Invoking Core._apply

[dalum]

This one has left me head scratching:

julia> hasmethod(Core._apply, Tuple{typeof(+), Tuple{Float64, Float64}})
true

julia> applicable(Core._apply, +, (2.0, 2.0))
true

julia> applicable(invoke, Core._apply, Tuple{typeof(+), Tuple{Float64, Float64}}, +, (2.0, 2.0))
true

julia> invoke(Core._apply, Tuple{typeof(+), Tuple{Float64, Float64}}, +, (2.0, 2.0))
ERROR: MethodError: no method matching _apply(::typeof(+), ::Tuple{Float64,Float64})
Stacktrace:
 [1] top-level scope at none:0

The issue is that invoke(Core._apply, ...) doesn't work for any arguments. The easiest fix would be to make applicable not lie about being able to call invoke with the given arguments, but I think it would be even better if invoke(Core._apply, ...) worked. 🙂

I discovered this when using Cassette to replace internal calls by invoke, but the fact that Core._apply can't be used with invoke breaks my assumption that f(x, y) == invoke(f, Tuple{typeof(x), typeof(y)}, x, y).

[JeffBezanson]

Making invoke just work is probably the best thing to do. _apply and other "builtins" aren't generic functions, but they can probably consistently behave as if they had a single Any... method for most purposes.

[laborg]

I assume this isn't relevant after #39115 anymore.

[vtjnash]

Fixed by #43865 actually!

julia> invoke(Core._apply_iterate, Tuple{typeof(iterate), typeof(+), Tuple{Float64, Float64}}, iterate, +, (2.0, 2.0))
4.0

julia> typeof(Core._apply_iterate).name.mt.defs.func.invokes
Core.TypeMapEntry(nothing, Tuple{typeof(Core._apply_iterate), typeof(iterate), typeof(+), Tuple{Float64, Float64}}, nothing, svec(), 0x0000000000000001, 0xffffffffffffffff, MethodInstance for _apply_iterate(...), true, true, false)