Ruby: decoding an empty protocol buffer allocates objects for every field in that protobuf's descriptor

[fables-tales]

require 'google/protobuf'

Google::Protobuf::DescriptorPool.generated_pool.build do
  add_message "hi" do
    # vary this to see more or less allocations
    10.times do |i|
      optional :"resource_name_#{i}", :string, i+1
    end
  end
end

module Mod
  Hi = Google::Protobuf::DescriptorPool.generated_pool.lookup("hi").msgclass
end

h = Mod::Hi.new
data = Mod::Hi.encode(h)

# 4 GCs to force all allocated objects to either be removed
# or moved to the oldest generation
GC.start
GC.start
GC.start
GC.start

start_alloc = GC.stat
# uncomment either
#Mod::Hi.decode("")
#Mod::Hi.decode(data)
end_alloc = GC.stat

p end_alloc[:total_allocated_objects]-start_alloc[:total_allocated_objects]

[fables-tales]

(Just dropped this in here from a hangout with @haberman)

[haberman]

Confirmed. This is specific to string fields. I have a fix for this here: #6502