Just because I find it fun, this is why I don’t like Elixir’s protocol implementation… ^.^;
I implemented the same trivial protocol implementation with protocol_ex with full optimizations enabled (though not bothered setting priority of callbacks or anything, it’s almost a 1-to-1 copy of the orderable protocol and implementations) and this is the benchmark using the Rating object from the readme of that project (yes there is a test to ensure they generate identical output for identical input) in a Tuple of 1000 Ratings:
╰─➤ mix bench ordered
Compiling 1 file (.ex)
Operating System: Linux"
CPU Information: AMD Phenom(tm) II X6 1090T Processor
Number of Available Cores: 6
Available memory: 15.67 GB
Elixir 1.7.4
Erlang 21.1.1
Benchmark suite executing with the following configuration:
warmup: 5 s
time: 5 s
memory time: 0 μs
parallel: 1
inputs: tuples
Estimated total run time: 20 s
Benchmarking Orderable with input tuples...
Benchmarking OrderableEx with input tuples...
##### With input tuples #####
Name ips average deviation median 99th %
OrderableEx 2.67 K 375.12 μs ±12.98% 361 μs 459 μs
Orderable 1.22 K 821.59 μs ±9.05% 803 μs 993.54 μs
Comparison:
OrderableEx 2.67 K
Orderable 1.22 K - 2.19x slower
Elixir protocols could have such a more efficient implementation… ^.^;
With some slight optimizations of Rating itself (the rating_index part to be specific, using proper constant mappings instead of a map) I got them both up to:
##### With input tuples #####
Name ips average deviation median 99th %
OrderableEx 3.03 K 330.57 μs ±16.23% 313 μs 427 μs
Orderable 1.32 K 757.94 μs ±13.54% 742 μs 946 μs
Comparison:
OrderableEx 3.03 K
Orderable 1.32 K - 2.29x slower
I did multiple runs of each benchmark and always got the same instruction counts and ratio, no real variance.
And if you think that’s fast, you should see my ‘Access’ replacement. ^.^