Module: Rigor::Inference::ScopeIndexer

Defined in:
lib/rigor/inference/scope_indexer.rb

Overview

Builds a per-node scope index for a Prism program by running ‘Rigor::Inference::StatementEvaluator` over the root and recording the entry scope visible at every node. Expression-interior nodes the evaluator does not specialise (call receivers, arguments, array/hash elements, …) inherit their nearest statement-y ancestor’s recorded scope, so a downstream caller that looks up the scope for any Prism node in the tree always gets the scope that was effectively visible at that point.

The CLI commands ‘rigor type-of` and `rigor type-scan` consume the index so that local-variable bindings established earlier in the program are visible to the typer when probing later nodes. Without the index, both commands would type every node under an empty scope and miss the constant-folding / dispatch precision that Slice 3 phase 2’s StatementEvaluator unlocks.

The returned object is an identity-comparing Hash:

“‘ruby index = Rigor::Inference::ScopeIndexer.index(program, default_scope: Scope.empty) index #=> the Rigor::Scope visible at that node “`

Nodes that are not part of the program subtree (e.g. synthesised virtual nodes that the caller looks up after the fact) yield the ‘default_scope`. The returned Hash is mutable in principle but callers MUST treat it as read-only; the indexer itself never exposes a way to update it past construction. rubocop:disable Metrics/ModuleLength

Defined Under Namespace

Classes: MethodEffectResolver

Constant Summary collapse

TOP_LEVEL_DEF_KEY =

v0.0.3 A — sentinel key under which ‘record_def_node` files DefNodes that live outside any class / module body (top-level helpers, `def`s nested inside DSL blocks like `RSpec.describe … do; def helper; end`). Looked up by `Scope#top_level_def_for` to give implicit-self calls priority over RBS dispatch when the file defines a same-named local method.

"<toplevel>"
MIXIN_CALL_NAMES = 
%i[include prepend].freeze
VISIBILITY_MODIFIERS = 
%i[public private protected].freeze
ATTR_MACROS =

The ‘attr_*` accessor macros that introduce methods Rigor must treat as source-declared. Without this, a class that defines an accessor with `attr_reader :x` AND carries RBS that omits `x` (a common gap — the project ships an incomplete `sig/`) fires a false `call.undefined-method` on `obj.x`, because the undefined-method rule only suppressed `def` / `define_method` / `alias_method`-discovered methods. `attr_reader` defines readers, `attr_writer` writers (`x=`), `attr_accessor` both.

%i[attr_reader attr_writer attr_accessor].freeze

Class Method Summary collapse

Class Method Details

.accumulate_ivar_type(accumulator, class_name, ivar_name, type) ⇒ Object

Unions ‘type` into the class-ivar accumulator for `(class_name, ivar_name)`. Shared by the single-write and multi-write (parallel-assignment) collectors.



1138
1139
1140
1141
1142
1143
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1138

def accumulate_ivar_type(accumulator, class_name, ivar_name, type)
  accumulator[class_name] ||= {}
  existing = accumulator[class_name][ivar_name]
  accumulator[class_name][ivar_name] =
    existing ? Type::Combinator.union(existing, type) : type
end

.accumulate_project_index(acc, path, root) ⇒ Object

Folds one file’s class-keyed indexes into the cross-file accumulator. ‘method_visibilities` (ADR-35) is collected here so the override-visibility-reduced rule can read an ancestor’s visibility declared in a sibling file.



2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2295

def accumulate_project_index(acc, path, root)
  merge_discovered_defs(acc[:def_nodes], acc[:def_sources], path, root)
  build_discovered_singleton_def_nodes(root).each do |class_name, methods|
    (acc[:singleton_def_nodes][class_name] ||= {}).merge!(methods)
  end
  superclasses = build_discovered_superclasses(root)
  includes = build_discovered_includes(root)
  acc[:superclasses].merge!(superclasses)
  includes.each do |class_name, mods|
    acc[:includes][class_name] = ((acc[:includes][class_name] || []) + mods).uniq
  end
  record_class_sources(acc[:class_sources], path, root, superclasses, includes)
  merge_class_keyed_index_tables(acc, root)
  acc[:data_member_layouts].merge!(build_data_member_layouts(root))
end

.additional_initializer?(class_name, method_name, default_scope) ⇒ Boolean

ADR-38 — true when a loaded plugin declares ‘method_name` an additional initializer for `class_name` (or an ancestor). Reads the plugin registry off the pre-pass scope’s environment; the receiver-constraint match reuses ‘Environment#class_ordering` (the same mechanism ADR-16 Tier A’s ‘MacroBlockSelfType` uses). The whole lookup is wrapped so any resolution failure degrades to “no match” —since the gate only ever SUPPRESSES a nil contribution, a missed match is false-positive-safe (it merely leaves the existing nil widening in place).

Returns:

  • (Boolean) 


539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
 
# File 'lib/rigor/inference/scope_indexer.rb', line 539

def additional_initializer?(class_name, method_name, default_scope)
  return false if class_name.nil? || default_scope.nil?

  environment = default_scope.environment
  registry = environment&.plugin_registry
  return false if registry.nil?
  return false if registry.respond_to?(:empty?) && registry.empty?
  return false unless registry.respond_to?(:additional_initializers)

  registry.additional_initializers.any? do |entry|
    entry.covers_method?(method_name) &&
      class_matches_constraint?(class_name, entry.receiver_constraint, environment)
  end
rescue StandardError
  false
end

.always_raises?(node) ⇒ Boolean

True when ‘node` (a single statement or its last statement) is an unconditional `raise`/`fail` call that always terminates the path — used to treat raise-terminated branches as non-completing (they never observe the seed nil).

Returns:

  • (Boolean) 


1081
1082
1083
1084
1085
1086
1087
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1081

def always_raises?(node)
  node = top_level_statements(node).last if node.is_a?(Prism::StatementsNode)
  return false unless node.is_a?(Prism::CallNode)
  return false unless node.receiver.nil?

  %i[raise fail].include?(node.name)
end

.apply_alias_def_nodes(root, accumulator) ⇒ Object

Post-pass over the ‘def_nodes` accumulator: for every `alias` declaration inside a class body, if the original method name maps to a `Prism::DefNode`, register the new name pointing to the same node so inter-procedural return-type inference works for the aliased name.



2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2085

def apply_alias_def_nodes(root, accumulator)
  alias_map = collect_class_alias_map(root, [], {})
  alias_map.each do |class_name, aliases|
    class_defs = accumulator[class_name]
    next unless class_defs

    aliases.each do |new_name, old_name|
      def_node = class_defs[old_name]
      next unless def_node.is_a?(Prism::DefNode)

      (accumulator[class_name] ||= {})[new_name] = def_node
    end
  end
end

.apply_named_visibility(args, qualified_prefix, visibility, accumulator) ⇒ Object 



2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2048

def apply_named_visibility(args, qualified_prefix, visibility, accumulator)
  class_name = qualified_prefix.join("::")
  args.each do |arg|
    name = visibility_target_name(arg)
    next if name.nil?

    accumulator[class_name] ||= {}
    accumulator[class_name][name] = visibility
  end
end

.apply_visibility_call(call_node, qualified_prefix, current_visibility, accumulator) ⇒ Object

Recognises modifier calls on the implicit-self receiver inside a class body. Returns the (possibly updated) current visibility:

  • ‘private` / `public` / `protected` (no args) —switch the running default for subsequent defs.

  • ‘private :foo, :bar` — back-patch the named methods in the accumulator. Returns `current_visibility` unchanged because the running default does NOT change for this form.



2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2034

def apply_visibility_call(call_node, qualified_prefix, current_visibility, accumulator)
  return current_visibility unless call_node.receiver.nil?
  return current_visibility unless VISIBILITY_MODIFIERS.include?(call_node.name)
  return current_visibility if qualified_prefix.empty?

  args = call_node.arguments&.arguments || []
  if args.empty?
    call_node.name
  else
    apply_named_visibility(args, qualified_prefix, call_node.name, accumulator)
    current_visibility
  end
end

.bare_module_function?(node) ⇒ Boolean

Returns:

  • (Boolean) 


1765
1766
1767
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1765

def bare_module_function?(node)
  node.arguments.nil? || node.arguments.arguments.empty?
end

.block_initializer?(class_name, method_name, default_scope) ⇒ Boolean

ADR-38 block-form gate: true when a loaded plugin declares ‘method_name` a block-form initializer for `class_name` (or an ancestor). Mirrors `additional_initializer?` but queries `covers_block_method?` instead of `covers_method?`.

Returns:

  • (Boolean) 


478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
 
# File 'lib/rigor/inference/scope_indexer.rb', line 478

def block_initializer?(class_name, method_name, default_scope)
  return false if class_name.nil? || default_scope.nil?

  environment = default_scope.environment
  registry = environment&.plugin_registry
  return false if registry.nil?
  return false if registry.respond_to?(:empty?) && registry.empty?
  return false unless registry.respond_to?(:additional_initializers)

  registry.additional_initializers.any? do |entry|
    entry.covers_block_method?(method_name) &&
      class_matches_constraint?(class_name, entry.receiver_constraint, environment)
  end
rescue StandardError
  false
end

.branch_definitely_assigns?(branch, target, class_name, effects, depth, visiting) ⇒ Boolean

True when a branch body (a StatementsNode / single node) definitely assigns ‘target` non-nil on every path that completes the method through it, OR terminates every path by raise (vacuously safe — no completing path observes the seed nil). Returns false if any path can complete/return without the assignment.

Returns:

  • (Boolean) 


1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1031

def branch_definitely_assigns?(branch, target, class_name, effects, depth, visiting)
  stmts = top_level_statements(branch)
  return false if stmts.empty?

  stmts.each do |stmt|
    outcome = statement_assignment_outcome(stmt, target, class_name, effects, depth, visiting)
    return true if outcome == :assigned
    return false if outcome == :terminates_unassigned
  end
  # Reached the end of the branch without a definite assignment;
  # safe only if the branch's last statement always raises (no
  # completing path falls out of it).
  always_raises?(stmts.last)
end

.build_class_cvar_index(root, default_scope) ⇒ Object

Slice 7 phase 6 — class-cvar pre-pass. Same shape as the ivar pre-pass but collects ‘Prism::ClassVariableWriteNode` writes inside ANY def body (instance or singleton) of the enclosing class, because Ruby cvars are shared across both facets. The resulting table is seeded into both instance and singleton method bodies through `Scope#class_cvars_for`.



1243
1244
1245
1246
1247
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1243

def build_class_cvar_index(root, default_scope)
  accumulator = {}
  walk_class_cvars(root, [], default_scope, accumulator)
  accumulator.transform_values(&:freeze).freeze
end

.build_class_ivar_index(root, default_scope) ⇒ Object

Slice 7 phase 2. Builds the class-level ivar accumulator by walking every ‘Prism::ClassNode` / `Prism::ModuleNode` body, descending into each nested `Prism::DefNode`, and typing every `Prism::InstanceVariableWriteNode` rvalue under a scope that carries the appropriate `self_type` for that def (singleton vs instance). The rvalue is typed with NO local bindings — the pre-pass lacks statement-level threading — so `@x = 1` records `Constant` but `@x = some_local + 1` records `Dynamic` (since `some_local` is unbound at pre-pass time). Multiple writes to the same ivar union via `Type::Combinator.union`.



218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
 
# File 'lib/rigor/inference/scope_indexer.rb', line 218

def build_class_ivar_index(root, default_scope)
  accumulator = {}
  mutated_ivars = {}
  read_before_write = {}
  init_writes = {}
  # WD3 — per-class summary of `{class_name => {method_name =>
  # Set<ivar names definitely assigned non-nil on every
  # completing path>}}`, consulted by `dead_transient_nil_writes`
  # so a ctor that reassigns `@x` indirectly through an
  # unconditional same-class method call (`mask!`) credits the
  # overwrite. Built once per program here, memoised by class.
  method_assign_effects = build_method_assign_effects(root)
  walk_class_ivars(root, [], default_scope, accumulator, mutated_ivars,
                   read_before_write, init_writes, method_assign_effects)
  widen_mutated_ivar_entries!(accumulator, mutated_ivars)
  contribute_read_before_write_nil!(accumulator, read_before_write, init_writes)
  accumulator.transform_values(&:freeze).freeze
end

.build_data_member_layouts(root) ⇒ Object

ADR-48 — per qualified class name -> ordered ‘Data.define` member-name list, for both the named-subclass form (`class Point < Data.define(:x, :y)`) and the constant-assigned form (`Point = Data.define(:x, :y)`). Only `Data.define` is recorded: `Struct.new` instances are mutable, so member-value folding would be unsound (the Struct follow-up is deferred — see ADR-48 § “Struct follow-up”). Consumed by MethodDispatcher::DataFolding via Scope#data_member_layout.



1843
1844
1845
1846
1847
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1843

def build_data_member_layouts(root)
  accumulator = {}
  walk_data_member_layouts(root, [], accumulator)
  accumulator.freeze
end

.build_declaration_artifacts(root) ⇒ Object

Walks the program once for ‘Prism::ModuleNode` and `Prism::ClassNode`, recording the `Singleton` type for the outermost `constant_path` node of each declaration. Inner segments of a `class Foo::Bar::Baz` path remain real references (resolved through the ordinary lexical walk), so we annotate ONLY the topmost path node. Nested declarations contribute their fully qualified path: `class A::B; class C; …` produces `A::B` for the outer and `A::B::C` for the inner.



2442
2443
2444
2445
2446
2447
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2442

def build_declaration_artifacts(root)
  identity_table = {}.compare_by_identity
  discovered = {}
  record_declarations(root, [], identity_table, discovered)
  [identity_table.freeze, discovered.freeze]
end

.build_discovered_def_nodes(root) ⇒ Object

v0.0.2 #5 — instance-side def-node recording. Walks class bodies the same way as ‘build_discovered_methods` but records the actual `Prism::DefNode` for each instance method so `ExpressionTyper` can re-type the body at the call site for inter-procedural return inference. Singleton methods and `define_method` calls are intentionally skipped: the inference path needs a statically introspectable body, and singleton dispatch has its own complications (Class / Module ancestry) the first-iteration rule does not yet model.



1596
1597
1598
1599
1600
1601
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1596

def build_discovered_def_nodes(root)
  accumulator = {}
  walk_def_nodes(root, [], false, accumulator)
  apply_alias_def_nodes(root, accumulator)
  accumulator.transform_values(&:freeze).freeze
end

.build_discovered_includes(root) ⇒ Object

ADR-24 slice 2 — per-class/module table mapping a fully qualified user class or module to the list of module names it ‘include`s / `prepend`s, AS WRITTEN at the mixin call (`include Foo` / `include Foo::Bar`). Only constant arguments are recorded; dynamic mixins (`include some_method`) produce no entry. `prepend` is bucketed with `include` — both contribute instance methods to the ancestor chain. `extend` is NOT tracked (it adds singleton methods; ADR-24 slice 2 resolves the instance-side chain).



