Postfix notation not typing

[dlfivefifty]

I was under the impression f(g) and g|>f are synonyms, so the following is strange:

julia> foo(cfs)=length(cfs)
foo (generic function with 1 method)

julia> foo2(cfs)=cfs|>length
foo2 (generic function with 1 method)

julia> @code_typed foo([1,2,3])
1-element Array{Any,1}:
:($(Expr(:lambda, Any[:cfs], Any[Any[],Any[Any[:cfs,Array{Int64,1},0]],Any[]], :(begin # none, line 1:
return (top(arraylen))(cfs::Array{Int64,1})::Int64
end::Int64))))

julia> @code_typed foo2([1,2,3])
1-element Array{Any,1}:
:($(Expr(:lambda, Any[:cfs], Any[Any[],Any[Any[:cfs,Array{Int64,1},0]],Any[]], :(begin # none, line 1:
return (length::F)(cfs::Array{Int64,1})
end))))

[simonster]

|> is a function like any other; cfs |> length behaves the same way as g(cfs, length) where g(x, f) = f(x). We can't (yet) optimize cases where functions are passed to functions (see "function-valued argument inlining" in #3440), but this is planned.

[dlfivefifty]

OK I guess its best to avoid |> in library code for the time being.

Could a work around be to force |>(f,g)  to be inlined?  (If forced inlining is possible.)

On 31 Dec 2014, at 11:23 am, Simon Kornblith notifications@github.com wrote:

|> is a function

julia/base/operators.jl

Line 178 in e099867

|>(x, f::Callable) = f(x)

like any other; cfs |> length behaves the same way as g(cfs, length) where g(x, f) = f(x). We can't (yet) optimize cases where functions are passed to functions (see "function-valued argument inlining" in #3440 #3440), but this is planned.

—
Reply to this email directly or view it on GitHub #9509 (comment).

[simonster]

In the code above, |> is actually inlined into foo2, but inlining happens after type inference so it doesn't provide any type information. We really want to inline length into |> and then inline |> into foo2 so that the type information propagates, or else rerun type inference after inlining.

[vtjnash]

covered by #210