Project

General

Profile

Feature #14706

Atomic Integer incr/decr

Added by mperham (Mike Perham) 10 months ago. Updated 10 months ago.

Status:
Open
Priority:
Normal
Assignee:
-
Target version:
-
[ruby-core:86657]

Description

Ruby does not any thread-safe way to implement simple counters without a Mutex. Today Ruby provides Integer#succ but this funcalls "+", making it thread-unsafe as far as I know.

I'd propose adding Integer#incr(amount=1) and Integer#reset which would use atomic operations, giving us thread-safe, high-performance counters.

counter = 0
counter.incr # => 1
counter.incr(10) # => 11
counter.incr(-1) # => 10
counter.reset # => 10
counter # => 0

Related issues

Is duplicate of Ruby trunk - Feature #12607: Ruby needs an atomic integerFeedbackActions

History

#1

Updated by mperham (Mike Perham) 10 months ago

  • Backport deleted (2.3: UNKNOWN, 2.4: UNKNOWN, 2.5: UNKNOWN)
  • Tracker changed from Bug to Feature

Updated by rosenfeld (Rodrigo Rosenfeld Rosas) 10 months ago

I've always missed that feature as well, so I'm +1 to this. This and many other built-in thread-safe helpers by the way ;)

Updated by jeremyevans0 (Jeremy Evans) 10 months ago

I think this feature in general is a good candidate for addition to stdlib.

I don't think the Integer class is appropriate for this, because (some) Integers are immediate values, and all Integers are frozen, and you can't modify them. Integer#+ and Integer#succ are thread safe (they return new objects, they do not modify existing objects), it's counter += 1 that is not thread safe.

In terms of naming, maybe Integer::Counter since that is unlikely to conflict with existing code, with an API like:

counter = Integer::Counter.new(0) # maybe default to 0 if no argument?
counter.incr     # => 1
counter.incr(10) # => 11
counter.incr(-1) # => 10
counter.reset    # => 10
counter.to_int   # => 0
counter.to_i     # => 0

Updated by rafaelfranca (Rafael França) 10 months ago

Related ticket with some discussion already (and partially rejected) https://bugs.ruby-lang.org/issues/12607

Updated by mperham (Mike Perham) 10 months ago

Jeremy, good point. I would agree that the different semantics are enough to justify a different class. Namespacing it within the Thread class also seems reasonable, e.g. Thread::Counter so it would be available if you require 'thread'.

#6

Updated by shyouhei (Shyouhei Urabe) 10 months ago

Updated by shevegen (Robert A. Heiler) 10 months ago

I guess this may come up several times in the future. Would
this be a good candidate for discussion in the upcoming
ruby developer meeting? I don't want to suggest it myself
since I was not the one who started the thread (that was
Mike), but it seems to come up every now and then by
different people, so it may be of some relevance.

Rafael França pointed out that it was parially rejected/closed
but I think that did not come via matz primarily since he
did not comment on it; it may be that other ruby developers
pointed out some problems with it before, like Koichi; but
it may still be good to have matz comment on it so that
people can understand what the remaining problems may be,
if there are some.

Also available in: Atom PDF