For maybe 90% of use cases, you will only need to modify the mustache template files to create your own custom generated code. If you need to include additional files in your generated output, manipulate the OpenAPI document inputs, or implement your own vendor extensions or other logic, you'll want to read [customization](./customization.md) after you read this document. Be sure to start here first, because templating is the easier concept and you'll need it for more advanced use cases.
The generator workflow has [transforming logic](https://github.com/openapitools/openapi-generator/tree/master/modules/openapi-generator/src/main/java/org/openapitools/codegen/languages) as well as templates for each generation of code.
Each generator will create a data structure from the OpenAPI document; OpenAPI 2.0 and OpenAPI 3.x documents are normalized into the same API model within the generator. This model is then applied to the templates. While generators do not need to perform transformations, it's often necessary in order to add more advanced support for your language or framework. You may need to refer to the generator implementation to understand some of the logic while creating or customizing templates (see [ScalaFinchServerCodegen.java](https://github.com/OpenAPITools/openapi-generator/blob/master/modules/openapi-generator/src/main/java/org/openapitools/codegen/languages/ScalaFinchServerCodegen.java) for an advanced example).
The transform logic needs to implement [CodegenConfig.java](https://github.com/openapitools/openapi-generator/blob/master/modules/openapi-generator/src/main/java/org/openapitools/codegen/CodegenConfig.java) and is most easily done by extending [DefaultCodegen.java](https://github.com/openapitools/openapi-generator/blob/master/modules/openapi-generator/src/main/java/org/openapitools/codegen/DefaultCodegen.java). Take a look at the various implementations as a guideline while the instructions get more complete.
## Modifying Templates
> OpenAPI Generator applies user-defined templates via options:
Built-in templates are written in Mustache and processed by [jmustache](https://github.com/samskivert/jmustache). Beginning with version 4.0.0, we support experimental Handlebars and user-defined template engines via plugins.
OpenAPI Generator supports user-defined templates. This approach is often the easiest when creating a custom template. Our generators implement a combination of language and framework features, and it's fully possible to use an existing generator to implement a custom template for a different framework. Suppose you have internal utilities which you'd like to incorporate into generated code (e.g. logging, monitoring, fault-handling)... this is easy to add via custom templates.
> **Note:** You cannot use this approach to create new templates, only override existing ones. If you'd like to create a new generator to contribute back to the project, see `new.sh` in the repository root. If you'd like to create a private generator for more templating control, see the [customization](./customization.md) docs.
OpenAPI Generator not only supports local files for templating, but also templates defined on the classpath. This is a great option if you want to reuse templates across multiple projects. To load a template via classpath, you'll need to generate a little differently. For example, if you've created an artifact called `template-classpath-example` which contains extended templates for the `htmlDocs` generator with the following structure:
```
└── src
├── main
│ ├── java
│ └── resources
│ └── templates
│ └── htmlDocs
│ ├── index.mustache
│ └── style.css.mustache
```
You can define your classpath to contain your JAR and the openapi-generator-cli _fat jar_, then invoke main class `org.openapitools.codegen.OpenAPIGenerator`. For instance,
For this example, let's modify a Java client to use AOP via [jcabi/jcabi-aspects](https://github.com/jcabi/jcabi-aspects). We'll log API method execution at the `INFO` level. The jcabi-aspects project could also be used to implement method retries on failures; this would be a great exercise to further play around with templating.
The Java generator supports a `library` option. This option works by defining base templates, then applying library-specific template overrides. This allows for template reuse for libraries sharing the same programming language. Templates defined as a library need only modify or extend the templates concerning the library, and generation falls back to the root templates (the "defaults") when not extended by the library. Generators which support the `library` option will only support the libraries known by the generator at compile time, and will throw a runtime error if you try to provide a custom library name.
To get started, we will need to copy our target generator's directory in full. The directory will be located under `modules/opeanpi-generator/src/main/resources/{generator}`. In general, the generator directory matches the generator name (what you would pass to the `generator` option), but this is not a requirement-- if you are having a hard time finding the template directory, look at the `embeddedTemplateDir` option in your target generator's implementation.
