like /${a.c.g}
dependency arrows - Expressions like a.c.g refer to another part of the document, hence have a dependency
non-materialized nodes - These metadata nodes are created any time a reference creates a dependency on a node that does not exist in the JSON template. Such non-materialized nodes come in 2 flavors
1. results of expressions (since clearly results of expressions don't exist in the input template). In the diagram below,e.i is a local reference to a non-materialized node, from the expression ${e.1+100} found in node /b/f. Hence the absolute JSON Pointer for the non-materialized node is /b/e/i.
2. information that simply isn't present yet in the template. An example of this can be a dashboard whose expressions depend on a field that is expected to arrive at runtime, for instance query.result. Clearly the result of a query is not material in the template.

The MetaInfo graph for our example above is: Planning

DFS Tree Traversal

The core algorithm is in the method TemplateProcessor.topologicalSort. This method is passed a metaImnfo node, and recursively calls itself as it followDependencies and followChildren. A scope is used to detect circular dependencies within a given subgraph.

const listDependencies = (metaInfo) => {
   markAsVisited(metaInfo);
   addToScope(metaInfo);

   followDependencies(metaInfo);
   emit(metaInfo);
   followChildren(metaInfo);

   removeFromScope(metaInfo);
}

The only 'interesting' thing differentiating this from standard DFS tree-walking and dependency following, is that there is special treatment for non-materialized nodes. when a non-materialized node is encountered as a result of following a dependency, the algorithm travels to the parent of non-materialized node until it finds a node with an expression. The step labled 2 in the diagram below illustrates how the non-materialized node is followed up to its materialized parent expression.

Assuming we call findDeps on node /b/f the diagram below shows the order (1,2,3,4) in which links are traversed. Since the algorithm is DFS, expression nodes are added in reverse traversal order: ['/a/c/g/i', '/b/e', '/b/f']

Traversal

stated-js

Settings

Member Visibility

Theme

On This Page