Find Performance Bottleneck

[Skn0tt]

SuperJSON is somewhat slow - at least compared to not using SuperJSON. Our code currently is written for readability first, and not so much for optimal performance.

• Analyse the Code for it's bottlenecks
• Remove them

[tomhooijenga]

This project might be suffering from "death by a thousand cuts", aka there are a whole bunch of statements that in itself might not be very costly but it adds up.
Some quick examples

• array functions vs a plain old for loop
• lodash vs native (not always though!)
• isUndefined() vs ===

[KATT]

We've addressed this in tRPC by allowing different data transformers for upstream and downstream; upstream (client to server) are usually small payloads parsed by the server that needs to be secure and downstream are usually bigger payloads parsed by the client.

• https://trpc.io/docs/data-transformers#different-transformers-for-upload-and-download
• Original issue Use different transformers for client<->server trpc/trpc#378
• PR that addressed it: add combined transformers trpc/trpc#393

[Skn0tt]

That's a good idea! Still, I'm pretty sure there should be a way of speeding up SuperJSON to better compete with devalue. But I haven't found it yet 🤷‍♂️

[KATT]

Some ideas, algos isn't really my speciality, so these are mostly novice common sense

• You might not need to deep clone the object first (or have an option where this is skipped)
• Every setDeep()-seem to be traversing through the object
  • Potentially a faster strategy:
    • traverse through the whole object once and have a reference to path to update
    • use the reference to know "dead ends" and break; when there's nothing further down to update
• Make "slow" features optional
  • Maybe one of error/class/map/set/etc have a specific bottle-neck (or that you don't need them at all in your app) - allow for optionally disabling them (Promote best practices / clearer boundaries / do less #85)
• Rather than traversing yourself when encoding/serializing, you could try using JSON.stringify replacer & JSON.parse reviver.

For inspiration, you can look at https://github.com/yahoo/serialize-javascript where they have the isJSON-option that makes it "to be over 3x faster".