Witchcraft composition path is removed

.rubocop_todo.yml

@@ -45,7 +45,6 @@ Lint/ConstantDefinitionInBlock:
 Lint/EmptyBlock:
   Exclude:
     - 'spec/chewy/index/import/bulk_request_spec.rb'
-    - 'spec/chewy/index/witchcraft_spec.rb'
     - 'spec/chewy/minitest/helpers_spec.rb'
     - 'spec/chewy/rspec/update_index_spec.rb'
     - 'spec/chewy/search/scrolling_spec.rb'
@@ -136,7 +135,6 @@ Naming/VariableNumber:
 # Offense count: 3
 Style/DocumentDynamicEvalDefinition:
   Exclude:
-    - 'lib/chewy/index/witchcraft.rb'
     - 'lib/chewy/repository.rb'
     - 'lib/chewy/search/pagination/kaminari.rb'
 

CHANGELOG.md

@@ -8,6 +8,8 @@
 
 ### Changes
 
+* **Removal:** `Chewy::Index.witchcraft!` and the `Chewy::Index::Witchcraft` module have been removed. The compiled compose path (default since 8.3.0) delivers equivalent throughput without `method_source` / `parser` / `prism` / `unparser`. Remove any `witchcraft!` call from your indexes — no other changes are needed. ([@AlfonsoUceda][])
+
 ## 8.4.0 (2026-06-17)
 
 ### New Features

docs/indexing.md

@@ -278,55 +278,6 @@ The compiled path also handles hash inputs, join fields, and nested object field
 For a small number of edge cases the compiler falls back to the legacy path automatically: `ignore_blank` fields, `geo_point` fields, indexes with a custom root value proc, and fields whose name is not a valid Ruby identifier.
 The fallback is transparent — the compose result is identical either way.
 
-## Witchcraft technology (deprecated)
-
-> **Deprecated**.
-> `witchcraft!` is retained only for compatibility with older index definitions and will be removed in a future major release.
-> The compiled compose path above is the default for all indexes and delivers equivalent performance without `method_source` / `parser` / `prism` / `unparser` dependencies, with substantially lower boot-time memory (see [#644](https://github.com/toptal/chewy/issues/644)).
-> Remove the `witchcraft!` call from your index — no other changes are needed.
-
-Witchcraft was an experimental performance optimization that used `method_source` to read each field value proc's Ruby source, parse it with `parser` or `prism`, rewrite the AST, and `Unparser`-emit a single fused lambda per index.
-Concretely it compiled definitions like this:
-
-```ruby
-index_scope Product
-witchcraft!
-
-field :title
-field :tags, value: -> { tags.map(&:name) }
-field :categories do
-  field :name, value: -> (product, category) { category.name }
-  field :type, value: -> (product, category, crutch) { crutch.types[category.name] }
-end
-```
-
-into a single lambda roughly equivalent to:
-
-```ruby
--> (object, crutches) do
-  {
-    title: object.title,
-    tags: object.tags.map(&:name),
-    categories: object.categories.map do |object2|
-      {
-        name: object2.name,
-        type: crutches.types[object2.name]
-      }
-    end
-  }
-end
-```
-
-It came with several limitations the compiled path does not have:
-
-1. Required `method_source`-parseable proc source — reflowing a value proc could break the build.
-2. Did not support value procs with splat arguments.
-3. Required dynamically-generated fields to use procs with explicit arguments rather than argumentless `-> { ... }`.
-4. Required `method_source`, `parser` (or `prism`), and `unparser` at boot — together adding ~7 MiB of allocations and ~1 MiB of retained memory to every Rails boot, even for apps that did not call `witchcraft!`.
-
-Calling `witchcraft!` now prints a deprecation warning.
-Removing the call switches the index to the compiled path automatically; no other code changes are needed.
-
 ## Import context
 
 When importing, you often already have data in memory at the call site (precomputed embeddings, batched API responses, aggregations) that crutches would otherwise re-fetch from the database.
@@ -355,7 +306,7 @@ field :embedding, value: ->(object, crutches, context) {
 
 Both are backward-compatible: existing 1-arg crutch blocks and 1-2 arg field procs continue to work unchanged via arity-based dispatch.
 
