Optimize show(io::IO, m::Module)

[Sacha0]

This method appeared as a hotspot in code that stringifies many types.

               _
   _       _ _(_)_     |  Documentation: https://docs.julialang.org
  (_)     | (_) (_)    |
   _ _   _| |_  __ _   |  Type "?" for help, "]?" for Pkg help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 1.8.0-DEV.702 (2021-10-10)
 _/ |\__'_|_|_|\__'_|  |  master/82257229c9 (fork: 1 commits, 12 days)
|__/                   |

julia> using BenchmarkTools

julia> module Foo
       module Bar
       module Baz
       module Qux
       end
       end
       end
       end
Main.Foo

julia> @benchmark show(devnull, Foo.Bar.Baz.Qux) # before replacement
BenchmarkTools.Trial: 10000 samples with 26 evaluations.
 Range (min … max):  889.500 ns …  59.991 μs  ┊ GC (min … max): 0.00% … 97.99%
 Time  (median):     978.615 ns               ┊ GC (median):    0.00%
 Time  (mean ± σ):     1.052 μs ± 994.931 ns  ┊ GC (mean ± σ):  1.56% ±  1.70%

   ▇▆█▂
  ▆████▅▅▆▅▇▅▃▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▁▂▂▂▂▂▂▂▂▂▂▂▁▂▂▁▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂ ▃
  890 ns           Histogram: frequency by time         2.41 μs <

 Memory estimate: 640 bytes, allocs estimate: 17.

julia> @eval Base begin
           function show(io::IO, m::Module)
               if Base.is_root_module(m)
                   print(io, nameof(m))
               else
                   print_fullname(io, m)
               end
           end
           function print_fullname(io::IO, m::Module)
               _print_fullname(io, parentmodule(m))
               print(io, nameof(m))
           end
           function _print_fullname(io::IO, m::Module)
               mp = parentmodule(m)
               if m === Main || m === Base || m === Core || mp === m
                   print(io, nameof(m))
                   print(io, '.')
               else
                   _print_fullname(io, mp)
                   print(io, nameof(m))
                   print(io, '.')
               end
           end
       end
_print_fullname (generic function with 1 method)

julia> @benchmark show(devnull, Foo.Bar.Baz.Qux) # after replacement
BenchmarkTools.Trial: 10000 samples with 991 evaluations.
 Range (min … max):  39.268 ns … 227.557 ns  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     44.397 ns               ┊ GC (median):    0.00%
 Time  (mean ± σ):   45.804 ns ±   7.879 ns  ┊ GC (mean ± σ):  0.00% ± 0.00%

  ▂▆▆▁▁▄█▅ ▂▇▄
  █████████████▆▅▆▅▅▅▄▃▃▃▃▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▁▁▂▂▁▂▂▂▂▂▂▂▂▂▂ ▄
  39.3 ns         Histogram: frequency by time         80.1 ns <

 Memory estimate: 0 bytes, allocs estimate: 0.

[NHDaly]

LGTM modulo the suggested fix. (Maybe we need a small unit test for this?)

[Sacha0]

Tagged for backport to 1.6 and 1.7 as a low-risk optimization :).

[NHDaly]

LGTM

[Sacha0]

Thanks for reviewing/merging gentlemen! :)

[gustaphe]

Why not join(io, fullname(module), ".")?

[KristofferC]

Why not join(io, fullname(module), ".")?

The PR here was for performance as shown in the first post. Did you benchmark that version and find that it performs the same as the one proposed here?

[gustaphe]

My bad, fullname of course allocates. I didn't realize this recursive method didn't use the fullname tuple.

(the join(io...) version is inbetween the two methods in the PR for speed)