$x should become $(Ref(x))[]?

[stevengj]

It's a bit annoying when you have to use this convoluted Ref trick to prevent the compiler from constant-folding the whole calculation (ala #130):

julia> @btime exp(1);
  0.052 ns (0 allocations: 0 bytes)

julia> @btime exp($(Ref(1))[]);
  13.785 ns (0 allocations: 0 bytes)

It would be nicer if we could just use @btime exp($1) and the $ did the trick automatically.

(Another case just came up on discourse. cc @stillyslalom)

[ettersi]

+1 for a more convenient syntax, but I'm not sure whether $x is the right syntax. In my understanding of interpolation, x = 42; :(foo($x)) and foo(42) should be one and the same thing.

[stevengj]

Individual macros can do whatever they want with interpolation syntax; it doesn't have to be limited to expression interpolation.

[ettersi]

Of course you can write a macro such that @extremely_large_value_of_2(2+2) == 5, but that probably wouldn't be very useful unless you are running a T-shirt company.

A concrete example where (I believe) the proposed syntax would be problematic is x = 2.0; n = 2; @btime $(Ref(x))[]^$n. In this case, you might legitimately want n to be "hardwired" into the benchmark expression but not x.

[gdalle]

Would #130 alone solve this? Or can this pop up in more complex functions?

[stevengj]

#130 is just a warning. I'm suggesting that BenchmarkTools should just fix this automatically by automatically wrapping every interpolated expression in a Ref.

[gdalle]

Do you think @ettersi's counterexample is relevant @stevengj?

[stevengj]

A concrete example where (I believe) the proposed syntax would be problematic is x = 2.0; n = 2; @btime $(Ref(x))[]^$n. In this case, you might legitimately want n to be "hardwired" into the benchmark expression but not x.

In cases where you want to literally interpolate into the benchmark expression, you could use @eval @btime $(Ref(x))[]^$n … but the need for this seems pretty rare to me.

[willow-ahrens]

Is this solved via #30 and #265? Currently, here's the same example:

julia> @btime sin(1)
  1.000 ns (0 allocations: 0 bytes)
0.8414709848078965

julia> @btime sin($1)
  5.333 ns (0 allocations: 0 bytes)
0.8414709848078965

julia> @btime sin($(Ref(1))[])
  5.333 ns (0 allocations: 0 bytes)
0.8414709848078965