'MethodError' message doesn't show 'Closest candidates'

[samo-lin]

Problem

'MethodError' message doesn't show 'Closest candidates' if the function/method has more than one argument. Here is an illustration with floats:

$> julia

julia> using CUDAnative, CuArrays


julia> f1(a::Float32) = return
f1 (generic function with 1 method)

julia> f2(a::Float32, b::Float32) = return
f2 (generic function with 1 method)

julia> # Call with arguments of correct type.
       a = Float32(3.0)
3.0f0

julia> @cuda f1(a)

julia> @cuda f2(a, a)

julia> # Call with arguments of wrong type.
       a = Float64(3.0)
3.0

julia> @cuda f1(a)
ERROR: MethodError: no method matching f1(::Type{Float64})
Closest candidates are:
  f1(!Matched::Float32) at REPL[2]:1
Stacktrace:
 [1] method_age(::Function, ::Tuple{DataType}) at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:76
 [2] macro expansion at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:372 [inlined]
 [3] #cufunction#176(::Nothing, ::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}, ::typeof(cufunction), ::typeof(f1), ::Type{Tuple{Float64}}) at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:357
 [4] cufunction(::Function, ::Type) at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:357
 [5] top-level scope at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:174
 [6] top-level scope at gcutils.jl:87
 [7] top-level scope at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:171

julia> @cuda f2(a, a)
ERROR: MethodError: no method matching f2(::Type{Float64}, ::Type{Float64})
Stacktrace:
 [1] method_age(::Function, ::Tuple{DataType,DataType}) at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:76
 [2] macro expansion at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:372 [inlined]
 [3] #cufunction#176(::Nothing, ::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}, ::typeof(cufunction), ::typeof(f2), ::Type{Tuple{Float64,Float64}}) at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:357
 [4] cufunction(::Function, ::Type) at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:357
 [5] top-level scope at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:174
 [6] top-level scope at gcutils.jl:87
 [7] top-level scope at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:171

and here is an illustration with CuArrays:

$> julia

julia> using CUDAnative, CuArrays


julia> f3(A::CuDeviceArray{Float32}) = return
f3 (generic function with 1 method)

julia> f4(A::CuDeviceArray{Float32}, B::CuDeviceArray{Float32}) = return
f4 (generic function with 1 method)

julia> # Call with arguments of correct type.
       A = CuArrays.zeros(Float32, 3)
3-element CuArray{Float32,1}:
 0.0
 0.0
 0.0

julia> @cuda f3(A)

julia> @cuda f4(A, A)

julia> # Call with arguments of wrong type.
       A = CuArrays.zeros(Float64, 3)
3-element CuArray{Float64,1}:
 0.0
 0.0
 0.0

julia> @cuda f3(A)
ERROR: MethodError: no method matching f3(::Type{CuDeviceArray{Float64,1,CUDAnative.AS.Global}})
Closest candidates are:
  f3(!Matched::CuDeviceArray{Float32,N,A} where A where N) at REPL[2]:1
Stacktrace:
 [1] method_age(::Function, ::Tuple{DataType}) at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:76
 [2] macro expansion at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:372 [inlined]
 [3] #cufunction#176(::Nothing, ::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}, ::typeof(cufunction), ::typeof(f3), ::Type{Tuple{CuDeviceArray{Float64,1,CUDAnative.AS.Global}}}) at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:357
 [4] cufunction(::Function, ::Type) at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:357
 [5] top-level scope at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:174
 [6] top-level scope at gcutils.jl:87
 [7] top-level scope at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:171

julia> @cuda f4(A, A)
ERROR: MethodError: no method matching f4(::Type{CuDeviceArray{Float64,1,CUDAnative.AS.Global}}, ::Type{CuDeviceArray{Float64,1,CUDAnative.AS.Global}})
Stacktrace:
 [1] method_age(::Function, ::Tuple{DataType,DataType}) at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:76
 [2] macro expansion at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:372 [inlined]
 [3] #cufunction#176(::Nothing, ::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}, ::typeof(cufunction), ::typeof(f4), ::Type{Tuple{CuDeviceArray{Float64,1,CUDAnative.AS.Global},CuDeviceArray{Float64,1,CUDAnative.AS.Global}}}) at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:357
 [4] cufunction(::Function, ::Type) at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:357
 [5] top-level scope at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:174
 [6] top-level scope at gcutils.jl:87
 [7] top-level scope at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:171