If you've already cloned openapi-generator, find and copy the `modules/opeanpi-generator/src/main/resources/Java` directory. If you have the [Refined GitHub](https://github.com/sindresorhus/refined-github) Chrome or Firefox Extension, you can navigate to this directory on GitHub and click the "Download" button. Or, to pull the directory from latest master:
```bash
mkdir -p ~/.openapi-generator/templates/ && cd $_
curl -L https://api.github.com/repos/OpenAPITools/openapi-generator/tarball | tar xz
**Optional**: Before modifying your templates, you may want to `git init && git add . && git commit -am 'initial'` so you can easily revert to the base templates.
At this point, you have _every_ Java library's template locally. Let's delete all libraries except the `resteasy` library we'll be extending:
Execute `tree` in this Java directory and inspect the mustache files and directory structure. You'll notice there are quite a few templates in the directory root, but extending this root to support resteasy only requires modifying a handful of files:
```bash
tree libraries/resteasy/
libraries/resteasy/
├── ApiClient.mustache
├── JSON.mustache
├── api.mustache
├── build.gradle.mustache
├── build.sbt.mustache
└── pom.mustache
0 directories, 6 files
```
> NOTE: Some generators may be sensitive to _which_ files exist. If you're concerned with redundant files like `pom.mustache` and `build.sbt.mustache`, you can try deleting them. If the generator you're customizing fails at runtime, just `touch` these files to create an empty file.
First, let's add our new dependency to `libraries/resteasy/build.gradle.mustache`:
Next, we'll find the code which generates API methods. You'll see `{{#operations}}{{#operation}}` which is a mustache "loop" which executes the template logic if the model applied to the template has an `operations` array, and a non-null `operation` instance in that array. You can pass `-DdebugOpenAPI` when generating via CLI to inspect the full object model.
Further down in `api.mustache`, find implementation of the method call, and add the `@Loggable` annotation. This template is easy because it has a single method implementation.
> NOTE: This example includes log4j-slf4j-impl to demonstrate that our new code is working. Generally you'll want to leave logging implementations up to your consumers.
And because the java client generates with an outdated Gradle 2.6, let's update the gradle version in the default template (`Java/gradle-wrapper.properties.mustache`):
Now we're ready to generate the client with our simple changes. When we pass the template directory option to our toolset, we _must_ pass the generator's root directory and _not_ the library-only directory.
Execute `./gradlew build` and then `cat target/rolling/rollingtest.log`. You should see messages logged for every call in PetApi with a stubbed unit test.
Congratulations! You've now modified one of the built-in templates to meet your client code's needs.
Adding/modifying template logic simply requires a little bit of [mustache](https://mustache.github.io/), for which you can use existing templates as a guide.
> Custom template engine support is *experimental*
If Mustache or the experimental Handlebars engines don't suit your needs, you can define an adapter to your templating engine of choice. To do this, you'll need to define a new project which consumes the `openapi-generator-core` artifact, and at a minimum implement `TemplatingEngineAdapter`.
This example:
* creates an adapter providing the fundamental logic to compile [Pebble Templates](https://pebbletemplates.io)
* will be implemented in Kotlin to demonstrate ServiceLoader configuration specific to Kotlin (Java will be similar)
* requires Gradle 5.0+
* provides project setup instructions for IntelliJ
To begin, create a [new Gradle project](https://www.jetbrains.com/help/idea/getting-started-with-gradle.html) with Kotlin support. To do this, go to `File` ➞ `New` ➞ `Project`, choose "Gradle" and "Kotlin". Specify groupId `org.openapitools.examples` and artifactId `pebble-template-adapter`.
Ensure the new project uses Gradle 5.0. Navigate to the newly created directory and execute:
```bash
gradle wrapper --gradle-version 5.0
```
In `build.gradle`, we'll add a dependency for OpenAPI Tools core which defines the interface and an abstract helper type for implementing the adapter. We'll also pull in the Pebble artifact. We'll be evaluating this new artifact locally, so we'll also add the Maven plugin for installing to the local maven repository. We'll also create a fatjar using the `shadow` plugin to simplify our classpath.
Modifications to the new project's `build.gradle` should be made in the `plugins` and `dependencies` nodes:
```diff
plugins {
id 'org.jetbrains.kotlin.jvm' version '1.3.11'
id "com.github.johnrengelman.shadow" version "5.0.0"
The above configuration for the `shadow` plugin is strictly optional. It is not needed, for instance, if you plan to publish your adapter and consume it via the Maven or Gradle plugins.
