JUMP and JUMPI instructions are only allowed to jump onto destinations that are (1) occupied by a JUMPDEST opcode, and (2) are not inside PUSH data. We make sure we update the destinations we are JUMPing to to account for the extra opcodes we are adding in by appending an assembly switch/case to all transpiled bytecode.

The Approach

Here's how we deal with it, at a high level: 1. Create a map of all pre-transpilation JUMPDEST locations to post-transpilation JUMPDEST locations 2. Add some footer bytecode that acts as a switch statement, reading the pre-transpilation JUMPDEST location and JUMPing to the associated post-transpilation JUMPDEST 3. Replace all JUMPs and JUMPIs to JUMP to the footer bytecode switch statement.

JUMP Transpilation Detail: Replacements

Note: operations will be listed as [operation] -- [resulting stack]

JUMP:

• Expected Stack: <dest>

• Replacement:

  • PUSH32 <JUMPDEST of footer switch statement> -- <JUMPDEST of footer switch statement> <dest>

  • JUMP -- <dest>

• Total Replacement:

  • JUMP => PUSH32 <JUMPDEST of footer switch statement> JUMP

JUMPI:

• Expected Stack: <dest> <condition>

• Replacement:

  • SWAP1 -- <condition> <dest>

  • PUSH32 <JUMPDEST of footer switch statement> -- <JUMPDEST of footer switch statement> <condition> <dest>

  • JUMPI -- <dest>

  • POP -- empty

• Total Replacement:

  • JUMPI => SWAP1 PUSH32 <JUMPDEST of footer switch statement> JUMPI POP

JUMPDEST:

• Expected Stack: <prev jumpdest from footer switch> (footer switch statement results in 1 excess stack element)

• Replacement:

  • JUMPDEST -- <prev jumpdest from footer switch>

  • POP -- empty

• Total Replacement:

  • JUMPDEST => JUMPDEST POP

JUMP Transpilation Detail: Footer Switch Statement

• Expected Stack: <prev jumpdest>

• Single comparison:

  • DUP1 -- <prev jumpdest> <prev jumpdest>

  • PUSH32 <first compare jumpdest> -- <first compare jumpdest> <prev jumpdest> <prev jumpdest>

  • EQ -- <true/false> <prev jumpdest>

  • PUSH32 <post-transpile jumpdest> -- <post-transpile jumpdest> <true/false> <prev jumpdest>

  • JUMPI -- <prev jumpdest>

Note on bytecode interpretation

Note that properly processing these conditions requires preprocessing the code; a particularly pathological use case is PUSH2 JUMPDEST PUSH1 PUSH2 JUMPDEST PUSH1 PUSH2 JUMPDEST PUSH1 ..., as this code has all JUMPDESTs invalid but an alternative piece of code equivalent to this but only with the leading PUSH2 replaced with another op (eg. BALANCE) will have all JUMPDESTS valid. We appropriately deal with this, both in our transpiler and purity checker.