Code suggestions for #1210

[hyrodium]

Overview

This PR adds suggestions for #1210.

• Add support for @SArray rand(rng, 3)
• Simple output for @macroexpand @SArray rand(rng, Float32, 4)
• Better error handling for @SArray rand(rng, Float32, Float32)
• Add support for randn and randexp

Details

Add support for @SArray rand(rng, 3)

On the original PR #1210

julia> using Random, StaticArrays

julia> rng = Random.MersenneTwister(42)
MersenneTwister(42)

julia> @SArray rand(rng, 3)
ERROR: MethodError: Cannot `convert` an object of type Vector{Float64} to an object of type Int64
[...]

On this PR

julia> using Random, StaticArrays

julia> rng = Random.MersenneTwister(42)
MersenneTwister(42)

julia> @SArray rand(rng, 3)
3-element SVector{3, Float64} with indices SOneTo(3):
 0.5331830160438613
 0.4540291355871424
 0.017686826714964354

Simple output for @macroexpand @SArray rand(rng, Float32, 4)

On the original PR #1210

julia> @SArray rand(rng, Float32, 4)
4-element SVector{4, Float32} with indices SOneTo(4):
 0.9603034
 0.76554394
 0.46937835
 0.2332462

julia> @macroexpand @SArray rand(rng, Float32, 4)
quote
    #= /home/hyrodium/.julia/dev/StaticArrays/src/SArray.jl:214 =#
    if rng isa StaticArrays.DataType
        #= /home/hyrodium/.julia/dev/StaticArrays/src/SArray.jl:215 =#
        StaticArrays.rand((SArray){(Tuple){Float32, 4}, rng})
    elseif #= /home/hyrodium/.julia/dev/StaticArrays/src/SArray.jl:216 =# rng isa StaticArrays.Integer
        #= /home/hyrodium/.julia/dev/StaticArrays/src/SArray.jl:217 =#
        StaticArrays.rand((SArray){(Tuple){rng, Float32, 4}})
    elseif #= /home/hyrodium/.julia/dev/StaticArrays/src/SArray.jl:218 =# rng isa (StaticArrays.Random).AbstractRNG
        #= /home/hyrodium/.julia/dev/StaticArrays/src/SArray.jl:220 =#
        (StaticArrays.StaticArrays)._rand(rng, Float32, StaticArrays.Size(4), (SArray){StaticArrays.Tuple{4}, (StaticArrays.Random).gentype(Float32)})
    else
        #= /home/hyrodium/.julia/dev/StaticArrays/src/SArray.jl:231 =#
        (StaticArrays.StaticArrays)._rand((StaticArrays.Random).GLOBAL_RNG, rng, StaticArrays.Size(Float32, 4), (SArray){StaticArrays.Tuple{Float32, 4}, (StaticArrays.Random).gentype(rng)})
    end
end

On this PR

julia> @SArray rand(rng, Float32, 4)
4-element SVector{4, Float32} with indices SOneTo(4):
 0.22553396
 0.18603194
 0.16922617
 0.9104786

julia> @macroexpand @SArray rand(rng, Float32, 4)
:(StaticArrays._rand_with_Val(SArray, rng, Float32, StaticArrays._int2val(rng), StaticArrays._int2val(Float32), StaticArrays.Val(StaticArrays.tuple(4))))

Add support for randn and randexp

On the original PR #1210

julia> @SArray randn(rng, 5)
ERROR: TypeError: in Tuple, in parameter, expected Type, got a value of type MersenneTwister
Stacktrace:
[...]

julia> @SArray randexp(rng, 5)
ERROR: TypeError: in Tuple, in parameter, expected Type, got a value of type MersenneTwister
Stacktrace:
[...]

On this PR

julia> @SArray randn(rng, 5)
5-element SVector{5, Float64} with indices SOneTo(5):
  0.18702790710363
  0.5181487878771377
  1.4913791170403063
  0.3675627461748204
 -0.8862052960481365

julia> @SArray randexp(rng, 5)
5-element SVector{5, Float64} with indices SOneTo(5):
 1.11449101941971
 1.473948306820646
 0.06666365902783734
 0.39638956977933426
 0.5090311768294857

Note

• I added more test cases.
• I also updated functions for @SVector and SMatrox.
• The function _rand_with_Val is based on the comment (Extend macros with rand to support custom samplers #1210 (comment)).

[mateuszbaran]

I think this is a good idea, thanks for proposing that. Could you add some tests for the new functionality and safeguards for ex.args indexing (like the one I've proposed)?

[hyrodium]

Should I update @SVector and @SMatrix too? (x-ref: #1214)

[mateuszbaran]

Yes, I think it would be good to update these two macros too.

[hyrodium]

Hmm, I'm thinking of reusing static_array_gen in @SVector macro etc., but that may takes a while...
Replacing static_vector_gen in @SVector definition with

macro SVector(ex)
    ex = static_array_gen(SArray, ex, __module__)
    :(SVector($ex))
end

is almost fine, but some tests fail..

[hyrodium]

I have updated static_vector_gen. I will update static_matrix_gen and tests tomorrow.

[hyrodium]

Oooh, I didn't know that n=3; @SVector rand(n) is supported.

julia> n=3; @SVector rand(n)  # current master branch
3-element SVector{3, Float64} with indices SOneTo(3):
 0.41421419554720196
 0.8417028007187902
 0.05502353350873945

I assumed all of the array sizes must be integers but not variables in this PR.

julia> n=3; @SVector rand(n)  # on this PR
ERROR: LoadError: @SVector got bad expression: rand(n)
Stacktrace:
[...]

I noticed this behavior when testing @SMatrix...

StaticArrays.jl/test/eigen.jl

Lines 283 to 284 in 67d9c36

	for n=1:4
	m = @SMatrix randn(n,n)

[mateuszbaran]

Yes, that's why some macro variants generate if statements.

[hyrodium]

I'll update this PR this weekend:muscle:

[hyrodium]

Sorry for the late. I think this is ready to review after CI results turn green!
I have also updated the PR description.

[mateuszbaran]

Thanks! I will check this PR.

[mateuszbaran]

It looks great, thanks!

[hyrodium]

Thanks for the quick review! 😄