libopenapi - enterprise grade OpenAPI tools for golang.

libopenapi has full support for Swagger (OpenAPI 2), OpenAPI 3, and OpenAPI 3.1.

Introduction - Why?

There is already a really great OpenAPI library for golang, it's called kin-openapi.

Why does libopenapi exist?

kin-openapi is great, and you should go and use it.

If you're still reading, here is why libopenapi might be useful.

kin-openapi missing a few critical features... They are so important, this entire toolset was created to address those gaps.

When building tooling that needs to analyze OpenAPI specifications at a low level, kin-openapi runs out of power when you need to know the original line numbers and columns, or comments within all keys and values in the specification.

All that data is lost when the OpenAPI specification is loaded in by kin-openapi. Mainly because the library will unmarshal data directly into structs, which works great - if you don't need access to the original specification low level details.

Why not just modify kin-openapi?

It would require a fundamental re-build of the entire library, with a different design to expose the same functionality.

---

libopenapi retains everything.

libopenapi has been designed to retain all of that really low-level detail about the AST, line numbers, column numbers, comments, original AST structure - everything you could need.

libopenapi has a porcelain (high-level) and a plumbing (low-level) API. Every high level struct, has the ability to GoLow() and dive from the high-level model, down to the low-level model and look-up any detail about the underlying raw data backing that model.

This library exists because this very need existed inside VMware. The company built an internal version of libopenapi, which isn't something that can be released as it's customized for VMware (and it's incomplete).

libopenapi is the result of years of learning and battle testing OpenAPI in golang. This library represents what would have been created, if we knew then - what we know now.

Need to know which line, or column number a key or value for something is? libopenapi has you covered.

Comes with an Index and a Resolver

Want a lightning fast way to look up any element in an OpenAPI swagger spec? libopenapi has you covered.

Need a way to 'resolve' OpenAPI documents that are exploded out across multiple files, remotely or locally? libopenapi has you covered.

Read the full docs at https://pkg.go.dev

---

Installing

Grab the latest release of libopenapi

go get github.com/pb33f/libopenapi

Load an OpenAPI 3+ spec into a model

// import the library
import "github.com/pb33f/libopenapi"

func readSpec() {
    
    // load an OpenAPI 3 specification from bytes
    petstore, _ := ioutil.ReadFile("test_specs/petstorev3.json")
    
    // create a new document from specification bytes
    document, err := libopenapi.NewDocument(petstore)
    
    // if anything went wrong, an error is thrown
    if err != nil {
        panic(fmt.Sprintf("cannot create new document: %e", err))
    }
    
    // because we know this is a v3 spec, we can build a ready to go model from it.
    v3Model, errors := document.BuildV3Model()
    
    // if anything went wrong when building the v3 model, a slice of errors will be returned
    if len(errors) > 0 {
        for i := range errors {
            fmt.Printf("error: %e\n", errors[i])
        }
        panic(fmt.Sprintf("cannot create v3 model from document: %d errors reported", 
            len(errors)))
    }
    
    // get a count of the number of paths and schemas.
    paths := len(v3Model.Model.Paths.PathItems)
    schemas := len(v3Model.Model.Components.Schemas)
    
    // print the number of paths and schemas in the document
    fmt.Printf("There are %d paths and %d schemas in the document", paths, schemas)
}

This will print:

There are 13 paths and 8 schemas in the document

Load a Swagger (OpenAPI 2) spec into a model

// import the library
import "github.com/pb33f/libopenapi"

func readSpec() {

    // load a Swagger specification from bytes
    petstore, _ := ioutil.ReadFile("test_specs/petstorev2.json")
    
    // create a new document from specification bytes
    document, err := libeopnapi.NewDocument(petstore)
    
    // if anything went wrong, an error is thrown
    if err != nil {
        panic(fmt.Sprintf("cannot create new document: %e", err))
    }
    
    // because we know this is a v2 spec, we can build a ready to go model from it.
    v2Model, errors := document.BuildV2Model()
    
    // if anything went wrong when building the v3 model, a slice of errors will be returned
    if len(errors) > 0 {
        for i := range errors {
            fmt.Printf("error: %e\n", errors[i])    
        }
        panic(fmt.Sprintf("cannot create v3 model from document: %d errors reported", 
            len(errors)))
    }
    
    // get a count of the number of paths and schemas.
    paths := len(v2Model.Model.Paths.PathItems)
    schemas := len(v2Model.Model.Definitions.Definitions)
    
    // print the number of paths and schemas in the document
    fmt.Printf("There are %d paths and %d schemas in the document", paths, schemas)
}

This will print:

There are 14 paths and 6 schemas in the document

Dropping down from the high-level API to the low-level one

This example shows how after loading an OpenAPI spec into a document, navigating to an Operation is pretty simple. It then shows how to drop-down (using GoLow()) to the low-level API and query the line and start column of the RequestBody description.

