[ruby-core:63029] [CommonRuby - Feature #8259] Atomic attributes accessors

Issue #8259 has been updated by Vit Z.


In my opinion an ideal API would have "atomic" attributes declared at the class level,
yet can be used as normal `@foo` i-vars:

~~~
class MyNode
  atomic :item, :successor
  
  def initialize(item, successor)
    @foo = :foo # this is plain old i-var
    @item = item # this is atomic
    @successor = successor # this is atomic
    puts @foo # this is a non-atomic read
    puts @item # this is an atomic read
  end
  
  # this creates reader and writer that are atomic, because :item is atomic
  attr_accessor :item
  # this is not atomic
  attr_accessor :foo
  
  def cas_new_item(expected_old_item, new_item)
    # atomic :item also generates a new private cas_item method
    # that implements compare-and-swap semantics
    cas_item(expected_old_item, new_item)
  end
  
  class << self
    # atomic i-vars can also be declared on Modules/Classes
    atomic :head
  end
  
  @head = nil
  # cas_head is also added as a private module method
  cas_head(nil, MyNode.new(:item, nil))
end
~~~

Another thing, and I know that this might be controversial, is that `atomic` attributes
should only be "declare-able" before a first instance of the class is created:

~~~
class OtherNode
  new
  # this should raise an ArgumentError, since an instance of OtherNode
  # has already been created
  atomic :item

  class << self
    # not a problem on Module/Class level
    atomic :head
  end  
end
~~~

I believe this is absolutely necessary in order to have a proper and performant
implementation of "atomic" on concurrent Ruby VMs. If I turn out to be wrong
the restriction can be always be removed, however if it is not in place and
I am right, this forever puts an upper-limit constraint on Ruby performance in
concurrent environments.

The auto-generated `cas_item` methods need to be "dumb" pointer swapping CASes, ie
no clever trickery checking wether passed-in arguments are `Fixnum`s or something
else, this is again necessary to not hamstring future upper limit on performance.

The other thing that is needed is an "atomic" fixed-size array, lets call it `AtomicTuple`.
It has to be fixed-size again because otherwise this would create problems for
concurrent Ruby VMs.

~~~
tuple = AtomicTuple.new(16) # creates a new 16-slotted atomic array
# an atomic read
tuple[0] # => nil
# an atomic write
tuple[0] = :foo
# a cas
tuple.cas(0, :foo, :bar) # => true
tuple[0] # => :bar
~~~

I'm fairly certain that adding this to MRI should be quite easy, because of the
GVL all the "atomicness" is already present. For example declaring an i-var
atomic (via atomic `:item`) doesn't have to do anything except to add a `cas_item`
method, after all i-vars are already atomic. `cas_item` also in the first implementation
doesn't have to use native cmpxchg assembly a simple pointer check-then-swap
should be fine (as long as this is in C GVL has our back).

Eric Wong:
Adding "atomic" operations to built in `Array`s, `Hash`es etc. is a very bad idea
because that would force that *all* of the methods on `Array`s and `Hash`es be
concurrency friendly and would result in irreparable performance problems for
truly concurrent Ruby VMs (and this again would put an upper limit on Ruby
performance).

----------------------------------------
Feature #8259: Atomic attributes accessors
https://bugs.ruby-lang.org/issues/8259#change-47113

* Author: Yura Sokolov
* Status: Open
* Priority: Normal
* Assignee: 
* Category: 
* Target version: Ruby 2.1.0
----------------------------------------
=begin
Motivated by this gist ((<URL:https://gist.github.com/jstorimer/5298581>)) and atomic gem

I propose Class.attr_atomic which will add methods for atomic swap and CAS:

  class MyNode
    attr_accessor :item
    attr_atomic :successor

    def initialize(item, successor)
      @item = item
      @successor = successor
    end
  end
  node = MyNode.new(i, other_node)

  # attr_atomic ensures at least #{attr} reader method exists. May be, it should
  # be sure it does volatile access.
  node.successor

  # #{attr}_cas(old_value, new_value) do CAS: atomic compare and swap
  if node.successor_cas(other_node, new_node)
    print "there were no interleaving with other threads"
  end

  # #{attr}_swap atomically swaps value and returns old value.
  # It ensures that no other thread interleaves getting old value and setting
  # new one by cas (or other primitive if exists, like in Java 8)
  node.successor_swap(new_node)

It will be very simple for MRI cause of GIL, and it will use atomic primitives for
other implementations.

Note: both (({#{attr}_swap})) and (({#{attr}_cas})) should raise an error if instance variable were not explicitly set before.

Example for nonblocking queue: ((<URL:https://gist.github.com/funny-falcon/5370416>))

Something similar should be proposed for Structs. May be override same method as (({Struct.attr_atomic}))

Open question for reader:
should (({attr_atomic :my_attr})) ensure that #my_attr reader method exists?
Should it guarantee that (({#my_attr})) provides 'volatile' access?
May be, (({attr_reader :my_attr})) already ought to provide 'volatile' semantic?
May be, semantic of (({@my_attr})) should have volatile semantic (i doubt for that)?
=end




-- 
https://bugs.ruby-lang.org/