[WIP] fix(fuzz): strip metadata when collecting push bytes

[grandizzy]

Motivation

Closes #8115
When we collect values from push bytes we don't strip metadata from contract bytecode. This results in irrelevant values added to fuzz dictionary and it also affects forge snapshot (each time an comment is added / removed the snapshot values are changed)

Would be nice to have such method for getting bytecode without metadata in Bytecode revm primitives (and use in verify code as well)

Solution

• proposed solution is looking up the metadata len as last 2 bytes from original_byte_slice (prev we were using bytes_slice which is padded with 32 zero bytes) and then remove metadata
• the bytecode len is checked to be greater than 2 (cheatcode address have a len of 1) and greater than extracted metadata (the default create2 deployer does not contain metadata so it panics).
  TBD: should we ignore these addresses? check should be preserved in order to avoid panics from other edge cases
• tested with https://github.com/cruzdanilo/foundry-snapshot-repro by adding comments to SnapshotTest contract (test is too big to be included as unit test and I wasn't able to simplify it and reproduce issue)

[mattsse]

not an expert on bytecode layout, but this explainer made sense to me.

if this is something that holds true, we could upstream this to revm perhaps

pending @DaniPopes

[mattsse]

is this always true?

[mds1]

For initcode, no. Sometimes e.g. long strings in a constructor are after the metadata hash.

For runtime code, yes, IF the metadata hash is present. When bytecode_hash = none there is still a metadata hash containing the solidity version, so the last two bytes are still the length. When cbor_metadata = false then there is no bytecode hash at all and the last two bytes are code

To know for certain if what you have is a metadata hash, without knowing the compilation settings, you'd assume the last two bytes are the length of the metadata hash, then try CBOR decoding that hash. If it decodes successfully, it's a metadata hash, otherwise it's code.

Here is some old, probably bad, rust code I had that did this using ciborium: https://github.com/ScopeLift/cove-backend/blob/d8c875c5a64bf89a874892bc8ed57751c47b1233/src/bytecode.rs#L53-L97

[mattsse]

this could still fail due to the -2, but not sure if possible

perhaps metadata_offset = metadata_len + 2

&code[..code.len() - metadata_offset];