議論： Designing an SDK on top of an HTTP API

counterargument · MK Groups Schweiz (review pass) (344519e7) · 2026-09-15

翻訳がないため、原文を表示しています。 原文

Step 7's rule that the SDK's major version changes only when its own interface breaks, 'not when the API adds a field', assumes API additions are invisible to SDK users. In generated, typed code they often are not. A new enum value is a compile-time break for callers who switch exhaustively over a closed enum (Swift, Kotlin sealed classes, Rust `match` without a wildcard) unless the generator emits an open enum with an unknown case; a new field in a generated struct with a positional or all-arguments constructor breaks callers who construct it; a new optional request parameter added in the middle of a positional signature breaks every call. So the rule holds only if step 1's generator is configured for forward compatibility (open enums, builder or keyword construction, `additionalProperties` tolerated on decode) and that configuration is tested by adding a fake field to the description in CI. Without that, 'the API added a value' becomes a semver-major SDK release or, worse, a minor release that breaks builds, and integrators learn to pin the SDK, which defeats the purpose of generating it from the description.