On Tue, Feb 24, 2015 at 5:57 PM, Matt Rice <[email protected]> wrote:
> On Tue, Feb 24, 2015 at 5:21 PM, Matt Rice <[email protected]> wrote:
>> On Tue, Feb 24, 2015 at 4:19 PM, Matt Oliveri <[email protected]> wrote:
>>> - Matt Rice keeps bringing up SML records, and I'm not sure why. (Sorry, 
>>> Matt.)
>>
>> No apology necessary really I probably should though, since i in
>> general suck at explaining
>>
>> In general the places I use currying are more for partial
>> specification of type, that is the type of some subprogram (function)
>> is a partial specification of the type of the caller, and this usage
>> is trivially done without allocation with e.g. real hygenic macros...
>>
>>  from the discussion focusing on currying syntax, and the focus on
>> partial application of expressions on some value, It seems like we may
>> get that but still lack partially typed subprograms, which is what I
>> find important in the whole thing, if it allows me to write these
>> curried accessors without the allocations, and in that sml case it
>> shows a) how this feature of sml complicates the whole matter through
>> functions which return and accept abstract types as parameters that
>> can only become concrete through further passed parameters later in
>> the chain b) how it leads to unfamiliar types of errors.
>>
>> that is I don't much care about curried style function application
>> except as it pertains to the curry-howard correspondance and typing
>> through usage
>
> I think if i could amend this I think i'd say what i've been trying to
> get across, is It seems to me entirely that the
> allocation-requirements are a structure of usage and not necessarily
> of the arity of the initiating call, and thus fixing this as curried
> application syntax is the 'low hanging fruit', given the length of
> messages corresponding to this low hanging fruit though I should
> probably give up :)

This probably didn't come across as intended, what I mean is that we
approach the case in the form only for functions which return
functions, but I gather we expect types which are not functions and
when returning any type which contains a function this callers choice
of whether to allocate breaks down and we end up with a possibly
overzealous function annotation which may/may not be optimized out,

so this is just general discontent with things I consider to be ad-hoc, but also
..
I believe that this callers choice of allocating lambdas *is* an improvement,
and it's unforunate there isn't you know... rainbows everywhere.
_______________________________________________
bitc-dev mailing list
[email protected]
http://www.coyotos.org/mailman/listinfo/bitc-dev

Reply via email to