Improve support for constructing zero-step float ranges (#29056)

* Improve support for constructing zero-step float ranges

Fixes `0.0 * (1:4)`, for example. From https://github.com/JuliaLang/julia/issues/29052#issuecomment-418825887

* simplify

base/twiceprecision.jl

@@ -357,7 +357,7 @@ StepRangeLen(ref::TwicePrecision{T}, step::TwicePrecision{T},
 
 # Construct range for rational start=start_n/den, step=step_n/den
 function floatrange(::Type{T}, start_n::Integer, step_n::Integer, len::Integer, den::Integer) where T
-    if len < 2
+    if len < 2 || step_n == 0
         return steprangelen_hp(T, (start_n, den), (step_n, den), 0, Int(len), 1)
     end
     # index of smallest-magnitude value

test/ranges.jl

@@ -1374,6 +1374,40 @@ end # module NonStandardIntegerRangeTest
     end
 end
 
+@testset "constant-valued ranges (issues #10391 and #29052)" begin
+    for r in ((1:4), (1:1:4), (1.0:4.0))
+        if eltype(r) === Int
+            @test_broken @inferred(0 * r) == [0.0, 0.0, 0.0, 0.0]
+            @test_broken @inferred(0 .* r) == [0.0, 0.0, 0.0, 0.0]
+            @test_broken @inferred(r + (4:-1:1)) == [5.0, 5.0, 5.0, 5.0]
+            @test_broken @inferred(r .+ (4:-1:1)) == [5.0, 5.0, 5.0, 5.0]
+        else
+            @test @inferred(0 * r) == [0.0, 0.0, 0.0, 0.0]
+            @test @inferred(0 .* r) == [0.0, 0.0, 0.0, 0.0]
+            @test @inferred(r + (4:-1:1)) == [5.0, 5.0, 5.0, 5.0]
+            @test @inferred(r .+ (4:-1:1)) == [5.0, 5.0, 5.0, 5.0]
+        end
+        @test @inferred(r .+ (4.0:-1:1)) == [5.0, 5.0, 5.0, 5.0]
+        @test @inferred(0.0 * r) == [0.0, 0.0, 0.0, 0.0]
+        @test @inferred(0.0 .* r) == [0.0, 0.0, 0.0, 0.0]
+        @test @inferred(r / Inf) == [0.0, 0.0, 0.0, 0.0]
+        @test @inferred(r ./ Inf) == [0.0, 0.0, 0.0, 0.0]
+    end
+
+    @test_broken @inferred(range(0, step=0, length=4)) == [0, 0, 0, 0]
+    @test @inferred(range(0, stop=0, length=4)) == [0, 0, 0, 0]
+    @test @inferred(range(0.0, step=0.0, length=4)) == [0.0, 0.0, 0.0, 0.0]
+    @test @inferred(range(0.0, stop=0.0, length=4)) == [0.0, 0.0, 0.0, 0.0]
+    @test @inferred(range(0, step=0.0, length=4)) == [0.0, 0.0, 0.0, 0.0]
+    @test @inferred(range(0.0, step=0, length=4)) == [0.0, 0.0, 0.0, 0.0]
+    @test @inferred(range(0, stop=0.0, length=4)) == [0.0, 0.0, 0.0, 0.0]
+    @test @inferred(range(0.0, stop=0, length=4)) == [0.0, 0.0, 0.0, 0.0]
+
+    z4 = 0.0 * (1:4)
+    @test @inferred(z4 .+ (1:4)) === 1.0:1.0:4.0
+    @test @inferred(z4 .+ z4) === z4
+end
+
 @testset "getindex" begin
     @test getindex((typemax(UInt64)//one(UInt64):typemax(UInt64)//one(UInt64)), 1) == typemax(UInt64)//one(UInt64)
 end