From: "Eregon (Benoit Daloze)" <noreply@...>
Date: 2021-11-01T15:59:27+00:00
Subject: [ruby-core:105890] [Ruby master Bug#18243] Ractor.make_shareable does not freeze the receiver of a Proc but allows accessing ivars of it

Issue #18243 has been updated by Eregon (Benoit Daloze).


Dan0042 (Daniel DeLorme) wrote in #note-7:
> It's not uncommon for a proc to have behavior that is independent of `self`.

Yes, but it's not so trivial to assert it is the case for any non-trivial block/Proc.
Changing the self is breaking the context in which it was written, and I think that should be avoided in most cases.
`define_method(name) { ... }` is clear that the self is not the surrounding self, but that's about the only core method with those semantics and it needs to be like that of course.

> From the posts by @eregon it was unclear to me if "self must be shareable" was intended only for an explicit reference to `self`.

All Proc capture `self`, there is no other way currently.

I think we need `define_method(name, make_shareable: true) { ... }` (name up for discussion) and there we know it does not matter what the surrounding `self` is.

`Ractor.make_shareable(some_proc)` must make `some_proc` and all objects reachable from it shareable, hence the `self` captured by the Proc must be shareable, or it breaks the entire concept of Ractor isolation and shareable.
So either it does `Ractor.make_shareable(proc_self)` or it raises an exception.

Changing the proc's `self` on `make_shareable` seems way too confusing, I don't even consider it a reasonable possibility.

----------------------------------------
Bug #18243: Ractor.make_shareable does not freeze the receiver of a Proc but allows accessing ivars of it
https://bugs.ruby-lang.org/issues/18243#change-94433

* Author: Eregon (Benoit Daloze)
* Status: Open
* Priority: Normal
* Backport: 2.6: UNKNOWN, 2.7: UNKNOWN, 3.0: UNKNOWN
----------------------------------------
```ruby
class C
  attr_accessor :foo
  def setter_proc
    Ractor.make_shareable(-> v { @foo = v })
  end
end

c = C.new
c.foo = 1
p c
proc = c.setter_proc
p c.frozen?
Ractor.new(proc) { |s| s.call(42) }.take
p c
```
gives
```
#<C:0x00007f2a3aae7a98 @foo=1>
false
#<C:0x00007f2a3aae7a98 @foo=42> # BUG
```

But that must be a bug, it means the non-main Ractor can directly mutate an object from the main Ractor.

I found this while thinking about https://github.com/ruby/ostruct/pull/29/files and
whether `Ractor.make_shareable` would freeze `@table` and the `OpenStruct` instance (I think it needs to).

Repro code for ostruct and changing ostruct.rb to `$setter = ::Ractor.make_shareable(setter_proc)`:
```ruby
require 'ostruct'

os = OpenStruct.new
os.foo = 1

$setter.call(2)
p os

Ractor.new($setter) { |s| s.call(42) }.take
p os
```
gives
```
#<OpenStruct foo=2>
<internal:ractor>:267: warning: Ractor is experimental, and the behavior may change in future versions of Ruby! Also there are many implementation issues.
#<OpenStruct foo=42> # BUG
```



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

Unsubscribe: <mailto:ruby-core-request@ruby-lang.org?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>