kotlin multiplatform json5 for kotlinx.serialization

complete multiplatform support: jvm/js/native

// latest -> https://github.com/lisonge/kotlin-json5/releases
implementation("li.songe:json5:latest")

Json5String -> JsonElement

// import li.songe.json5.Json5

val element = Json5.parseToJsonElement("{a:1}")

Json5String -> Object

// import kotlinx.serialization.json.Json
// import kotlinx.serialization.Serializable
// import li.songe.json5.decodeFromJson5String

val json = Json {
    // add your json config
    ignoreUnknownKeys = true
}

@Serializable
data class A(val id:Int)

val a = json.decodeFromJson5String<A>("{id:0, b:''}")

JSON5 text, including incomplete input, to a source-aware document

val document = Json5.parseToDocument("{a:1,]{")
val tokens = document.tokens
val diagnostics = document.diagnostics

For fast, tolerant syntax highlighting, scan the input without building an AST, decoding values, or producing diagnostics:

Json5.scanSyntax(source) { kind, start, end ->
    highlight(start, end, kind)
}

The callback overload emits each span synchronously without retaining a token result. Whitespace is scanned for recovery but omitted by default; pass ignoreWhitespace = false when a complete token tape is needed. To scan ahead and cache theme-independent spans, use the materialized overload:

val syntax = Json5.scanSyntax(source)
for (index in syntax.indices) {
    highlight(
        syntax.startAt(index),
        syntax.endAt(index),
        syntax.kindAt(index),
    )
}

scanSyntax distinguishes property names, null and boolean literals, numbers, strings, comments, and invalid fragments. It accepts incomplete or syntactically invalid input. Materialized results are compact and read-only, backed by one integer array.

Parse JSON5 into a source-aware document. The document keeps property order, duplicate properties, comments, tokens, source ranges, and recoverable diagnostics.

val document = Json5.parseToDocument(
    """
    {
      // server port
      port: 8080,
    }
    """.trimIndent(),
)

val root = document.root
val comments = document.comments
val element = document.toJsonElement()
val port = document[Json5Path["port"]]

parseToDocument recovers incomplete JSON5 and reports syntax diagnostics. Formatting, semantic conversion, and structural editing require a valid document. Duplicate object properties stay distinct in the AST; conversion to JsonElement keeps the last value for compatibility.

JsonElement -> Json5String

// import li.songe.json5.Json5

val formatted: String = Json5.encodeToString(element)

Encode an individual string literal or property name without wrapping it in a JsonElement or JsonObject:

val config = Json5EncoderConfig(quoteStrategy = Json5QuoteStrategy.PreferSingle)
val stringLiteral = Json5.encodeString("it's JSON5", config)
val propertyName = Json5.encodeKey("server-port")

Object -> Json5String

// import kotlinx.serialization.json.Json
// import kotlinx.serialization.Serializable
// import li.songe.json5.encodeToJson5String

val json = Json {
  // add your json config
}

@Serializable
data class A(val id:Int)

val formatted: String = json.encodeToJson5String(A(id=0))

or use Json5EncoderConfig

Formatting preserves comments and raw literals such as quote style and hexadecimal numbers by default.

val formatted = Json5.format(
    "{port:0x1f90,// default\nenabled:true,}",
)

Use Json5FormatConfig to configure indentation, line separators, trailing commas, the final newline, and optional string and property-name re-encoding. Set quoteStrategy to normalize quotes and unquotedKey to control whether identifier-compatible property names remain unquoted. Re-encoding may normalize escape sequences as well as quote characters. Editor integrations can call Json5.formatToEdits and apply the result with source.applyJson5Edits(edits).

Source-aware edits only replace the selected ranges and preserve unrelated comments and formatting.

val path = Json5Path["server"]["port"]
val result = Json5.parseToDocument(source).set(path, JsonPrimitive(8080))

val editedText = result.text
val updatedDocument = result.document

The editing API also supports removing values, renaming or adding properties, inserting array elements, and adding or removing comments. A property path can specify an occurrence index when an object contains duplicate property names; without one, the last matching property is selected.

Add a comment before or after the node selected by a path:

val document = Json5.parseToDocument(source)
val path = Json5Path["server"]["port"]

val before = document.addCommentBefore(path, "TCP port")
val after = before.document.addCommentAfter(
    path,
    "May be overridden",
    kind = Json5CommentKind.Block,
)

Use removeComment with a comment from the same document:

val comment = document.comments.first()
val result = document.removeComment(comment)

Adding a node and attaching a comment are separate operations. Run the comment operation on the document returned by the insertion:

val inserted = document.putProperty(
    Json5Path.Root,
    "timeout",
    JsonPrimitive(30),
)
val commented = inserted.document.addCommentBefore(
    Json5Path["timeout"],
    "Seconds",
)
val editedText = commented.text

The high-level API does not currently update comment content. Replace the full comment range with a Json5TextEdit, then parse the updated text:

val comment = document.comments.first()
val replacement = when (comment.kind) {
    Json5CommentKind.Line -> "// Updated comment"
    Json5CommentKind.Block -> "/* Updated comment */"
}
val edit = Json5TextEdit(
    offset = comment.range.start,
    length = comment.range.length,
    content = replacement,
)
val editedText = document.source.applyJson5Edits(listOf(edit))
val updatedDocument = Json5.parseToDocument(editedText)

Each Json5EditResult.edits list is relative to the document used for that operation. Do not concatenate edits from chained operations. Use the final text, or apply each edit list before starting the next operation.

Json5Object and Json5Array are iterable, and arrays also support indexed access with array[index].

parseToJsonElement intentionally ignores comments because JsonElement cannot represent them. Use parseToDocument whenever comments or original source syntax must be preserved.