Time macros: use tryfinally macro approach

[IanButterworth]

@KristofferC's suggestion in #39133 of the @tryfinally macro approach is a neat way to maintain scope while ensuring the finalizer is run

I verified that the finalizer is run in each of these:

julia> @time y = 1
  0.000005 seconds
1

julia> y
1

julia> @time error()
ERROR: 
Stacktrace:
 [1] error()
   @ Base ./error.jl:42
 [2] top-level scope
   @ ~/Documents/GitHub/julia/base/timing.jl:213 [inlined]
 [3] top-level scope
   @ ./REPL[31]:0

julia> x = rand(3,3);

julia> Threads.@threads for i in 1:20 @time x * x; end
  0.565114 seconds (2.36 M allocations: 126.880 MiB, 4.93% gc time, 99.95% compilation time)
  0.566743 seconds (2.36 M allocations: 126.893 MiB, 4.92% gc time, 0.00% compilation time)
  0.000002 seconds (1 allocation: 160 bytes)
  0.565122 seconds (2.36 M allocations: 126.882 MiB, 4.93% gc time, 0.00% compilation time)
  0.568685 seconds (2.36 M allocations: 126.907 MiB, 4.90% gc time, 0.00% compilation time)
  0.566732 seconds (2.36 M allocations: 126.890 MiB, 4.92% gc time, 0.00% compilation time)
  0.000001 seconds (1 allocation: 160 bytes)
  0.000001 seconds (1 allocation: 160 bytes)
  0.000001 seconds (1 allocation: 160 bytes)
  0.000001 seconds (1 allocation: 160 bytes)
  0.000001 seconds (1 allocation: 160 bytes)
  0.000001 seconds (1 allocation: 160 bytes)
  0.000001 seconds (1 allocation: 160 bytes)
  0.000001 seconds (1 allocation: 160 bytes)
  0.000001 seconds (1 allocation: 160 bytes)
  0.565109 seconds (2.36 M allocations: 126.879 MiB, 4.93% gc time, 0.01% compilation time)
  0.000001 seconds (1 allocation: 160 bytes)
  0.000000 seconds (1 allocation: 160 bytes)
  0.000001 seconds (2 allocations: 192 bytes)
  0.000001 seconds (1 allocation: 160 bytes)

This new @tryfinally macro could be pretty useful! Perhaps in a future PR @trycatch could be added and both exported

julia> Base.@tryfinally(xx = 1, println("finalizer"))
finalizer
1

julia> xx
1

julia> Base.@tryfinally(error(), println("finalizer"))
finalizer
ERROR: 
Stacktrace:
 [1] error()
   @ Base ./error.jl:42
 [2] top-level scope
   @ REPL[37]:1

[IanButterworth]

If this is preferable, it should probably go into the next 1.6 beta etc. @JeffBezanson ?

[JeffBezanson]

I don't think it's worth switching to this. tryfinally is an intermediate form that we should not use in source ASTs.

[IanButterworth]

Ok!
Just to tie the two together this intended to use the same mechanism #39217

[NHDaly]

Interesting.

This looks great. Makes sense that we want to wait for #39217.

In the meantime, just to be sure everyone knows, we realized that the current @time macro implementation turns off compilation measurement whenever anyone throws an exception, even one that's handled:
https://github.com/JuliaLang/julia/pull/41733/files#r681256138

julia> @time begin
           # The following exception disables compilation time measurement
           try throw("ha HAH!") catch end
           @eval module M ; f(x) = 2*x ; end
           @eval M.f(2)
       end
  0.006116 seconds (2.17 k allocations: 156.874 KiB)
4

So this fix would definitely be nice. 😊

Does it still make sense to wait for #39217? How's the progress there?