Location

Mozilla Foundation 1981 Landings Drive Building K Mountain View, CA 94043-0801 USA

See a street map of our location for driving directions. Once there, see this satellite image to locate Building K within the complex.

Times

Sept. 21: 12pm-5pm PDT (we will dine out around 7pm). Sept. 22: 10am-5pm PDT

Attendees

  • Jeff Dyer, Adobe
  • Steven Johnson, Adobe
  • Dan Smith, Adobe
  • Ed Smith, Adobe
  • Michael O’Brien, Mbedthis
  • Brendan Eich, Mozilla
  • Graydon Hoare, Mozilla
  • Dave Herman, Northeastern
  • Lars Hansen, Opera
  • Iain Lamb, Yahoo!
  • Julien Lecomte, Yahoo!

Agenda

Notes

  • eval
    • method on global object
    • do we need eval(s, [oN, ..., o1])
      • where o1 is the head of the replacement scope chain
  • global self-name
    • just a property name, can be bound by sandboxing code as it wishes
  • yield
    • should make yield e and let (h) e use the same nonterminal for e
    • either: over-parenthesize
    • or: non-terminal for e is AssignmentExpression
    • resolved: use AssignmentExpression
  • documentation comments
    • lars: why not use a comment?
    • doug’s proposal does not reflect as doc
    • how does this relate to decorators?
      • need a unified proposal
      • want real-world use-cases so we don’t miss anything that should be ES4
  • dave/brendan shorter/more-compositional function expression forms
    • let function f(args) { body } ⇒ let f(args) { body } OR let f(args) expr
    • we have resolved to eliminate reference types in the spec
    • should we restrict optional reference types in the grammar? no
    • should we make the spec’s normative grammar be LL(1) or LR(1)? yes
    • jeff to take on formalizing the grammar, mob will help

proposals review

    • class A.<T> extends T { ... } should not be allowed
    • similar such questions may arise
    • dave: C# type non-erasure vs. Java erasure
    • do we allow any TypeExpression after : in initialiser annotation? yes
    • do we allow any Exprsssion after : in initialiser annotation? uhhh...
      • dave:
        • we want static type checker recognizing static constraint
        • casts for dynamic constraints (less common)
        • more common static constraint case should have lightweight syntax
      • agreement on these two points:
        • want {p: 42} : {p: int} where the annotated type is a TypeExpression
        • want cast T (E)
      • what about to?
        • want x to T where grammatically T is TypeExpression
        • confusion about foo().to(x) being backwards - should be from?
        • entertain proposals in the wiki for nice dynamic-to/is API syntax
      • resolved: want infix operator syntax for static case: TyExpr on right
      • what about is?
        • alternative is to match to and require static TyExpr on right
        • allowing any Expr means structural types must be named to be used
        • if we require TyExpr on right of is, we may break AS3 users
    • in good shape apart from wiki page title
    • dave to update based on recent type system work and previous item
    • agreement on nullability by default
    • discussion brought up need to:
      • update the spec before re-exporting
      • respond to es4-discuss list with pointers to new export
    • in good shape now (see recent numbers)
    • mob is doing int64 as extension; seems to fit
    • raised issues of spec language and completeness
    • build on E3 or try to improve it w/ a significantly different metalang?
    • take E3 metalang and clean it up a bit (a la ECMA-357)
    • dave to try writing a few more accessible spec styles for some productions

proposals, continued

  • enumerability
  • switch type
  • block expression
  • proper tail calls
  • type definitions
  • syntax for type expressions – fix ? to be postfix
    • all good
    • StopIteration is of type StopIterationClass
    • intrinsic::iterator must become iterator::get or some such
  • resurrected eval
  • expression closures, still reviewing
  • multiple compilation units - need to prove the two propositions
  • security wrappers - does it do enough to be worth its cost?
    • meaning: does enough? costs a lot?
  • issue with instrinsic::global
    • is it bound to the caller’s global in the “sandbox” (sic) object passed to the resurrected eval? lars said yes earlier, brendan said no; revised answer is no.
  • catchalls, hashcodes, operators, destructuring
    • All good
    • Note to self:
    for ([k] in o) => SyntaxError
    for ([k,v,u] in o) => SyntaxError
    for ([k,,,] in o) => ok
    for ([k,,] in o) => ok
    for ([k,v] in o) => ok
  • bug fixes
    • brendan: remove eval bug fixes, it has its own page
    • brendan: generic statics for Array and String should be split out
    • jeff: escaped newlines in string literals ok
    • graydon: pragma syntax update
    • otherwise looks good
    • update to leave typeof null === “object”
    • update to change typeof class === “object”
    • BUT: typeof String === “function” for backward compat
    • informative words expressing regret

day two

    • should we do as js1.7 and allow function delete? no
    • should we allow reserved identifiers after ::? yes
    • ok (discussion around clarity of implementation choice, how choice is one way or the other for all inputs, depending on input).
    • resolved: format conrol chars are not stripped from source input, therefore are preserved in string and regexp literals
    • Updated to note per yesterday’s discussion that typeof /re/ === “object”.
    • Also adopting the IE quirk that Opera and Mozilla do: /[/]/ matches “/”.
    • This means that #... line comments in /very-long/x regexps must balance [].
    • Brendan to clean up, move most to discussion, present minimal proposal
    • Iain: why not define a range generator function?
    • Discussion about +, <, == etc. for Array – put them in a new namespace that
    • new code can use: use namespace operators.
    • move to reflection library
    • use javadoc style comments (precedent: asdoc tool from adobe)
    • singularize intrinsic::globals
    • all good but nanotime:
      • discussion about accuracy needs – want delta-t for benchmarking, really
      • so don’t need nanoseconds, or want to impose them on all impls
      • ptw’s tick/tickScale proposal from es4-discuss considered too hardware-ish
        • not good if tickScale isn’t constant; if constant, it may have to be too large a number of nanoseconds in order for tick to be cheaply computed
    • dave: social psych reaction time research
    • lars/graydon: use nanoseconds since creation of Date object: d.nanoAge()
    • nanoAge to be drafted
    • Array.prototype.toJSONString
    • Object.prototype.toJSONString
    • String.prototype.parseJSON
    • String.prototype.trim (free-riding on JSON here)
  • the module perplex – packages don’t solve naming and loading issues
    • lars/dave: namespace() and name() should be in ClassType not Type
      • ditto for supertypes() and subtypes() (rename to super/subClasses())
    • dave: use iterators instead of arrays
    • dave: note to use [T] instead of no-longer-proposed Array.<T>
    • dave: should reflect public methods and fields, structural fields, etc.
    • InterfaceType? sure; InterfaceType.implementedBy()
  • iain: version reflection? object detection and try-eval rule
  • brendan: dict syntax for null-proto object initialisers
  • TODO:
    • brendan: slice
    • iain: json encoding and decoding, trim
    • lars: documentation
    • graydon: date and time
    • graydon and brendan: security wrappers
 
meetings/minutes_sep_21_2006.txt · Last modified: 2007/03/13 21:55 by graydon
 
Recent changes RSS feed Creative Commons License Donate Powered by PHP Valid XHTML 1.0 Valid CSS Driven by DokuWiki