EIP-2315 "Simple Subroutines for the EVM" - Analysis

[holiman]

Some quick notes/questions, I’ll probably have more when I’ve digested it a bit further.

@chfast

Fallthrough to next subroutine

This one is pretty simple, and I agree. Implementation-wise, it can be done either as you describe it or something else, but it’s not a big change.

@chfast

JUMPs across subroutines

Problem

The EIP intentionally does not modify the semantics of JUMP / JUMPI s. They are still only restricted to targeting JUMPDEST s. It is allowed to jump from any point in one subroutine to any JUMPDEST -marked point in another subroutine.

I agree this is unfortunate.

@chfast

Benefits

1. When checking if a jump destination is valid search is performed on a smaller collection (limited to current subroutine scope only).

This is not necessarily true. Geth does a one-pass analysis, and the lookup is O(1) on the bitmap. If we have N different (small) analysis:es for each contract, it will probably actually be slower.

@chfast

JUMPDEST analysis (collecting valid jump destinations) can be performed lazily per subroutine. Optimized EVM implementation may perform advanced code analysis and/or translation beyond collecting JUMPDEST offsets.

It can be done lazily even now. Not until an actual jump is there any need to do the analysis (geth does this, AFAIR). In theory, the analysis could stop after going past the desired jump location, and continue later if needed.

I agree with the other benefits, and the concept in general.

@chfast

This restriction can potentially break existing contracts. E.g.

Oh, that’s a bit nasty, nice catch!