// load an OpenAPI 3 specification from bytes
petstore, _ := ioutil.ReadFile("test_specs/petstorev3.json")

// create a new document from specification bytes 
// (ignore errors for the same of the example)
document, _ := libopenapi.NewDocument(petstore)

// because we know this is a v3 spec, we can build a ready to go model from it 
// (ignoring errors for the example)
v3Model, _ := document.BuildV3Model()

// extract the RequestBody from the 'put' operation under the /pet path
reqBody := document.Paths.PathItems["/pet"].Put.RequestBody

// dropdown to the low-level API for RequestBody
lowReqBody := reqBody.GoLow() 

// print out the value, the line it appears on and the 
// start columns for the key and value of the nodes.
fmt.Printf("value is %s, the value is on line %d, " + 
    "starting column %d, the key is on line %d, column %d", 
    reqBody.Description, 
    lowReqBody.Description.ValueNode.Line, 
    lowReqBody.Description.ValueNode.Column,
    lowReqBody.Description.KeyNode.Line, 
    lowReqBody.KeyNode.Column)

---

But wait, there's more - Mutating the model

Having a read-only model is great, but what about when we want to modify the model and mutate values, or even add new content to the model? What if we also want to save that output as an updated specification - but we don't want to jumble up the original ordering of the source.

marshaling and unmarshalling to and from structs into JSON/YAML is not ideal.

When we straight up use json.Marshal or yaml.Marshal to send structs to be rendered into the desired format, there is no guarantee as to the order in which each component will be rendered. This works great if...

• We don't care about the spec being randomly ordered.
• We don't care about code-reviews.
• We don't actually care about this very much.

But if we do care...

Then libopenpi provides a way to mutate the model, that keeps the original yaml.Node API tree in-tact. It allows us to make changes to values in place, and serialize back to JSON or YAML without any changes to other content order.

// create very small, and useless spec that does nothing useful, except showcase this feature.
spec := `
openapi: 3.1.0
info:
  title: This is a title
  contact:
    name: Some Person
    email: [email protected]
  license:
    url: http://some-place-on-the-internet.com/license
`
// create a new document from specification bytes
document, err := libopenapi.NewDocument([]byte(spec))

// if anything went wrong, an error is thrown
if err != nil {
    panic(fmt.Sprintf("cannot create new document: %e", err))
}

// because we know this is a v3 spec, we can build a ready to go model from it.
v3Model, errors := document.BuildV3Model()

// if anything went wrong when building the v3 model, a slice of errors will be returned
if len(errors) > 0 {
    for i := range errors {
        fmt.Printf("error: %e\n", errors[i])
    }
    panic(fmt.Sprintf("cannot create v3 model from document: %d errors reported", len(errors)))
}

// mutate the title, to do this we currently need to drop down to the low-level API.
v3Model.Model.GoLow().Info.Value.Title.Mutate("A new title for a useless spec")

// mutate the email address in the contact object.
v3Model.Model.GoLow().Info.Value.Contact.Value.Email.Mutate("[email protected]")

// mutate the name in the contact object.
v3Model.Model.GoLow().Info.Value.Contact.Value.Name.Mutate("Buckaroo")

// mutate the URL for the license object.
v3Model.Model.GoLow().Info.Value.License.Value.URL.Mutate("https://pb33f.io/license")

// serialize the document back into the original YAML or JSON
mutatedSpec, serialError := document.Serialize()

// if something went wrong serializing
if serialError != nil {
    panic(fmt.Sprintf("cannot serialize document: %e", serialError))
}

// print our modified spec!
fmt.Println(string(mutatedSpec))

Which will output:

openapi: 3.1.0
info:
    title: A new title for a useless spec
    contact:
         name: Buckaroo
         email: [email protected]
    license:
         url: https://pb33f.io/license

It's worth noting that the original line numbers and column numbers won't be respected when calling Serialize(), A new Document needs to be created from that raw YAML to continue processing after serialization.

Creating an index of an OpenAPI Specification

An index is really useful when a map of an OpenAPI spec is needed. Knowing where all the references are and where they point, is very useful when resolving specifications, or just looking things up.

