Fix type inference for multiple assignment of generic Array[T]

[sekito1107]

Problem

Previously, multiple assignment from instances of generic arrays (e.g., Array[Integer] defined in RBS) resulted in incorrect type inference. MAsgnBox#run0 only handled literal Type::Array correctly. For other types, it assigned the entire array object to the first variable on the left-hand side, leaving others as untyped.

Example

# sig: def get_ints: -> Array[Integer]
a, b = get_ints 
# Before: a: Array[Integer] (the array itself), b: untyped
# After:  a: Integer, b: Integer

Fix

Handled Type::Instance with mod_ary in MAsgnBox#run0 by distributing its element type (ty.args[0]) to all left-hand side variables and the rest element.

Changes

• Update lib/typeprof/core/graph/box.rb to handle Type::Instance in MAsgnBox#run0.
• Add scenario/variable/masgn-generic-array.rb as a regression test.

How to reproduce

Save the following as repro.rb in the typeprof root and run with ruby -Ilib repro.rb

Note:
In repro.rb, update_rbs_file, and update_rb_file are called twice. I observed that this was necessary to correctly trigger the type inference in this standalone script (possibly related to the diff-based analysis). If this is not the recommended way to use the internal API, please let me know with your advice.

require "typeprof"

core = TypeProf::Core::Service.new({})

rbs = <<~RBS
  class Foo
    def get_ints: () -> Array[Integer]
  end
RBS

rb = <<~RUBY
  class Foo
    def test_masgn
      a, b, c = get_ints
      [a, b, c]
    end
    def test_star
      a, *rest = get_ints
      [a, rest]
    end
  end
RUBY

2.times { core.update_rbs_file("test.rbs", rbs) }
2.times { core.update_rb_file("test.rb", rb) }

puts core.dump_declarations("test.rb")

Output without fix

class Foo
  def test_masgn: -> [Array[Integer], untyped, untyped]
  def test_star: -> [Array[Integer], Array[untyped]]
end

Output with fix

class Foo
  def test_masgn: -> [Integer, Integer, Integer]
  def test_star: -> [Integer, Array[Integer]]
end

[mame]

Thank you! (I had just made a similar patch in my working copy a few days ago :-)