1898
1899
1900
1901
1902
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1898

def build_discovered_includes(root)
  accumulator = {}
  walk_class_includes(root, [], nil, accumulator)
  accumulator.transform_values { |mods| mods.uniq.freeze }.freeze
end

.build_discovered_method_visibilities(root) ⇒ Object

v0.1.2 — per-class method-visibility table for the ‘def.method-visibility-mismatch` CheckRule.

Tracks two visibility-changing forms:

  • **Modifier blocks**: a bare ‘private` / `protected` / `public` call inside a class body switches the “current default” visibility for every subsequent `def` until another modifier flips it again.

  • **Named-argument form**: ‘private :foo, :bar` (or the same with `protected` / `public`) marks specific names already-recorded under the class. Symbol-only args are recognised; `private def foo; end` (the wrap-around form) is not yet — it would need tracking the def-call’s return-value visibility, which is a separate slice.

Top-level (no surrounding class) defs do not contribute — Ruby’s top-level visibility nuances (private at top-level marks the method on ‘Object`) are out of scope for v0.1.2.



1957
1958
1959
1960
1961
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1957

def build_discovered_method_visibilities(root)
  accumulator = {}
  walk_method_visibilities(root, [], false, :public, accumulator)
  accumulator.transform_values(&:freeze).freeze
end

.build_discovered_methods(root) ⇒ Object

Slice 7 phase 12 — in-source method discovery pre-pass. Walks every class/module body and records the methods introduced via ‘Prism::DefNode` (instance + singleton) and via recognised `define_method(:name) { … }` calls. The returned table maps qualified class name to a `Hash[Symbol, :instance | :singleton]`.



1392
1393
1394
1395
1396
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1392

def build_discovered_methods(root)
  accumulator = {}
  walk_methods(root, [], false, accumulator)
  accumulator.transform_values(&:freeze).freeze
end

.build_discovered_singleton_def_nodes(root) ⇒ Object

Module-singleton call resolution (ADR-57 follow-up) — the SINGLETON-side mirror of ‘build_discovered_def_nodes`. Records the `Prism::DefNode` for every singleton-side method (`def self.x`, `def Foo.x`, a `class << self` body, and a `module_function` method) keyed by qualified class/module name → method → node, so `ExpressionTyper` can re-type the body when a `Singleton` receiver dispatches `Foo.x`. The instance-side table is kept singleton-free on purpose (its ancestor walk binds `self` as `Nominal`), so the two never overlap except for `module_function` defs, which are genuinely callable on both sides and so appear in both tables. Top-level singleton defs (`def self.x` outside any class — `self` is `main`) are not recorded; they have no constant receiver to dispatch through.



1667
1668
1669
1670
1671
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1667

def build_discovered_singleton_def_nodes(root)
  accumulator = {}
  walk_singleton_def_nodes(root, [], false, accumulator)
  accumulator.transform_values(&:freeze).freeze
end

.build_discovered_superclasses(root) ⇒ Object

ADR-24 slice 2 — per-class table mapping a fully qualified user class to its superclass name AS WRITTEN at the ‘class Foo < Bar` declaration. Only constant superclasses are recorded (`class Foo < Struct.new(…)` and other non-constant superclasses produce no entry). The as-written name is resolved to a qualified class at the call site against the subclass’s lexical nesting —see ‘ExpressionTyper#resolve_ancestor_class_name`.



1802
1803
1804
1805
1806
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1802

def build_discovered_superclasses(root)
  accumulator = {}
  walk_class_superclasses(root, [], accumulator)
  accumulator.freeze
end

.build_in_source_constants(root, default_scope) ⇒ Object

Slice 7 phase 9 — in-source constant value pre-pass. Walks the entire program (top-level AND inside class / module / def bodies) for ‘Prism::ConstantWriteNode` and `Prism::ConstantPathWriteNode`, types each rvalue, and accumulates by qualified name. Constants defined inside a class body are qualified with the surrounding class path; constants written via a path (`Foo::BAR = …`) use the rendered path as-is.



1329
1330
1331
1332
1333
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1329

def build_in_source_constants(root, default_scope)
  accumulator = {}
  walk_constant_writes(root, [], default_scope, accumulator)
  accumulator.freeze
end

.build_method_assign_effects(root) ⇒ Object

WD3 — builds the per-class definite-assignment summary ‘=> {method_name => Set<ivar names assigned non-nil on every completing path>}`. Used so a ctor’s ‘dead_transient_nil_writes` can credit an indirect overwrite through an unconditionally-called same-class method (ipaddr’s ‘initialize` reassigns `@mask_addr` via `mask!`).

Each method’s set is computed by the same suffix definite-assignment analysis used for the ctor seed, run from the method body’s first statement for every ivar the method writes anywhere. Same-class calls inside a method are credited transitively (depth-capped, cycle-guarded) so the resulting FLAT table is correct at depth 0 for the ctor lookup.



822
823
824
825
826
827
828
829
830
831
832
833
 
# File 'lib/rigor/inference/scope_indexer.rb', line 822

def build_method_assign_effects(root)
  defs = collect_class_method_defs(root)
  effects = {}
  memo = {}.compare_by_identity
  defs.each do |class_name, methods|
    methods.each do |method_name, def_node|
      assigns = method_definite_assigns(class_name, method_name, def_node, defs, effects, memo, 0)
      (effects[class_name] ||= {})[method_name] = assigns unless assigns.empty?
    end
  end
  effects.freeze
end

.build_program_global_index(root, default_scope) ⇒ Object

Slice 7 phase 6 — program-global pre-pass. Globals are process-wide so the accumulator is a flat ‘Hash[Symbol, Type::t]` populated from every `Prism::GlobalVariableWriteNode` in the program (top-level AND inside method bodies). The same accumulator is seeded into every method body and the top-level scope.



1301
1302
1303
1304
1305
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1301

def build_program_global_index(root, default_scope)
  accumulator = {}
  gather_global_writes(root, default_scope, accumulator)
  accumulator.freeze
end

.case_assignment_outcome(node, target, class_name, effects, depth, visiting) ⇒ Object

‘case` is a definite assignment only when there is a real `else` clause AND every `when`/`in` body plus the else body definitely assigns (or raises-out). A missing else lets an unmatched subject fall through unassigned.



1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1065

def case_assignment_outcome(node, target, class_name, effects, depth, visiting)
  else_clause = node.else_clause
  return :falls_through_unassigned unless else_clause.is_a?(Prism::ElseNode)

  branches = node.conditions.map { |c| c.respond_to?(:statements) ? c.statements : nil }
  branches << else_clause.statements
  all_ok = branches.all? do |b|
    branch_definitely_assigns?(b, target, class_name, effects, depth, visiting)
  end
  all_ok ? :assigned : :falls_through_unassigned
end

.class_matches_constraint?(class_name, constraint, environment) ⇒ Boolean

Returns:

  • (Boolean) 


556
557
558
559
560
561
562
563
564
 
# File 'lib/rigor/inference/scope_indexer.rb', line 556

def class_matches_constraint?(class_name, constraint, environment)
  return true if class_name == constraint
  return false if environment.nil?

  ordering = environment.class_ordering(class_name, constraint)
  %i[equal subclass].include?(ordering)
rescue StandardError
  false
end

.class_new_call?(node) ⇒ Boolean

Recognises ‘Class.new`, `Class.new(super_class)`, and the block form `Class.new { … }`. Like `module_new_call?`, the block body is walked as the anonymous class’s body. The optional ‘super_class` positional is accepted but does NOT route through `ancestor` discovery in this slice — the synthesised class still answers method lookups via its own body’s defs, mirroring how ‘Struct.new` / `Data.define` are handled.

Returns:

  • (Boolean) 


2553
2554
2555
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2553

def class_new_call?(node)
  meta_call_with_name?(node, :Class, :new)
end

.class_new_superclass_name(call_node, qualified_prefix, accumulator) ⇒ Object

Lexically-qualified name of a ‘Class.new(Super)` superclass argument, or nil when there is no positional superclass (a bare `Class.new` / `Module.new`). When the unqualified super name is a class already discovered under an enclosing-prefix segment, the qualified form is returned (so `Class.new(Error)` inside `module M` resolves to `M::Error`); otherwise the literal name is returned (covering a core / RBS-known superclass spelled bare).



2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2416

def class_new_superclass_name(call_node, qualified_prefix, accumulator)
  arg = call_node.arguments&.arguments&.first
  return nil if arg.nil?

  raw = qualified_name_for(arg)
  return nil if raw.nil?

  prefix = qualified_prefix.dup
  until prefix.empty?
    candidate = (prefix + [raw]).join("::")
    return candidate if accumulator.key?(candidate)

    prefix.pop
  end
  raw
end

.collect_block_ivar_writes(block_node, qualified_prefix, default_scope, accumulator, mutated_ivars, init_writes) ⇒ Object

ADR-38 block-form: collects ivar writes from a CallNode’s block body (e.g. RSpec ‘before { @x = … }` / `let(:x) { … }`) and folds them into `init_writes`, suppressing the read-before-write nil contribution the same way a def-form initializer does. The block body is always treated as an initializer (the caller has already verified the method name is declared as a block_method initializer), so there is no read-before-write evidence collection step here.



456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
 
# File 'lib/rigor/inference/scope_indexer.rb', line 456

def collect_block_ivar_writes(block_node, qualified_prefix, default_scope, accumulator,
                              mutated_ivars, init_writes)
  return if block_node.body.nil? || qualified_prefix.empty?

  class_name = qualified_prefix.join("::")
  self_type = Type::Combinator.nominal_of(class_name)
  body_scope = default_scope.with_self_type(self_type)

  gather_ivar_writes(block_node.body, body_scope, class_name, accumulator,
                     EMPTY_GUARDED_IVARS, mutated_ivars)

  seen_writes = Set.new
  read_first = Set.new
  detect_read_before_write(block_node.body, seen_writes, read_first)
  init_set = (init_writes[class_name] ||= Set.new)
  seen_writes.each { |name| init_set << name }
end

.collect_class_alias_map(node, qualified_prefix, accumulator) ⇒ Object

Builds a map ‘=> {new_name_sym => old_name_sym}` by walking the tree for `AliasMethodNode` nodes inside class bodies.



2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2102

def collect_class_alias_map(node, qualified_prefix, accumulator)
  return accumulator unless node.is_a?(Prism::Node)

  case node
  when Prism::ClassNode, Prism::ModuleNode
    name = qualified_name_for(node.constant_path)
    if name
      collect_class_alias_map(node.body, qualified_prefix + [name], accumulator) if node.body
      return accumulator
    end
  when Prism::SingletonClassNode
    return accumulator
  when Prism::AliasMethodNode
    record_alias_map_entry(node, qualified_prefix, accumulator)
    return accumulator
  end

  node.compact_child_nodes.each { |child| collect_class_alias_map(child, qualified_prefix, accumulator) }
  accumulator
end

.collect_class_body_ivar_writes(node, class_name, init_writes) ⇒ Object

Walks class-body level statements (i.e. NOT inside any nested DefNode / ClassNode / ModuleNode) and records every ‘@x = …` write target as a class-body init. Consumed by `contribute_read_before_write_nil!` to exempt ivars the author already knows might be nil (the `@x = nil` at class-body level is the canonical nullability acknowledgement; the instance @x is technically a separate store, but the pragmatic intent is unambiguous).



583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
 
# File 'lib/rigor/inference/scope_indexer.rb', line 583

def collect_class_body_ivar_writes(node, class_name, init_writes)
  return unless node.is_a?(Prism::Node)
  return if IVAR_BARRIER_NODES.any? { |klass| node.is_a?(klass) }

  if node.is_a?(Prism::InstanceVariableWriteNode) ||
     node.is_a?(Prism::InstanceVariableOrWriteNode) ||
     node.is_a?(Prism::InstanceVariableAndWriteNode) ||
     node.is_a?(Prism::InstanceVariableOperatorWriteNode)
    (init_writes[class_name] ||= Set.new) << node.name
  end

  node.compact_child_nodes.each do |child|
    collect_class_body_ivar_writes(child, class_name, init_writes)
  end
end

.collect_class_decls(node, qualified_prefix, accumulator) ⇒ Object

Class-only variant of ‘record_declarations` — descends into nested module bodies (so `module Foo; class Bar` registers `Foo::Bar`) but never registers the module itself in `accumulator`.



2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2362

def collect_class_decls(node, qualified_prefix, accumulator)
  return unless node.is_a?(Prism::Node)

  case node
  when Prism::ClassNode
    name = qualified_name_for(node.constant_path)
    if name
      full = (qualified_prefix + [name]).join("::")
      accumulator[full] = Type::Combinator.singleton_of(full)
      return collect_class_decls(node.body, qualified_prefix + [name], accumulator) if node.body
    end
  when Prism::ModuleNode
    name = qualified_name_for(node.constant_path)
    return collect_class_decls(node.body, qualified_prefix + [name], accumulator) if name && node.body
  when Prism::ConstantWriteNode
    record_class_new_constant_decl(node, qualified_prefix, accumulator)
  end

  node.compact_child_nodes.each { |child| collect_class_decls(child, qualified_prefix, accumulator) }
end

.collect_class_method_defs(root, prefix = [], acc = {}) ⇒ Object

Collects ‘=> {method_name => DefNode}` for every instance-method def in the program. Singleton defs (`def self.x`) are excluded — the ctor-call crediting only follows instance-method calls on `self`. Last def wins on redefinition.



