refactor: bubble up PrecompileError

[Wodann]

Precompile errors are represented as a String, but there doesn't seem to be anything limiting us from bubbling up the PrecompileError type. This gives us better static typing for precompile errors.

[codspeed-hq]

CodSpeed Performance Report

Merging #2191 will degrade performances by 4.78%

_{Comparing NomicFoundation:refactor/precompile-error (48f6528) with main (f2a2eb7)}

Summary

❌ 1 regressions
✅ 7 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
❌	precompile bench | ecrecover precompile	194.7 µs	204.5 µs	-4.78%

[rakita]

PrecompileError is an internal error similar to Halt, so it is not possible to bubble it up to outside. It is converted to InterpreterResult.

[Wodann]

PrecompileError is an internal error similar to Halt, so it is not possible to bubble it up to outside. It is converted to InterpreterResult.

I could be wrong, but we actually do convert PrecompileError to an EVMError here:
https://github.com/bluealloy/revm/blob/main/crates/precompile/src/interface.rs#L69

It's also exposed in EvmError::Precompile as a String. So we are actually exposing the error, as far as I can tell.

[Wodann]

CodSpeed Performance Report

Merging #2191 will degrade performances by 4.78%

Comparing NomicFoundation:refactor/precompile-error (48f6528) with main (f2a2eb7)

Summary

❌ 1 regressions ✅ 7 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

Benchmark BASE HEAD Change
❌ precompile bench | ecrecover precompile 194.7 µs 204.5 µs -4.78%

I expect the performance regression to be a result of the usage of a nested enum with String. This could be avoided by using enum variants for the BLS12_381 precompile. Something like:

pub enum Error {
    ...
    #[error("G1ADD input should be {G1_ADD_INPUT_LENGTH} bytes, was {length}")]
    InvalidG1AddInputLength { length: usize },
    ...
}