Helix Pipeline

Last modified

(formerly known as Hypermedia Pipeline)

This project provides helper functions and default implementations for creating Hypermedia Processing Pipelines.

It uses reducers and continuations to create a simple processing pipeline that can pre-and post-process HTML, JSON, and other hypermedia.


codecov CircleCI GitHub license GitHub issues npm Greenkeeper badge Known Vulnerabilities

Anatomy of a Pipeline

A pipeline consists of following main parts:

  • pre-processing functions
  • the main response generating function
  • an optional wrapper function
  • post-processing functions
  • error handling functions

Each step of the pipeline is processing a single context object, which eventually can be used to return the http response.

See below for the anatomy of the context.

Typically, there is one pipeline for each content type supported and pipeline are identified by file name, e.g.

  • html.pipe.js – creates HTML documents with the text/html content-type
  • json.pipe.js – creates JSON documents with the application/json content-type

Building a Pipeline

A pipeline builder can be created by creating a CommonJS module that exports a function pipe which accepts following arguments and returns a Pipeline function.

  • cont: the main function that will be executed as a continuation of the pipeline
  • context: the context (formerly known as payload) that is accumulated during the pipeline.
  • action: the action that serves as holder for extra pipeline invocation argument.

This project’s main entry provides a helper function for pipeline construction and a few helper functions, so that a basic pipeline can be constructed like this:

// the pipeline itself
const pipeline = require("@adobe/hypermedia-pipeline");

module.exports.pipe = function(cont, context, action) {
    action.logger.log("debug", "Constructing Custom Pipeline");

    return pipeline()
        .use(cont)            // execute the continuation function

In a typical pipeline, you will add additional processing steps as .use(require('some-module')).

The Main Function

The main function is typically a pure function that converts the request and content properties of the context into a response object.

In most scenarios, the main function is compiled from a template in a templating language like HTL, JST, or JSX.

Typically, there is one template (and thus one main function) for each content variation of the file type. Content variations are identified by a selector (the piece of the file name before the file extension, e.g. in example.navigation.html the selector would be navigation). If no selector is provided, the template is the default template for the pipeline.

Examples of possible template names include:

  • html.jsx (compiled to html.js) – default for the HTML pipeline
  • html.navigation.jst (compiled to html.navigation.js) – renders the navigation
  • dropdown.json.js (not compiled) – creates pure JSON output
  • dropdown.html.htl (compiled to dropdown.html.js) – renders the dropdown component

(Optional) The Wrapper Function

Sometimes it is necessary to pre-process the context in a template-specific fashion. This wrapper function (often called “Pre-JS” for brevity sake) allows the full transformation of the pipeline’s context.

Compared to the pipeline-specific pre-processing functions which handle the request, content, and response, the focus of the wrapper function is implementing business logic needed for the main template function. This allows for a clean separation between:

  1. presentation (in the main function, often expressed in declarative templates)
  2. business logic (in the wrapper function, often expressed in imperative code)
  3. content-type specific implementation (in the pipeline, expressed in functional code)

A simple implementation of a wrapper function would look like this:

// All wrapper functions must export `pre`
// The functions takes following arguments:
// - `cont` (the continuation function, i.e. the main template function)
// - `context` (the context of the pipeline)
// - `action` (the action of the pipeline)
module.exports.pre = (cont, context, action) => {
    const {request, content, response} = context;
    // modifying the context content before invoking the main function
    content.hello = 'World';
    const modifiedcontext = {request, content, response};

    // invoking the main function with the new context. Capturing the response
    // context for further modification

    const responsecontext = cont(modifiedcontext, action);

    // Adding a value to the context response
    const modifiedresponse = modifiedcontext.response;
    modifiedresponse.hello = 'World';

    return Object.assign(modifiedcontext, modifiedresponse);

Pre-Processing Functions

Pre-Processing functions are meant to:

  • parse and process request parameters
  • fetch and parse the requested content
  • transform the requested content

Post-Processing Functions

Post-Processing functions are meant to:

  • process and transform the response

Error Handlers

In default state, the pipeline will process all normal functions but will skip error handlers (.error()). When the pipeline is in the error state normal processing functions are suspended until the end of the pipeline is reached, or the error state is cleared. While in the error state, error handlers will be executed. The pipeline execution is in an error state if context.error is defined. An Error state can happen when a processing function throws an Exception, or if it sets the context.error object directly.


new pipeline()

If in the above example, the doSomething causes an error, subsequently, render and cleanup will not be invoked. but handleError will. If handleError clears the error state (i.e. sets context.error = null), the done function will be invoked again.

