frictionless.js is a lightweight, standardized "stream-plus-metadata" interface for accessing files and datasets, especially tabular ones (CSV, Excel).
frictionless.js follows the "Frictionless Data Lib Pattern".
- Open it fast: simple
openmethod for data on disk, online and inline - Data plus: data plus metadata (size, path, etc) in standardized way
- Stream it: raw streams and object streams
- Tabular: open CSV, Excel or arrays and get a row stream
- Frictionless: compatible with Frictionless Data standards
A line of code is worth a thousand words ...
const {open} = require('frictionless.js')
var file = open('path/to/ons-mye-population-totals.xls')
file.descriptor
{
path: '/path/to/ons-mye-population-totals.xls',
pathType: 'local',
name: 'ons-mye-population-totals',
format: 'xls',
mediatype: 'application/vnd.ms-excel',
encoding: 'windows-1252'
}
file.size
67584
file.rows() => stream object for rows
// keyed by header row by default ...
{ 'col1': 1, 'col2': 2, ... }
{ 'col1': 10, 'col2': 20, ... }
Table of Contents
frictionless.js is motivated by the following use cases:
- Data "plus": when you work with data you always find yourself needing the data itself plus a little bit more -- things like where the data came from on disk (or is going to), or how large it is. This library gives you that information in a standardized way.
- Convenient open: the same simple
openmethod whether you are accessing data on disk, from a URL or inline data from a string, buffer or array. - Streams (or strings): standardized iterator / object stream interface to data wherever you've loaded from online, on disk or inline
- Building block for data pipelines: provides a standardized building block for more complex data processing. For example, suppose you want to load a csv file then write to JSON. That's simple enough. But then suppose you want to delete the first 3 rows, delete the 2nd column. Now you have a more complex processing pipeline. This library provides a simple set of "data plus metadata" objects that you can pass along your pipeline.
- Easy: a single common API for local, online and inline data
- Micro: The whole project is ~400 lines of code
- Simple: Oriented for single purpose
- Explicit: No hidden behaviours, no extra magic
- Frictionlesss: uses and supports (but does not require) Frictionless Data specs such as Data Package so you can leverage Frictionless tooling
- Minimal glue: Use on its own or as a building block for more complex data tooling (thanks to its common minimal metadata)
npm install frictionless.js
If you want to use the it in the browser, first you need to build the bundle.
Run the following command to generate the bundle for the necessary JS targets
yarn build
This will create two bundles in the dist folder. node sub-folder contains build for node environment, while browser sub-folder contains build for the browser. In a simple html file you can use it like this:
<head>
<script src="./dist/browser/bundle.js"></script>
<script>
// Global data lib is available here...
const file = data.open('path/to/file')
...
</script>
</head>
<body></body>With a simple file:
const data = require('frictionless.js')
// path can be local or remote
const file = data.open(path)
// descriptor with metadata e.g. name, path, format, (guessed) mimetype etc
console.log(file.descriptor)
// returns promise with raw stream
const stream = await file.stream()
// let's get an object stream of the rows
// (assuming it is tabular i.e. csv, xls etc)
const rows = await file.rows()
// entire file as a buffer
const buffer = await file.buffer
//for large files you can return in chunks
await file.bufferInChunks((chunk, progress)=>{
console.log(progress, chunk)
})
With a Dataset:
const { Dataset } = require('frictionless.js')
const path = '/path/to/directory/' // must have datapackage.json in the directory atm
Dataset.load(path).then(dataset => {
// get a data file in this dataset
const file = dataset.resources[0]
const data = file.stream()
})Load a file from a path or descriptor.
load(pathOrDescriptor, {basePath, format}={})
There are 3 types of file source we support:
- Local path
- Remote url
- Inline data
const data = require('frictionless.js')
const file = data.open('/path/to/file.csv')
const file = data.open('https://example.com/data.xls')
// loading raw data
const file = data.open({
name: 'mydata',
data: { // can be any javascript - an object, an array or a string or ...
a: 1,
b: 2
}
})
// Loading with a descriptor - this allows more fine-grained configuration
// The descriptor should follow the Frictionless Data Resource model
// http://specs.frictionlessdata.io/data-resource/
const file = data.open({
// file or url path
path: 'https://example.com/data.csv',
// a Table Schema - https://specs.frictionlessdata.io/table-schema/
schema: {
fields: [
...
]
}
// CSV dialect - https://specs.frictionlessdata.io/csv-dialect/
dialect: {
// this is tab separated CSV/DSV
delimiter: '\t'
}
})basePath: use in cases where you want to create a File with a path that is relative to a base directory / path e.g.
const file = data.open('data.csv', {basePath: '/my/base/path'})
Will open the file: /my/base/path/data.csv
This functionality is primarily useful when using Files as part of Datasets where it can be convenient for a File to have a path relative to the directory of the Dataset. (See also Data Package and Data Resource in the Frictionless Data specs).
A single data file - local or remote.
DEPRECATED. Use simple open.
