Compiler optimization example outdated?

[johannes-fischer]

I was just trying to get a feeling for which expressions the compiler optimizes away, to not get wrong timings. When running the example in the docs, it seems it is not optimized away by the compiler anymore:

On Julia 1.8.4:

julia> a = 1; b = 2;

julia> @btime $a + $b
  2.660 ns (0 allocations: 0 bytes)
3

julia> @btime $(Ref(a))[] + $(Ref(b))[]
  2.455 ns (0 allocations: 0 bytes)
3

julia> @btime 3
  1.450 ns (0 allocations: 0 bytes)
3

On Julia 1.7.3:

julia> a = 1; b = 2
2

julia> a = 1; b = 2;

julia> @btime $a + $b
  1.451 ns (0 allocations: 0 bytes)
3

julia> @btime $(Ref(a))[] + $(Ref(b))[]
  1.451 ns (0 allocations: 0 bytes)
3

julia> @btime 3
  0.023 ns (0 allocations: 0 bytes)
3

Does the compiler not anymore optimize code away in @btime? Or is the trick with referencing and de-referencing not working anymore?

[vtjnash]

The Ref is still written correctly, but the compiler is now disallowed from removing the loop

[johannes-fischer]

Thanks for the quick answer! But are there still pitfalls to stay aware of that the compiler is allowed to optimize, which can lead to wrong benchmark results? If yes, I would suggest to update the example in the docs with something where it actually shows this behavior.