Async/await semantics vary widely across seven major language runtimes for identical code

Async/await was introduced to simplify concurrent programming by making asynchronous code resemble sequential logic. However, this exploration reveals that the abstraction leaks in practice. The same pseudocode, writing to a log in the background, yields different outputs across languages like Python, Rust, JavaScript, and Swift. These discrepancies arise from underlying design choices, such as whether tasks start eagerly or lazily, and how long they persist after creation.

The nine identified dimensions categorize these choices into three lifecycle phases: task initiation, termination, and cancellation. For example, ‘eagerness’ determines whether an async function begins execution immediately (e.g., JavaScript) or waits for an await (e.g., Rust). ‘Extent’ defines whether tasks survive their spawning scope (indefinite) or are bound to it (dynamic). These dimensions interact in non-obvious ways, making it difficult to predict behavior without deep language-specific knowledge.

Cancellation semantics further complicate portability. Some languages (e.g., Rust) make tasks unaware of cancellation, while others (e.g., Swift) allow tasks to respond but enforce persistent cancellation. These differences affect error handling and resource cleanup. For engineers, the takeaway is that async/await is not a monolithic abstraction, its behavior is shaped by runtime-specific trade-offs, and assumptions from one language may not hold in another.

The findings underscore the need for clearer documentation and tooling to expose these dimensions. Without explicit guidance, developers risk writing code that works in one runtime but fails silently or unpredictably in another. This is particularly critical for libraries or frameworks targeting multiple languages, where subtle semantic mismatches could introduce bugs or performance issues. The paper’s framework provides a structured way to compare implementations, but adoption remains a challenge.