#CSVTOJSON

Nodejs csv to json converter. Fully featured:

• Pipe in / Pipe out
• Use as Command Line tool or a Node.js lib
• Parse CSV to JSON or CSV column arrays
• Support all types of CSV
• Non-blocking parsing / multi core support
• Extremely fast: 4 - 6 times faster than other csv parsers on node.js
• Streaming data / low memory usage on large CSV data source

Version 1.1.0 has added new features and optimised lib performance. It also introduced simpler APIs to use. Thus readme is re-written to adapt the preferred new APIs. The lib will support old APIs. To review the old readme please click here.

• Performance Optimisation: V1.1.0 is 30%-50% faster
• Better error tolerance
• Simplified API (see below)

All changes are backward compatible.

Here is a free online csv to json service ultilising latest csvtojson module.

• Quick Start
• API

• As Library
• As Command Line Tool

npm i --save csvtojson

/**
csvStr:
1,2,3
4,5,6
7,8,9
*/
const csv=require('csvtojson')
csv({noheader:true})
.fromString(csvStr)
.on('csv',(csvRow)=>{ // this func will be called 3 times
	console.log(csvRow) // => [1,2,3] , [4,5,6]  , [7,8,9]
})
.on('done',()=>{
	//parsing finished
})

/** csv file
a,b,c
1,2,3
4,5,6
*/
const csvFilePath='<path to csv file>'
const csv=require('csvtojson')
csv()
.fromFile(csvFilePath)
.on('json',(jsonObj)=>{
	// combine csv header row and csv line to a json object
	// jsonObj.a ==> 1 or 4
})
.on('done',(error)=>{
	console.log('end')
})

//const csvReadStream -- Readable stream for csv source
const csv=require('csvtojson')

csv()
.fromStream(csvReadStream)
.on('csv',(csvRow)=>{
	// csvRow is an array
})
.on('done',(error)=>{

})

/**
csvStr:
a,b,c
1,2,3
4,5,6
*/

const csv=require('csvtojson')
csv()
.fromString(csvStr)
.on('csv',(csvRow)=>{ //this func will be called twice. Header row will not be populated
	// csvRow =>  [1,2,3] and [4,5,6]
})
.on('done',()=>{
	console.log('end')
})

/**
csvStr:
1,2,3
4,5,6
7,8,9
*/

const csv=require('csvtojson')
csv({noheader:true})
.fromString(csvStr)
.on('json',(json)=>{ //this func will be called 3 times
	// json.field1 => 1,4,7
	// json.field2 => 2,5,8
	// json.field3 => 3,6,9
})
.on('done',()=>{
	console.log('end')
})

$ npm i -g csvtojson

$ csvtojson [options] <csv file path>

Convert csv file and save result to json file:

$ csvtojson source.csv > converted.json

Use multiple cpu-cores:

$ csvtojson --workerNum=4 source.csv > converted.json

Pipe in csv data:

$ cat ./source.csv | csvtojson > converted.json

Print Help:

$ csvtojson

const csv=require('csvtojson')
const converter=csv(params) //params see below Parameters section

In above, converter is an instance of Converter which is a subclass of node.js Transform class.

• Parameters
• Events
• Hook / Transform
• Nested JSON Structure
• Header Row
• Multi CPU Core Support

require('csvtojson') returns a constructor function which takes 2 arguments:

1. parser parameters
2. Stream options

const csv=require('csvtojson')
const converter=csv(parserParameters, streamOptions)

Both arguments are optional.

For Stream Options please read Stream Option from Node.JS

parserParameters is a JSON object like:

const converter=csv({
	noheader:true,
	trim:true,
})

Following parameters are supported:

• delimiter: delimiter used for seperating columns. Use "auto" if delimiter is unknown in advance, in this case, delimiter will be auto-detected (by best attempt). Use an array to give a list of potential delimiters e.g. [",","|","$"]. default: ","
• quote: If a column contains delimiter, it is able to use quote character to surround the column content. e.g. "hello, world" wont be split into two columns while parsing. Set to "off" will ignore all quotes. default: " (double quote)
• trim: Indicate if parser trim off spaces surrounding column content. e.g. " content " will be trimmed to "content". Default: true
• checkType: This parameter turns on and off whether check field type. default is true.
• toArrayString: Stringify the stream output to JSON array. This is useful when pipe output to a file which expects stringified JSON array. default is false and only stringified JSON (without []) will be pushed to downstream.
• ignoreEmpty: Ignore the empty value in CSV columns. If a column value is not giving, set this to true to skip them. Defalut: false.
• workerNum: Number of worker processes. The worker process will use multi-cores to help process CSV data. Set to number of Core to improve the performance of processing large csv file. Keep 1 for small csv files. Default 1.
• noheader:Indicating csv data has no header row and first row is data row. Default is false. See header row
• headers: An array to specify the headers of CSV data. If --noheader is false, this value will override CSV header row. Default: null. Example: ["my field","name"]. See header row
• flatKeys: Don't interpret dots (.) and square brackets in header fields as nested object or array identifiers at all (treat them like regular characters for JSON field identifiers). Default: false.
• maxRowLength: the max character a csv row could have. 0 means infinite. If max number exceeded, parser will emit "error" of "row_exceed". if a possibly corrupted csv data provided, give it a number like 65535 so the parser wont consume memory. default: 0
• checkColumn: whether check column number of a row is the same as headers. If column number mismatched headers number, an error of "mismatched_column" will be emitted.. default: false
• eol: End of line character. If omitted, parser will attempt retrieve it from first chunk of CSV data. If no valid eol found, then operation system eol will be used.
• escape: escape character used in quoted column. Default is double quote (") according to RFC4108. Change to back slash (\) or other chars for your own case.
• includeColumns: This parameter instructs the parser to include only those columns as specified by an array of column indexes. Example: [0,2,3] will parse and include only columns 0, 2, and 3 in the JSON output.
• ignoreColumns: This parameter instructs the parser to ignore columns as specified by an array of column indexes. Example: [1,3,5] will ignore columns 1, 3, and 5 and will not return them in the JSON output.

