CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog.

[1.3.0] - 2026-09-20

Added

• Retry.withMaxDelay(Duration) - caps every backoff delay (applied after the multiplier and jitter). Without a cap the delay grows unboundedly and, for a large multiplier or many attempts, multiplier^n overflows to an endless wait; the cap keeps waits bounded.
• Retry.onRetry(RetryListener) and the new RetryListener functional interface - called right before each retry with the 1-based failed attempt number, its failure and the delay about to be waited. Works for both execute and executeAsync; not called after the final attempt or for non-retryable failures. A listener that throws aborts retrying (propagated by execute, completes the future exceptionally for executeAsync).

[1.2.0] - 2026-09-07

Added

• Retry.executeAsync(Supplier<CompletableFuture<T>>): an async counterpart to execute(Callable<T>) for tasks that are already asynchronous.
  • Applies the exact same retry policy as the sync path: it reuses the existing isRetryable (exception-type matching via retryOn) and computeDelayMillis (backoff multiplier + jitter) logic rather than duplicating it.
  • Never blocks a thread while waiting between attempts: retries are scheduled via CompletableFuture.delayedExecutor(...) instead of Thread.sleep.
  • A synchronous throw from the supplier itself (before it returns a future) is treated the same as an exceptionally-completed future.
  • Failure causes are unwrapped from CompletionException before the retryability check, so retryOn-configured types match the same way they would on the synchronous path.
  • Returns a CompletableFuture<T> that completes with the eventual result, or completes exceptionally once retries are exhausted or the failure isn't retryable.
• README: new "## Async Retry" section with a runnable example, and an additional realistic example (a reused policy across multiple calls) in "## Usage".
• Tests (RetryTest) covering the new async path: eventual success after N failures with the exact invocation count asserted, immediate failure on a non-retryable exception, exhaustion after maxAttempts, success on the first try, and a supplier that throws synchronously before returning a future.

[1.1.0] - 2026-09-06

Added

• Test coverage for documented-but-untested Retry edge cases:
  • The constructor rejects maxAttempts < 1, a null initialDelay, and a negative initialDelay.
  • withBackoffMultiplier() rejects a negative multiplier.
  • withJitter() rejects a fraction outside [0, 1].
  • Explicitly calling retryOn() with zero arguments retries any exception, the same as never calling it.
  • retryOn() matches a subtype of a configured exception class (assignability via isInstance), not just an exact type match.
  • maxAttempts = 1 means no retries at all: the first failure propagates immediately and the task is called exactly once.
  • An Error (not Exception) thrown by the task propagates as the same Error instance once retries are exhausted, rather than being wrapped.

No behavioral changes were needed — all new edge-case tests passed against the existing implementation.