Valhalla Spec Experts Group 20190814
Attendees: Dan S, Simms, Frederic, Brian, John, Tobi, Dan H
Feel free to add any color, but the topics we touched were...
* (John) Circularity with static fields, SPEC impact
** There is a fix on its way into LW2, probably has JVMS spec impact
** Init error handling might need some attention
** Clarify which phase static default values
** DanS and Fred can sync up some more
** ...needs more thought.
* Collapsing the requirements
** (DanH) Interesting proposal
*** Nailing down the relationship between the two pieces, especially
from the outside.
** (Brian) Couple of implementation choices (inner classes, interfaces,
abstract)
*** either way impact on existing spec / code
*** Relates to RefObject / ValObject
*** Depends on sealing
* DanH: all the fields push down ?
** Brian: No, fields stay in inline-class, bring public methods
** John: yeah in inline-class, belongs to "concrete type"
*** The object type is a "view" on the inline-type
* DanS: what we are saying disallow LInlineClass; ? (Yes)
** What does that by us ?
** DanH: difficult to differentiate, e.g. reflection, method types, VM
needs two representations of the same thing.
** DanS: will "species" suffer the same problem ?
*** John: no, type will differ...but yeah, design discussion surrounding
templates need to happen yet
** DanS: from spec perspective does seem like a gain...
** Fred ...impl wise, yeah we need extra work to extract and transport
signature into the VM
* Brian: question to DanH, what you think ?
** DanH: yeah, it seems like we won't need to be tweak invoke byte-codes
(was a potential concern)
** Brian: so we are looking into how to use the mechanism for "records"
** John: method handles will what to know the relation
* DanS: so single inner class might cut it ?
* Brian: Yeah, there is a user model question: need new syntax, or is
inference enough
* Fred: statics, fields and methods...is there any magic there, or does
the static compiler take of it
** Brian: yeah...let's put that on the list. <TBD>
** Fred: any such thing in VBC today...Brian: yeah factories
** Fred: so need new factories ?
** DanS: there is a mix between lang/vm... lang doesn't allow covariant
return / vm might
* John: "garbage descriptors", if L-type is trash (eclair model)
** DanS: reflection is eager, but indy you could fallover
* "Equality for values"...we are out of time
** John: let's get that benchmarking summary for more information
* Simms: "Collapsing the requirements" no serious rejection, clearly
worth digging into
Cheers
/Simms
