[#70252] Re: [ruby-cvs:58640] nobu:r51492 (trunk): node.c: NODE_ALLOCA for ALLOCV — Eric Wong <normalperson@...>
Besides possible backwards compatibility, can we drop volatile
3 messages
2015/08/05
[#70257] [Ruby trunk - Feature #11420] [Open] Introduce ID key table into MRI — ko1@...
Issue #11420 has been reported by Koichi Sasada.
11 messages
2015/08/06
[#70337] Re: [Ruby trunk - Feature #11420] [Open] Introduce ID key table into MRI
— Eric Wong <normalperson@...>
2015/08/11
Nice. Thank you guys for looking into this.
[#70349] Re: [Ruby trunk - Feature #11420] [Open] Introduce ID key table into MRI
— Eric Wong <normalperson@...>
2015/08/12
Btw, did you consider using flexible array to avoid extra malloc
[#70355] Re: [Ruby trunk - Feature #11420] [Open] Introduce ID key table into MRI
— Юрий Соколов <funny.falcon@...>
2015/08/12
I thought to suggest to embed hash_id_table directly into places when it is
[#70356] Re: [Ruby trunk - Feature #11420] [Open] Introduce ID key table into MRI
— SASADA Koichi <ko1@...>
2015/08/12
On 2015/08/13 4:29, Юрий Соколов wrote:
[#70358] Re: [Ruby trunk - Feature #11420] [Open] Introduce ID key table into MRI
— Eric Wong <normalperson@...>
2015/08/12
SASADA Koichi <ko1@atdot.net> wrote:
[#70509] [Ruby trunk - Misc #11276] [RFC] compile.c: convert to use ccan/list — ko1@...
Issue #11276 has been updated by Koichi Sasada.
3 messages
2015/08/21
[#70639] the undefined behavior of an iterator if it is modified inside of the block to which it yields — Daniel Doubrovkine <dblock@...>
(this is my first time e-mailing list list, so apologies for any misstep :)
4 messages
2015/08/31
[ruby-core:70545] [Ruby trunk - Feature #10600] [PATCH] Queue#close
From:
ko1@...
Date:
2015-08-22 09:09:52 UTC
List:
ruby-core #70545
Issue #10600 has been updated by Koichi Sasada.
Assignee set to Koichi Sasada
Thank you for your great survey. I want to introduce Queue#close in Ruby 2.3.
Just now I'm not sure it is okay to provide think Queue#close(token) API because there are no similar examples in Ruby.
The followings are summary of your survey.
| language | API | deq from empty queue after close? |
|----------|----------|-------------------------------------------------------------------|
| Java | N/A | |
| go | Close | Run-time panics (return error) https://golang.org/ref/spec#Close |
| C++ | close() | return queue_op_status::closed (element is returned by reference) |
| closure | close! | return nil |
| Ruby's similar operation | What happen after read from empty stream? |
|-------------------------------------|-------------------------------------------|
| File#read | return nil |
| File#read_nonblock() | raise EOFError |
| File#read_nonblock(exception: false)| return nil |
| File#gets | return nil |
Options:
1. Queue#close(token)
2. Queue#close() and raise on deq from empty closed Queue
3. Queue#close() and return nil from empty closed Queue (raise by deq(nonblock=true))
4. Queue#close(exc) -> (2) if exc is not nil, (3) if exc is nil
5. Queue#close(exception: true/false) -> (2) if exception is true (specific exception, such as ClosedQueueError < StopIteration), (3) if exception is false
6. Queue#close() and provide Queue#deq(exception: false)
(3) is similar to IO's gets/read/...
(6) is similar to IO's read_nonblock.
I think (1) is over-spec. (4) should be nice than (1). But I like (5) because it is more simple.
----------------------------------------
Feature #10600: [PATCH] Queue#close
https://bugs.ruby-lang.org/issues/10600#change-53952
* Author: John Anderson
* Status: Open
* Priority: Normal
* Assignee: Koichi Sasada
----------------------------------------
In a multiple-producer / multiple-consumer situation using blocking enq and deq, closing a queue cleanly is difficult. It's possible using a queue poison token, but unpleasant because either producers have to know how to match up number of poison tokens with number of consumers, or consumers have to keep putting the poison back into the queue which complicates testing for empty and not blocking on deq.
This patch (from trunk at b2a128f) implements Queue#close which will close the queue to producers, leaving consumers to deq the remaining items. Once the queue is both closed and empty, consumers will not block. When an empty queue is closed, all consumers blocking on deq will be woken up and given nil.
With Queue#close, clean queue shutdown is simple:
~~~ ruby
queue = SizedQueue.new 1000
consumer_threads = lots_of.times.map do
Thread.new do
while item = queue.pop
do_work item
end
end
end
source = somewhat_async_enumerator
producer_threads = a_few.times.map do
Thread.new do
loop{queue << source.next}
end
end
producer_threads.each &:join
queue.close
consumer_threads.each &:join
~~~
---Files--------------------------------
queue-close.diff (5.18 KB)
queue-close-2.diff (10.2 KB)
patch-25f99aef.diff (25.2 KB)
queue_benchmark.rb (2.95 KB)
--
https://bugs.ruby-lang.org/