Wow, things are taking shape faster than I anticipated! I finalized the PUSH opcode, which required providing the opcode runner with a reference to the execution code and program counter, which needs to be incremented for the number of bytes corresponding to the push instruction. For now I am using a simple MachineState context struct for this.
I also implemented a few more simple opcodes such as SUB and POP. For each instruction I am also making sure that the overflow/underflow cases are handled correctly. The challenge tests help a lot with this.
An incredibly useful resource I forgot to mention is the EVM illustrated which does a great job of explaining the state transitions of the EVM and exactly what data dependencies are involved in each different operation.
From reading these slides, I already know that the two major challenges will be implementing the CREATE and CALL operations, but these will surely be the subject of future posts.
I am noticing that many opcodes have repeated functionality, such as the PUSH operation which has 32 different sizes, or the POP operation which has 16. I think it's worth to generalize the implementation of these runners to avoid repeated code, with a helper function that chooses the correct runner based on the opcode range.
Update: I did this with a simple util called buildOpcodeRangeObjects and I love how elegant & clear the result looks.
I am following the evm-from-scratch challenge which offers a test suite of the evm functionality, but I am thinking of starting to write proper unit tests to isolate the more complex parts of the project. Perhaps I will postpone this until I have to deal with the more difficult parts like CREATE and CALL.