If in the above example, none of the functions causes an error, the handleError will never be invoked.

Extension Points

In addition to the (optional) wrapper function which can be invoked prior to the once function, pipeline creators can expose named extension points. These extension points allow users of a pipeline to inject additional functions that will be called right before, right after or instead of an extension point. To keep the extension points independent from the implementation (i.e. the name of the function), pipeline authors should use the expose(name) function to expose a particular extension point.


new pipeline()

In this example, two extension points, init and cleanup have been defined. Note how the name of the extension point can be the same as the name of the function (i.e. cleanup), but does not have to be the same (i.e. init vs. doSomething).

Common Extension Points

The creation of extension points is the responsibility of the pipeline author, but in order to standardize extension points, the following common names have been established:

  • fetch for the pipeline step that retrieves raw content, i.e. a Markdown document
  • parse for the pipeline step that parses the raw content and transforms it into a document structure such as a Markdown AST
  • meta for the pipeline step that extracts metadata from the content structure
  • html for the pipeline step that turns the Markdown document into a (HTML) DOM
  • esi for the pipeline step that scans the generated output for ESI markers and sets appropriate headers

Using Extension Points

The easiest way to use extension points is by expanding on the Wrapper Function described above. Instead of just exporting a pre function, the wrapper can also export:

  • a before object
  • an after object
  • a replace object

Each of these objects can have keys that correspond to the named extension points defined for the pipeline.


module.exports.before = {
  html: (context, action) => {
    // will get called before the "html" pipeline step

module.exports.after = {
  fetch: (context, action) => {
    // will get called after the "fetch" pipeline step

module.exports.replace = {
  meta: (context, action) => {
    // will get called instead of the "meta" pipeline step

All functions that are using the before and after extension points need to follow the same interface that all other pipeline functions follow, i.e. they have access to context and action and they should return a modified context object.

A more complex example of using these extension points to implement custom markdown content nodes and handle 404 errors can be found in the helix-cli integration tests.

Anatomy of the Context

Following main properties exist:

  • request
  • content
  • response
  • error

also see context schema

The request object

  • params: a map of request parameters
  • headers: a map of HTTP headers

also see request schema

The content object

  • body: the unparsed content body as a string
  • mdast: the parsed Markdown AST
  • meta: a map metadata properties, including
    • title: title of the document
    • intro: a plain-text introduction or description
    • type: the content type of the document
    • image: the URL of the first image in the document
  • document: a DOM-compatible Document representation of the (HTML) document (see below)
  • sections[]: The main sections of the document, as an enhanced MDAST (see below)
  • html: a string of the content rendered as HTML
  • children: an array of top-level elements of the HTML-rendered content

also see content schema

content.document in Detail

For developers that prefer using the rendered HTML over the input Markdown AST, content.document provides a representation of the rendered HTML that is API-compatible to the window.document object you would find in a browser.

The most common way of using it is probably calling content.document.innerHTML, which gives the full HTML of the page, but other functions like

  • content.document.getElementsByClassName
  • content.document.querySelector
  • content.document.querySelectorAll

are also available. Please note that some functions like

  • content.document.getElementsByClassName
  • content.document.getElementByID

are less useful because the HTML generated by the default pipeline does not inject class name or ID attributes.

The tooling for generating (Virtual) DOM nodes from Markdown AST is made available as a utility class, so that it can be used in custom pre.js scripts, and described below.

content.sections in Detail

The default pipeline extracts sections from a Markdown document, using both “thematic breaks” like *** or --- and embedded YAML blocks as section markers. If no sections can be found in the document, the entire content.mdast will be identically to content.sections[0].

content.sections is an Array of section nodes, with type (String) and children (array of Node) properties. In addition, each section has a types attribute, which is an array of derived content types. Project Helix (and Hypermedia Pipeline) uses implied typing over declared content typing, which means it is not the task of the author to explicitly declare the content type of a section or document, but rather have the template interpret the contents of a section to understand the type of content it is dealing with.

