Exciting news!Announcing our $4.6M Series Seed. Learn more
Log in


Package Overview
File Explorer

Advanced tools


SAX based evented streaming JSON parser in Typescript (browser and node)


Version published
Yearly downloads
decreased by-65.37%

Weekly downloads




NPM Downloads NPM Version

bass-clarinet is a JSON parser.

It was forked from clarinet but the API has been changed significantly. In addition to the port to TypeScript, the following changes have been made:

  • The parser was made as robust as possible. It will try to continue parsing after an unexpected event. This is useful for editors as documents will often be in an invalid state during editing.
  • onopenobject no longer includes the first key
  • JSONTestSuite is added to the test set. All tests pass.
  • line and column information is fixed
  • the parser accepts multiple subscribers per event type
  • trim and normalize options have been dropped. This can be handled by the consumer in the onsimplevalue callback
  • there is a stack based wrapper named createStackedDataSubscriber which pairs onopenobject/oncloseobject and onopenarray/onclosearray events in a callback
  • the following features have been added (to disallow them, attach the strictJSON validator attachStictJSONValidator to the parser):
    • angle brackets instead of brackets
    • apostrophes instead of quotation marks
    • comments
    • compact
    • missing commas
    • parens instead of braces
    • schema
    • trailing commas
    • tagged unions
    • tokenizer option: spaces_per_tab
  • stream support has been dropped for now. Can be added back upon request
  • There is an 'ExpectContext' class that helps processing documents that should conform to an expected structure.

bass-clarinet is a sax-like streaming parser for JSON. works in the browser and node.js. just like you shouldn't use sax when you need dom you shouldn't use bass-clarinet when you need JSON.parse.

Clear reasons to use bass-clarinet over the built-in JSON.parse:

  • you want location info
  • you want the parser to continue after it encountered an error
  • you work with very large files
  • you want a syntax that is less strict than JSON. This might be desirable when the file needs to be edited manually. See the options below

design goals

bass-clarinet is very much like yajl but written in TypeScript:

  • written in TypeScript
  • portable
  • no runtime dependency on other modules
  • robust (around 400 tests)
  • data representation independent
  • fast
  • generates verbose, useful error messages including context of where the error occurs in the input text.
  • simple to use
  • tiny



  1. install npm
  2. npm install bass-clarinet
  3. add this to your .ts file: import * as bc from "bass-clarinet"


high level

//a simple pretty printer import * as p from "pareto" import * as fs from "fs" import * as bc from "bass-clarinet" const [, , path] = process.argv if (path === undefined) { console.error("missing path") process.exit(1) } const dataAsString = fs.readFileSync(path, { encoding: "utf-8" }) function createRequiredValuesPrettyPrinter(indentation: string, writer: (str: string) => void): bc.RequiredValueHandler { return { onValue: createValuesPrettyPrinter(indentation, writer), onMissing: () => { //write out an empty string to fix this missing data? }, } } function createValuesPrettyPrinter(indentation: string, writer: (str: string) => void): bc.OnValue { return () => { return { array: (beginRange, beginMetaData) => { writer(beginMetaData.openCharacter) return { element: () => createValuesPrettyPrinter(`${indentation}\t`, writer), end: _endRange => { writer(`${indentation}${bc.printRange(beginRange)}`) }, } }, object: (_beginRange, data) => { writer(data.openCharacter) return { property: (_keyRange, key) => { writer(`${indentation}\t"${key}": `) return p.value(createRequiredValuesPrettyPrinter(`${indentation}\t`, writer)) }, end: endRange => { writer(`${indentation}${bc.printRange(endRange)}`) }, } }, simpleValue: (_range, data) => { if (data.quote !== null) { writer(`${JSON.stringify(data.value)}`) } else { writer(`${data.value}`) } return p.value(false) }, taggedUnion: () => { return { option: (_range, option) => { writer(`| "${option}" `) return createRequiredValuesPrettyPrinter(`${indentation}`, writer) }, missingOption: () => { // }, } }, } } } export function createPrettyPrinter(indentation: string, writer: (str: string) => void): bc.ParserEventConsumer<null, null> { const datasubscriber = bc.createStackedDataSubscriber<null, null>( { onValue: createValuesPrettyPrinter(indentation, writer), onMissing: () => { // }, }, error => { console.error("FOUND STACKED DATA ERROR", error.message) }, () => { //onEnd //no need to return an value, we're only here for the side effects, so return 'null' return p.success(null) } ) return datasubscriber } const pp = createPrettyPrinter("\r\n", str => process.stdout.write(str)) bc.parseString( dataAsString, () => { return pp }, () => { return pp }, err => { console.error("FOUND ERROR", err) }, () => { return p.value(false) }, ).handle( () => { //we're only here for the side effects, so no need to handle the error }, () => { //we're only here for the side effects, so no need to handle the result (which is 'null' anyway) } )

