Reduce Array object allocation during parsing

[ksss]

The most basic function for parsing a type, parse_type, always generates two Array objects each time it is executed.
However, for simple types, this Array object is not always needed, and the object allocation becomes unnecessary.

In this PR, I will make modifications to minimize unnecessary object allocations in parse_type as much as possible.

Profile script

require 'rbs'
require 'rbs/cli'
require 'memory_profiler'

loader = RBS::EnvironmentLoader.new
Dir["stdlib/*"].each do |path|
  lib = File.basename(path).to_s
  loader.add(library: lib, version: nil)
end

MemoryProfiler.report do
  RBS::Environment.from_loader(loader)
end.pretty_print

Before

Total allocated: 49914684 bytes (453299 objects)
allocated memory by class: 3219256  Array
allocated objects by class: 70648  Array

After

Total allocated: 48873724 bytes (427275 objects)
allocated memory by class: 2178296  Array
allocated objects by class: 44624  Array

[soutaro]

We can early return before allocating intersection_types, and I think it's better code organization.

...

VALUE type = parse_optional(state);

if (state->next_token.type != pAMP) {
  return type;
}

VALUE intersection_types = rb_ary_new();
rb_ary_push(intersection_types, type);

...

[ksss]

Cool! Let's try it.

[soutaro]

👍