Handle non-Hash to Hash transformation in before(:key_coercer)

[ojab]

Extract from #353.
Fixes #350

Instead of passing only part of the output to the scoped results we're passing whole output and path as a scope, this was result can replace scoped value without any assumptions on the initial and replacement types.

Performance for the common case (steps execution only on the outer schema) is not affected, running benchmarks/params_valid_vs_invalid.rb (10 times each because of the noise)
Before:

       valid input 8:    32011.2 i/s
       valid input 0:    31832.2 i/s - same-ish: difference falls within error
       valid input 7:    31779.7 i/s - same-ish: difference falls within error
       valid input 1:    31690.8 i/s - same-ish: difference falls within error
       valid input 2:    31623.7 i/s - same-ish: difference falls within error
       valid input 4:    31438.5 i/s - same-ish: difference falls within error
       valid input 6:    31369.0 i/s - same-ish: difference falls within error
       valid input 5:    31180.7 i/s - same-ish: difference falls within error
       valid input 3:    31118.2 i/s - same-ish: difference falls within error
       valid input 9:    30681.5 i/s - same-ish: difference falls within error
     invalid input 9:    20153.8 i/s - 1.59x  (± 0.00) slower
     invalid input 8:    20103.2 i/s - 1.59x  (± 0.00) slower
     invalid input 7:    20067.7 i/s - 1.60x  (± 0.00) slower
     invalid input 1:    19941.1 i/s - 1.61x  (± 0.00) slower
     invalid input 2:    19937.2 i/s - 1.61x  (± 0.00) slower
     invalid input 3:    19563.4 i/s - 1.64x  (± 0.00) slower
     invalid input 0:    19548.0 i/s - 1.64x  (± 0.00) slower
     invalid input 6:    19313.2 i/s - 1.66x  (± 0.00) slower
     invalid input 5:    18576.3 i/s - 1.72x  (± 0.00) slower
     invalid input 4:    17281.0 i/s - 1.85x  (± 0.00) slower

After:

Comparison:
       valid input 0:    32465.3 i/s
       valid input 2:    32141.4 i/s - same-ish: difference falls within error
       valid input 3:    31842.5 i/s - same-ish: difference falls within error
       valid input 5:    31600.6 i/s - same-ish: difference falls within error
       valid input 6:    31525.6 i/s - same-ish: difference falls within error
       valid input 4:    31306.8 i/s - same-ish: difference falls within error
       valid input 7:    31173.7 i/s - same-ish: difference falls within error
       valid input 8:    31123.6 i/s - same-ish: difference falls within error
       valid input 9:    30858.5 i/s - same-ish: difference falls within error
       valid input 1:    30832.5 i/s - same-ish: difference falls within error
     invalid input 5:    20427.4 i/s - 1.59x  (± 0.00) slower
     invalid input 4:    20413.8 i/s - 1.59x  (± 0.00) slower
     invalid input 6:    20385.7 i/s - 1.59x  (± 0.00) slower
     invalid input 7:    20368.8 i/s - 1.59x  (± 0.00) slower
     invalid input 9:    20276.1 i/s - 1.60x  (± 0.00) slower
     invalid input 8:    20184.7 i/s - 1.61x  (± 0.00) slower
     invalid input 1:    20084.0 i/s - 1.62x  (± 0.00) slower
     invalid input 2:    20078.9 i/s - 1.62x  (± 0.00) slower
     invalid input 3:    19898.7 i/s - 1.63x  (± 0.00) slower
     invalid input 0:    19893.5 i/s - 1.63x  (± 0.00) slower

dry-v specs are passing, didn't measured performance of the nested steps.

[ojab]

I'd prefer default: -> { Path::EMPTY }, but we're losing 5% on this:

       valid input 1:    30483.6 i/s
       valid input 6:    30419.9 i/s - same-ish: difference falls within error
       valid input 7:    30366.9 i/s - same-ish: difference falls within error
       valid input 0:    30279.9 i/s - same-ish: difference falls within error
       valid input 5:    30050.6 i/s - same-ish: difference falls within error
       valid input 3:    29944.3 i/s - same-ish: difference falls within error
       valid input 8:    29839.7 i/s - same-ish: difference falls within error
       valid input 9:    29766.8 i/s - same-ish: difference falls within error
       valid input 2:    29754.7 i/s - same-ish: difference falls within error
       valid input 4:    29656.9 i/s - same-ish: difference falls within error
     invalid input 4:    19371.6 i/s - 1.57x  (± 0.00) slower
     invalid input 1:    19308.5 i/s - 1.58x  (± 0.00) slower
     invalid input 0:    19255.6 i/s - 1.58x  (± 0.00) slower
     invalid input 7:    19084.0 i/s - 1.60x  (± 0.00) slower
     invalid input 9:    19066.2 i/s - 1.60x  (± 0.00) slower
     invalid input 5:    19024.8 i/s - 1.60x  (± 0.00) slower
     invalid input 6:    18973.9 i/s - 1.61x  (± 0.00) slower
     invalid input 8:    18956.7 i/s - 1.61x  (± 0.00) slower
     invalid input 3:    18102.1 i/s - 1.68x  (± 0.00) slower
     invalid input 2:    18000.0 i/s - 1.69x  (± 0.00) slower

