dirty

d3rty/json parses JSON from sources you can't fully control or trust: third-party APIs with schema flakiness, legacy feeds, AI-generated output. Stringified numbers, "on"/"off" booleans, drifting key casing - instead of failing on the first mismatch, it decodes strictly first and falls back to a forgiving dirty schema you define per model.

[!NOTE] Work in progress: the API may still change before v1.

data := []byte(`{"id": "123"}`)

var plain struct{ ID int `json:"id"` }
json.Unmarshal(data, &plain)  // error: cannot unmarshal string into Go struct field .id of type int

var event Event               // same shape, dirty-enabled (see Quick Start)
dirty.Unmarshal(data, &event) // nil; event.ID == 123

Install

go get github.com/d3rty/json

import dirty "github.com/d3rty/json"

Quick Start

Embed dirty.Enabled into your clean model and describe what may be dirty in a Dirty() counterpart. Everything else stays strict.

package main

import (
	"fmt"
	"log"

	dirty "github.com/d3rty/json"
)

// Event is a regular "clean" model with strict types.
type Event struct {
	dirty.Enabled // opts this model into dirty unmarshalling

	Name string `json:"name"`

	// Covered by the dirty schema below: stringified numbers,
	// sloppy booleans, etc. are forgiven.
	ID       int  `json:"id"`
	IsActive bool `json:"is_active"`

	// Not covered by the dirty schema: must be a real bool or it's lost.
	MustBool bool `json:"must_bool"`
}

// Dirty links the clean model to its dirty counterpart.
func (e *Event) Dirty() any {
	return &struct {
		ID       dirty.Number `json:"id"`
		IsActive dirty.Bool   `json:"is_active"`
	}{}
}

func main() {
	data := []byte(`{"id": "123", "name": "Sample Event", "is_active": "on", "must_bool": true}`)

	var event Event
	if err := dirty.Unmarshal(data, &event); err != nil {
		log.Fatal(err)
	}

	fmt.Println(event.ID)       // 123
	fmt.Println(event.IsActive) // true
	fmt.Println(event.MustBool) // true
}

The dirty types are dirty.Number, dirty.String, dirty.Bool, dirty.Array and dirty.Object. Swap dirty.Enabled for dirty.Disabled to switch a model back to strict stdlib behavior without touching anything else.

Flexible Keys

Key naming drifts too: isActive today, IsActive tomorrow, is-active from the mobile team. The FlexKeys option (off by default) matches keys case-insensitively and across camelCase/snake_case/kebab-case/PascalCase:

dirty.ConfigSetGlobal(func(cfg *dirty.Config) {
	cfg.FlexKeys.Disabled = false
	cfg.FlexKeys.CaseInsensitive = true
	cfg.FlexKeys.ChameleonCase = true
})

// key arrives as PascalCase, value as a string - both forgiven
data := []byte(`{"IsActive": "on"}`)

The case-detection machinery is exposed as its own package, cases.

Configuration

Every forgiveness rule is a switch in a TOML config: which strings count as true, whether "1 000 000" or "1e6" are numbers, which date layouts to try, and so on. The full annotated list of defaults lives in default.toml.

// tweak the global config in place
dirty.ConfigSetGlobal(func(cfg *dirty.Config) {
	cfg.Bool.FromStrings.CustomListForTrue = []string{"true", "yes", "on", "da"}
})

// or load a full config from TOML bytes
cfg := dirty.ConfigFromBytes(tomlBytes)

Unmarshalling results are classified by color: 🟢 green (decoded cleanly), 🟡 yellow (decoded via forgiving conversions), 🔴 red (data was lost). Reporting this back to the caller is not implemented yet.

Playground

The WASM demo runs the real decoder in your browser: paste dirty JSON, toggle config switches, see what gets forgiven.

Feedback

d3rty/json is a solo, opinionated project - but if you stumbled upon it and have ideas, questions, or bug reports, an issue is always welcome :)

License

MIT © d3rty