Skip to main content

Why is Wing a language, not just another library or framework? 🤔

New language for a new programming model

Cloud applications are fundamentally different from applications that run on a single machine - they are distributed systems that rely on cloud infrastructure to achieve their goals.

To express both infrastructure and application logic in a safe and unified programming model (aka "cloud-oriented"), Winglang has two execution phases: Preflight for infrastructure definitions and inflight for runtime code.

Preflight code is executed during compilation and produces the infrastructure configuration for your app (e.g. Terraform, CloudFormation, etc). Inflight code is compiled to JavaScript and executed within cloud compute platforms in Node.js environments.

Let's look at a simple example:

bring cloud;

let queue = new cloud.Queue();
let counter = new cloud.Counter();
let bucket = new cloud.Bucket();

queue.setConsumer(inflight (message: str) => {
let i =;
bucket.put("file-${i}.txt", message);

cloud.Queue, cloud.Counter, and cloud.Bucket are preflight objects. They represent cloud infrastructure resources. When compiled to a specific cloud provider, such as AWS, a Terraform file will be produced with the provider's implementation of these resources. The queue.setConsumer() method is a preflight method that configures the infrastructure to invoke a particular inflight function for each message in the queue.

Now comes the cool part: the code that runs inside the inflight function interacts with the counter and the bucket objects through their inflight methods ( and bucket.put()). These methods can only be called from inflight scopes.

Very cool, but what here cannot be done by a library or compiler extension?

In existing languages, where there is no way to distinguish between multiple execution phases, it is impossible to naturally represent this idea that an object has methods that can only be executed from within a specific execution phase.

In addition, other parts of Wing's cloud-oriented cannot be represented naturally without native support from the language. We've seen some worthy efforts in projects like Pulumi's Function Serialization and Functionless. But although these efforts come close, they either have to make compromises in safety (e.g. it is possible to capture mutable objects) or in the programming model (e.g. type system is too complex).

We believe that Wing, a language that natively supports this paradigm, will make using it much easier by streamlining common patterns, in a way that is impossible to accomplish with existing ones. Kind of like what C++ did for object orientation. You could write object-oriented code in C, but you had to work hard to make it work without native support from the language.

So how does writing code in Wing compare to other solutions?

By natively supporting the cloud-oriented programming paradigm, Wing allows developers to write very few lines of code and let the compiler take care of IAM policies and other cloud mechanics.

Below we've written the same simple demo app in Wing and other solutions. This simple app uses a Function to upload a text file to a Bucket, even without the Counter seen above that further complicates matters in other solutions:

import CodeComparison from '../src/components/CodeComparison';

The below table contains the main differences that you can see in the code examples:

FeatureWingPulumiTerraformAWS CDK
LanguageWingYAML + your choiceHCL + your choiceYour choice
Line of code79812272
Cloud Mechanics (IAM, Networking)Generated automatically from intentManualManualManual

This table contains some more differences that cannot fit in a small code sample :)

FeatureWingPulumiTerraformAWS CDK
ExtensibilityCustom resourcesCustom/dynamic providersCustom modulesCustom Constructs / Resource Providers
Customizing lower levelsCompiler plugins to customize generated TerraformDirect access to cloud-provider APIsNo need since working at low levelEscape hatches to modify underlying CloudFormation
Local simulationBuilt-in functional local simulator with instant hot reloadingNoNoNo
Cross-cloud supportSame code compiles to different clouds (for supported resources)No (need to write different code for different clouds)No (need to write different code for different clouds)No (need to write different code for different clouds)
Provisioning engineTerraform / CloudFormation through AWS-CDK (Extensible solution that supports adding more engines)ProprietaryTerraformCloudFormation (CDKTF is used for Terraform, but you cannot compile the same code to different engines)
TestingSame tests run on local simulator and cloud, without mocksNeed mocks for local testingNeed mocks for local testingNeed mocks for local testing

Anything else?

Creating a new language that is tailored to the cloud from the ground up also allows us to assemble a variety of features (some of which exist in other languages, of course) that, when put together, offer a delightful cloud development experience. You can read more about what these features are here.