Use `invokelatest(timeit_debug_enabled())`

[theogf]

I wanted to build a wrapper to use @timeit_debug quickly like

function with_timer(f::Function, modules::Vector{Module})
  reset_timer!(timer)
  try
    for mod in modules
      enable_debug_timings(mod)
    end
    f()
  finally
    print_timer(timer)
    for mod in modules
      disable_debug_timings(mod)
    end
  end
end

However I realized that doing so creates method version issues (this is printed just before calling f()):

invokelatest(Main.timeit_debug_enabled) = true
Main.timeit_debug_enabled() = false

Would that makes sense to use invokelatest here? Or is there a better solution to my problem?

[fredrikekre]

I don't think the branch would be compiled away then. Does invokelatest(f) work?

[theogf]

After trying multiple times in a row, it does not look deterministic.

I have the following

@testset "TimerOutput related functions" begin
  function timed_function()
    @timeit_debug timer "foo" sleep(0.1)
  end
  # By default nothing is run.
  with_timer(timed_function, Module[])
  @test tottime(timer) == 0
  with_timer(timed_function, [Main])
  @test tottime(timer) > 0
end

And inspecting the output of timeit_debug_enabled() directly it seems to be either false, true or true, true...

[theogf]

So using this version:

function with_timer(f::Function, modules::Vector{Module})
  reset_timer!(timer)
  for mod in modules
    enable_debug_timings(mod)
  end
  try
    invokelatest(f)
  finally
    print_timer(timer)
    for mod in modules
      disable_debug_timings(mod)
    end
  end
end

It's quite weird, The tests pass the first time, fail the second time, pass the third time etc...

[theogf]

The issue seems to be with the conditional flow of disable_debug_timings. Some invokelatest would be required there.
Changing the loop to

    for mod in modules
      Core.eval(mod, :(timeit_debug_enabled() = false))
    end

solves the issue.