839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
 
# File 'lib/rigor/inference/scope_indexer.rb', line 839

def collect_class_method_defs(root, prefix = [], acc = {})
  return acc unless root.is_a?(Prism::Node)

  case root
  when Prism::ClassNode, Prism::ModuleNode
    name = qualified_name_for(root.constant_path)
    if name && root.body
      child = prefix + [name]
      collect_class_method_defs(root.body, child, acc)
    end
    return acc
  when Prism::DefNode
    (acc[prefix.join("::")] ||= {})[root.name] = root unless prefix.empty? || root.receiver
    return acc
  end

  root.compact_child_nodes.each { |c| collect_class_method_defs(c, prefix, acc) }
  acc
end

.collect_def_cvar_writes(def_node, qualified_prefix, default_scope, accumulator) ⇒ Object 



1270
1271
1272
1273
1274
1275
1276
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1270

def collect_def_cvar_writes(def_node, qualified_prefix, default_scope, accumulator)
  return if def_node.body.nil? || qualified_prefix.empty?

  class_name = qualified_prefix.join("::")
  body_scope = default_scope.with_self_type(Type::Combinator.nominal_of(class_name))
  gather_cvar_writes(def_node.body, body_scope, class_name, accumulator)
end

.collect_def_ivar_writes(def_node, qualified_prefix, default_scope, accumulator, mutated_ivars, read_before_write = nil, init_writes = nil, method_assign_effects = nil) ⇒ Object

rubocop:disable Metrics/ParameterLists



404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
 
# File 'lib/rigor/inference/scope_indexer.rb', line 404

def collect_def_ivar_writes(def_node, qualified_prefix, default_scope, accumulator, mutated_ivars, # rubocop:disable Metrics/ParameterLists
                            read_before_write = nil, init_writes = nil, method_assign_effects = nil)
  return if def_node.body.nil? || qualified_prefix.empty?

  class_name = qualified_prefix.join("::")
  singleton = def_node.receiver.is_a?(Prism::SelfNode) ||
              def_receiver_targets_lexical_self?(def_node.receiver, qualified_prefix)
  self_type =
    if singleton
      Type::Combinator.singleton_of(class_name)
    else
      Type::Combinator.nominal_of(class_name)
    end
  body_scope = default_scope.with_self_type(self_type)

  # C2 — transient `@x = nil` dead-write elimination. When a
  # method body opens with an unconditional `@x = nil`
  # (defensive init) and then *definitely* reassigns `@x` to a
  # non-nil value on every completing path (a later
  # unconditional statement-level write, OR an `if/else` whose
  # both branches write `@x`), the opening nil is dead — it can
  # never be observed at method exit. Recording it anyway folds
  # a spurious `nil` constituent into the flow-insensitive
  # class-ivar union, which then poisons reads in OTHER methods
  # (e.g. ipaddr `IN4MASK ^ @mask_addr` rejects the resulting
  # `Integer | nil`). The set holds the `object_id`s of the
  # transient write nodes to skip; soundness is post-domination
  # at the top statement level, so dropping the nil never hides
  # a real runtime-nil read.
  dead_writes = dead_transient_nil_writes(def_node.body, class_name, method_assign_effects)
  gather_ivar_writes(def_node.body, body_scope, class_name, accumulator,
                     EMPTY_GUARDED_IVARS, mutated_ivars, dead_writes)

  # B2.3 — collect per-method evidence for the read-before-
  # write nil contribution. The accumulator-level decision
  # ("is this ivar truly read-before-write across the
  # class lifetime?") is finalised at
  # `contribute_read_before_write_nil!` after the whole
  # class body has been walked, using `init_writes` as
  # the soundness gate (an ivar written in `initialize`
  # is initialised before any other method body runs).
  collect_read_before_write_evidence(def_node, class_name, read_before_write, init_writes, default_scope)
end

.collect_defined_test_ivars(node, names) ⇒ Object 



753
754
755
756
757
758
759
760
761
762
763
764
 
# File 'lib/rigor/inference/scope_indexer.rb', line 753

def collect_defined_test_ivars(node, names)
  return unless node.is_a?(Prism::Node)

  case node
  when Prism::DefinedNode
    target = node.value
    names << target.name if target.is_a?(Prism::InstanceVariableReadNode)
  when Prism::AndNode, Prism::OrNode
    collect_defined_test_ivars(node.left, names)
    collect_defined_test_ivars(node.right, names)
  end
end

.collect_nil_test_ivars(node, names) ⇒ Object 



766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
 
# File 'lib/rigor/inference/scope_indexer.rb', line 766

def collect_nil_test_ivars(node, names)
  return unless node.is_a?(Prism::Node)

  case node
  when Prism::CallNode
    receiver = node.receiver
    if receiver.is_a?(Prism::InstanceVariableReadNode) &&
       %i[nil? !].include?(node.name)
      names << receiver.name
    end
  when Prism::AndNode, Prism::OrNode
    collect_nil_test_ivars(node.left, names)
    collect_nil_test_ivars(node.right, names)
  end
end

.collect_read_before_write_evidence(def_node, class_name, read_before_write, init_writes, default_scope = nil) ⇒ Object

Walks the method body in AST (== execution) order tracking ivar names whose first reference is a read. The set is unioned into the class-wide ‘read_before_write` accumulator. For `initialize` def bodies, every write target is unioned into `init_writes` instead — used by the finalisation step to suppress nil contribution for ivars the constructor guarantees are initialised.



503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
 
# File 'lib/rigor/inference/scope_indexer.rb', line 503

def collect_read_before_write_evidence(def_node, class_name, read_before_write, init_writes, default_scope = nil)
  return if read_before_write.nil? || init_writes.nil?

  seen_writes = Set.new
  read_first = Set.new
  detect_read_before_write(def_node.body, seen_writes, read_first)

  # ADR-38 — `initialize` is the built-in initializer gate;
  # a plugin may declare additional `def`-form initializer
  # methods (minitest `setup`, Rails `after_initialize`, DI
  # setters) on a constrained class. Both fold their writes
  # into `init_writes`, suppressing the read-before-write nil
  # contribution for sibling readers.
  if def_node.name == :initialize ||
     additional_initializer?(class_name, def_node.name, default_scope)
    init_set = (init_writes[class_name] ||= Set.new)
    seen_writes.each { |name| init_set << name }
    return
  end

  return if read_first.empty?

  rbw_set = (read_before_write[class_name] ||= Set.new)
  read_first.each { |name| rbw_set << name }
end

.collect_truthy_test_ivars(node, names) ⇒ Object 



741
742
743
744
745
746
747
748
749
750
751
 
# File 'lib/rigor/inference/scope_indexer.rb', line 741

def collect_truthy_test_ivars(node, names)
  return unless node.is_a?(Prism::Node)

  case node
  when Prism::InstanceVariableReadNode
    names << node.name
  when Prism::AndNode, Prism::OrNode
    collect_truthy_test_ivars(node.left, names)
    collect_truthy_test_ivars(node.right, names)
  end
end

.conditional_assignment_outcome(node, target, class_name, effects, depth, visiting) ⇒ Object

‘if`/`unless` is a definite assignment of `target` only when BOTH the then and else arms definitely assign (or raise-out). A missing else arm means the fall-through path skips the assignment -> not definite. Modifier-form `if`/`unless` (no else, single predicate’d statement) likewise.



1051
1052
1053
1054
1055
1056
1057
1058
1059
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1051

def conditional_assignment_outcome(node, target, class_name, effects, depth, visiting)
  else_branch = node.is_a?(Prism::IfNode) ? node.subsequent : node.else_clause
  return :falls_through_unassigned unless else_branch.is_a?(Prism::ElseNode)
  return :falls_through_unassigned unless node.statements

  then_ok = branch_definitely_assigns?(node.statements, target, class_name, effects, depth, visiting)
  else_ok = branch_definitely_assigns?(else_branch.statements, target, class_name, effects, depth, visiting)
  then_ok && else_ok ? :assigned : :falls_through_unassigned
end

.contribute_read_before_write_nil!(accumulator, read_before_write, init_writes) ⇒ Object

B2.3 — finalize the read-before-write nil contribution. For each class, for each ivar where SOME method body observed a read-before-write AND no ‘initialize` write exists for that ivar, contribute `Constant` to the class-wide accumulator.

The ‘initialize` filter is the soundness gate: Ruby semantics guarantee `initialize` runs first (via `Class.new`), so a write there reaches every other method body’s read. Read-before-write in a non-init method is then NOT a nil-at-runtime case — it’s just AST-order coincidence. Without this filter a normal ‘def initialize; @x = … end` / `def use; @x.foo end` class would have `@x` widened with nil, producing FPs at every `@x.foo` call.



252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
 
# File 'lib/rigor/inference/scope_indexer.rb', line 252

def contribute_read_before_write_nil!(accumulator, read_before_write, init_writes)
  nil_t = Type::Combinator.constant_of(nil)
  read_before_write.each do |class_name, ivar_set|
    init_set = init_writes[class_name] || EMPTY_GUARDED_IVARS
    per_class = accumulator[class_name]
    next if per_class.nil?

    ivar_set.each do |ivar_name|
      # Soundness gates (in order):
      # (1) `initialize` writes the ivar → it's set
      #     before any other method runs, so the
      #     read-before-write in a sibling method is
      #     NOT a runtime nil case.
      # (2) The accumulator has NO entry for the ivar
      #     → some write was deliberately skipped (the
      #     falsey-default `@x = nil unless @x` slice's
      #     no-seed behaviour). Adding nil here would
      #     defeat that skip and re-introduce the
      #     `Constant[nil]` FP the skip silenced.
      next if init_set.include?(ivar_name)
      next unless per_class.key?(ivar_name)

      existing = per_class[ivar_name]
      per_class[ivar_name] = Type::Combinator.union(existing, nil_t)
    end
  end
end

.data_define_call?(node) ⇒ Boolean

Recognises ‘Data.define(*Symbol)` and `Data.define(*Symbol) do … end` at constant-write rvalue position. The receiver MUST be the bare `Data` constant (or `::Data`); other receivers (a local variable, a method call return) are rejected because their identity is not statically known.

Returns:

  • (Boolean) 


2506
2507
2508
2509
2510
2511
2512
2513
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2506

def data_define_call?(node)
  return false unless node.is_a?(Prism::CallNode)
  return false unless node.name == :define
  return false unless meta_constant_receiver?(node.receiver, :Data)

  args = node.arguments&.arguments || []
  args.all?(Prism::SymbolNode)
end

.dead_transient_nil_writes(body, class_name = nil, method_assign_effects = nil) ⇒ Object 



930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
 
# File 'lib/rigor/inference/scope_indexer.rb', line 930

def dead_transient_nil_writes(body, class_name = nil, method_assign_effects = nil)
  statements = top_level_statements(body)
  return nil if statements.length < 2

  dead = nil

  statements.each_with_index do |stmt, i|
    next unless stmt.is_a?(Prism::InstanceVariableWriteNode) && nil_literal_value?(stmt.value)

    # The opening `@x = nil` is dead when every completing path
    # of the SUFFIX after it (normal end OR early `return`,
    # never a `raise`-terminated path) definitely reassigns
    # `@x` non-nil. The suffix analysis credits an
    # unconditionally-called same-class method's own definite
    # assignments via `method_assign_effects`.
    if suffix_definitely_assigns?(statements, i + 1, stmt.name, class_name, method_assign_effects)
      (dead ||= Set.new) << stmt.object_id
    end
  end

  dead
end

.decompose_multi_write_rhs(rhs_type, front_count, back_count, rest_present:) ⇒ Object 



1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1179

def decompose_multi_write_rhs(rhs_type, front_count, back_count, rest_present:)
  if rhs_type.is_a?(Type::Tuple)
    elements = rhs_type.elements
    fronts = Array.new(front_count) { |i| multi_write_slot_type(elements, i) }
    if rest_present
      middle_end = [elements.size - back_count, front_count].max
      backs = Array.new(back_count) { |i| multi_write_slot_type(elements, middle_end + i) }
      [fronts, Type::Combinator.untyped, backs]
    else
      backs = Array.new(back_count) { |i| multi_write_slot_type(elements, front_count + i) }
      [fronts, nil, backs]
    end
  else
    # Unanalyzable / non-tuple RHS: every slot is the unknown floor.
    floor = Type::Combinator.untyped
    [Array.new(front_count) { floor }, rest_present ? floor : nil, Array.new(back_count) { floor }]
  end
end

.deep_merge_class_methods(base, overlay) ⇒ Object

Merges two ‘class_name => { method => kind }` tables, unioning the per-class method maps (so a seeded cross-file table and the current file’s table combine instead of clobbering).



1401
1402
1403
1404
1405
1406
1407
1408
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1401

def deep_merge_class_methods(base, overlay)
  return overlay if base.nil? || base.empty?
  return base if overlay.empty?

  base.merge(overlay) do |_class_name, base_methods, overlay_methods|
    base_methods.merge(overlay_methods)
  end
end

.def_receiver_targets_lexical_self?(receiver, qualified_prefix) ⇒ Boolean

Only ‘Prism::ConstantReadNode` is observed in real Ruby —Prism mis-parses `def C::P.method` as `def C.P` (Ruby itself rejects the form as a SyntaxError). The ConstantPathNode branch stays defensive in case Prism’s grammar widens.

Returns:

  • (Boolean) 


1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1569

def def_receiver_targets_lexical_self?(receiver, qualified_prefix)
  return false if qualified_prefix.empty?

  case receiver
  when Prism::ConstantReadNode
    receiver.name.to_s == qualified_prefix.last
  when Prism::ConstantPathNode
    rendered = render_constant_path(receiver)
    return false unless rendered

    path = rendered.split("::")
    qualified_prefix.last(path.length) == path
  else
    false
  end
end

.def_singleton?(def_node, qualified_prefix, in_singleton_class) ⇒ Boolean

‘def Foo.bar` inside `module Foo` (or `def Meta.init` inside `module Meta`) is semantically equivalent to `def self.bar`: at the def-site, the runtime value of the constant `Foo` is the module itself (== `self`). Recognise the form so the method registers as singleton on the enclosing class.

The cross-class form ‘def Bar.baz` inside `module Foo` —where the receiver names a constant other than the enclosing class — is not supported at this slice; falls through to `:instance` (current behaviour) rather than silently re-routing the registration.

Returns:

  • (Boolean) 


1559
1560
1561
1562
1563
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1559

def def_singleton?(def_node, qualified_prefix, in_singleton_class)
  return true if def_node.receiver.is_a?(Prism::SelfNode) || in_singleton_class

  def_receiver_targets_lexical_self?(def_node.receiver, qualified_prefix)
end

.detect_read_before_write(node, seen_writes, read_first) ⇒ Object 



599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
 
# File 'lib/rigor/inference/scope_indexer.rb', line 599

def detect_read_before_write(node, seen_writes, read_first)
  return unless node.is_a?(Prism::Node)
  return if IVAR_BARRIER_NODES.any? { |klass| node.is_a?(klass) }

  read_first << node.name if node.is_a?(Prism::InstanceVariableReadNode) && !seen_writes.include?(node.name)

  # Descend BEFORE recording a write — `@x = @x + 1`'s
  # RHS is an `InstanceVariableReadNode` that runs before
  # the write is committed; the read is therefore
  # read-before-write semantically. Prism's
  # `compact_child_nodes` returns the value child before
  # the lvalue target, matching this order.
  node.compact_child_nodes.each do |c|
    detect_read_before_write(c, seen_writes, read_first)
  end

  seen_writes << node.name if IVAR_WRITE_NODES.any? { |klass| node.is_a?(klass) }
end

.discovered_classes_for_paths(paths, buffer: nil) ⇒ Hash{String => Rigor::Type::Singleton}

Walks every file in ‘paths` (each path is parsed once with `Prism.parse_file`) and returns the unioned project-wide `discovered_classes` Hash: `=> Singleton`. Used by `Analysis::Runner` to seed each file’s ‘default_scope.discovered_classes` so that lexical constant lookup in one file resolves a `class Foo` declared in a sibling file. Per-file collisions are last-write-wins (matches the existing in-file merge semantics). Parse failures fail-soft to an empty contribution. The `buffer` argument, when present, redirects reads for the bound logical path to the buffer’s physical path so editor-mode pre-passes see the in-flight bytes.

**Modules are intentionally excluded** from the project-wide seed: a ‘module M; module_function; def x; end; end` body, when surfaced as `singleton(M)` to the dispatcher, falls through to `Kernel#x` (or any Module ancestor method) when the project’s per-file ‘discovered_methods` doesn’t know ‘M.x` — leading to surprising types like `Kernel.select → Array`. Until cross-file `discovered_methods` follows the same project-wide seed, registering modules here would introduce regressions in modules-with-module_function idioms that previously resolved to `Dynamic`. Class declarations are safe because per-file `discovered_methods` already tracks `def self.x` / `def x` instance and singleton methods consistently.