low level

import * as p from "pareto" import * as p20 from "pareto-20" import * as fs from "fs" import * as bc from "bass-clarinet" function assertUnreachable<RT>(_x: never): RT { throw new Error("unreachable") } const [, , path] = process.argv if (path === undefined) { console.error("missing path") process.exit(1) } const dataAsString = fs.readFileSync(path, { encoding: "utf-8" }) export const parserEventConsumer: bc.ParserEventConsumer<null, null> = { onData: data => { switch (data.type[0]) { case bc.BodyEventType.CloseArray: { //const $ = data.type[1] //place your code here break } case bc.BodyEventType.CloseObject: { //const $ = data.type[1] //place your code here break } case bc.BodyEventType.Colon: { //const $ = data.type[1] //place your code here break } case bc.BodyEventType.Comma: { //const $ = data.type[1] //place your code here break } case bc.BodyEventType.OpenArray: { //const $ = data.type[1] //place your code here break } case bc.BodyEventType.OpenObject: { //const $ = data.type[1] //place your code here break } case bc.BodyEventType.Overhead: { const $ = data.type[1] switch ($.type[0]) { case bc.OverheadTokenType.BlockComment: { //const $ = data.type[1] //place your code here break } case bc.OverheadTokenType.LineComment: { //const $ = data.type[1] //place your code here break } case bc.OverheadTokenType.NewLine: { //const $ = data.type[1] //place your code here break } case bc.OverheadTokenType.WhiteSpace: { //const $ = data.type[1] //place your code here break } default: assertUnreachable($.type[0]) } break } case bc.BodyEventType.SimpleValue: { //const $ = data.type[1] //place your code here //in strict JSON, the value is a string, a number, null, true or false break } case bc.BodyEventType.TaggedUnion: { //const $ = data.type[1] //place your code here break } default: assertUnreachable(data.type[0]) } return p.value(false) }, onEnd: () => { //place your code here return p.success(null) }, } const parserStack = bc.createParserStack( () => { return parserEventConsumer }, () => { return parserEventConsumer }, err => { console.error("FOUND ERROR", err) }, () => { return p.value(false) } ) p20.createArray([dataAsString]).streamify().handle( null, parserStack )


pass the following argument to the tokenizer function:

  • spaces_per_tab - number. needed for proper column info.: Rationale: without knowing how many spaces per tab base-clarinet is not able to determine the colomn of a character. Default is 4 (ofcourse)

pass the following arguments to the parser function. all are optional.

opt - object bag of settings.


write - write bytes to the tokenizer. you don't have to do this all at once. you can keep writing as much as you want.

end - ends the stream. once ended, no more data may be written, it signals the onend event.

additional features

the parser supports the following additional (to JSON) features

  • optional commas - No comma's are required. Rationale: When manually editing documents, keeping track of the comma's is cumbersome. With this option this is no longer an issue
  • trailing commas - Allows commas before the } or the ]. Rationale: for serializers it is easier to write a comma for every property/element instead of keeping a state that tracks if a property/element is the first one.
  • comments - Allows both line comments // and block comments /* */. Rationale: when using JSON-like documents for editing, it is often useful to add comments
  • apostrophes instead of quotation marks - Allows ' in place of ". Rationale: In an editor this is less intrusive (although only slightly)
  • angle brackets instead of brackets - Allows < and > in place of [ and ]. Rationale: a semantic distinction can be made between fixed length arrays (ArrayType) and variable length arrays (lists)
  • parens instead of braces - Allows ( and ) in place of { and }. Rationale: a semantic distinction can be made between objctes with known properties (Type) and objects with dynamic keys (dictionary)
  • schema - The document may start with a ! followed by a value (object, string etc), followed by an optional # (indicating compact).
      • The schema value can be used by a processor for schema validation. For example a string can indicate a URL of the schema.
      • compact is an indicator for a processor (code that uses bass-clarinet's API) that the data is compact. base-clarinet only sends the compact flag but does not change any other behaviour. Rationale: If a schema is known, the keys of a Type are known at design time. these types can therefor be converted to ArrayTypes and thus omit the keys without losing information. This trades in readability in favor of size. This option indicates that this happened in this document. The file can only be properly interpreted by a processor in combination with the schema.
  • tagged unions - This allows an extra value type that is not present in JSON but is very useful. tagged unions are also known as sum types or choices, see taggedunion. The notation is a pipe, followed by a string, followed by any other value. eg: | "the chosen option" { "my data": "foo" }. The same information can ofcourse also be written in strict JSON with an array with 2 elements of which the first element is a string.


onerror (passed as argument to the constructor) - indication that something bad happened. The parser will continue as good as it can

the data subscriber can be seen in the example code above


The stack consists of the following chain: Stream -(string chunks)-> PreTokenizer -(PreToken's)-> Tokenizer -(Token's)-> Parser -(BodyEvent)-> ParserEventConsumer -(Resulting Type)-> ...

PreTokens are low level token parts. For example BlockCommentBegin

Tokens are higher level. For example BlockComment

an example of a BodyEvent is OpenArray


check issues


everyone is welcome to contribute. patches, bug-fixes, new features

  1. create an issue so the community can comment on your idea
  2. fork bass-clarinet
  3. create a new branch git checkout -b my_branch
  4. create tests for the changes you made
  5. make sure you pass both existing and newly inserted tests
  6. commit your changes
  7. push to your branch git push origin my_branch
  8. create an pull request





Subscribe to our newsletter

Get open source security insights delivered straight into your inbox. Be the first to learn about new features and product updates.

  • Terms
  • Privacy
  • Security

Made with ⚡️ by Socket Inc