Performance Optimization in serde_derive Macro

[nurmohammed840]

Context

The quote2 crate, which is an alternative lightweight version of quote, offers some improvements over quote

Proposal

If requested, I would be happy to submit a pull request to implement these optimizations. My proposal involves refactoring existing code to leverage the improved design in quote2

Please let me know if this contribution would be valuable. I can provide more details or send a pull request directly for review.

[oli-obk]

We don't have the capacity for reviewing an entire crate for feasability.

If you want to make a case, I recommend:

1. creating benchmarks, ideally by swapping out the quote usage in serde and showing concrete numbers
2. Running crater-at-home to see if swapping it out will break any crates (this requires extending crater-at-home to allow injecting patched crates)
3. Writing a paragraph or two explaining the differences between the crates

All of these are significant effort and may not pan out, so only do this if you think there is a strong argument for switching and want to convince us.

[nurmohammed840]

So, At the moment I only refactor serialization part, And the result is very promising 👌

Here’s a link to the prototype implementation

System Info

• CPU: Ryzen 7 5700 (3.8GHz - 4.6GHz)

Benchmark Result

Iteration	File	serde_derive	serde_derive_next	Improvement	Faster
1	simple.rs	253.5µs	120µs	52.66%	2.11x
	user.rs	339.9µs	210.9µs	37.96%	1.61x
10	simple.rs	536µs	331.8µs	38.06%	1.61x
	user.rs	2.8184ms	1.6602ms	41.09%	1.70x
100	simple.rs	4.6072ms	2.8672ms	37.77%	1.61x
	user.rs	27.6456ms	15.1838ms	45.08%	1.82x
10,000	simple.rs	447.8222ms	282.0575ms	37.03%	1.59x
	user.rs	2.3379857s	1.3935152s	40.43%	1.68x

@oli-obk, let me know, If I should continue refactoring (deserialization part) rest of the codebase.