Main metadata is available via the descriptor:
file.descriptor
This metadata is a combination of the metadata passed in at File creation (if you created the File with a descriptor object) and auto-inferred information from the File path. This is the info that is auto-inferred:
path: path this was instantiated with - may not be same as file.path (depending on basePath)
pathType: remote | local
name: file name (without extension)
format: the extension
mediatype: mimetype based on file name and extension
In addition to this metadata there are certain properties which are computed on demand:
// the full path to the file (using basepath)
const path = file.path
const size = file.size
// md5 hash of the file
const hash = file.hash()
// sha256 hash of the file
const hash256 = file.hash(hashType='sha256')
// file encoding
const encoding = file.encodingNote: size, hash are not available for remote Files (those created from urls).
stream()
Get readable stream
@returns Promise with readable stream object on resolve
File.buffer
Get this file as a buffer (async)
@returns: promise which resolves to the buffer
rows({keyed}={})
Get the rows for this file as a node object stream (assumes underlying data is tabular!)
@returns Promise with rows as parsed JS objects (depends on file format)
keyed: iffalse(default) returns rows as arrays. Iftruereturns rows as objects.
TODO: casting (does data get cast automatically for you or not ...)
What formats are supported?
The rows functionality is currently available for CSV and Excel files. The Tabular support incorporates supports for Table Schema and CSV Dialect e.g. you can do:
// load a CSV with a non-standard dialect e.g. tab separated or semi-colon separated
const file = data.open({
path: 'mydata.tsv'
// Full support for http://specs.frictionlessdata.io/csv-dialect/
dialect: {
delimiter: '\t' // for tabs or ';' for semi-colons etc
}
})
// open a CSV with a Table Schema
const file = data.open({
path: 'mydata.csv'
// Full support for Table Schema https://specs.frictionlessdata.io/table-schema/
schema: {
fields: [
{
name: 'Column 1',
type: 'integer'
},
...
]
}
})A collection of data files with optional metadata.
Under the hood it heavily uses Data Package formats and it natively supports Data Package formats including loading from datapackage.json files. However, it does not require knowledge or use of Data Packages.
A Dataset has four primary properties:
descriptor: key metadata. The descriptor follows the Data Package specresources: an array of the Files contained in this Datasetidentifier: the identifier encapsulates the location (or origin) of this Datasetreadme: the README for this Dataset (if it exists). The readme content is taken from the README.md file located in the Dataset root directory, or, if that does not exist from thereadmeproperty on the descriptor. If neither of those exist the readme will be undefined or null.
In addition we provide the convenience attributes:
path: the path (remote or local) to this datasetdataPackageJsonPath: the path to thedatapackage.jsonfor this Dataset (if it exists)
To create a new Dataset object use Dataset.load. It takes descriptor Object or identifier string:
async Dataset.load(pathOrDescriptor, {owner = null} = {})
pathOrDescriptor- can be one of:- local path to Dataset
- remote url to Dataset
- descriptor object
- @returns: a fully loaded Dataset (parsed and used
datapackage.jsonandREADME.md-- if README exists)
For example:
const data = require('frictionless.js')
const pathOrDescriptor = 'https://raw.githubusercontent.com/datasets/co2-ppm/master/datapackage.json'
const dataset = await data.Dataset.load(pathOrDescriptor)Add a resource to the Dataset:
addResource(resource)resource: may be an already instantiated File object or it is a resource descriptor- @returns: null
// seeks to guess whether a given path is the path to a Dataset or a File
// (i.e. a directory or datapackage.json)
data.isDataset(path)// parses dataset path and returns identifier dictionary
// handles local paths, remote URLs as well as DataHub and GitHub specific URLs
// (e.g., https://datahub.io/core/finance-vix or https://github.com/datasets/finance-vix
const identifier = data.parseDatasetIdentifier(path)
console.log(identifier)and it prints out:
{
name: <name>,
owner: <owner>,
path: <path>,
type: <type>,
original: <path>,
version: <version>
}
Requirements:
- NodeJS >= v8.10.0
- NPM >= v5.2.0
We have two type of tests Karma based for browser testing and Mocha with Chai for Node. All node tests are in datajs/test folder. Since Mocha is sensitive to test namings, we have separate the folder /browser-test for only Karma.
- To run browser test, first you need to build the library in order to have the bundle in
dist/browserfolder. Run:yarn build:browserto achieve this, then for browser testing use the commandyarn test:browser, this will run Karma tests. - To test in Node:
yarn test:node - To run all tests including Node and browser run
yarn test - To watch Node test run:
yarn test:node:watch
-
Git clone the repo
-
Install dependencies:
yarn -
To make the browser and node test work, first run the build:
yarn build -
Run tests:
yarn test -
Do some dev work
-
Once done, make sure tests are passing. Then build distribution version of the app -
yarn build.Run
yarn buildto compile using webpack and babel for different node and web target. To watch the build run:yarn build:watch. -
Now proceed to "Deployment" stage
- Update version number in
package.json. - Git commit:
git commit -m "some message, eg, version". - Release:
git tag -a v0.12.0 -m "some message". - Push:
git push origin master --tags - Publish to NPM:
npm publish