Go library for the TOML format.

This library supports TOML v1.0.0.

🐞 Bug Reports

💬 Anything else

Full API, examples, and implementation notes are available in the Go documentation.

import "github.com/pelletier/go-toml/v2"

See Modules.

As much as possible, this library is designed to behave similarly as the standard library's encoding/json.

While go-toml favors usability, it is written with performance in mind. Most operations should not be shockingly slow. See benchmarks.

Decoder can be set to "strict mode", which makes it error when some parts of the TOML document was not present in the target structure. This is a great way to check for typos. See example in the documentation.

When most decoding errors occur, go-toml returns DecodeError, which contains a human readable contextualized version of the error. For example:

1| [server]
2| path = 100
 |        ~~~ cannot decode TOML integer into struct field toml_test.Server.Path of type string
3| port = 50

TOML supports native local date/times. It allows to represent a given date, time, or date-time without relation to a timezone or offset. To support this use-case, go-toml provides LocalDate, LocalTime, and LocalDateTime. Those types can be transformed to and from time.Time, making them convenient yet unambiguous structures for their respective TOML representation.

Since TOML is often used for configuration files, go-toml can emit documents annotated with comments and commented-out values. For example, it can generate the following file:

# Host IP to connect to.
host = '127.0.0.1'
# Port of the remote server.
port = 4242

# Encryption parameters (optional)
# [TLS]
# cipher = 'AEAD-AES128-GCM-SHA256'
# version = 'TLS 1.3'

Given the following struct, let's see how to read it and write it as TOML:

type MyConfig struct {
	Version int
	Name    string
	Tags    []string
}

Unmarshal reads a TOML document and fills a Go structure with its content. For example:

doc := `
version = 2
name = "go-toml"
tags = ["go", "toml"]
`

var cfg MyConfig
err := toml.Unmarshal([]byte(doc), &cfg)
if err != nil {
	panic(err)
}
fmt.Println("version:", cfg.Version)
fmt.Println("name:", cfg.Name)
fmt.Println("tags:", cfg.Tags)

// Output:
// version: 2
// name: go-toml
// tags: [go toml]

Marshal is the opposite of Unmarshal: it represents a Go structure as a TOML document:

cfg := MyConfig{
	Version: 2,
	Name:    "go-toml",
	Tags:    []string{"go", "toml"},
}

b, err := toml.Marshal(cfg)
if err != nil {
	panic(err)
}
fmt.Println(string(b))

// Output:
// Version = 2
// Name = 'go-toml'
// Tags = ['go', 'toml']

This API does not yet follow the backward compatibility guarantees of this library. They provide early access to features that may have rough edges or an API subject to change.

Parser is the unstable API that allows iterative parsing of a TOML document at the AST level. See https://pkg.go.dev/github.com/pelletier/go-toml/v2/unstable.

Execution time speedup compared to other Go TOML libraries:

go-toml uses Go's standard modules system.

Installation instructions:

• Go ≥ 1.16: Nothing to do. Use the import in your code. The go command deals with it automatically.
• Go ≥ 1.13: GO111MODULE=on go get github.com/pelletier/go-toml/v2.

In case of trouble: Go Modules FAQ.

Go-toml provides three handy command line tools:

• tomljson: Reads a TOML file and outputs its JSON representation.

  $ go install github.com/pelletier/go-toml/v2/cmd/tomljson@latest
  $ tomljson --help
  

• jsontoml: Reads a JSON file and outputs a TOML representation.

  $ go install github.com/pelletier/go-toml/v2/cmd/jsontoml@latest
  $ jsontoml --help
  

• tomll: Lints and reformats a TOML file.

  $ go install github.com/pelletier/go-toml/v2/cmd/tomll@latest
  $ tomll --help
  

Those tools are also available as a Docker image. For example, to use tomljson:

docker run -i ghcr.io/pelletier/go-toml:v2 tomljson < example.toml

Multiple versions are available on ghcr.io.

This section describes the differences between v1 and v2, with some pointers on how to get the original behavior when possible.

When unmarshaling to a struct, if a key in the TOML document does not exactly match the name of a struct field or any of the toml-tagged field, v1 tries multiple variations of the key (code).

V2 instead does a case-insensitive matching, like encoding/json.

This could impact you if you are relying on casing to differentiate two fields, and one of them is a not using the toml struct tag. The recommended solution is to be specific about tag names for those fields using the toml struct tag.

When decoding into a non-nil interface{}, go-toml v1 uses the type of the element in the interface to decode the object. For example:

type inner struct {
	B interface{}
}
type doc struct {
	A interface{}
}

d := doc{
	A: inner{
		B: "Before",
	},
}

data := `
[A]
B = "After"
`

toml.Unmarshal([]byte(data), &d)
fmt.Printf("toml v1: %#v\n", d)

// toml v1: main.doc{A:main.inner{B:"After"}}

In this case, field A is of type interface{}, containing a inner struct. V1 sees that type and uses it when decoding the object.

When decoding an object into an interface{}, V2 instead disregards whatever value the interface{} may contain and replaces it with a map[string]interface{}. With the same data structure as above, here is what the result looks like:

toml.Unmarshal([]byte(data), &d)
fmt.Printf("toml v2: %#v\n", d)

// toml v2: main.doc{A:map[string]interface {}{"B":"After"}}

This is to match encoding/json's behavior. There is no way to make the v2 decoder behave like v1.

