Some alternatives (compiled Zod, Ajv, Valibot and Typebox) were suggested in the replies for reducing Zod's memory usage, so I tested them against a workload similar to our own service, where a JSON request can reach roughly 5 MB. I pushed the benchmark further using a fully synthetic ~50 MB payload containing 10k reports. Takeaways from this workload: → Ajv and compiled Zod with separate normalization performed best; both reduced validation time by roughly two-thirds and used about 60% less adjusted peak memory than current Zod’s native-transform path. → Native Zod transforms were slower and used more memory than validating first and normalizing in a separate pass. → TypeBox's native-transform path was the biggest outlier: roughly x32 slower than TypeBox with separate normalization. Probably I've messed up there with something, but did not bother to check; I want footgun risk also be shown. → Security, unrelated to performance: Ajv and Typebox leak prototype keys while validating loose objects. While this is not considered a direct vulnerability, it does increase the risk of prototype pollution. This type of protection should be enabled by default in JS runtime validators, as they can be placed in front of external APIs.

Formischかっこいい #workers_tech #ashibinaa

I still genuinely think that if you're using Hono, you should be using Valibot instead of Zod The point of Hono is to be really small, and Zod defeats the purpose

i ranked some validation libraries this is very official and the results are final

I am excited to announce the release of Valibot v1, a 1 kB alternative to Zod with a type-safe and easy-to-remember API. Check out the announcement post now: valibot.dev/blog/valibot-v1-…