Next, create a new class file called `PebbleTemplateEngineAdapter` under `src/kotlin`. We'll define the template adapter's name as `pebble` and we'll also list this as the only supported file extension. We'll implement the adapter by extending `AbstractTemplatingEngineAdapter`, which includes reusable logic, such as retrieving a list of all possible template names for our provided template extensions(s).
override fun getFileExtensions(): Array<String> = arrayOf("pebble")
}
```
Lastly, create a file `resources/META-INF/services/org.openapitools.codegen.api.TemplatingEngineAdapter`, containing the full class path to the above class:
This allows the adapter to load via ServiceLoader, and to be referenced via the identifier `pebble`. This is optional; if you don't provide the above file and contents, you'll only be able to load the engine via full class name (explained in a bit).
Now, build the fatjar for this new adapter:
```bash
./gradlew shadowJar
```
To test compilation of some templates, we'll need to first create one or more template files. Create a temp directory at `/tmp/pebble-example/templates` and add the following files.
Notice how we've targeted our custom template engine adapter via `-e pebble`. If you don't include the SPI file under `META-INF/services`, you'll need to specify the exact classpath: `org.openapitools.examples.templating.PebbleTemplateAdapter`. Notice that the target class here matches the Kotlin class name. This is because of the `@file:JvmName` annotation.
Congratulations on creating a custom templating engine adapter!
Aside from transforming an API document, the implementing class gets to decide how to apply the data structure to templates. We can decide which data structure to apply to which template files. You have the following structures at your disposal.
Examples for the following structures will be presented using the following spec document:
There is a data structure which represents all the operations that are defined in the OpenAPI specification. A single API file is created for each `OperationGroup`, which is essentially a grouping of different operations. See the `addOperationToGroup` in `DefaultCodegen.java` for details on this operation.
You can have many files created for each `OperationGroup` by processing multiple templates and assigning a different file naming pattern to them. So for a single file per operation:
```java
// process the `api.mustache` template and output a single file with suffix `.java`:
apiTemplateFiles.put("api.mustache", ".java");
```
For C-like languages which also require header files, you may create two files per operation.
Here, an Operation with tag `Pet` will generate two files: `SWGPetApi.h` and `SWGPetApi.m`. The `SWG` prefix and `Api` suffix are options specific to the Objective-C geneator.
### Models
> Inspect models passed to templates with system property `-DdebugModels`
Each model identified inside the generator will be passed into the `Models` data structure and will generate a new model file (or files) for each model.
A `Pet` model with three properties will provide a _lot_ of information about the type and properties. The output from `-DdebugModels` is presented in truncated format here.
We expose the same properties in multiple sets because this allows us to conditionally iterate over properties based on some condition ("is it required" or "is it readonly"). This is driven by the use of the logic-less Mustache templates. It is possible that models passed to the templating engine may be cleaned up as we support more template engines, but such an effort will go through a deprecation phase and would be communicated at runtime through log messages.
This is a "catch-all" which gives you the entire structure--operations, model, etc--so you can create "single-file" code from them.
Supporting files can either be processed through the templating engine or copied as-is. When creating your own templates, you're limited to the files and extensions expected by the generator implementation. For more control over the supporting files produced by a generator, see our [customization](./customization.md) documentation.
## Variables
> This is a very limited list of variable name explanations. Feel free to [open a pull request](https://github.com/OpenAPITools/openapi-generator/pull/new/master) to add to this documentation!
Many generators (*those extending DefaultCodegen*) come with a small set of lambda functions available under the key `lambda`:
-`lowercase` - Converts all of the characters in this fragment to lower case using the rules of the `ROOT` locale.
-`uppercase` - Converts all of the characters in this fragment to upper case using the rules of the `ROOT` locale.
-`titlecase` - Converts text in a fragment to title case. For example `once upon a time` to `Once Upon A Time`.
-`camelcase` - Converts text in a fragment to camelCase. For example `Input-text` to `inputText`.
-`indented` - Prepends 4 spaces indention from second line of a fragment on. First line will be indented by Mustache.
-`indented_8` - Prepends 8 spaces indention from second line of a fragment on. First line will be indented by Mustache.
-`indented_12` - Prepends 12 spaces indention from second line of a fragment on. First line will be indented by Mustache.
-`indented_16` -Prepends 16 spaces indention from second line of a fragment on. First line will be indented by Mustache.