Parameters:

Returns:



2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2213

def discovered_classes_for_paths(paths, buffer: nil)
  accumulator = {}
  paths.each do |path|
    physical = buffer ? buffer.resolve(path) : path
    source = File.read(physical)
    root = Prism.parse(source, filepath: path).value
    collect_class_decls(root, [], accumulator)
  rescue StandardError
    # Skip files that fail to parse or read; the per-file
    # analyzer surfaces the parse error separately.
    next
  end
  accumulator.freeze
end

.discovered_def_index_for_paths(paths, buffer: nil) ⇒ Hash{Symbol => Hash}

ADR-24 slice 2 — cross-file companion to ‘discovered_classes_for_paths`. Walks every project file once and returns both the merged `discovered_def_nodes` table (a class reopened across files has its method tables merged) and the merged class -> superclass-name map. The engine consults these so an implicit-self call inside a subclass resolves against a superclass `def` declared in a sibling file (`Mastodon::CLI::Accounts` calling a helper defined in `Mastodon::CLI::Base`).

The returned ‘def_sources` map mirrors `def_nodes` but stores a `“path:line”` String per `(class_name, method_name)` instead of the `Prism::DefNode`. A `Prism::Location` does not expose its source file through public API, so the source site is captured here, in the pre-pass loop that still holds `path`. `CheckRules#undefined_method_diagnostic` consults the seeded copy to name the defining file when a project monkey-patch on a core/stdlib/gem class is called cross-file (ADR-17). First write wins, matching `def_nodes`’ own merge order.

Parameters:

Returns:

  • (Hash{Symbol => Hash}) —

    ‘{ def_nodes:, def_sources:, superclasses:, includes:, class_sources: }`



2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2253

def discovered_def_index_for_paths(paths, buffer: nil)
  acc = { def_nodes: {}, singleton_def_nodes: {}, def_sources: {}, superclasses: {}, includes: {},
          method_visibilities: {}, methods: {}, class_sources: {}, data_member_layouts: {} }
  paths.each do |path|
    physical = buffer ? buffer.resolve(path) : path
    root = Prism.parse(File.read(physical), filepath: path).value
    accumulate_project_index(acc, path, root)
  rescue StandardError
    # Skip files that fail to parse or read; the per-file
    # analyzer surfaces the parse error separately.
    next
  end
  # Cross-file method suppression is for the project's OWN
  # accessors (attr_* / define_method / alias) — NOT for plain
  # `def`s. A cross-file `def` on a class is exactly the ADR-17
  # monkey-patch case the undefined-method rule deliberately
  # surfaces (fire + def-site annotation, nudging `pre_eval:`),
  # so dropping the `def`-declared names keeps that contract
  # intact while still letting `attr_reader :x` in one file
  # suppress a false undefined-method for `obj.x` in another.
  acc[:methods] = subtract_def_methods(acc[:methods], acc[:def_nodes])
  %i[def_nodes singleton_def_nodes def_sources includes method_visibilities methods class_sources].each do |key|
    acc[key].each_value(&:freeze)
  end
  acc.transform_values(&:freeze)
end

.falsey_constant?(type) ⇒ Boolean

Returns:

  • (Boolean) 


1232
1233
1234
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1232

def falsey_constant?(type)
  type.is_a?(Type::Constant) && (type.value.nil? || type.value == false)
end

.gather_cvar_writes(node, scope, class_name, accumulator) ⇒ Object 



1278
1279
1280
1281
1282
1283
1284
1285
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1278

def gather_cvar_writes(node, scope, class_name, accumulator)
  return unless node.is_a?(Prism::Node)

  record_cvar_write(node, scope, class_name, accumulator) if node.is_a?(Prism::ClassVariableWriteNode)
  return if IVAR_BARRIER_NODES.any? { |klass| node.is_a?(klass) }

  node.compact_child_nodes.each { |c| gather_cvar_writes(c, scope, class_name, accumulator) }
end

.gather_global_writes(node, scope, accumulator) ⇒ Object 



1307
1308
1309
1310
1311
1312
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1307

def gather_global_writes(node, scope, accumulator)
  return unless node.is_a?(Prism::Node)

  record_global_write(node, scope, accumulator) if node.is_a?(Prism::GlobalVariableWriteNode)
  node.compact_child_nodes.each { |c| gather_global_writes(c, scope, accumulator) }
end

.gather_ivar_writes(node, scope, class_name, accumulator, guarded_ivars = EMPTY_GUARDED_IVARS, mutated_ivars = nil, dead_writes = nil) ⇒ Object 



624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
 
# File 'lib/rigor/inference/scope_indexer.rb', line 624

def gather_ivar_writes(node, scope, class_name, accumulator, guarded_ivars = EMPTY_GUARDED_IVARS,
                       mutated_ivars = nil, dead_writes = nil)
  return unless node.is_a?(Prism::Node)

  if node.is_a?(Prism::InstanceVariableWriteNode) &&
     !(dead_writes && dead_writes.include?(node.object_id))
    record_ivar_write(node, scope, class_name, accumulator,
                      guarded: guarded_ivars.include?(node.name))
  end

  # N1 — parallel / multiple assignment (`old, @cb = @cb, block`,
  # `@i, @o, @e, @thr = Open3.popen3(cmd)`). A direct
  # `InstanceVariableWriteNode` is the only write form this
  # collector handled, so an ivar appearing as a `MultiWriteNode`
  # target was silently dropped from the class-ivar union — leaving
  # it to seed as pure `nil` (from a sibling `@cb = nil` ctor write,
  # or absent entirely) and false-fire `if @cb` always-falsey /
  # `@thr.alive?` undefined-for-nil. Record each ivar target with
  # its tuple-position RHS type where the RHS is array/tuple-shaped,
  # else the unanalyzable floor (the same `Dynamic[top]` a single
  # write to an unknown RHS records — an unanalyzable multi-write
  # means unknown, not nil).
  record_multi_write_ivars(node, scope, class_name, accumulator)

  record_ivar_mutator_call(node, class_name, mutated_ivars) if mutated_ivars && node.is_a?(Prism::CallNode)

  # Don't recurse into nested defs, classes, or modules; their
  # ivars belong to their own enclosing class.
  return if IVAR_BARRIER_NODES.any? { |klass| node.is_a?(klass) }

  if node.is_a?(Prism::IfNode) || node.is_a?(Prism::UnlessNode)
    walk_conditional_ivar_writes(node, scope, class_name, accumulator, guarded_ivars,
                                 mutated_ivars, dead_writes)
    return
  end

  node.compact_child_nodes.each do |c|
    gather_ivar_writes(c, scope, class_name, accumulator, guarded_ivars, mutated_ivars, dead_writes)
  end
end

.index(root, default_scope:, converged_loop_recording: false) ⇒ Hash{Prism::Node => Rigor::Scope}

Build the scope index for a Prism program subtree.

Parameters:

  • root (Prism::Node) —

    usually a ‘Prism::ProgramNode`, but any subtree the caller wants the indexer to walk works.

  • default_scope (Rigor::Scope) —

    the scope used for the root, and the fallback returned for any Prism node not contained in ‘root`’s subtree.

  • converged_loop_recording (Boolean) (defaults to: false) —

    display-path flag —when true the evaluator re-records fixpoint-tracked loop bodies from their CONVERGED bindings so per-line probes (‘rigor annotate`) reflect the post-writeback state, not the cap-N intermediate constants. Off for the check path.

Returns:

  • (Hash{Prism::Node => Rigor::Scope}) —

    identity-comparing table whose default value is ‘default_scope`.



59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
 
# File 'lib/rigor/inference/scope_indexer.rb', line 59

def index(root, default_scope:, converged_loop_recording: false) # rubocop:disable Metrics/AbcSize
  # Slice A-declarations. Build the declaration overrides
  # first so every scope handed to the StatementEvaluator
  # already carries the table; structural sharing through
  # `Scope#with_local` / `#with_fact` / `#with_self_type`
  # propagates it across every derived scope.
  declared_types, discovered_classes = build_declaration_artifacts(root)
  # Merge the indexer's findings on top of whatever the
  # base scope already carries so callers that seed
  # cross-file class knowledge (e.g. the ADR-14
  # `SigGen::ObservationCollector` pre-walking project
  # `lib/` before scanning `spec/`) keep their seeds
  # alongside the per-file declarations the indexer
  # itself discovers. Indexer-found entries win on
  # collision — same-file declarations are the most
  # specific authority.
  merged_classes = default_scope.discovered_classes.merge(discovered_classes)
  seeded_scope = default_scope.with_discovery(
    default_scope.discovery.with(declared_types: declared_types,
                                 discovered_classes: merged_classes)
  )

  # Slice 7 phase 2. Pre-pass over every class/module body
  # to collect the per-class ivar accumulator. Seeded after
  # declared_types so the rvalue typer in the pre-pass can
  # see declaration overrides.
  class_ivars = build_class_ivar_index(root, seeded_scope)
  seeded_scope = seeded_scope.with_discovery(seeded_scope.discovery.with(class_ivars: class_ivars))

  # Slice 7 phase 6. Same pre-pass shape for cvars (per
  # class) and globals (program-wide). Globals are also
  # materialised into the top-level scope's `globals` map
  # so reads at the top level (and in CLI probes that do
  # not enter a method body) observe the precise type
  # without consulting the accumulator on every lookup.
  class_cvars = build_class_cvar_index(root, seeded_scope)
  seeded_scope = seeded_scope.with_discovery(seeded_scope.discovery.with(class_cvars: class_cvars))
  program_globals = build_program_global_index(root, seeded_scope)
  seeded_scope = seeded_scope.with_discovery(seeded_scope.discovery.with(program_globals: program_globals))
  program_globals.each { |name, type| seeded_scope = seeded_scope.with_global(name, type) }

  # Slice 7 phase 9. In-source constant value tracking.
  # Walks every ConstantWriteNode/ConstantPathWriteNode in
  # the program and types its rvalue under a scope that
  # carries the surrounding qualified prefix as
  # `self_type`, so the rvalue typer sees in-class
  # references resolve correctly. Multiple writes to the
  # same qualified name union via `Type::Combinator.union`.
  in_source_constants = build_in_source_constants(root, seeded_scope)
  seeded_scope = seeded_scope.with_discovery(
    seeded_scope.discovery.with(in_source_constants: in_source_constants)
  )

  # Slice 7 phase 12. In-source method discovery. Walks
  # every class/module body for `Prism::DefNode` and
  # recognised `define_method` calls and records the
  # introduced method names. `rigor check` consults the
  # table to suppress false positives for methods the
  # user has defined but no RBS sig describes.
  # Merged UNDER any cross-file pre-pass seed (like the def-node
  # / include tables below) so a method `def`/`attr_reader`-
  # declared in one file suppresses a false `undefined-method`
  # for a call in another — `rigor check` seeds the project-wide
  # table via `Runner#seed_project_scope`.
  discovered_methods = deep_merge_class_methods(
    default_scope.discovered_methods, build_discovered_methods(root)
  )
  seeded_scope = seeded_scope.with_discovery(seeded_scope.discovery.with(discovered_methods: discovered_methods))

  # v0.0.2 #5 + ADR-24 slice 2 — record per-instance-method
  # def nodes, the class -> superclass map, and the
  # class/module -> included-modules map, each merged under
  # the cross-file pre-pass seed (see below).
  # v0.1.2 — per-class table of method visibilities
  # (`:public` / `:private` / `:protected`). The
  # `def.method-visibility-mismatch` and ADR-35
  # `def.override-visibility-reduced` CheckRules consult the
  # table. Seeded inside `merge_project_method_indexes` so the
  # per-file visibilities merge OVER the cross-file project seed
  # rather than overwriting it.
  seeded_scope = merge_project_method_indexes(seeded_scope, default_scope, root)

  table = {}.compare_by_identity
  table.default = seeded_scope

  # Last-visit-wins, not first: when `StatementEvaluator`
  # internally re-evaluates a subtree (notably `eval_begin`'s
  # retry-edge widening pass), the LATER visit carries the
  # corrected entry scope (e.g. a `tries` widened to
  # `Nominal[Integer]` after the rescue body's `tries += 1;
  # retry` is observed). The diagnostic layer reads
  # `table[node]` to type predicates; the second pass's
  # entry is the one that reflects all flow-derived
  # rebinds, so it MUST overwrite the first.
  on_enter = ->(node, scope) { table[node] = scope }
  StatementEvaluator.new(scope: seeded_scope, on_enter: on_enter,
                         converged_loop_recording: converged_loop_recording).evaluate(root)

  propagate(root, table, seeded_scope)
  table
end

.ivar_write_targets(node, acc = Set.new) ⇒ Object

Every ivar this body assigns a non-nil value to ANYWHERE (the candidate set for the method’s definite-assignment scan).



884
885
886
887
888
889
890
 
# File 'lib/rigor/inference/scope_indexer.rb', line 884

def ivar_write_targets(node, acc = Set.new)
  return acc unless node.is_a?(Prism::Node)

  acc << node.name if node.is_a?(Prism::InstanceVariableWriteNode) && !nil_literal_value?(node.value)
  node.compact_child_nodes.each { |c| ivar_write_targets(c, acc) }
  acc
end

.literal_method_name(node) ⇒ Object 



2175
2176
2177
2178
2179
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2175

def literal_method_name(node)
  return nil unless node.is_a?(Prism::SymbolNode) || node.is_a?(Prism::StringNode)

  node.unescaped&.to_sym
end

.merge_class_keyed_index_tables(acc, root) ⇒ Object

Folds the per-class method-visibility and method-existence tables of one file into the cross-file accumulator (kept out of #accumulate_project_index to hold its ABC budget).



2314
2315
2316
2317
2318
2319
2320
2321
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2314

def merge_class_keyed_index_tables(acc, root)
  build_discovered_method_visibilities(root).each do |class_name, table|
    (acc[:method_visibilities][class_name] ||= {}).merge!(table)
  end
  build_discovered_methods(root).each do |class_name, table|
    (acc[:methods][class_name] ||= {}).merge!(table)
  end
end

.merge_discovered_defs(def_nodes, def_sources, path, root) ⇒ Object

Merges one file’s ‘class → method → DefNode` map into the cross-file `def_nodes` index and records each method’s first- seen ‘“path:line”` definition site in `def_sources` (ADR-17 —the un-registered-project-patch signal `call.undefined-method` and `rigor triage` key on).



2348
2349
2350
2351
2352
2353
2354
2355
2356
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2348

def merge_discovered_defs(def_nodes, def_sources, path, root)
  build_discovered_def_nodes(root).each do |class_name, methods|
    (def_nodes[class_name] ||= {}).merge!(methods)
    sources = (def_sources[class_name] ||= {})
    methods.each do |method_name, def_node|
      sources[method_name] ||= "#{path}:#{def_node.location&.start_line || 1}"
    end
  end
end

.merge_project_method_indexes(seeded_scope, default_scope, root) ⇒ Object

v0.0.2 #5 + ADR-24 slice 2 — seeds the three project-method indexes onto ‘seeded_scope`: the per-instance-method def-node table, the class -> superclass map, and the class/module -> included-modules map. Each per-file table is merged UNDER the cross-file `discovered_def_index_for_paths` seed carried on `default_scope` — same-file declarations win per entry, the cross-file seed supplies sibling-file ancestors.



