APW2014 GLR discussion highlight summary

gistfile1.txt

Since I know many people are eager for it, here's a very brief summary of GLR-related 
decisions and accomplishments from #apw2014.  In the next 2..3 days I plan to create 
blog posts with much greater discussion of each. If any seem particularly confusing 
or troubling to you, perhaps wait for the blog post(s) before panicking.  :-)

* flattening rules have been revised somewhat
  + list assignment and array constructor continue to flatten their arguments (no change)
  + for (args) { ... }  flattens args (no change)
  + .[] never flattens its invocant (consistency, capability, and performance wins)
  + function calls tend to flatten their arguments (e.g. "sort", "map", "pick")
  + method calls tend to not flatten their invocants (e.g., ".sort", ".map", ".pick")
    + .map will no longer flatten its invocant
    + .for will be introduced as an equivalent of .flat.map (corresponds to 'for' 
      statement semantic)
  + operator argument flattening depends on operator (typically grouped by 
    operator precedence)
  + operator return values tend to not be automatically flattened
* Parcel will likely disappear as a type
  + infix:<,> will create (immutable) List objects
  + many benefits:
    + eliminate ongoing confusion between Parcel and List (programmer win)
    + eliminating need to convert between Parcels and Lists (likely big 
      performance win)
    + .perl output won't need "(1,2,3).list"`
* MapIter will become LoopIter
  + for/while/until/repeat/loop will be able to produce lazy lists
    + many existing instances of gather+take will be able to be rewritten as 
      inexpensive lazy loops (big performance win)
    + more list comprehensions will work (language spec compliance win)
  + LoopIter will be able to operate directly from Iterator chains without an
    intermediate List (big performance win)
    + significant step towards enabling GC collect of "already processed 
      values" (memory win, but some further work needed)
* sink context made significantly smarter, much less codegen and method calls
  required (performance win)
* roadmap for hyper/race processing of Lists
* better-defined semantics of .list, .flat, .items, .values
* clear/clean API definition for List, Iterable, Iterator, LoL, etc. 
  (foundation for S07 rewrite)
* Nil will be updated to match current language spec