YP semantics for RETURNDATACOPY are incorrect

[Arachnid]

The YP says that RETURNDATACOPY is zero padded, but EIP 211 specifies it's an error to request data past the end of the buffer. geth's implementation concurs.

[recmo]

It is actually covered in section 9.4.2."Exceptional Halting". This is consistent with the specification for invalid JUMPs and other potentially reverting instructions, none of which have a mention of this in the instructions table.

That being said, the zero-padding in the expression is misleading. These values can never be created without reverting. Perhaps we can replace the value zero in the out of bounds case with "exceptional halt otherwise" ?

parity's implementation

[recmo]

Twitter discussion leading up to this report: https://twitter.com/recmo/status/1150326278022729729