169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
 
# File 'lib/rigor/inference/scope_indexer.rb', line 169

def merge_project_method_indexes(seeded_scope, default_scope, root)
  def_nodes = default_scope.discovered_def_nodes.merge(
    build_discovered_def_nodes(root)
  ) { |_class, cross_file, per_file| cross_file.merge(per_file) }
  singleton_def_nodes = default_scope.discovered_singleton_def_nodes.merge(
    build_discovered_singleton_def_nodes(root)
  ) { |_class, cross_file, per_file| cross_file.merge(per_file) }
  superclasses = default_scope.discovered_superclasses.merge(
    build_discovered_superclasses(root)
  )
  includes = default_scope.discovered_includes.merge(
    build_discovered_includes(root)
  ) { |_class, cross_file, per_file| (cross_file + per_file).uniq }
  # ADR-35 — per-file visibilities merged OVER the cross-file
  # seed (the current file is authoritative for its own classes;
  # sibling-file ancestors are preserved from the project seed).
  method_visibilities = default_scope.discovered_method_visibilities.merge(
    build_discovered_method_visibilities(root)
  ) { |_class, cross_file, per_file| cross_file.merge(per_file) }
  # ADR-48 — per-file Data member layouts merged OVER the cross-file
  # seed (same-file declaration is authoritative for its own classes).
  data_member_layouts = default_scope.data_member_layouts.merge(
    build_data_member_layouts(root)
  )

  seeded_scope.with_discovery(
    seeded_scope.discovery.with(
      discovered_def_nodes: def_nodes,
      discovered_singleton_def_nodes: singleton_def_nodes,
      discovered_superclasses: superclasses,
      discovered_includes: includes,
      discovered_method_visibilities: method_visibilities,
      data_member_layouts: data_member_layouts
    )
  )
end

.meta_call_with_name?(node, receiver_name, method_name) ⇒ Boolean

Returns:

  • (Boolean) 


2557
2558
2559
2560
2561
2562
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2557

def meta_call_with_name?(node, receiver_name, method_name)
  return false unless node.is_a?(Prism::CallNode)
  return false unless node.name == method_name

  meta_constant_receiver?(node.receiver, receiver_name)
end

.meta_constant_receiver?(node, expected_name) ⇒ Boolean

Returns:

  • (Boolean) 


2568
2569
2570
2571
2572
2573
2574
2575
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2568

def meta_constant_receiver?(node, expected_name)
  case node
  when Prism::ConstantReadNode
    node.name == expected_name
  when Prism::ConstantPathNode
    node.parent.nil? && node.name == expected_name
  end
end

.meta_member_names(call_node) ⇒ Object

The Symbol member names of a ‘Data.define(*Symbol)` / `Struct.new(*Symbol [, keyword_init:])` call. For `Struct.new` the optional leading String name and trailing `keyword_init:` hash are stripped by #struct_new_positionals; `Data.define` args are all Symbols already.



1532
1533
1534
1535
1536
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1532

def meta_member_names(call_node)
  raw = call_node.arguments&.arguments || []
  symbols = struct_new_call?(call_node) ? (struct_new_positionals(raw) || []) : raw
  symbols.filter_map { |arg| arg.unescaped.to_sym if arg.is_a?(Prism::SymbolNode) }
end

.meta_new_block_body(node) ⇒ Object

v0.1.2 — when a ‘Const = Data.define(*sym) do … end` / `Const = Struct.new(*sym) do … end` constant write carries a block, the block body holds method overrides whose canonical class is `Const`. Survey item (e) extended the recognition to `Const = Module.new do … end` and `Const = Class.new(?super) do … end` — the ADR-16 Tier A “block-as-method” idiom at constant-write position. Returns the block body node (a `Prism::StatementsNode`) when the rvalue matches; nil otherwise. Used by `walk_methods` / `walk_def_nodes` to push `Const` onto the qualified prefix before recursing.



1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1492

def meta_new_block_body(node)
  return nil unless node.is_a?(Prism::ConstantWriteNode)

  rvalue = node.value
  return nil unless data_define_call?(rvalue) ||
                    struct_new_call?(rvalue) ||
                    module_new_call?(rvalue) ||
                    class_new_call?(rvalue)

  rvalue.block&.body
end

.meta_new_constant_type(node, full) ⇒ Object

Survey item (e): when the rvalue is a recognised ‘Module.new do … end` / `Class.new do … end` / `Struct.new(*sym) do … end` / `Data.define(*sym) do … end` form, type the named constant as `Singleton` so the discovered-method table registered under `full` becomes reachable through singleton-side dispatch (`Const.[]=` etc.). Returns nil for non-meta-new rvalues so the caller falls back to the default `body_scope.type_of(node.value)` shape.



1380
1381
1382
1383
1384
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1380

def meta_new_constant_type(node, full)
  return nil unless meta_new_block_body(node)

  Type::Combinator.singleton_of(full)
end

.method_definite_assigns(class_name, _method_name, def_node, defs, effects, memo, depth) ⇒ Object

Computes the definite-assignment set for one method, memoised per def node. The ‘memo` cycle-guards: a method re-entered while its own summary is in progress contributes nothing (sound under-approximation), so mutual recursion terminates.



863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
 
# File 'lib/rigor/inference/scope_indexer.rb', line 863

def method_definite_assigns(class_name, _method_name, def_node, defs, effects, memo, depth)
  return Set.new if def_node.body.nil?
  return memo[def_node] if memo.key?(def_node)
  return Set.new if depth >= SAME_CLASS_CALL_DEPTH_CAP

  memo[def_node] = Set.new # in-progress sentinel (cycle guard)
  statements = top_level_statements(def_node.body)
  candidates = ivar_write_targets(def_node.body)
  # A transient `@x = nil` opener whose own method reassigns it
  # later must still count `@x` as assigned for callers, so the
  # crediting is computed at the BUILD-time depth.
  resolver = MethodEffectResolver.new(self, class_name, defs, effects, memo, depth)
  assigns = Set.new
  candidates.each do |ivar|
    assigns << ivar if suffix_definitely_assigns_with_resolver?(statements, 0, ivar, class_name, resolver, depth)
  end
  memo[def_node] = assigns
end

.module_function_toggle?(node) ⇒ Boolean

A bare ‘module_function` (no arguments) flips every following `def` in the module body to module-function (instance + singleton) mode.

Returns:

  • (Boolean) 


1761
1762
1763
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1761

def module_function_toggle?(node)
  node.name == :module_function && node.receiver.nil?
end

.module_new_call?(node) ⇒ Boolean

Recognises ‘Module.new` and `Module.new(&block)` / `Module.new do … end` at constant-write rvalue position. The block body is the anonymous module’s ‘module_eval` body; defs inside it bind methods on the named constant (`Const = Module.new do …; def foo; …; end; end`). Arguments are NOT inspected because `Module.new` accepts no positionals — Ruby raises ArgumentError if any are passed — so a malformed call falls through the walker without affecting analysis.

Returns:

  • (Boolean) 


2541
2542
2543
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2541

def module_new_call?(node)
  meta_call_with_name?(node, :Module, :new)
end

.multi_write_slot_type(elements, index) ⇒ Object

The per-slot type for index ‘i` of a tuple RHS. A missing slot (over-destructure) is `nil` at runtime; a present slot keeps its type. Unlike the local-variable binder we do NOT soften an optional slot here — a class-ivar seed deliberately preserves a genuine `T | nil`, and any spurious nil is removed by the flow-side narrowing, not by dropping it at collection time.



1204
1205
1206
1207
1208
1209
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1204

def multi_write_slot_type(elements, index)
  element = elements[index]
  return Type::Combinator.constant_of(nil) if element.nil?

  element
end

.nil_literal_value?(node) ⇒ Boolean

Returns:

  • (Boolean) 


960
961
962
 
# File 'lib/rigor/inference/scope_indexer.rb', line 960

def nil_literal_value?(node)
  node.is_a?(Prism::NilNode)
end

.propagate(node, table, parent_scope) ⇒ Object

Walks ‘node`’s subtree DFS and fills in scope entries for every Prism node the StatementEvaluator did not visit (i.e. expression- interior nodes like the receiver/args of a CallNode). Those nodes inherit their nearest recorded ancestor’s scope.

‘IfNode` / `UnlessNode` are special-cased: the truthy and falsey branches each get their predicate’s narrowed scope before recursing. This handles expression-position conditionals (e.g. ‘cache = if cond; t; else; e; end` and conditionals nested as call arguments) which are typed by ExpressionTyper without going through `eval_if`’s narrowing path.



2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2607

def propagate(node, table, parent_scope)
  return unless node.is_a?(Prism::Node)

  current_scope =
    if table.key?(node)
      table[node]
    else
      table[node] = parent_scope
      parent_scope
    end

  case node
  when Prism::IfNode
    propagate_if_branches(node, table, current_scope)
  when Prism::UnlessNode
    propagate_unless_branches(node, table, current_scope)
  else
    node.compact_child_nodes.each { |child| propagate(child, table, current_scope) }
  end
end

.propagate_if_branches(node, table, current_scope) ⇒ Object 



2628
2629
2630
2631
2632
2633
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2628

def propagate_if_branches(node, table, current_scope)
  truthy_scope, falsey_scope = Narrowing.predicate_scopes(node.predicate, current_scope)
  propagate(node.predicate, table, current_scope) if node.predicate
  propagate(node.statements, table, truthy_scope) if node.statements
  propagate(node.subsequent, table, falsey_scope) if node.subsequent
end

.propagate_unless_branches(node, table, current_scope) ⇒ Object 



2635
2636
2637
2638
2639
2640
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2635

def propagate_unless_branches(node, table, current_scope)
  truthy_scope, falsey_scope = Narrowing.predicate_scopes(node.predicate, current_scope)
  propagate(node.predicate, table, current_scope) if node.predicate
  propagate(node.statements, table, falsey_scope) if node.statements
  propagate(node.else_clause, table, truthy_scope) if node.else_clause
end

.qualified_name_for(constant_path_node) ⇒ Object 



2577
2578
2579
2580
2581
2582
2583
2584
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2577

def qualified_name_for(constant_path_node)
  case constant_path_node
  when Prism::ConstantReadNode
    constant_path_node.name.to_s
  when Prism::ConstantPathNode
    render_constant_path(constant_path_node)
  end
end

.record_alias_map_entry(alias_node, qualified_prefix, accumulator) ⇒ Object 



2123
2124
2125
2126
2127
2128
2129
2130
2131
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2123

def record_alias_map_entry(alias_node, qualified_prefix, accumulator)
  return if qualified_prefix.empty?
  return unless alias_node.new_name.is_a?(Prism::SymbolNode) && alias_node.old_name.is_a?(Prism::SymbolNode)

  class_name = qualified_prefix.join("::")
  new_name = alias_node.new_name.unescaped.to_sym
  old_name = alias_node.old_name.unescaped.to_sym
  (accumulator[class_name] ||= {})[new_name] = old_name
end

.record_alias_method(alias_node, qualified_prefix, in_singleton_class, accumulator) ⇒ Object

Registers the alias name in the ‘discovered_methods` table so `undefined-method` diagnostics are not emitted for calls to the aliased name. The kind mirrors the surrounding class context (instance inside a regular class body, singleton inside `class << self`).



2070
2071
2072
2073
2074
2075
2076
2077
2078
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2070

def record_alias_method(alias_node, qualified_prefix, in_singleton_class, accumulator)
  return if qualified_prefix.empty?
  return unless alias_node.new_name.is_a?(Prism::SymbolNode)

  class_name = qualified_prefix.join("::")
  new_name = alias_node.new_name.unescaped.to_sym
  kind = in_singleton_class ? :singleton : :instance
  (accumulator[class_name] ||= {})[new_name] = kind
end

.record_attr_methods(call_node, qualified_prefix, in_singleton_class, accumulator) ⇒ Object 



2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2156

def record_attr_methods(call_node, qualified_prefix, in_singleton_class, accumulator)
  return if qualified_prefix.empty?
  return unless call_node.receiver.nil? # only the implicit-self macro defines on the lexical class
  return if call_node.arguments.nil?

  kind = in_singleton_class ? :singleton : :instance
  reader = call_node.name != :attr_writer
  writer = call_node.name != :attr_reader
  class_name = qualified_prefix.join("::")
  call_node.arguments.arguments.each do |arg|
    base = literal_method_name(arg)
    next if base.nil?

    accumulator[class_name] ||= {}
    accumulator[class_name][base] = kind if reader
    accumulator[class_name][:"#{base}="] = kind if writer
  end
end

.record_class_new_constant_decl(node, qualified_prefix, accumulator) ⇒ Object

T1 (template-corpora survey) — record a ‘Const = Class.new(Super)` (and the bare `Class.new` / `Module.new`) class-creating constant in the cross-file discovery table so a reference to `Const` from ANOTHER file under the same namespace resolves to the project class instead of falling through to a core same-named class (`Liquid::SyntaxError = Class.new(Error)` referenced in a sibling file’s ‘rescue SyntaxError => e`, which otherwise resolved to core `::SyntaxError`). Mirrors the single-file `in_source_constants` answer, which types `Class.new(Super)` as `Singleton` (the constructed class answers method lookups through Super’s chain). The superclass name is resolved lexically against the enclosing prefix; a bare ‘Class.new` with no superclass (or `Module.new`) types as `Singleton` itself. The block form is left to the existing `meta_new_block_body` machinery — only the plain `Class.new(Super)` constant (the namespaced-sibling-error idiom) is added here.



2399
2400
2401
2402
2403
2404
2405
2406
2407
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2399

def record_class_new_constant_decl(node, qualified_prefix, accumulator)
  rvalue = node.value
  return unless class_new_call?(rvalue) || module_new_call?(rvalue)
  return if rvalue.block # block form: handled by meta_new_block_body walks

  full = (qualified_prefix + [node.name.to_s]).join("::")
  super_name = class_new_superclass_name(rvalue, qualified_prefix, accumulator)
  accumulator[full] = Type::Combinator.singleton_of(super_name || full)
end

.record_class_or_module?(node, qualified_prefix, identity_table, discovered) ⇒ Boolean

Returns:

  • (Boolean) 


2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2464

def record_class_or_module?(node, qualified_prefix, identity_table, discovered)
  name = qualified_name_for(node.constant_path)
  return false unless name

  full = (qualified_prefix + [name]).join("::")
  singleton = Type::Combinator.singleton_of(full)
  identity_table[node.constant_path] = singleton
  discovered[full] = singleton
  child_prefix = qualified_prefix + [name]
  record_declarations(node.body, child_prefix, identity_table, discovered) if node.body
  true
end

.record_class_sources(class_sources, path, root, superclasses, includes) ⇒ Object

ADR-46 slice 1 — accumulates, per qualified user class/module name, the set of files that declare it. A class’s declaration shape (its body ‘def`s, its `class Foo < Bar` superclass, its `include`s) lives wherever the class is opened, so every file that contributes a def / superclass / include for a name is a source of that name’s ancestry edges. Scope#superclass_of / Scope#includes_of record this set when resolving the edge during dependency recording (ADR-46). The class-declaration walk (‘collect_class_decls`) catches bodyless / def-less reopenings the other three builders miss.



