gistfile1.txt

* In the s-table, only REPR is REPR-dependent. :)
* All the s-table caches are optional. A meta-object publishes those
  it can sanely use. Lack of doing so just implies - in the case you
  asked about - a fallback call to .^typecheck. So the type check
  cache simply isn't applicable to subset types - or at least the
  routiney bit. At some point it'll check against a non-subset type
  and that can take advantage of it though.
* If you find a system with non 8-bit bytes that the rest of Perl 6 
  could otherwise run on, let me know. ;-)
* The missing string mark is a bug, thanks. Though in nqp-rx we
  never use that REPR directly, but it always is inlined into a
  P6opaque, which does handle the marking...but it would be a problem
  if the repr was used directly, which would be possible.
* native type declarator just implies using NativeHoW. That in itself
  does little; it demenads you specify a repr explicitly though. In
  the future it may sanity-check that it'll probably sanity check that
  it's an inlineable one. It also doesn't allow methods and attrs and
  the like. Perhaps the "what does this box to" should live on this
  meta-object too. I'm still undecided on that area.
* At the moment the compiler must call the correct get_attribute
  [_int|_num|_str]? variant and emit its own boxing. So that those
  implementing REPRs don't have to repeat that. Note that I didn't add
  safety checks into those REPR functions yet in P6opaque...partly
  because I'm not 100% decided on this area and want to play with the
  compilation of things like you asked about before I decide.