All parameters can be used in Command Line tool.

Converter class defined a series of events.

json event is emitted for each parsed CSV line. It passes JSON object and the row number of the CSV line in its callback function.

const csv=require('csvtojson')
csv()
.on('json',(jsonObj, rowIndex)=>{
	//jsonObj=> {header1:cell1,header2:cell2}
	//rowIndex=> number
})

csv event is emitted for each CSV line. It passes an array object which contains cells content of one csv row.

const csv=require('csvtojson')
csv()
.on('csv',(csvRow, rowIndex)=>{
	//csvRow=> [cell1, cell2, cell3]
	//rowIndex=> number
})

csvRow is always an array of strings no matter checkType value.

csv event is the fastest parse event while json and data event is about 2 times slower. Thus if csv is enough, for best performance, just use it without json and data event.

data event is emitted for each parsed CSV line. It passes buffer of strigified JSON unless objectMode is set true in stream option.

const csv=require('csvtojson')
csv()
.on('data',(data)=>{
	//data is a buffer object
	const jsonStr= data.toString('utf8')
})

error event is emitted if there is any errors happened during parsing.

const csv=require('csvtojson')
csv()
.on('error',(err)=>{
	console.log(err)
})

Note that if error being emitted, the process will stop as node.js will automatically unpipe() upper-stream and chained down-stream1. This will cause end / end_parsed event never being emitted because end event is only emitted when all data being consumed 2.

1. Node.JS Readable Stream
2. Writable end Event

record_parsed event is emitted for each parsed CSV line. It is combination of json and csv events. For better performance, try to use json and csv instead.

const csv=require('csvtojson')
csv()
.on('record_parsed',(jsonObj, row, index)=>{
})

end event is emitted when all CSV lines being parsed.

end_parsed event is emitted when all CSV lines being parsed. The only difference between end_parsed and end events is end_parsed will pass in a JSON array which contains all JSON objects. For better performance, try to use end event instead.

const csv=require('csvtojson')
csv()
.on('end_parsed',(jsonArrObj)=>{
})

done event is emitted either after end or error. This indicates the processor has stopped.

const csv=require('csvtojson')
csv()
.on('done',(error)=>{
	//do some stuff
})

if any error during parsing, it will be passed in callback.

const csv=require('csvtojson')
csv()
.preRawData((csvRawData,cb)=>{
	var newData=csvRawData.replace('some value','another value')
	cb(newData);
})
.on('json',(jsonObj)=>{

});

the function in preRawData will be called directly with the string from upper stream.

const csv=require('csvtojson')
csv()
.preFileLine((fileLineString, lineIdx)=>{
	if (lineIdx === 2){
		return fileLineString.replace('some value','another value')
	}
	return fileLineString
})
.on('json',(jsonObj)=>{

});

the function is called each time a file line being found in csv stream. the lineIdx is the file line number in the file. The function should return a string to processor.

const csv=require('csvtojson')
csv()
.transf((jsonObj,csvRow,index)=>{
	jsonObj.myNewKey='some value'
})
.on('json',(jsonObj)=>{
	console.log(jsonObj.myNewKey) // some value
});

Transform happens after CSV being parsed before result being emitted or pushed to downstream. This means if jsonObj is changed, the corresponding field in csvRow will not change. Vice versa. The events will emit changed value and downstream will receive changed value.

Transform will cause some performance panelties because it voids optimisation mechanism. Try to use Node.js Transform class as downstream for transformation instead.

One of the powerful feature of csvtojson is the ability to convert csv line to a nested JSON by correctly defining its csv header row. This is default out-of-box feature.

Here is an example. Original CSV:

fieldA.title, fieldA.children.0.name, fieldA.children.0.id,fieldA.children.1.name, fieldA.children.1.employee.0.name,fieldA.children.1.employee.1.name, fieldA.address.0,fieldA.address.1, description
Food Factory, Oscar, 0023, Tikka, Tim, Joe, 3 Lame Road, Grantstown, A fresh new food factory
Kindom Garden, Ceil, 54, Pillow, Amst, Tom, 24 Shaker Street, HelloTown, Awesome castle

The data above contains nested JSON including nested array of JSON objects and plain texts.