2333
2334
2335
2336
2337
2338
2339
2340
2341
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2333

def record_class_sources(class_sources, path, root, superclasses, includes)
  names = Set.new
  collect_class_decls(root, [], decls = {})
  names.merge(decls.keys)
  names.merge(superclasses.keys)
  names.merge(includes.keys)
  names.merge(build_discovered_def_nodes(root).keys)
  names.each { |name| (class_sources[name] ||= Set.new) << path }
end

.record_constant_write(node, qualified_prefix, default_scope, accumulator, base_name) ⇒ Object 



1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1360

def record_constant_write(node, qualified_prefix, default_scope, accumulator, base_name)
  full = qualified_prefix.empty? ? base_name : "#{qualified_prefix.join('::')}::#{base_name}"
  body_scope = default_scope
  unless qualified_prefix.empty?
    body_scope = body_scope.with_self_type(Type::Combinator.singleton_of(qualified_prefix.join("::")))
  end
  rvalue_type = meta_new_constant_type(node, full) || body_scope.type_of(node.value)
  existing = accumulator[full]
  accumulator[full] = existing ? Type::Combinator.union(existing, rvalue_type) : rvalue_type
end

.record_cvar_write(node, scope, class_name, accumulator) ⇒ Object 



1287
1288
1289
1290
1291
1292
1293
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1287

def record_cvar_write(node, scope, class_name, accumulator)
  rvalue_type = scope.type_of(node.value)
  accumulator[class_name] ||= {}
  existing = accumulator[class_name][node.name]
  accumulator[class_name][node.name] =
    existing ? Type::Combinator.union(existing, rvalue_type) : rvalue_type
end

.record_data_member_layout(accumulator, qualified_parts, expr) ⇒ Object

Records ‘qualified -> [members]` when `expr` is a `Data.define(*Symbol)` call with at least one literal-Symbol member.



1877
1878
1879
1880
1881
1882
1883
1884
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1877

def record_data_member_layout(accumulator, qualified_parts, expr)
  return unless data_define_call?(expr)

  members = meta_member_names(expr)
  return if members.empty?

  accumulator[qualified_parts.join("::")] = members.freeze
end

.record_declarations(node, qualified_prefix, identity_table, discovered) ⇒ Object 



2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2449

def record_declarations(node, qualified_prefix, identity_table, discovered)
  return unless node.is_a?(Prism::Node)

  case node
  when Prism::ModuleNode, Prism::ClassNode
    return if record_class_or_module?(node, qualified_prefix, identity_table, discovered)
  when Prism::ConstantWriteNode
    return if record_meta_new_constant?(node, qualified_prefix, identity_table, discovered)
  end

  node.compact_child_nodes.each do |child|
    record_declarations(child, qualified_prefix, identity_table, discovered)
  end
end

.record_def_method(def_node, qualified_prefix, in_singleton_class, accumulator) ⇒ Object 



1538
1539
1540
1541
1542
1543
1544
1545
1546
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1538

def record_def_method(def_node, qualified_prefix, in_singleton_class, accumulator)
  return if qualified_prefix.empty?

  class_name = qualified_prefix.join("::")
  singleton = def_singleton?(def_node, qualified_prefix, in_singleton_class)
  kind = singleton ? :singleton : :instance
  accumulator[class_name] ||= {}
  accumulator[class_name][def_node.name] = kind
end

.record_def_node(def_node, qualified_prefix, in_singleton_class, accumulator) ⇒ Object 



1646
1647
1648
1649
1650
1651
1652
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1646

def record_def_node(def_node, qualified_prefix, in_singleton_class, accumulator)
  return if def_singleton?(def_node, qualified_prefix, in_singleton_class)

  class_name = qualified_prefix.empty? ? TOP_LEVEL_DEF_KEY : qualified_prefix.join("::")
  accumulator[class_name] ||= {}
  accumulator[class_name][def_node.name] = def_node
end

.record_def_visibility(def_node, qualified_prefix, in_singleton_class, current_visibility, accumulator) ⇒ Object

rubocop:enable Metrics/CyclomaticComplexity, Metrics/MethodLength, Metrics/PerceivedComplexity, Metrics/AbcSize



2015
2016
2017
2018
2019
2020
2021
2022
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2015

def record_def_visibility(def_node, qualified_prefix, in_singleton_class, current_visibility, accumulator)
  return if def_node.receiver.is_a?(Prism::SelfNode) || in_singleton_class
  return if qualified_prefix.empty?

  class_name = qualified_prefix.join("::")
  accumulator[class_name] ||= {}
  accumulator[class_name][def_node.name] = current_visibility
end

.record_define_method(call_node, qualified_prefix, in_singleton_class, accumulator) ⇒ Object 



2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2133

def record_define_method(call_node, qualified_prefix, in_singleton_class, accumulator)
  return if qualified_prefix.empty?
  return if call_node.arguments.nil? || call_node.arguments.arguments.empty?

  first_arg = call_node.arguments.arguments.first
  method_name = literal_method_name(first_arg)
  return if method_name.nil?

  class_name = qualified_prefix.join("::")
  accumulator[class_name] ||= {}
  accumulator[class_name][method_name] = in_singleton_class ? :singleton : :instance
end

.record_global_write(node, scope, accumulator) ⇒ Object 



1314
1315
1316
1317
1318
1319
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1314

def record_global_write(node, scope, accumulator)
  rvalue_type = scope.type_of(node.value)
  existing = accumulator[node.name]
  accumulator[node.name] =
    existing ? Type::Combinator.union(existing, rvalue_type) : rvalue_type
end

.record_ivar_mutator_call(node, class_name, mutated_ivars) ⇒ Object

Records ‘@ivar.<method>(…)` calls whose method is in `MutationWidening::ARRAY_MUTATORS` or `HASH_MUTATORS`. The class-ivar pre-pass uses the resulting set to widen the post-collected accumulator entries (see widen_mutated_ivar_entries!). Always-safe to over- collect: any name that the widening primitive declines is ignored at finalization.



672
673
674
675
676
677
678
679
680
681
 
# File 'lib/rigor/inference/scope_indexer.rb', line 672

def record_ivar_mutator_call(node, class_name, mutated_ivars)
  receiver = node.receiver
  return unless receiver.is_a?(Prism::InstanceVariableReadNode)
  return unless MutationWidening::ARRAY_MUTATORS.include?(node.name) ||
                MutationWidening::HASH_MUTATORS.include?(node.name)

  per_class = (mutated_ivars[class_name] ||= {})
  per_ivar = (per_class[receiver.name] ||= Set.new)
  per_ivar << node.name
end

.record_ivar_write(node, scope, class_name, accumulator, guarded: false) ⇒ Object 



1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1108

def record_ivar_write(node, scope, class_name, accumulator, guarded: false)
  rvalue_type = scope.type_of(node.value)

  # `@x = nil unless @x` / `@y = false unless @y` —
  # follow-up to the polarity-aware defensive-init guard
  # fix (ROADMAP § Future cycles — "Defensive ivar-init
  # with nil / false rvalue"). When the rvalue is itself a
  # falsey Constant, `union(rvalue, Constant[nil])`
  # collapses (for `nil`) or doesn't widen the type's
  # truthiness profile (for `false`) — the predicate
  # `unless @x` then folds to a single `Constant[nil]` /
  # `Constant[false]` and the
  # `flow.always-truthy-condition` / `-always-falsey-`
  # rule false-fires on the no-op-but-documented-default
  # idiom. Skip the seed contribution for this write
  # (matches the existing skip for `@x ||= v`, which the
  # pre-pass also does not seed). Other writes to the
  # same ivar still contribute; the falsey-default write
  # carries no useful precision the predicate hasn't
  # already given us. See tdiary-core HEAD `ee40c2b`
  # `lib/tdiary/configuration.rb:157` for the worked site.
  return if guarded && falsey_constant?(rvalue_type)

  rvalue_type = Type::Combinator.union(rvalue_type, Type::Combinator.constant_of(nil)) if guarded
  accumulate_ivar_type(accumulator, class_name, node.name, rvalue_type)
end

.record_meta_new_constant?(node, qualified_prefix, identity_table, discovered) ⇒ Boolean

Recognises class-creating meta calls at constant-write rvalue position and registers ‘Const` (qualified by the surrounding class/module path) as a discovered class. `Const.new(…)` then resolves to a fresh `Nominal` via `meta_new`, instead of the un-narrowed `Dynamic` returned by the default `Class#new` envelope.

Two recognised meta forms:

  • ‘Const = Data.define(*Symbol) [do … end]`

  • ‘Const = Struct.new(*Symbol [, keyword_init: …]) [do … end]`

The block body, if present, is recursed into so any nested class/module declarations in the override block (rare but legal) still feed the discovered table.

Returns:

  • (Boolean) 


2492
2493
2494
2495
2496
2497
2498
2499
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2492

def record_meta_new_constant?(node, qualified_prefix, identity_table, discovered)
  return false unless data_define_call?(node.value) || struct_new_call?(node.value)

  full = (qualified_prefix + [node.name.to_s]).join("::")
  discovered[full] = Type::Combinator.singleton_of(full)
  record_declarations(node.value, qualified_prefix, identity_table, discovered)
  true
end

.record_meta_superclass_members(class_node, qualified_prefix, accumulator) ⇒ Object

‘class Foo < Data.define(:a, :b)` / `class Bar < Struct.new(:x)` synthesizes reader methods (`a`, `b`, `x`) on the subclass that no `def` / `attr_*` declares. Register them in the discovered-methods existence table so an implicit-self read of a member inside the class body is known to exist — both for the existing undefined-method suppression and for the ADR-24 slice-4 self-call recorder, which must treat a synthesized member as an existing method, not an unresolved call. The block-form (`Const = Data.define(:a) do … end`) is handled by the `ConstantWriteNode` branch’s block recursion; its members type ‘self` as `Object`, out of scope here.



1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1515

def record_meta_superclass_members(class_node, qualified_prefix, accumulator)
  superclass = class_node.superclass
  return unless data_define_call?(superclass) || struct_new_call?(superclass)

  members = meta_member_names(superclass)
  return if members.empty?

  class_name = qualified_prefix.join("::")
  table = (accumulator[class_name] ||= {})
  members.each { |member| table[member] ||= :instance }
end

.record_mixin_call(node, current_class, accumulator) ⇒ Object 



1924
1925
1926
1927
1928
1929
1930
1931
1932
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1924

def record_mixin_call(node, current_class, accumulator)
  return unless current_class && node.receiver.nil?
  return unless MIXIN_CALL_NAMES.include?(node.name)

  node.arguments&.arguments&.each do |arg|
    mod = qualified_name_for(arg)
    (accumulator[current_class] ||= []) << mod if mod
  end
end

.record_module_function_names(node, qualified_prefix, body, accumulator) ⇒ Object

‘module_function :a, :b` retro-marks named siblings (defined earlier OR later in the same body) as module-functions. Resolves each symbol-literal argument against the body’s own ‘def`s and registers the matching `DefNode` on the module’s singleton side. Non-symbol arguments and names with no matching ‘def` are skipped (a miss degrades to today’s ‘Dynamic`, never a false resolution).



1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1775

def record_module_function_names(node, qualified_prefix, body, accumulator)
  return if qualified_prefix.empty?

  defs_by_name = statements_of(body).each_with_object({}) do |stmt, acc|
    acc[stmt.name] = stmt if stmt.is_a?(Prism::DefNode) && stmt.receiver.nil?
  end
  class_name = qualified_prefix.join("::")
  node.arguments&.arguments&.each do |arg|
    name = symbol_argument_name(arg)
    def_node = name && defs_by_name[name]
    (accumulator[class_name] ||= {})[name] = def_node if def_node
  end
end

.record_multi_ivar_rest(splat_node, _type, class_name, accumulator) ⇒ Object 



1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1220

def record_multi_ivar_rest(splat_node, _type, class_name, accumulator)
  return unless splat_node.is_a?(Prism::SplatNode)

  expression = splat_node.expression
  return unless expression.is_a?(Prism::InstanceVariableTargetNode)

  # A splat collects the middle slots into an Array; the precise
  # element type is not worth recovering here. Record the
  # unanalyzable floor (an Array of unknown), never nil.
  accumulate_ivar_type(accumulator, class_name, expression.name, Type::Combinator.untyped)
end

.record_multi_ivar_target(target, type, class_name, accumulator) ⇒ Object 



1211
1212
1213
1214
1215
1216
1217
1218
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1211

def record_multi_ivar_target(target, type, class_name, accumulator)
  case target
  when Prism::InstanceVariableTargetNode
    accumulate_ivar_type(accumulator, class_name, target.name, type)
  when Prism::MultiTargetNode
    record_multi_target_ivars(target, type, class_name, accumulator)
  end
end

.record_multi_target_ivars(node, rhs_type, class_name, accumulator) ⇒ Object

Walks a ‘MultiWriteNode` / `MultiTargetNode` target tree against `rhs_type`, recording ivar targets per slot. Mirrors `MultiTargetBinder`’s tuple decomposition but for ivar (rather than local-variable) targets.



1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1167

def record_multi_target_ivars(node, rhs_type, class_name, accumulator)
  lefts = node.lefts || []
  rest = node.rest
  rights = node.rights || []
  fronts, rest_type, backs =
    decompose_multi_write_rhs(rhs_type, lefts.size, rights.size, rest_present: !rest.nil?)

  lefts.each_with_index { |t, i| record_multi_ivar_target(t, fronts[i], class_name, accumulator) }
  record_multi_ivar_rest(rest, rest_type, class_name, accumulator) if rest
  rights.each_with_index { |t, i| record_multi_ivar_target(t, backs[i], class_name, accumulator) }
end

.record_multi_write_ivars(node, scope, class_name, accumulator) ⇒ Object

N1 — records each ‘InstanceVariableTargetNode` of a `MultiWriteNode` (parallel / multiple assignment) into the class-ivar union, with the best cheap per-slot type. When the RHS is array/tuple-shaped (`Type::Tuple`) the ivar at position `i` records the type of element `i`; otherwise — an unanalyzable RHS such as `Open3.popen3(cmd)` typing to `Dynamic` — every ivar slot records that unanalyzable floor (NOT `nil`: a multi-write we cannot decompose means the value is unknown, and `Dynamic` is the sound union constituent, mirroring what a single write to an unknown RHS records). Nested targets (`(@a, @b), @c = …`) recurse with the slot’s type as the new RHS type.



