Tab Atkins Jr.
jackalmage at gmail.com
Wed Aug 21 15:16:08 PDT 2013
On Wed, Aug 21, 2013 at 1:57 PM, Mark S. Miller <erights at google.com> wrote:
> Excellent excellent! It seems we are in agreement and clarity on all
> terminology issue, and on almost complete agreement on substantive issues.
> The only remaining issue is whether either of the following expressions
> always return a fresh promise, and if they are exactly equivalent:
> a) Promise.resolve(v1)
> b) new Promise(r=>r.resolve(v1))
> When v1 is a promise,
> x) I prefer that #a return v1 and #b return a fresh promise that adopts v1.
> y) You prefer that both #a and #b return a fresh promise
> z) The fewest feasible allocations would have both return v1.
> w) I do not expect anyone will want #a to return a fresh promise and #b to
> return v1.
> We both agree that if we settle on either #y or #z, then #a and #b are
> equivalent. Otherwise they are equivalent up to this issue.
> The reason I prefer that #b return a fresh promise is that it is surprising
> for a "new" expression to return something other than a fresh object.
I think it's super bizarre for a class static to act differently from
the trivial desugaring to the same-named instance method. (In this
case, it's not *quite* an instance method, but close enough. ^_^ It
would be an instance method if we weren't caring about capability
> The reason I prefer #a is that the primary use case for #a is coercion --
> given possible promise v1, gimme a guaranteed genuine promise that
> designates the same (possibly eventual) non-promise. See calls to the Q()
> function at http://wiki.ecmascript.org/doku.php?id=strawman:concurrency and
> http://es-lab.googlecode.com/svn/trunk/src/ses/contract/ (which corresponds
> to http://research.google.com/pubs/pub40673.html ). These calls are numerous
> and necessary, and sufficiently frequent that the extra transient allocation
> in the typical case will be a noticeable cost.
That's reasonable, but we should do it with a different function that
just promotes things into promises, or leave it to library code when
they need it. (We don't, for example, have such an operation for
Array/Map/Set, though I have used such an operation before.)
More information about the es-discuss