stated-js

v0.1.46

Published

4 days ago

JSONata embedded in JSON

Downloads

715

0High
0Medium
0Low

geoffhendrey

zhirafovod

jsonata json config state dag

Stated

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 /

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'}|"
}

Published

Vulnerabilities

Links

Maintainers

Keywords

Readme

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

SVG command

YAML

Setting Values in the stated REPL

Expression Scope

Rerooting Expressions

DAG

visualizing the plan with .svg command

Concurrency

Serialized Mutations

Atomic State Updates

Multi Version Concurrency Control (MVCC) and $forked

YAML

Com