1156
1157
1158
1159
1160
1161
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1156

def record_multi_write_ivars(node, scope, class_name, accumulator)
  return unless node.is_a?(Prism::MultiWriteNode)

  rhs_type = scope.type_of(node.value)
  record_multi_target_ivars(node, rhs_type, class_name, accumulator)
end

.record_singleton_def_node(def_node, qualified_prefix, in_singleton_class, module_function_on, accumulator) ⇒ Object 



1750
1751
1752
1753
1754
1755
1756
1757
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1750

def record_singleton_def_node(def_node, qualified_prefix, in_singleton_class, module_function_on, accumulator)
  singleton = def_singleton?(def_node, qualified_prefix, in_singleton_class) || module_function_on
  return unless singleton
  return if qualified_prefix.empty?

  class_name = qualified_prefix.join("::")
  (accumulator[class_name] ||= {})[def_node.name] = def_node
end

.render_constant_path(node) ⇒ Object 



2586
2587
2588
2589
2590
2591
2592
2593
2594
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2586

def render_constant_path(node)
  prefix =
    case node.parent
    when Prism::ConstantReadNode then "#{node.parent.name}::"
    when Prism::ConstantPathNode then "#{render_constant_path(node.parent)}::"
    else ""
    end
  "#{prefix}#{node.name}"
end

.singleton_class_prefix(node, qualified_prefix) ⇒ Object

Resolves a ‘class << X` body’s qualified prefix.

- `class << self` keeps `qualified_prefix` (the enclosing class).
- `class << Foo` inside `class Foo` collapses to the same prefix
  (semantically `class << self`).
- `class << Foo` not nested in `class Foo` returns `[Foo]`
  so methods defined inside register on Foo's singleton.
- Any other expression (variable, method call) returns nil
  so the walker falls through and skips the body.



1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1464

def singleton_class_prefix(node, qualified_prefix)
  case node.expression
  when Prism::SelfNode
    qualified_prefix
  when Prism::ConstantReadNode, Prism::ConstantPathNode
    rendered = qualified_name_for(node.expression)
    return nil unless rendered

    if !qualified_prefix.empty? && qualified_prefix.last == rendered
      qualified_prefix
    else
      rendered.split("::")
    end
  end
end

.statement_assignment_outcome(stmt, target, class_name, effects, depth, visiting) ⇒ Object

Classifies a single statement’s effect on ‘target`:

:assigned                 — every path through the statement
                            that continues OR returns assigns
                            `target` non-nil (suffix is done);
:terminates_unassigned    — the statement ends the method
                            (return/raise) on some path
                            without a definite assignment, so
                            a completing path escaped;
:falls_through_unassigned — control may continue past it
                            without the assignment (keep
                            scanning the suffix).



998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
 
# File 'lib/rigor/inference/scope_indexer.rb', line 998

def statement_assignment_outcome(stmt, target, class_name, effects, depth, visiting)
  case stmt
  when Prism::InstanceVariableWriteNode
    return :falls_through_unassigned if stmt.name != target

    nil_literal_value?(stmt.value) ? :falls_through_unassigned : :assigned
  when Prism::CallNode
    if unconditional_call_assigns?(stmt, target, class_name, effects, depth, visiting)
      :assigned
    else
      :falls_through_unassigned
    end
  when Prism::IfNode, Prism::UnlessNode
    conditional_assignment_outcome(stmt, target, class_name, effects, depth, visiting)
  when Prism::CaseNode
    case_assignment_outcome(stmt, target, class_name, effects, depth, visiting)
  when Prism::ReturnNode
    :terminates_unassigned
  else
    # Any other statement — including a bare `raise`/`fail`,
    # which terminates without a completing path that observes
    # the seed nil — is neutral: control either continues or the
    # path never reaches method exit. Keep scanning the suffix.
    :falls_through_unassigned
  end
end

.statements_of(body) ⇒ Object

Direct statement children of a class/module body node (a ‘Prism::StatementsNode`, a `Prism::BeginNode` wrapping one, or a lone statement). Returns an empty list for an empty body.



1741
1742
1743
1744
1745
1746
1747
1748
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1741

def statements_of(body)
  case body
  when Prism::StatementsNode then body.body
  when Prism::BeginNode then statements_of(body.statements)
  when nil then []
  else [body]
  end
end

.struct_new_call?(node) ⇒ Boolean

Recognises ‘Struct.new(*Symbol)` and `Struct.new(*Symbol, keyword_init: <expr>)` at constant-write rvalue position. A trailing `KeywordHashNode` (the `keyword_init: …` form) is accepted but does not contribute to member discovery; every other argument MUST be a `Prism::SymbolNode`. At least one Symbol member is required —`Struct.new()` is a degenerate form callers don’t typically use.

Returns:

  • (Boolean) 


2522
2523
2524
2525
2526
2527
2528
2529
2530
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2522

def struct_new_call?(node)
  return false unless meta_call_with_name?(node, :Struct, :new)

  args = node.arguments&.arguments || []
  positional = struct_new_positionals(args)
  return false if positional.nil? || positional.empty?

  positional.all?(Prism::SymbolNode)
end

.struct_new_positionals(args) ⇒ Object 



2564
2565
2566
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2564

def struct_new_positionals(args)
  args.last.is_a?(Prism::KeywordHashNode) ? args[0..-2] : args
end

.subtract_def_methods(methods, def_nodes) ⇒ Object

Removes, per class, the method names that have a project ‘def` node, leaving only accessor/alias/define_method-introduced methods in the cross-file suppression table.



2283
2284
2285
2286
2287
2288
2289
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2283

def subtract_def_methods(methods, def_nodes)
  methods.each_with_object({}) do |(class_name, table), out|
    defs = def_nodes[class_name] || {}
    kept = table.reject { |method_name, _kind| defs.key?(method_name) }
    out[class_name] = kept unless kept.empty?
  end
end

.suffix_definitely_assigns?(statements, from, target, class_name, effects) ⇒ Boolean

True when, starting from ‘statements`, EVERY path that completes the method (falls off the end OR hits an early `return`) definitely assigns `target` a non-nil value first. Paths terminated by `raise` are not completing paths and are ignored (they never observe the ivar at method exit). A path that can fall through `statements` without assigning fails.

Returns:

  • (Boolean) 


970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
 
# File 'lib/rigor/inference/scope_indexer.rb', line 970

def suffix_definitely_assigns?(statements, from, target, class_name, effects)
  statements[from..].each do |stmt|
    outcome = statement_assignment_outcome(stmt, target, class_name, effects, 0, nil)
    # The statement assigned on every continuing path -> the
    # suffix is satisfied no matter what follows.
    return true if outcome == :assigned
    # The statement terminates control here (return/raise) and
    # the value it carried did not assign on every path -> some
    # completing path reached exit without the assignment.
    return false if outcome == :terminates_unassigned
    # Otherwise (:falls_through_unassigned) keep scanning the
    # remaining statements.
  end
  # Fell off the end with no definite assignment.
  false
end

.suffix_definitely_assigns_with_resolver?(statements, from, target, class_name, resolver, depth) ⇒ Boolean

Build-time variant of ‘suffix_definitely_assigns?` that resolves same-class calls through the lazy `resolver` (which recurses into `method_definite_assigns` for not-yet-computed callees) rather than the finished flat table.

Returns:

  • (Boolean) 


896
897
898
899
900
901
902
903
 
# File 'lib/rigor/inference/scope_indexer.rb', line 896

def suffix_definitely_assigns_with_resolver?(statements, from, target, class_name, resolver, depth)
  statements[from..].each do |stmt|
    outcome = statement_assignment_outcome(stmt, target, class_name, resolver, depth, nil)
    return true if outcome == :assigned
    return false if outcome == :terminates_unassigned
  end
  false
end

.symbol_argument_name(arg) ⇒ Object

The Symbol value of a ‘:name` / `“name”` literal argument, or nil.



1790
1791
1792
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1790

def symbol_argument_name(arg)
  arg.unescaped.to_sym if arg.is_a?(Prism::SymbolNode) || arg.is_a?(Prism::StringNode)
end

.then_body_guarded_ivars(node) ⇒ Object

Returns the set of ivar names that, in the THEN body of this conditional, are statically known to be in a nil / unset state — i.e. the body really IS the defensive-init half of the idiom. Conservative on purpose: only the shapes that idiomatically express “the ivar is missing” qualify.

For ‘unless P; body; end`, body runs when `P` is falsey:

- `P = @x` (or `@x && other` / `@x || other`)            → @x is falsey
- `P = defined?(@x)`                                     → @x is undefined

For ‘if P; body; …`, body runs when `P` is truthy:

- `P = @x.nil?`                                          → @x is nil
- `P = !@x` / `not @x`                                   → @x is falsey



730
731
732
733
734
735
736
737
738
739
 
# File 'lib/rigor/inference/scope_indexer.rb', line 730

def then_body_guarded_ivars(node)
  names = Set.new
  if node.is_a?(Prism::UnlessNode)
    collect_truthy_test_ivars(node.predicate, names)
    collect_defined_test_ivars(node.predicate, names)
  else
    collect_nil_test_ivars(node.predicate, names)
  end
  names
end

.top_level_statements(body) ⇒ Object 



953
954
955
956
957
958
 
# File 'lib/rigor/inference/scope_indexer.rb', line 953

def top_level_statements(body)
  return [] if body.nil?
  return body.body if body.is_a?(Prism::StatementsNode)

  [body]
end

.unconditional_call_assigns?(call, target, class_name, effects, depth, _visiting) ⇒ Boolean

True when ‘call` is an unconditional, statement-level, implicit-self (or `self.`) call to a SAME-CLASS method whose definite-assignment summary includes `target`. Calls through a block, on another receiver, or to an unresolved name contribute nothing (the seed nil stays).

Returns:

  • (Boolean) 


1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1094

def unconditional_call_assigns?(call, target, class_name, effects, depth, _visiting)
  return false if effects.nil? || class_name.nil?
  return false if depth >= SAME_CLASS_CALL_DEPTH_CAP
  return false unless call.is_a?(Prism::CallNode)
  return false unless call.block.nil?
  # Implicit self (`mask!(x)`) or explicit `self.mask!(x)` only.
  return false unless call.receiver.nil? || call.receiver.is_a?(Prism::SelfNode)

  assigns = effects.dig(class_name, call.name)
  return false if assigns.nil?

  assigns.include?(target)
end

.visibility_target_name(arg) ⇒ Object 



2059
2060
2061
2062
2063
 
# File 'lib/rigor/inference/scope_indexer.rb', line 2059

def visibility_target_name(arg)
  return arg.unescaped.to_sym if arg.is_a?(Prism::SymbolNode) || arg.is_a?(Prism::StringNode)

  nil
end

.walk_class_cvars(node, qualified_prefix, default_scope, accumulator) ⇒ Object 



1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1249

def walk_class_cvars(node, qualified_prefix, default_scope, accumulator)
  return unless node.is_a?(Prism::Node)

  case node
  when Prism::ClassNode, Prism::ModuleNode
    name = qualified_name_for(node.constant_path)
    if name
      child_prefix = qualified_prefix + [name]
      walk_class_cvars(node.body, child_prefix, default_scope, accumulator) if node.body
      return
    end
  when Prism::DefNode
    collect_def_cvar_writes(node, qualified_prefix, default_scope, accumulator)
    return
  end

  node.compact_child_nodes.each do |child|
    walk_class_cvars(child, qualified_prefix, default_scope, accumulator)
  end
end

.walk_class_includes(node, qualified_prefix, current_class, accumulator) ⇒ Object 



1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1904

def walk_class_includes(node, qualified_prefix, current_class, accumulator)
  return unless node.is_a?(Prism::Node)

  case node
  when Prism::ClassNode, Prism::ModuleNode
    name = qualified_name_for(node.constant_path)
    if name
      full = (qualified_prefix + [name]).join("::")
      walk_class_includes(node.body, qualified_prefix + [name], full, accumulator) if node.body
      return
    end
  when Prism::CallNode
    record_mixin_call(node, current_class, accumulator)
  end

  node.compact_child_nodes.each do |child|
    walk_class_includes(child, qualified_prefix, current_class, accumulator)
  end
end

.walk_class_ivars(node, qualified_prefix, default_scope, accumulator, mutated_ivars, read_before_write = nil, init_writes = nil, method_assign_effects = nil) ⇒ Object

rubocop:disable Metrics/CyclomaticComplexity,Metrics/ParameterLists



354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
 
# File 'lib/rigor/inference/scope_indexer.rb', line 354

def walk_class_ivars(node, qualified_prefix, default_scope, accumulator, mutated_ivars, # rubocop:disable Metrics/CyclomaticComplexity,Metrics/ParameterLists
                     read_before_write = nil, init_writes = nil, method_assign_effects = nil)
  return unless node.is_a?(Prism::Node)

  case node
  when Prism::ClassNode, Prism::ModuleNode
    name = qualified_name_for(node.constant_path)
    if name
      child_prefix = qualified_prefix + [name]
      if node.body
        # Class-body level `@x = nil` writes don't
        # initialise instance ivars at runtime (the
        # class's own singleton ivars and the instance's
        # ivars are separate stores), but they signal
        # "the author KNOWS @x could be nil" and extend
        # the B2.3 soundness gate: an ivar with a
        # class-body write is exempted from the
        # read-before-write nil contribution because the
        # seed already reflects the author's acknowledged
        # nullability via the def-body writes' union.
        # Without this exemption, code that explicitly
        # `@x = nil`s at class-body level then writes
        # `@x = SomeClass.new` inside an instance method
        # gains an unjustified nil widening at every
        # read.
        collect_class_body_ivar_writes(node.body, child_prefix.join("::"), init_writes) if init_writes
        walk_class_ivars(node.body, child_prefix, default_scope, accumulator,
                         mutated_ivars, read_before_write, init_writes, method_assign_effects)
      end
      return
    end
  when Prism::DefNode
    collect_def_ivar_writes(node, qualified_prefix, default_scope, accumulator,
                            mutated_ivars, read_before_write, init_writes, method_assign_effects)
    return
  when Prism::CallNode
    if init_writes && !qualified_prefix.empty? &&
       node.block.is_a?(Prism::BlockNode) &&
       block_initializer?(qualified_prefix.join("::"), node.name, default_scope)
      collect_block_ivar_writes(node.block, qualified_prefix, default_scope,
                                accumulator, mutated_ivars, init_writes)
    end
  end

  node.compact_child_nodes.each do |child|
    walk_class_ivars(child, qualified_prefix, default_scope, accumulator,
                     mutated_ivars, read_before_write, init_writes, method_assign_effects)
  end
end

.walk_class_superclasses(node, qualified_prefix, accumulator) ⇒ Object 



1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1808

def walk_class_superclasses(node, qualified_prefix, accumulator)
  return unless node.is_a?(Prism::Node)

  case node
  when Prism::ClassNode
    name = qualified_name_for(node.constant_path)
    if name
      full = (qualified_prefix + [name]).join("::")
      superclass = node.superclass && qualified_name_for(node.superclass)
      accumulator[full] = superclass if superclass
      walk_class_superclasses(node.body, qualified_prefix + [name], accumulator) if node.body
      return
    end
  when Prism::ModuleNode
    name = qualified_name_for(node.constant_path)
    if name
      walk_class_superclasses(node.body, qualified_prefix + [name], accumulator) if node.body
      return
    end
  end

  node.compact_child_nodes.each do |child|
    walk_class_superclasses(child, qualified_prefix, accumulator)
  end
end

.walk_conditional_ivar_writes(node, scope, class_name, accumulator, guarded_ivars, mutated_ivars = nil, dead_writes = nil) ⇒ Object

Walk an ‘IfNode` / `UnlessNode` so writes inside the THEN body that look like defensive ivar initialisation gain a `nil` union in the seeded type. Without this, `@x = v unless @x` records `Constant` for `@x`, then the predicate folds to that same constant and `flow.always-truthy-condition` fires against a working program. Mirrors the existing skip for `@x ||= v` (`Prism::InstanceVariableOrWriteNode`, which the pre-pass does not seed at all).