The types property is an array of string values that describes the type of the section based on the occurrence of child nodes. This makes it easy to copy the value of types into the class attribute of an HTML element, so that CSS expressions matching types of sections can be written with ease. Following patterns of type values can be found:

  • has-<type>: for each type of content that occurs at least once in the section, e.g. has-heading
  • has-only-<type>: for sections that only have content of a single type, e.g. has-only-image
  • is-<type-1>-<type-2>-<type3>, is-<type-1>-<type-2>, and is-<type-1> for the top 3 most frequent types of children in the section. For instance a gallery with a heading and description would be is-image-text-heading. You can infer additional types using utils.types.
  • nb-<type>-<occurences>: number of occurences of each type in the section

Each section has additional content-derived metadata properties, in particular:

  • title: the value of the first headline in the section
  • intro: the value of the first paragraph in the section
  • image: the URL of the first image in the section
  • meta: the parsed YAML metadata of the section (as an object)

The response object

  • body: the unparsed response body as a string
  • headers: a map of HTTP response headers
  • status: the HTTP status code

also see response schema

The error object

This object is only set when there has been an error during pipeline processing. Any step in the pipeline may set the error object. Subsequent steps should simply skip any processing if they encounter an error object.

Alternatively, steps can attempt to handle the error object, for instance by generating a formatted error message and leaving it in response.body.

The only known property in error is

  • message: the error message


Generate a Virtual DOM with utils.vdom

VDOM is a helper class that transforms MDAST Markdown into DOM nodes using customizable matcher functions or expressions.

It can be used in scenarios where:

  • you need to represent only a section of the document in HTML
  • you have made changes to content.mdast and want them reflected in HTML
  • you want to customize the HTML output for certain Markdown elements

Getting Started

In most cases, you can simply access a pre-configured transformer from the action.transformer property. It will be available in the pipeline after the parse step and used in the html step, so registering after.parse and before.html are the ideal points to adjust the generated HTML.

Alternatively, load the VDOM helper through:

const VDOM = require('@adobe/helix-pipeline').utils.vdom;

Simple Transformations

content.document = new VDOM(content.mdast).getDocument();

This replaces content.document with a re-rendered representation of the Markdown AST. It can be used when changes to content.mdast have been made.

When using action.transformer, this manual invocation is not required.

content.document = new VDOM(content.sections[0]).getDocument();

This uses only the content of the first section to render the document.

Matching Nodes

Nodes in the Markdown AST can be matched in two ways: either using a select-statement or using a predicate function.

action.transformer.match('heading', () => '<h1>This text replaces your heading</h1>');
content.document = vdom.getDocument();

Every node with the type heading will be rendered as <h1>This text replaces your heading</h1>;

action.transformer.match(function test(node) {
  return node.type === 'heading';
}, () => '<h1>This text replaces your heading</h1>');
content.document = vdom.getDocument();

Instead of the select-statement, you can also provide a predicate function that returns true or false. The two examples above will have the same behavior.

Creating DOM Nodes

The second argument to match is a node-generating function that should return one of the following three options:

  1. a DOM Node
  2. a String containing HTML tags.
action.transformer.match('link', (_, node) => {
  return {
    type: 'element',
    tagName: 'a',
    properties: {
      href: node.url,
      rel: 'nofollow'
    children: [
        type: 'text',
        value:{ value }) => value)

Above: injecting rel="nofollow" using HTAST.

const h = require('hyperscript');

action.transformer.match('link', (_, node) => h(
    { href: node.url, rel: 'nofollow' },{ value }) => value),

Above: doing the same using Hyperscript (which creates DOM elements) is notably shorter.

action.transformer.match('link', (_, node) => 
  `<a href="${node.url}" rel="nofollow">$({ value }) => value)).join('')</a>`;

Above: Plain Strings can be constructed using String Templates in ES6 for the same result.

Dealing with Child Nodes

The examples above have only been processing terminal nodes (or almost terminal nodes). In reality, you need to make sure that all child nodes of the matched node are processed too. For this, the signature of the handler function provides you with a handlechild function.

// match all "emphasis" nodes
action.transformer.match('emphasis', (h, node, _, handlechild) => {
  // create a new HTML tag <i class="its-not-semantic…">
  const i = h(node, 'i', { className: 'its-not-semantic-html-but-i-like-it' });
  // make sure all child nodes of the markdown node are processed
  node.children.forEach((childnode) => handlechild(h, childnode, node, i));
  // return the i HTML element
  return i;

The handlechild function is called with:

  • h: a DOM-node producing function
  • childnode: the child node to be processed
  • node: the parent node of the node to be processed (usually the current node)
  • i: the DOM node will be the new parent node for newly created DOM nodes

Infer Content Types with utils.types

In addition to the automatically inferred content types for each section, utils.types provides a TypeMatcher utility class that allows matching section content against a simple expression language and thus enrich the section[].types values.

const TypeMatcher = require('@adobe/hypermedia-pipeline').utils.types;

const matcher = new TypeMatcher(content.sections);
matcher.match('^heading', 'starts-with-heading');
content.sections = matcher.process();

In the example above, all sections that have a heading as the first child will get the value starts-with-heading appended to the types array. ^heading is an example of the content expression language, which allows matching content against a simple regular expression-like syntax.

