This is a stop-gap. The long term solution is to make anonymous functions faster and to make higher order function specialize on their function arguments.
On Thu, Sep 4, 2014 at 11:54 AM, Jake Bolewski <[email protected]> wrote: > See the discussion @ https://github.com/JuliaLang/julia/issues/1864 for > the trade-offs of this approach. > > > On Thursday, September 4, 2014 11:34:44 AM UTC-4, Job van der Zwan wrote: >> >> Cool. >> >> So as someone who doesn't use anonymous functions enough to understand >> the fine details: is this a stopgap solution, or a "won't be part of base >> because it makes a certain trade-off"-solution? >> >> On Thursday, 4 September 2014 00:29:31 UTC+2, Tim Holy wrote: >>> >>> I'm pleased to announce the availability of the FastAnonymous package: >>> https://github.com/timholy/FastAnonymous.jl >>> >>> Those of you who have used Julia for a while may know that things such >>> as >>> map(myfunction, collection) >>> do not have ideal performance, and particularly the anonymous-function >>> version >>> map(x->x^2, collection) >>> is quite slow. FastAnonymous provides a simple mechanism for getting >>> inlined >>> performance for arbitrary functions, even those that require additional >>> parameters. Please see the README for details. >>> >>> In this context it's also worth reminding folks of the existence of >>> Dahua's >>> excellent NumericFuns, which provides partially-overlapping >>> functionality with >>> a different API. >>> >>> Best, >>> --Tim >>> >>>