[ojab]

heh, getting rid of dry-initializer

--- a/lib/dry/schema/result.rb
+++ b/lib/dry/schema/result.rb
@@ -1,6 +1,5 @@
 # frozen_string_literal: true
 
-require "dry/initializer"
 require "dry/core/equalizer"
 
 require "dry/schema/path"
@@ -15,21 +14,14 @@ module Dry
     class Result
       include Dry::Equalizer(:output, :errors)
 
-      extend Dry::Initializer[undefined: false]
+      attr_reader :message_compiler, :result_ast, :path
 
-      # @api private
-      param :output, reader: false
-
-      # A list of failure ASTs produced by rule result objects
-      #
-      # @api private
-      option :result_ast, default: -> { EMPTY_ARRAY.dup }
-
-      # @api private
-      option :message_compiler
-
-      # @api private
-      option :path, optional: true, reader: false
+      def initialize(output, result_ast: nil, message_compiler:, path: Path::EMPTY)
+        @output = output
+        @result_ast = result_ast || EMPTY_ARRAY.dup
+        @message_compiler = message_compiler
+        @path = path
+      end
 
       # @api private
       def self.new(*, **)
@@ -66,11 +58,6 @@ module Dry
         self
       end
 
-      # @api private
-      def path
-        @path || Path::EMPTY
-      end
-
       # Dump result to a hash returning processed and validated data
       #
       # @return [Hash]

leads to

       valid input 9:    49575.0 i/s
       valid input 3:    49255.8 i/s - same-ish: difference falls within error
       valid input 5:    49251.2 i/s - same-ish: difference falls within error
       valid input 6:    48890.9 i/s - same-ish: difference falls within error
       valid input 4:    48641.2 i/s - same-ish: difference falls within error
       valid input 7:    48640.9 i/s - same-ish: difference falls within error
       valid input 0:    48610.4 i/s - same-ish: difference falls within error
       valid input 8:    48258.1 i/s - same-ish: difference falls within error
       valid input 1:    48055.8 i/s - same-ish: difference falls within error
       valid input 2:    47932.6 i/s - same-ish: difference falls within error
     invalid input 8:    27419.9 i/s - 1.81x  (± 0.00) slower
     invalid input 9:    27418.2 i/s - 1.81x  (± 0.00) slower
     invalid input 2:    27125.1 i/s - 1.83x  (± 0.00) slower
     invalid input 1:    27051.1 i/s - 1.83x  (± 0.00) slower
     invalid input 0:    27025.2 i/s - 1.83x  (± 0.00) slower
     invalid input 7:    26545.8 i/s - 1.87x  (± 0.00) slower
     invalid input 3:    25785.0 i/s - 1.92x  (± 0.00) slower
     invalid input 6:    23449.3 i/s - 2.11x  (± 0.00) slower
     invalid input 4:    23342.6 i/s - 2.12x  (± 0.00) slower
     invalid input 5:    22610.3 i/s - 2.19x  (± 0.00) slower

[solnic]

@ojab so, it's faster with dry-initializer, am I reading this right?

[ojab]

nope, 32k/19k with dry-initializer, 49k/27k without.

[solnic]

@ojab oh damn this is a gigantic difference. We could replace it in a separate PR because it'd be a standalone performance improvement. BTW I'd like to re-implement dry-initializer as a simpler API in dry-core eventually. It's a very useful API that we have in a ton of gems, including rom-rb, but it could be simplified internally and most likely it would become faster, too.

[ojab]

Not ideal placement, but dunno where else should I put it.

[solnic]

Super minor but maybe EMPTY = new(EMPTY_ARRAY).freeze?

[ojab]

done

[solnic]

Super minor but maybe EMPTY = new(EMPTY_ARRAY).freeze?

[solnic]

@ojab so, it's faster with dry-initializer, am I reading this right?