Polarity-aware on purpose: only the THEN body picks up the guard. The ELSE branch of ‘if @x; …; else; @x = init; end` would otherwise be marked too — but that pattern (write @x in the else of `if @x`) is a separate idiom whose surrounding reads of `@x` would then surface a nil-receiver FP. The ELSE branch is left ungarded so those reads continue to type as they did before this fix.



699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
 
# File 'lib/rigor/inference/scope_indexer.rb', line 699

def walk_conditional_ivar_writes(node, scope, class_name, accumulator, guarded_ivars,
                                 mutated_ivars = nil, dead_writes = nil)
  then_guards = then_body_guarded_ivars(node)
  then_guarded = then_guards.empty? ? guarded_ivars : (guarded_ivars | then_guards)

  gather_ivar_writes(node.predicate, scope, class_name, accumulator, guarded_ivars,
                     mutated_ivars, dead_writes)
  if node.statements
    gather_ivar_writes(node.statements, scope, class_name, accumulator, then_guarded,
                       mutated_ivars, dead_writes)
  end
  branch = node.is_a?(Prism::IfNode) ? node.subsequent : node.else_clause
  return unless branch

  gather_ivar_writes(branch, scope, class_name, accumulator, guarded_ivars,
                     mutated_ivars, dead_writes)
end

.walk_constant_writes(node, qualified_prefix, default_scope, accumulator) ⇒ Object 



1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1335

def walk_constant_writes(node, qualified_prefix, default_scope, accumulator)
  return unless node.is_a?(Prism::Node)

  case node
  when Prism::ClassNode, Prism::ModuleNode
    name = qualified_name_for(node.constant_path)
    if name
      child_prefix = qualified_prefix + [name]
      walk_constant_writes(node.body, child_prefix, default_scope, accumulator) if node.body
      return
    end
  when Prism::ConstantWriteNode
    record_constant_write(node, qualified_prefix, default_scope, accumulator, node.name.to_s)
    return
  when Prism::ConstantPathWriteNode
    full = qualified_name_for(node.target)
    record_constant_write(node, [], default_scope, accumulator, full) if full
    return
  end

  node.compact_child_nodes.each do |child|
    walk_constant_writes(child, qualified_prefix, default_scope, accumulator)
  end
end

.walk_data_member_layouts(node, qualified_prefix, accumulator) ⇒ Object 



1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1849

def walk_data_member_layouts(node, qualified_prefix, accumulator)
  return unless node.is_a?(Prism::Node)

  case node
  when Prism::ClassNode
    name = qualified_name_for(node.constant_path)
    if name
      record_data_member_layout(accumulator, qualified_prefix + [name], node.superclass)
      walk_data_member_layouts(node.body, qualified_prefix + [name], accumulator) if node.body
      return
    end
  when Prism::ModuleNode
    name = qualified_name_for(node.constant_path)
    if name
      walk_data_member_layouts(node.body, qualified_prefix + [name], accumulator) if node.body
      return
    end
  when Prism::ConstantWriteNode
    record_data_member_layout(accumulator, qualified_prefix + [node.name.to_s], node.value)
  end

  node.compact_child_nodes.each do |child|
    walk_data_member_layouts(child, qualified_prefix, accumulator)
  end
end

.walk_def_nodes(node, qualified_prefix, in_singleton_class, accumulator) ⇒ Object 



1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1603

def walk_def_nodes(node, qualified_prefix, in_singleton_class, accumulator)
  return unless node.is_a?(Prism::Node)

  case node
  when Prism::ClassNode, Prism::ModuleNode
    name = qualified_name_for(node.constant_path)
    if name
      child_prefix = qualified_prefix + [name]
      walk_def_nodes(node.body, child_prefix, false, accumulator) if node.body
      return
    end
  when Prism::SingletonClassNode
    if node.body
      singleton_prefix = singleton_class_prefix(node, qualified_prefix)
      if singleton_prefix
        walk_def_nodes(node.body, singleton_prefix, true, accumulator)
        return
      end
    end
  when Prism::ConstantWriteNode
    if meta_new_block_body(node)
      child_prefix = qualified_prefix + [node.name.to_s]
      walk_def_nodes(meta_new_block_body(node), child_prefix, false, accumulator)
      return
    end
  when Prism::DefNode
    record_def_node(node, qualified_prefix, in_singleton_class, accumulator)
    return
  end

  node.compact_child_nodes.each do |child|
    walk_def_nodes(child, qualified_prefix, in_singleton_class, accumulator)
  end
end

.walk_method_visibilities(node, qualified_prefix, in_singleton_class, current_visibility, accumulator) ⇒ Object

rubocop:disable Metrics/CyclomaticComplexity, Metrics/MethodLength, Metrics/PerceivedComplexity, Metrics/AbcSize



1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1964

def walk_method_visibilities(node, qualified_prefix, in_singleton_class, current_visibility, accumulator)
  return current_visibility unless node.is_a?(Prism::Node)

  case node
  when Prism::ClassNode, Prism::ModuleNode
    name = qualified_name_for(node.constant_path)
    if name
      child_prefix = qualified_prefix + [name]
      walk_method_visibilities(node.body, child_prefix, false, :public, accumulator) if node.body
      return current_visibility
    end
  when Prism::SingletonClassNode
    if node.body
      singleton_prefix = singleton_class_prefix(node, qualified_prefix)
      if singleton_prefix
        walk_method_visibilities(node.body, singleton_prefix, true, :public, accumulator)
        return current_visibility
      end
    end
  when Prism::ConstantWriteNode
    if meta_new_block_body(node)
      child_prefix = qualified_prefix + [node.name.to_s]
      walk_method_visibilities(meta_new_block_body(node), child_prefix, false, :public, accumulator)
      return current_visibility
    end
  when Prism::DefNode
    record_def_visibility(node, qualified_prefix, in_singleton_class, current_visibility, accumulator)
    return current_visibility
  when Prism::CallNode
    updated = apply_visibility_call(node, qualified_prefix, current_visibility, accumulator)
    return updated unless updated.equal?(current_visibility)
  end

  # Statement-position StatementsNode preserves
  # left-to-right visibility flow; everything else
  # recurses with the entry visibility unchanged.
  if node.is_a?(Prism::StatementsNode)
    local_visibility = current_visibility
    node.compact_child_nodes.each do |child|
      local_visibility = walk_method_visibilities(child, qualified_prefix, in_singleton_class,
                                                  local_visibility, accumulator)
    end
  else
    node.compact_child_nodes.each do |child|
      walk_method_visibilities(child, qualified_prefix, in_singleton_class, current_visibility, accumulator)
    end
  end
  current_visibility
end

.walk_methods(node, qualified_prefix, in_singleton_class, accumulator) ⇒ Object

rubocop:disable Metrics/CyclomaticComplexity, Metrics/MethodLength, Metrics/AbcSize, Metrics/PerceivedComplexity



1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1411

def walk_methods(node, qualified_prefix, in_singleton_class, accumulator)
  return unless node.is_a?(Prism::Node)

  case node
  when Prism::ClassNode, Prism::ModuleNode
    name = qualified_name_for(node.constant_path)
    if name
      child_prefix = qualified_prefix + [name]
      record_meta_superclass_members(node, child_prefix, accumulator) if node.is_a?(Prism::ClassNode)
      walk_methods(node.body, child_prefix, false, accumulator) if node.body
      return
    end
  when Prism::SingletonClassNode
    if node.body
      singleton_prefix = singleton_class_prefix(node, qualified_prefix)
      if singleton_prefix
        walk_methods(node.body, singleton_prefix, true, accumulator)
        return
      end
    end
  when Prism::ConstantWriteNode
    if meta_new_block_body(node)
      child_prefix = qualified_prefix + [node.name.to_s]
      walk_methods(meta_new_block_body(node), child_prefix, false, accumulator)
      return
    end
  when Prism::DefNode
    record_def_method(node, qualified_prefix, in_singleton_class, accumulator)
    return
  when Prism::AliasMethodNode
    record_alias_method(node, qualified_prefix, in_singleton_class, accumulator)
    return
  when Prism::CallNode
    record_define_method(node, qualified_prefix, in_singleton_class, accumulator) if node.name == :define_method
    if ATTR_MACROS.include?(node.name)
      record_attr_methods(node, qualified_prefix, in_singleton_class,
                          accumulator)
    end
  end

  node.compact_child_nodes.each do |child|
    walk_methods(child, qualified_prefix, in_singleton_class, accumulator)
  end
end

.walk_singleton_body(body, qualified_prefix, in_singleton_class, accumulator) ⇒ Object

Walks a class/module/singleton-class body’s direct statements in source order, threading the bare-‘module_function` toggle: once a bare `module_function` is seen, every subsequent `def` in the body registers as a singleton method. Nested classes/modules/defs and `module_function :a, :b` named forms recurse / record through the general walker so the toggle stays scoped to its own body.



1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1719

def walk_singleton_body(body, qualified_prefix, in_singleton_class, accumulator)
  module_function_on = false
  statements_of(body).each do |stmt|
    if stmt.is_a?(Prism::CallNode) && module_function_toggle?(stmt)
      if bare_module_function?(stmt)
        module_function_on = true
      else
        record_module_function_names(stmt, qualified_prefix, body, accumulator)
      end
      next
    end
    if stmt.is_a?(Prism::DefNode)
      record_singleton_def_node(stmt, qualified_prefix, in_singleton_class, module_function_on, accumulator)
      next
    end
    walk_singleton_def_nodes(stmt, qualified_prefix, in_singleton_class, accumulator)
  end
end

.walk_singleton_def_nodes(node, qualified_prefix, in_singleton_class, accumulator) ⇒ Object

Walks every node, entering class/module/singleton-class bodies via #walk_singleton_body so a bare ‘module_function` toggle threads correctly across the body’s sibling statements (a child-by-child recursion would reset it). At the top level / inside an arbitrary node there is no ‘module_function` state to carry, so descent is a plain per-child walk.



1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
 
# File 'lib/rigor/inference/scope_indexer.rb', line 1679

def walk_singleton_def_nodes(node, qualified_prefix, in_singleton_class, accumulator)
  return unless node.is_a?(Prism::Node)

  case node
  when Prism::ClassNode, Prism::ModuleNode
    name = qualified_name_for(node.constant_path)
    if name
      walk_singleton_body(node.body, qualified_prefix + [name], false, accumulator) if node.body
      return
    end
  when Prism::SingletonClassNode
    if node.body
      singleton_prefix = singleton_class_prefix(node, qualified_prefix)
      if singleton_prefix
        walk_singleton_body(node.body, singleton_prefix, true, accumulator)
        return
      end
    end
  when Prism::ConstantWriteNode
    if meta_new_block_body(node)
      child_prefix = qualified_prefix + [node.name.to_s]
      walk_singleton_body(meta_new_block_body(node), child_prefix, false, accumulator)
      return
    end
  when Prism::DefNode
    record_singleton_def_node(node, qualified_prefix, in_singleton_class, false, accumulator)
    return
  end

  node.compact_child_nodes.each do |child|
    walk_singleton_def_nodes(child, qualified_prefix, in_singleton_class, accumulator)
  end
end

.widen_member_for_observed_mutators(member, observed_methods) ⇒ Object 



338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
 
# File 'lib/rigor/inference/scope_indexer.rb', line 338

def widen_member_for_observed_mutators(member, observed_methods)
  case member
  when Type::Tuple
    return member unless observed_methods.any? { |m| MutationWidening::ARRAY_MUTATORS.include?(m) }

    Type::Combinator.nominal_of("Array", type_args: [Type::Combinator.untyped])
  when Type::HashShape
    return member unless observed_methods.any? { |m| MutationWidening::HASH_MUTATORS.include?(m) }

    Type::Combinator.nominal_of("Hash",
                                type_args: [Type::Combinator.untyped, Type::Combinator.untyped])
  else
    member
  end
end

.widen_mutated_ivar_entries!(accumulator, mutated_ivars) ⇒ Object

Walks the post-collected accumulator and widens any Tuple / HashShape entry for an ivar that observed a mutator call anywhere in the same class body. The mutation evidence comes from ‘gather_ivar_writes` recording every `@ivar.<method>(…)` call whose method is in `MutationWidening::ARRAY_MUTATORS` or `HASH_MUTATORS`.

The widening uses ‘MutationWidening.widen_for_mutator` —the same primitive `Inference::StatementEvaluator#eval_call` applies for per-method-body widening on a local / ivar receiver. The class-level pass extends that primitive’s reach so a ‘Tuple`-seeded ivar in `initialize` is observed as `Nominal` at the entry of every OTHER method body in the class — closing the cross-method gap noted in ROADMAP § Future cycles / Type-language / engine (“Tuple / HashShape widening for ivar-seeded literals after mutation”; Redmine 6.1.2 `Redmine::Views::Builders::Structure` is the canonical worked site).

Always-safe: the widening can only LOSE precision; the underlying nominal (‘Array` / `Hash`) and the element union are preserved.



303
304
305
306
307
308
309
310
311
312
313
314
315
 
# File 'lib/rigor/inference/scope_indexer.rb', line 303

def widen_mutated_ivar_entries!(accumulator, mutated_ivars)
  accumulator.each do |class_name, ivars|
    observed = mutated_ivars[class_name]
    next if observed.nil? || observed.empty?

    ivars.each do |ivar_name, type|
      methods = observed[ivar_name]
      next if methods.nil? || methods.empty?

      ivars[ivar_name] = widen_type_for_observed_mutators(type, methods)
    end
  end
end

.widen_type_for_observed_mutators(type, observed_methods) ⇒ Object

Walks a class-ivar accumulator entry (which may be a ‘Union` of multiple write rvalues) and widens any `Tuple` or `HashShape` member whose corresponding mutator family was observed against the ivar somewhere in the class. Class-level widening is more aggressive than the per-method-body `MutationWidening` primitive: it widens both the SHAPE carrier (Tuple → Array, HashShape → Hash) AND the element types to `Dynamic`. The justification — once any method mutates the ivar, its post-mutation contents are statically unknown across method boundaries, so preserving the seed-write’s element precision would be an unsound over-claim (e.g. ‘@struct = [{}]; somewhere: `Constant`).



332
333
334
335
336
 
# File 'lib/rigor/inference/scope_indexer.rb', line 332

def widen_type_for_observed_mutators(type, observed_methods)
  members = type.is_a?(Type::Union) ? type.members : [type]
  widened = members.map { |m| widen_member_for_observed_mutators(m, observed_methods) }
  Type::Combinator.union(*widened)
end