Using csvtojson to convert, the result would be like:

[{
    "fieldA": {
        "title": "Food Factory",
        "children": [{
            "name": "Oscar",
            "id": "0023"
        }, {
            "name": "Tikka",
            "employee": [{
                "name": "Tim"
            }, {
                "name": "Joe"
            }]
        }],
        "address": ["3 Lame Road", "Grantstown"]
    },
    "description": "A fresh new food factory"
}, {
    "fieldA": {
        "title": "Kindom Garden",
        "children": [{
            "name": "Ceil",
            "id": "54"
        }, {
            "name": "Pillow",
            "employee": [{
                "name": "Amst"
            }, {
                "name": "Tom"
            }]
        }],
        "address": ["24 Shaker Street", "HelloTown"]
    },
    "description": "Awesome castle"
}]

In case to not produce nested JSON, simply set flatKeys:true in parameters.

/**
csvStr:
a.b,a.c
1,2
*/
csv({flatKeys:true})
.fromString(csvStr)
.on('json',(jsonObj)=>{
	//{"a.b":1,"a.c":2}  rather than  {"a":{"b":1,"c":2}}
});

csvtojson uses csv header row as generator of JSON keys. However, it does not require the csv source containing a header row. There are 4 ways to define header rows:

1. First row of csv source. Use first row of csv source as header row. This is default.
2. If first row of csv source is header row but it is incorrect and need to be replaced. Use headers:[] and noheader:false parameters.
3. If original csv source has no header row but the header definition can be defined. Use headers:[] and noheader:true parameters.
4. If original csv source has no header row and the header definition is unknow. Use noheader:true. This will automatically add fieldN header to csv cells

// replace header row (first row) from original source with 'header1, header2'
csv({
	noheader: false,
	headers: ['header1','header2']
})

// original source has no header row. add 'field1' 'field2' ... 'fieldN' as csv header
csv({
	noheader: true
})

// original source has no header row. use 'header1' 'header2' as its header row
csv({
	noheader: true
	headers: ['header1','header2']
})

csvtojson has built-in workers to allow CSV parsing happening on another process and leave Main Process non-blocked. This is very useful when dealing with large csv data on a webserver so that parsing CSV will not block the entire server due to node.js being single threaded.

It is also useful when dealing with tons of CSV data on command line. Multi-CPU core support will dramatically reduce the time needed.

To enable multi-cpu core, simply do:

csv({
	workerNum:4  // workerNum>=1
})

or in command line:

$ csvtojson --workerNum=4

This will create 3 extra workers. Main process will only be used for delegating data / emitting result / pushing to downstream. Just keep in mind, those operations on Main process are not free and it will still take a certain amount CPU time.

See here for how csvtojson leverages CPU usage when using multi-cores.

There are some limitations when using multi-core feature:

• Does not support if a column contains line break.

#Change Log

• Fix bugs: preProcessLine is not emitted
• Changed array definition in nested json structure to follow [lodash set] (https://lodash.com/docs/4.17.2#set)
• Only use first line of csv body for type inference
• added done event
• added hooks section
• removed parserMgr

• Remove support of new Converter(true)
• Optimised Performance
• Added new APIs

• supported ndjson format as per #113 and #87
• issue: #120

• Add Stream Options
• Change version syntax to follow x.y.z

• Added support for scientific notation number support (#100)
• Added "off" option to quote parameter

• Added new feature: accept special delimiter "auto" and array

• Changed type separator from # to #!
• Fixed bugs

• Fixed some bugs
• Performance improvement
• Implicity type for numbers now use RegExp:/^[-+]?[0-9]*.?[0-9]+$/. Previously 00131 is a string now will be recognised as number type
• If a column has no head, now it will use current column index as column name: 'field'. previously parser uses a fixed index starting from 1. e.g. csv data: 'aa,bb,cc' with head 'a,b'. previously it will convert to {'a':'aa','b':'bb','field1':'cc'} and now it is {'a':'aa','b':'bb','field3':'cc'}*

• ignoreEmpty now ignores empty rows as well
• optimised performance
• added fromFile method

• Add error handling for corrupted CSV data
• Exposed "eol" param

• Added header configuration
• Refactored worker code
• Number type field now returns 0 if parseFloat returns NaN with the value of the field. Previously it returns original value in string.

• Added Multi-core CPU support to increase performance
• Added "fork" option to delegate csv converting work to another process.
• Refactoring general flow

• Refactored Command Line Tool.
• Added ignoreEmpty parameter.

• Fixed double qoute parse as per CSV standard.

• Added field type support
• Fixed some minor bugs

• Empowered built-in JSON parser.
• Change: Use JSON parser as default parser.
• Added parameter trim in constructor. default: true. trim will trim content spaces.

• Added fromString method to support direct string input

• Added more parameters to command line tool.

• Added quote in parameter to support quoted column content containing delimiters
• Changed row index starting from 0 instead of 1 when populated from record_parsed event

• Removed all dependencies
• Deprecated applyWebServer
• Added construct parameter for Converter Class
• Converter Class now works as a proper stream object