On Wed, Feb 24, 2016 at 8:46 AM, Lars Mescheder <[email protected]> wrote:
> Hi,
> first thanks for all the amazing work on julia! I am currently trying to
> write a very general version of some optimization algorithm. Ideally, it
> would take an array of linear maps as input and then perform optimization
> depending on those.
> So, is it possible to make a function of the form
>
> function doostuff(Vector{Function} funcs)
>     ....
> end
>
> fast?
>
> Since I know the signature of all these function, it would be amazing if
> something like the following would be possible
>
> function doostuff(Vector{Method{InT, OutT}} funcs)
>     ....
> end
>
> and the compiler could use it to infer types.
>
> Maybe some metaprogramming magic could do the trick, i.e. I dynamically
> generate a function that takes the function index and its arguments as input
> and then calls the right function depending on the index? I believe with the
> recent changes to julia that make function arguments fast  this would result
> in a fast function.
> Before getting into the details of this approach though, I wanted to make
> sure there is no "easy" solution.

Ref https://github.com/JuliaLang/julia/issues/13984
See also my hacked version in the issue.

>
> Thanks in advance!
> Lars

Reply via email to