Printing always the 'Closest candidates' would be particularly useful for debugging something like this:

julia> using CUDAnative, CuArrays


julia> f5(A::CuArray{Float32}, B::CuArray{Float32}) = return  # Function declaration uses the wrong type (should use CuDeviceArray)
f5 (generic function with 1 method)

julia> # Call with arguments of correct type, but the function declaration uses the wrong type.
       A = CuArrays.zeros(Float32, 3)
3-element CuArray{Float32,1}:
 0.0
 0.0
 0.0

julia> @cuda f5(A, A)
ERROR: MethodError: no method matching f5(::Type{CuDeviceArray{Float32,1,CUDAnative.AS.Global}}, ::Type{CuDeviceArray{Float32,1,CUDAnative.AS.Global}})
Stacktrace:
 [1] method_age(::Function, ::Tuple{DataType,DataType}) at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:76
 [2] macro expansion at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:372 [inlined]
 [3] #cufunction#176(::Nothing, ::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}, ::typeof(cufunction), ::typeof(f5), ::Type{Tuple{CuDeviceArray{Float32,1,CUDAnative.AS.Global},CuDeviceArray{Float32,1,CUDAnative.AS.Global}}}) at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:357
 [4] cufunction(::Function, ::Type) at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:357
 [5] top-level scope at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:174
 [6] top-level scope at gcutils.jl:87
 [7] top-level scope at /apps/dom/UES/jenkins/7.0.UP01/gpu/easybuild/software/Julia/1.2.0-CrayGNU-19.10-cuda-10.1/extensions/packages/CUDAnative/Lr0yj/src/execution.jl:171

Suggestion

Printing always the 'Closest candidates'.

Appendix

Code for reproduction without output (to copy and paste to REPL):

using CUDAnative, CuArrays

f1(a::Float32) = return
f2(a::Float32, b::Float32) = return

# Call with arguments of correct type.
a = Float32(3.0)
@cuda f1(a)
@cuda f2(a, a)

# Call with arguments of wrong type.
a = Float64(3.0)
@cuda f1(a)
@cuda f2(a, a)

using CUDAnative, CuArrays

f3(A::CuDeviceArray{Float32}) = return
f4(A::CuDeviceArray{Float32}, B::CuDeviceArray{Float32}) = return

# Call with arguments of correct type.
A = CuArrays.zeros(Float32, 3)
@cuda f3(A)
@cuda f4(A, A)

# Call with arguments of wrong type.
A = CuArrays.zeros(Float64, 3)
@cuda f3(A)
@cuda f4(A, A)

using CUDAnative, CuArrays

f5(A::CuArray{Float32}, B::CuArray{Float32}) = return  # Function declaration uses the wrong type (should use CuDeviceArray)

# Call with arguments of correct type, but the function declaration uses the wrong type.
A = CuArrays.zeros(Float32, 3)
@cuda f5(A, A)

[maleadt]

You don't get these on the CPU either:

julia> f1(a)
ERROR: MethodError: no method matching f1(::Float64)
Closest candidates are:
  f1(::Float32) at REPL[22]:1
Stacktrace:
 [1] top-level scope at REPL[30]:1

julia> f2(a, a)
ERROR: MethodError: no method matching f2(::Float64, ::Float64)
Stacktrace:
 [1] top-level scope at REPL[31]:1

And we just reuse the MethodError infrastructure from there. Note that the output here shows an error with the kind of MethodError generated, which has been fixed in #472. But if you want better output for the MethodError suggestions, please file an issue upstream (on the Julia issue tracker).

[samo-lin]

OK, sorry, I didn't realize that this is not an issue specific to CUDAnative. I will therefore open an issue there. Thanks!!