Issue #20608 has been updated by mame (Yusuke Endoh). Status changed from Assigned to Closed I merged https://github.com/ruby/ruby/pull/11110. ---------------------------------------- Bug #20608: Hash#find always allocates each iterated pair https://bugs.ruby-lang.org/issues/20608#change-109047 * Author: ivoanjo (Ivo Anjo) * Status: Closed * Assignee: ko1 (Koichi Sasada) * ruby -v: ruby 3.4.0preview1 (2024-05-16 master 9d69619623) [x86_64-linux] * Backport: 3.1: UNKNOWN, 3.2: UNKNOWN, 3.3: UNKNOWN ---------------------------------------- Hey there! Recently I ran into this sharp edge in `Hash#find`: ```ruby puts RUBY_DESCRIPTION def allocated_now = GC.stat(:total_allocated_objects) dummy_hash = 10_000.times.each_with_index.to_h before_allocs = allocated_now dummy_hash.any? { |k, v| } puts "Allocated #{allocated_now - before_allocs} for #{File.read(__FILE__).lines[__LINE__-2]}" before_allocs = allocated_now dummy_hash.each { |k, v| } puts "Allocated #{allocated_now - before_allocs} for #{File.read(__FILE__).lines[__LINE__-2]}" before_allocs = allocated_now dummy_hash.find { |k, v| } puts "Allocated #{allocated_now - before_allocs} for #{File.read(__FILE__).lines[__LINE__-2]}" before_allocs = allocated_now dummy_hash.any? { |k| } puts "Allocated #{allocated_now - before_allocs} for #{File.read(__FILE__).lines[__LINE__-2]}" ``` Result: ``` ruby 3.4.0preview1 (2024-05-16 master 9d69619623) [x86_64-linux] Allocated 0 for dummy_hash.any? { |k, v| } Allocated 0 for dummy_hash.each { |k, v| } Allocated 10002 for dummy_hash.find { |k, v| } Allocated 10000 for dummy_hash.any? { |k| } ``` That is, while `Hash#any?`, `Hash#each`, etc avoid doing any allocations during iteration, `Hash#find` does not hit the `rb_block_pair_yield_optimizable` => `each_pair_i_fast` fast path, and so is massively costly compared to the others. This is very surprising, as I'd expect a `find` to have a comparable cost to `any?` (and I ended up [redoing some code to avoid find](https://github.com/DataDog/dd-trace-rb/pull/3757)). Also while experimenting a bit, it was surprising to me that the allocation optimization only kicks in when `|k, v|` are declared, and thus `.any? { |k| }` is also more expensive. -- https://bugs.ruby-lang.org/