-Context flows through the default compiled compose path automatically, and is also honored under the legacy `witchcraft!` path and under parallel imports (`parallel: N`).
+Context flows through the default compiled compose path automatically, and is also honored under parallel imports (`parallel: N`).
 With `parallel:`, the same context hash is shared across all workers — precompute once on the main process, reuse in every worker.
 
 ```ruby

docs/troubleshooting.md

@@ -79,7 +79,6 @@ See [configuration.md](configuration.md#import-scope-clean-up-behavior) for deta
 Some Chewy features require additional gems that are not listed as hard dependencies:
 
 - **`parallel`** — required for `chewy:parallel:*` rake tasks. Install it with `gem 'parallel'` in your Gemfile.
-- **`method_source`**, **`parser`** (or **`prism`** on Ruby 3.3+), **`unparser`** — required only for the deprecated `witchcraft!` path. Not needed for the default compiled compose path. Install them only if you still call `witchcraft!` from an index.
 
 If these gems are missing you'll get a `LoadError` when the relevant feature is used.
 

gemfiles/base.gemfile

@@ -1,7 +1,6 @@
 gem 'database_cleaner'
 gem 'elasticsearch-extensions'
 gem 'kaminari-core', require: false
-gem 'method_source'
 gem 'parallel', require: false
 gem 'rake'
 gem 'redis', require: false
@@ -14,4 +13,3 @@ gem 'rubocop-rake'
 gem 'sidekiq', require: false
 gem 'sqlite3'
 gem 'timecop'
-gem 'unparser', '~> 0.6.15'

lib/chewy/index.rb

@@ -10,7 +10,6 @@
 require 'chewy/index/settings'
 require 'chewy/index/specification'
 require 'chewy/index/syncer'
-require 'chewy/index/witchcraft'
 require 'chewy/index/compiled'
 require 'chewy/index/wrapper'
 
@@ -32,7 +31,6 @@ class Index
     include Mapping
     include Observe
     include Crutch
-    include Witchcraft
     include Compiled
     include Wrapper
 

lib/chewy/index/compiled.rb

@@ -7,8 +7,7 @@ class Index
     # per-field arity into the call site and inlines hash construction,
     # removing the per-field iteration + dispatch overhead of the legacy
     # plain compose path. Proc bodies are NOT inlined — procs are called
-    # via `.call`. In exchange there are no parser/unparser/method_source
-    # deps and no Ruby/Prism version coupling.
+    # via `.call`.
     #
     # `fields:` selection is supported: a dedicated method is generated
     # and memoized per unique fields set.
@@ -28,7 +27,6 @@ module ClassMethods
         # Returns true when this index can use the compiled path for the
         # given compose call. Builds the per-fields method if needed.
         def compiled_compose_available?(fields)
-          return false if witchcraft?
           return false unless root&.children&.present?
 
           ensure_compiled_compose_for!(fields)

lib/chewy/index/import.rb

@@ -117,7 +117,7 @@ def bulk(**options)
           payload_errors(error_items)
         end
 
-        # Composes a single document from the passed object. Uses either witchcraft
+        # Composes a single document from the passed object. Uses either the compiled
         # or normal composing under the hood.
         #
         # @param object [Object] a data source object
@@ -140,8 +140,6 @@ def compose(object, crutches = nil, fields: [], context: {})
             else
               compiled_compose(object, crutches, context, fields)
             end
-          elsif witchcraft? && root.children.present?
-            cauldron(fields: fields).brew(object, crutches, context)
           else
             root.compose(object, crutches, fields: fields, context: context)
           end