Lambda is invoked by `lambda.[lambda name]` expression. For example: `{{#lambda.lowercase}}FRAGMENT TO LOWERCASE{{/lambda.lowercase}}` to lower case text between `lambda.lowercase`.
OpenAPI supports a concept called "Extensions". These are called "Specification Extensions" [in 3.x](https://github.com/OAI/OpenAPI-Specification/blob/master/versions/3.0.2.md#specificationExtensions) and "Vendor Extensions" [in 2.0](https://github.com/OAI/OpenAPI-Specification/blob/master/versions/2.0.md#vendorExtensions).
You'll see them referred to as "Vendor Extensions" in most places in this project.
Vendor extensions allow you to provide vendor-specific configurations to your specification document.
For example, suppose you use your specification document for code generation with a (hypothetical) C# OpenAPI generator supporting a desired operationId prefix where the extension is `x-csharp-operationid`, you can define this property alongside the object you'd like to extend (which would be a Path Object in this case). You could then apply additional extensions alongside this property, whether they're for another language or other tooling.
> Well-defined vendor extensions don't cause conflicts with other tooling.
<!-- TODO: Auto-generate this list using generator metadata -->
The following are vendor extensions supported by OpenAPI Generator. The list may not be up-to-date, the best way is to look for "x-" in the built-in mustache templates.
`x-enum-varnames` can be used to have an other enum name for the corresponding value.
This is used to define names of the enum items.
`x-enum-descriptions` can be used to provide an individual description for each value.
This is used for comments in the code (like javadoc if the target language is java).
`x-enum-descriptions` and `x-enum-varnames` are each expected to be list of items containing the same number of items as `enum`.
The order of the items in the list matters: their position is used to group them together.
Example:
```yaml
WeatherType:
type: integer
format: int32
enum:
- 42
- 18
- 56
x-enum-descriptions:
- 'Blue sky'
- 'Slightly overcast'
- 'Take an umbrella with you'
x-enum-varnames:
- Sunny
- Cloudy
- Rainy
```
In the example for the integer value `42`, the description will be `Blue sky` and the name of the enum item will be `Sunny` (some generators changes it to `SUNNY` to respect some coding convention).
MySQL schema generator creates vendor extensions based on openapi `dataType` and `dataFormat`. When user defined extensions with same key already exists codegen accepts those as is. It means it won't validate properties or correct it for you. Every model in `definitions` can contain table related and column related extensions like in example below:
description: This should be most common InnoDB table
type: object
properties:
id:
description: >-
This column should be unsigned BIGINT with AUTO_INCREMENT
type: integer
format: int64
x-mysqlSchema:
columnDefinition:
colName: id
colDataType: DECIMAL
colDataTypeArguments:
- argumentValue: 16
isString: false
hasMore: true
- argumentValue: 4
isString: false
hasMore: false
colUnsigned: true
colNotNull: true
colDefault:
defaultValue: AUTO_INCREMENT
isString: false
isNumeric: false
isKeyword: true
colComment: >-
Column comment. This column should be unsigned BIGINT with AUTO_INCREMENT
x-mysqlSchema:
tableDefinition:
tblName: orders
tblStorageEngine: InnoDB
tblComment: >-
Table comment. This should be most common InnoDB table
```
> There are properties that are not implemented by now(`tblStorageEngine`), but you can see how generator can be enhanced in future.
## Mustache Tips
Here are a few tips we've found useful for new template authors.
For more details on Mustache see [mustache.5](https://mustache.github.io/mustache.5.html). See also [samskivert/jmustache](https://github.com/samskivert/jmustache) for implementation-specific details.
### First/Last
To access the first or last element in a list using Mustache:
```mustache
{{#vars}}{{#-first}} this is the first element {{.}} {{/-first}}{{/vars}}
{{#vars}}{{#-last}} this is the last element {{.}} {{/-last}}{{/vars}}
```
### This
Mustache evaluates template variables contextually. If the variable isn't found in the immediate object, mustache will search the parent. This is similar to JavaScript's prototype object (if you're familiar with the concept).
You can inspect this entire context by outputting `{{this}}`. For example:
```mustache
{{#operations}}{{this}}{{/operations}}
```
### Index
If you'd like a 1-based index in your array traversal, you can use `{{-index}}`:
