- Dan: was there a another piece to Fred's cast translation ?
- Simms: yeah John had some issues, we aren't clear on
- Remi, there may be case where the user never wants a reference project
- Simms: so the proposal doesn't enforce a reference projection...
- Remi: it's a language level thing: `Object` is still there
- Remi, so John were talking about `indy` for casting inline type
- John: main concern is the forced classfile change and how much
that breaks tooling
- Remi: there are 2 new byte codes...
- John: so you are concerned about bootstrapping indy
- Simms: well that, hard for JIT opt, and it's a large hammer
- John: so if indy is too hard for now, Fred's proposal would be the way
to go
- John: do like doing a constant pool reworking for checkcast
- When we see "Q" in a descriptor, need to go and look (from last
meeting)
- Concerned with generic methods and specialization, need to
further information to optimize with
- Simms: so envelop for "Q", but not for "L" or is it symmetric, with
legacy support for without ?
- John, it is not clear, we could have flag ?
- Fred: have proto code without "L", CP sharing a user string is a
problem, prefer envelop always, both L and Q
- Remi: someone who spins bytecodes, prefer not to have a flag day exercise
- Prefer as few byte-codes as possible affected
...note-taker was pulled into other matters...eof
