Tell users that function barriers can help with dynamic properties performance

[Tortar]

While this being a general performance tip, I think that emphasizing this in the ReadMe could be beneficial. Consider:

julia> using DynamicStructs

julia> @dynamic struct A end

julia> a = A(x = [1]);

julia> b = A(x = [2]);

julia> f(a, b) = a.x .+ a.y;

julia> using BenchmarkTools

julia> f(a, b) = a.x .+ a.x
f (generic function with 1 method)

julia> @benchmark f($a, $b)
BenchmarkTools.Trial: 10000 samples with 984 evaluations.
 Range (min … max):   56.815 ns … 198.130 μs  ┊ GC (min … max):  0.00% … 99.92%
 Time  (median):      99.324 ns               ┊ GC (median):     0.00%
 Time  (mean ± σ):   111.972 ns ±   1.981 μs  ┊ GC (mean ± σ):  18.85% ±  2.37%

                                        ▄█▂▁                     
  ▂▃██▅▄▃▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▁▂▂▁▁▁▁▁▁▁▁▂▅▇████▅▄▄▃▃▂▂▂▂▂▂▂▂▂▂▂▂▂▂ ▃
  56.8 ns          Histogram: frequency by time          123 ns <

 Memory estimate: 96 bytes, allocs estimate: 2.

julia> f(a, b) = g(a.x, b.x);

julia> g(x1, x2) = x1 .+ x2;

julia> @benchmark f($a, $b)
BenchmarkTools.Trial: 10000 samples with 993 evaluations.
 Range (min … max):  34.486 ns … 431.558 ns  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     63.524 ns               ┊ GC (median):    0.00%
 Time  (mean ± σ):   63.087 ns ±   8.535 ns  ┊ GC (mean ± σ):  0.00% ± 0.00%

                                           ▇ █▂                 
  ▂▂▂▃▂▂▂▂▂▂▂▂▂▂▂▁▂▂▁▁▁▁▁▁▁▂▁▁▁▁▁▁▁▁▁▁▁▁▁▂▅█▄██▃▄▃▃▃▃▂▂▂▂▂▂▂▂▂ ▃
  34.5 ns         Histogram: frequency by time         74.4 ns <

 Memory estimate: 64 bytes, allocs estimate: 1.

[AntonOresten]

Ah, smart! I had a hunch something like this could help.

However, is it possible that in this case the first one is slower because Julia calls Broadcast.broadcastable on each argument, whereas the second one can compile more effectively because it dispatches on known types, and knows that two Vector{Int}s get broadcasted in a certain way? I could not reproduce this with Int types instead of Vector{Int}, which makes sense since + is just another function, which serves as a barrier, I guess?

Regardless, it is worth noting this type of optimization.

EDIT: Actually nevermind! this was the point you were proving!🤣

[Tortar]

EDIT: Actually nevermind! this was the point you were proving!🤣

yes, I think so :-)

Could I ask you if you can give me some privileges on the repo, because I think that I could probably maintain it for a fair amount of time? I won't do any changes without first discussing them with you anyway, it just would be cool :D

[AntonOresten]

Sure thing! You've had a lot of good ideas, and I'm happy to cut out the middleman.😋I've added you as a collaborator. You might need to check your email.

[Tortar]

Thanks a lot!