Content Expression Language
  • ^heading – the first element is a heading
  • paragraph$ – the last element is a paragraph
  • heading image+ – a heading followed by one or more images
  • heading? image – an optional heading followed by one image
  • heading paragraph* image – a heading followed by any number of paragraphs (also no paragraphs at all), followed by an image
  • (paragraph|list) – a paragraph or a list
  • ^heading (image paragraph)+$ – one heading, followed by pairs of image and paragraph, but at least one

Inspecting the Pipeline Context

When run in non-production, i.e. outside an OpenWhisk action, for example in hlx up, Pipeline Dumping is enabled. Pipeline Dumping allows developers to easily inspect the Context object of each step of the pipeline and can be used to debug pipeline functions and to generate realistic test cases.

Each stage of the pipeline processing will create a file like $PWD/logs/debug/context_dump_34161BE5KuR0nuFDp/context-20180902-1418-05.0635-step-2.json inside the debug directory. These dumps will be removed when the node process ends, so that after stopping hlx up the debug directory will be clean again. The -step-n in the filename indicates the step in the pipeline that has been logged.

A simple example might look like this:

Step 0:


Step 1:

  "request": {}

Step 2:

  "request": {},
  "content": {
    "body": "---\ntemplate: Medium\n---\n\n# Bill, Welcome to the future\n> Project Helix\n\n## Let's talk about Project Helix\n![](./moscow/assets/IMG_0167.jpg)\n",
    "sources": [

Step 3 (diff only):

@@ -1,6 +1,58 @@
   "content": {
-    "body": "Hello World"
+    "body": "Hello World",
+    "mdast": {
+      "type": "root",
+      "children": [
+        {
+          "type": "paragraph",
+          "children": [
+            {
+              "type": "text",
+              "value": "Hello World",
+              "position": {
+                "start": {
+                  "line": 1,
+                  "column": 1,
+                  "offset": 0
+                },
+                "end": {
+                  "line": 1,
+                  "column": 12,
+                  "offset": 11
+                },
+                "indent": []
+              }
+            }
+          ],
+          "position": {
+            "start": {
+              "line": 1,
+              "column": 1,
+              "offset": 0
+            },
+            "end": {
+              "line": 1,
+              "column": 12,
+              "offset": 11
+            },
+            "indent": []
+          }
+        }
+      ],
+      "position": {
+        "start": {
+          "line": 1,
+          "column": 1,
+          "offset": 0
+        },
+        "end": {
+          "line": 1,
+          "column": 12,
+          "offset": 11
+        }
+      }
+    }
   "request": {}

Step 5 (diff only):

@@ -52,7 +52,49 @@
           "offset": 11
-    }
+    },
+    "sections": [
+      {
+        "type": "root",
+        "children": [
+          {
+            "type": "paragraph",
+            "children": [
+              {
+                "type": "text",
+                "value": "Hello World",
+                "position": {
+                  "start": {
+                    "line": 1,
+                    "column": 1,
+                    "offset": 0
+                  },
+                  "end": {
+                    "line": 1,
+                    "column": 12,
+                    "offset": 11
+                  },
+                  "indent": []
+                }
+              }
+            ],
+            "position": {
+              "start": {
+                "line": 1,
+                "column": 1,
+                "offset": 0
+              },
+              "end": {
+                "line": 1,
+                "column": 12,
+                "offset": 11
+              },
+              "indent": []
+            }
+          }
+        ]
+      }
+    ]
   "request": {}

Step 6 (diff only):

@@ -92,9 +92,19 @@
              "indent": []
-        ]
+        ],
+        "title": "Hello World",
+        "types": [
+          "has-text",
+          "has-only-text"
+        ],
+        "intro": "Hello World",
+        "meta": {}
-    ]
+    ],
+    "meta": {},
+    "title": "Hello World",
+    "intro": "Hello World"
  "request": {}

Step 9 (diff only):

@@ -169,7 +169,11 @@
        "search": "",
        "hash": ""
-    }
+    },
+    "html": "<p>Hello World</p>",
+    "children": [
+      "<p>Hello World</p>"
+    ]
  "request": {}

Step 10 (diff only):

@@ -175,5 +175,9 @@
      "<p>Hello World</p>"
-  "request": {}
+  "request": {},
+  "response": {
+    "status": 201,
+    "body": "<p>Hello World</p>"
+  }