Creating an index from the Stripe OpenAPI Spec

// define a rootNode to hold our raw stripe spec AST.
var rootNode yaml.Node

// load in the stripe OpenAPI specification into bytes (it's pretty meaty)
stripeSpec, _ := ioutil.ReadFile("test_specs/stripe.yaml")

// unmarshal spec into our rootNode
yaml.Unmarshal(stripeSpec, &rootNode)

// create a new specification index.
index := index.NewSpecIndex(&rootNode)

// print out some statistics
fmt.Printf("There are %d references\n"+
    "%d paths\n"+
    "%d operations\n"+
    "%d schemas\n"+
    "%d enums\n"+
    "%d polymorphic references",
    len(index.GetAllCombinedReferences()),
    len(index.GetAllPaths()),
    index.GetOperationCount(),
    len(index.GetAllSchemas()),
    len(index.GetAllEnums()),
    len(index.GetPolyOneOfReferences())+len(index.GetPolyAnyOfReferences()))

Resolving an OpenAPI Specification

When creating an index, the raw AST that uses yaml.Node is preserved when looking up local, file-based and remote references. This means that if required, the spec can be 'resolved' and all the reference nodes will be replaced with the actual data they reference.

What this looks like from a spec perspective.

If the specification looks like this:

paths:
  "/some/path/to/a/thing":
    get:
      responses:
        "200":
          $ref: '#/components/schemas/MySchema'
components:
  schemas:
   MySchema:
     type: string
     description: This is my schema that is great!

Will become this (as represented by the root yaml.Node

paths:
  "/some/path/to/a/thing":
    get:
      responses:
        "200":
          type: string
          description: This is my schema that is great!
components:
  schemas:
   MySchema:
     type: string
     description: This is my schema that is great!

This is not a valid spec, it's just design to illustrate how resolving works.

The reference has been 'resolved', so when reading the raw AST, there is no lookup required anymore.

Resolving Example:

Using the Stripe API as an example, we can resolve all references, and then count how many circular reference issues were found.

import (
    "github.com/pb33f/libopenapi/index"
    "github.com/pb33f/libopenapi/resolver"
)

// create a yaml.Node reference as a root node.
var rootNode yaml.Node

//  load in the Stripe OpenAPI spec (lots of polymorphic complexity in here)
stripeBytes, _ := ioutil.ReadFile("../test_specs/stripe.yaml")

// unmarshal bytes into our rootNode.
_ = yaml.Unmarshal(stripeBytes, &rootNode)

// create a new spec index (resolver depends on it)
index := index.NewSpecIndex(&rootNode)

// create a new resolver using the index.
resolver := resolver.NewResolver(index)

// resolve the document, if there are circular reference errors, they are returned/
// WARNING: this is a destructive action and the rootNode will be 
// PERMANENTLY altered and cannot be unresolved
circularErrors := resolver.Resolve()

// The Stripe API has a bunch of circular reference problems, 
// mainly from polymorphism. So let's print them out.
fmt.Printf("There are %d circular reference errors, " +
    "%d of them are polymorphic errors, %d are not",
    len(circularErrors), 
    len(resolver.GetPolymorphicCircularErrors()), 
    len(resolver.GetNonPolymorphicCircularErrors()))

This will output:

There are 21 circular reference errors, 19 of them are polymorphic errors, 2 are not

Important to remember: Resolving is destructive and will permanently change the tree, it cannot be un-resolved.

Checking for circular errors without resolving

Resolving is destructive, the original reference nodes are gone and all replaced, so how do we check for circular references in a non-destructive way? Instead of calling Resolve(), we can call CheckForCircularReferences() instead.

The same code as Resolve() executes, except the tree is not actually resolved, nothing changes and no destruction occurs. A handy way to perform circular reference analysis on the specification, without permanently altering it.

import (
    "github.com/pb33f/libopenapi/index" 
    "github.com/pb33f/libopenapi/resolver"
)

// create a yaml.Node reference as a root node.
var rootNode yaml.Node

//  load in the Stripe OpenAPI spec (lots of polymorphic complexity in here)
stripeBytes, _ := ioutil.ReadFile("../test_specs/stripe.yaml")

// unmarshal bytes into our rootNode.
_ = yaml.Unmarshal(stripeBytes, &rootNode)

// create a new spec index (resolver depends on it)
index := index.NewSpecIndex(&rootNode)

// create a new resolver using the index.
resolver := resolver.NewResolver(index)

// extract circular reference errors without any changes to the original tree.
circularErrors := resolver.CheckForCircularReferences()

// The Stripe API has a bunch of circular reference problems, 
// mainly from polymorphism. So let's print them out.
fmt.Printf("There are %d circular reference errors, " +
    "%d of them are polymorphic errors, %d are not",
    len(circularErrors),
    len(resolver.GetPolymorphicCircularErrors()),
    len(resolver.GetNonPolymorphicCircularErrors()))

---

Read the full docs at https://pkg.go.dev

---

The library heavily depends on the fantastic (yet hard to get used to) yaml.Node API. This is what is exposed by the GoLow API.

It does not matter if the input material is JSON or YAML, the yaml.Node API supports both and creates a great way to navigate the AST of the document.

Logo gopher is modified, originally from egonelbre