Open Question: Bytecode compilation

[roninjin10]

Summary

Generating bytecode is as much as 100x slower than generating an ABI. Furthermore, most use cases do not need the bytecode only the abi because the bytecode is already deployed. For this reason, Tevm only compiles bytecode for files that explicitly have a .s.sol file extension. Users are expected to either name their files .s.sol or reexport their contract from a .s.sol file

This issue is to revisit that decision before going stable

Alternative solutions

0 Keep it the same

Pros

• Pretty simple solution that intuitively makes sense based on foundry conventions of what a script is
• Forces dev to be explicit about whether they want bytecode

Cons

• New users trying to use the bytecode can struggle to understand why it is not there

1 Always compile the bytecode

Pros

• simplist solution
• good devx

Cons

• Bloats bundle with unused bytecode
• Hurts performance of your production build and dev server (not lsp is unaffected by this issue as it always uses Hex as type)

2 Import attributes

Uses the new import attribute

import {MyContract} from '~/contracts/MyContract.sol' // by default no bytecode
import {MyContract} from '~/contracts/MyContract.sol' as { type: 'sol-script' } // this will have bytecode
import {MyContract} from '~/contracts/MyContract.sol' as { type: 'sol' } // this will not have bytecode

Pros

• uses a readable syntax
• don't need to think about or change file extensions

Cons

• Doesn't enforce file extension conventions
• Yet another API the developer needs to learn to use Tevm

3 Make the bytecode different import target

This solution borrows from typechain

import { MyContract, MyContract__bytecode } from '~/contracts/MyContract.sol'

const contractWithBytecode = MyContract.withBytecode(MyContract__bytecode)

Pros

• Explicit bytecode compilation
• Explicit bytecode bundling
• Bytecode can be autoimported

Cons

• Ugly looking subjectively
• Requires extra boilerplate to use the contract

[linear[bot]]

TEV-1394 Open Question: Bytecode compilation

[roninjin10]

Historically i initially made this change to not compiling bytecode by default well over a year ago. Users were suffering from tevm slowing their LSP to a crawl when they started adding more and more contracts and this change put out that fire.

Since then we have:

1. Added caching
2. Made the LSP and bundler code paths different such that this would no longer affect the LSP only the prodcution build and dev server

So even though I think the current API is good and perhaps my favorite API it's worth revisiting before going stable