When decoding into an array, v1 returns an error when the number of elements contained in the doc is superior to the capacity of the array. For example:

type doc struct {
	A [2]string
}
d := doc{}
err := toml.Unmarshal([]byte(`A = ["one", "two", "many"]`), &d)
fmt.Println(err)

// (1, 1): unmarshal: TOML array length (3) exceeds destination array length (2)

In the same situation, v2 ignores the last value:

err := toml.Unmarshal([]byte(`A = ["one", "two", "many"]`), &d)
fmt.Println("err:", err, "d:", d)
// err: <nil> d: {[one two]}

This is to match encoding/json's behavior. There is no way to make the v2 decoder behave like v1.

This method was not widely used, poorly defined, and added a lot of complexity. A similar effect can be achieved by implementing the encoding.TextUnmarshaler interface and use strings.

This feature adds complexity and a poorly defined API for an effect that can be accomplished outside of the library.

It does not seem like other format parsers in Go support that feature (the project referenced in the original ticket #202 has not been updated since 2017). Given that go-toml v2 should not touch values not in the document, the same effect can be achieved by pre-filling the struct with defaults (libraries like go-defaults can help). Also, string representation is not well defined for all types: it creates issues like #278.

The recommended replacement is pre-filling the struct before unmarshaling.

This structure was the initial attempt at providing a document model for go-toml. It allows manipulating the structure of any document, encoding and decoding from their TOML representation. While a more robust feature was initially planned in go-toml v2, this has been ultimately removed from scope of this library, with no plan to add it back at the moment. The closest equivalent at the moment would be to unmarshal into an interface{} and use type assertions and/or reflection to manipulate the arbitrary structure. However this would fall short of providing all of the TOML features such as adding comments and be specific about whitespace.

The API for retrieving the position (line, column) of a specific TOML element do not exist anymore. This was done to minimize the amount of concepts introduced by the library (query path), and avoid the performance hit related to storing positions in the absence of a document model, for a feature that seemed to have little use. Errors however have gained more detailed position information. Position retrieval seems better fitted for a document model, which has been removed from the scope of go-toml v2 at the moment.

V1 emits struct fields order alphabetically by default. V2 struct fields are emitted in order they are defined. For example:

type S struct {
	B string
	A string
}

data := S{
	B: "B",
	A: "A",
}

b, _ := tomlv1.Marshal(data)
fmt.Println("v1:\n" + string(b))

b, _ = tomlv2.Marshal(data)
fmt.Println("v2:\n" + string(b))

// Output:
// v1:
// A = "A"
// B = "B"

// v2:
// B = 'B'
// A = 'A'

There is no way to make v2 encoder behave like v1. A workaround could be to manually sort the fields alphabetically in the struct definition, or generate struct types using reflect.StructOf.

V1 automatically indents content of tables by default. V2 does not. However the same behavior can be obtained using Encoder.SetIndentTables. For example:

data := map[string]interface{}{
	"table": map[string]string{
		"key": "value",
	},
}

b, _ := tomlv1.Marshal(data)
fmt.Println("v1:\n" + string(b))

b, _ = tomlv2.Marshal(data)
fmt.Println("v2:\n" + string(b))

buf := bytes.Buffer{}
enc := tomlv2.NewEncoder(&buf)
enc.SetIndentTables(true)
enc.Encode(data)
fmt.Println("v2 Encoder:\n" + string(buf.Bytes()))

// Output:
// v1:
//
// [table]
//   key = "value"
//
// v2:
// [table]
// key = 'value'
//
//
// v2 Encoder:
// [table]
//   key = 'value'

V1 always uses double quotes (") around strings and keys that cannot be represented bare (unquoted). V2 uses single quotes instead by default ('), unless a character cannot be represented, then falls back to double quotes. As a result of this change, Encoder.QuoteMapKeys has been removed, as it is not useful anymore.

There is no way to make v2 encoder behave like v1.

Types that implement encoding.TextMarshaler can emit arbitrary TOML in v1. The encoder would append the result to the output directly. In v2 the result is wrapped in a string. As a result, this interface cannot be implemented by the root object.

There is no way to make v2 encoder behave like v1.

Emitting compact comments is now the default behavior of go-toml. This option is not necessary anymore.

V1 used to provide multiple struct tags: comment, commented, multiline, toml, and omitempty. To behave more like the standard library, v2 has merged toml, multiline, commented, and omitempty. For example:

type doc struct {
	// v1
	F string `toml:"field" multiline:"true" omitempty:"true" commented:"true"`
	// v2
	F string `toml:"field,multiline,omitempty,commented"`
}

Has a result, the Encoder.SetTag* methods have been removed, as there is just one tag now.

The new name is Encoder.SetArraysMultiline. The behavior should be the same.

The new name is Encoder.SetIndentSymbol. The behavior should be the same.

V1 defaults to merging embedded struct fields into the embedding struct. This behavior was unexpected because it does not follow the standard library. To avoid breaking backward compatibility, the Encoder.PromoteAnonymous method was added to make the encoder behave correctly. Given backward compatibility is not a problem anymore, v2 does the right thing by default: it follows the behavior of encoding/json. Encoder.PromoteAnonymous has been removed.

go-toml v1 provided the go-toml/query package. It allowed to run JSONPath-style queries on TOML files. This feature is not available in v2. For a replacement, check out dasel.

This package has been removed because it was essentially not supported anymore (last commit May 2020), increased the complexity of the code base, and more complete solutions exist out there.

Expect for parts explicitely marked otherwise, go-toml follows Semantic Versioning. The supported version of TOML is indicated at the beginning of this document. The last two major versions of Go are supported (see Go Release Policy).

The MIT License (MIT). Read LICENSE.