Skip to main content


Inflights (or "inflight functions") are Wing's distributed computing primitive. They are isolated code blocks which can be packaged and executed on compute platforms in the cloud (such as containers, CI/CD pipelines or FaaS).

Inflights can seamlessly interact with resources through the resource's inflight API.

The following example shows how my_inflight is hosted inside a cloud function and puts an object inside a bucket:

bring cloud;

let bucket = new cloud.Bucket();

new cloud.Function(inflight (_: str): str => {
bucket.put("hello.txt", "world");

But before we explain what inflight functions are (we also call them "inflights"), let's take a step back and talk about what are cloud applications anyway?

A cloud application is software that deeply leverages cloud resources in order to perform its duties. A cloud resource is any network-accessible service. These resources can be used to store data or files (like buckets), process HTTP requests (like API gateways), distribute content (like CDNs), publish notifications to mobile phones or run user-defined code (such as Lambda functions or Kubernetes deployments). Cloud providers such as AWS, GCP and Azure offer a basic set of resources (similar to how an operating system offers a basic set of resources to applications running on the machine), but that doesn't imply that those are the only resources an application might use.

These managed services offer not only functional capabilities to applications, but also many non-functional benefits such as high availability, elastic scalability and reliability. By relying on cloud resources, developers are able to innovate faster and focus on the direct value that their applications create for users.

Wing is designed around this new programming paradigm, which we call cloud-oriented programming.

Diving deeper into inflight functions

Taken from Inflight Magazine

Inflight functions are likely Wing's most significant language innovation. Since cloud applications are distributed systems and code is executed on multiple machines across various compute platforms, our language needs first-class support for this idea: code that executes in the cloud and interacts with cloud resources around it. This is where inflight functions come into play.

Asynchronous and concurrent programming have been a continuous source for programming language innovation and research. Since the early days of fork() in POSIX, multi-threaded programming, Goroutines, Erlang Processes and the recently popularized, cooperative concurrency with async/await, developers have been looking for ways to write applications that included more than a single thread of execution in a safe and intuitive manner.

Wing's inflight functions can be thought of as a type of asynchronous function that can be packaged and executed on a remote system, such as inside a Kubernetes Pod, AWS Lambda, Azure Functions, on Cloudflare Edge Workers or on any other cloud compute platform.

While inflights are typically compiled into machine code (like JavaScript), Wing may also be able to compile them into state machines, orchestrated workflows, and other formats in the future

Our canonic example is a function that wants to put an object inside a bucket:

bring cloud;

let my_bucket = new cloud.Bucket();

new cloud.Function(inflight () => {
my_bucket.put("hello", "world");

The magic happens when the inflight function is referencing my_bucket. The cloud.Bucket which is defined (with new) outside the function closure represents a cloud infrastructure resource, not an in-memory object. When the function code interacts with this bucket (calls my_bucket.put()), it represents an interaction across machine boundaries. The code is executed inside some compute service such as AWS Lambda, and the bucket is backed by something like Amazon S3. The compiler has to do quite a lot in order to make this a reality on the cloud: it needs to wire up the bucket information during deployment, add the right permissions to the Lambda's IAM policy, and create a code bundle with just the right set of dependencies.

Inflight functions can interact with resources through their inflight interface. This is not very different from e.g. async functions in JavaScript, where it doesn't really make sense to await an async function from synchronous contexts. It is also not very different from how Goroutines can only interact with the world through Channels, or child processes can interact through pipes.

Inflight functions are an isolation boundary enforced by the compiler. This means that there are certain limitations on what can be accessed from an inflight scope. Specifically, you can only access immutable data and interact with cloud resources through their inflight API (like the example above demonstrates). If you try to call my_bucket.put() outside of an inflight context the Wing compiler will emit this error:

Cannot call inflight function "my_bucket.put" while in preflight phase

Many folks are asking us why is Wing a language and not a library. Inflight functions and resources are a great example. We believe it is impossible to ensure this level of isolation and interaction semantics in existing languages through a library, or even through a compiler extension. We've seen some worthy efforts in projects like Pulumi's Function Serialization, Punchcard and Functionless, and in all of these cases, there are either 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 inflight functions are a fundamental building block of cloud-oriented programming, and we are on a journey to get them right. From a syntactic perspective, we recently decided to remove the short hand ~> for representing inflight closures, and instead use a simple inflight modifier. We've been debating this for a while, and a comment from a Wingnut, Adam Ruka, tilted the scale in favor of the less quirky syntax. Don't be surprised, though, if ~> will come back, in addition to the inflight keyword, at some point when things mature.

So now, defining inflight functions in Wing is very similar to defining async functions, only that instead of the async modifier, you can use the inflight modifier. Here's an inflight function closure:

let handler = inflight (e: str): str => {
// on air (apologies, the puns are unavoidable, really)

The next major step for inflights is to establish a model that allows them to be composed together, while tracking the operations performed by each one for the purpose of permission inference and other wiring. We've implemented the first stage in the SDK (#869), and we are working on implementing this in the compiler (#542), which will enable the creation of resources with inflight methods.

To see what this work will enable, check out this user story which talks about declaring a user-defined resource in Wing with a set of inflight methods. This is going to be fabulous!