unicorn Ruby/Rack server user+dev discussion/patches/pulls/bugs/help
 help / color / Atom feed
* Re: [ruby-core:99184] [Ruby master Bug#17023] How to prevent String memory to be relocated in ruby-ffi
       [not found] ` <redmine.journal-86563.20200715200039.5550@ruby-lang.org>
@ 2020-07-15 23:35   ` Eric Wong
  2020-07-15 23:49     ` [ruby-core:99185] " Aaron Patterson
  0 siblings, 1 reply; 3+ messages in thread
From: Eric Wong @ 2020-07-15 23:35 UTC (permalink / raw)
  To: ruby-core; +Cc: unicorn-public

tenderlove@ruby-lang.org wrote:
> Right, that makes sense.  I really need to document this (and
> I apologize for not doing so already), but
> `rb_gc_register_address` will pin your objects.  When you know
> you're done with the reference, you can release it with
> `rb_gc_unregister_address`.  Of course if you don't call the
> unregister function, the reference will stay alive forever.

Btw, does rb_gc_register_mark_object pin?  A quick glance at
gc.c tells me it doesn't, and I'll need to revert commit
2a6cb76d5010cb763ef5a2c305728465d15eb7c9 in unicorn:
https://yhbt.net/unicorn-public/20181226050857.6413-1-e@80x24.org/

Anyways, it takes me too long to compile Ruby so I'm back to
running whatever my distro ships.  I haven't been able to
test GC.compact at all.

> https://bugs.ruby-lang.org/issues/17023#change-86563

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [ruby-core:99185] [Ruby master Bug#17023] How to prevent String memory to be relocated in ruby-ffi
  2020-07-15 23:35   ` [ruby-core:99184] [Ruby master Bug#17023] How to prevent String memory to be relocated in ruby-ffi Eric Wong
@ 2020-07-15 23:49     ` Aaron Patterson
  2020-07-16  0:04       ` Eric Wong
  0 siblings, 1 reply; 3+ messages in thread
From: Aaron Patterson @ 2020-07-15 23:49 UTC (permalink / raw)
  To: Ruby developers; +Cc: unicorn-public



> On Jul 15, 2020, at 4:35 PM, Eric Wong <normalperson@yhbt.net> wrote:
> 
> tenderlove@ruby-lang.org wrote:
>> Right, that makes sense.  I really need to document this (and
>> I apologize for not doing so already), but
>> `rb_gc_register_address` will pin your objects.  When you know
>> you're done with the reference, you can release it with
>> `rb_gc_unregister_address`.  Of course if you don't call the
>> unregister function, the reference will stay alive forever.
> 
> Btw, does rb_gc_register_mark_object pin?  A quick glance at
> gc.c tells me it doesn't, and I'll need to revert commit
> 2a6cb76d5010cb763ef5a2c305728465d15eb7c9 in unicorn:
> https://yhbt.net/unicorn-public/20181226050857.6413-1-e@80x24.org/

Yes, it does pin.  I’m not super proud of this code, but here is where objects passed to rb_gc_register_mark_object get pinned:

  https://github.com/ruby/ruby/blob/c2a6295ec04a191c689d22254ac1ad5d665e27ad/vm.c#L2307-L2320

I don’t know why the mark object array is an array of arrays (I assume so as not to waste space in the array buffer?).  Maybe this could be a more friendly data structure.

I created a pinned list in compile.c so that objects allocated and used at compile time don’t move (they become free to move once iseq assembly is finished). It seems that might be a more generally useful thing, but so far I’ve only seen two places that need this feature.



^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [ruby-core:99185] [Ruby master Bug#17023] How to prevent String memory to be relocated in ruby-ffi
  2020-07-15 23:49     ` [ruby-core:99185] " Aaron Patterson
@ 2020-07-16  0:04       ` Eric Wong
  0 siblings, 0 replies; 3+ messages in thread
From: Eric Wong @ 2020-07-16  0:04 UTC (permalink / raw)
  To: Aaron Patterson; +Cc: ruby-core, unicorn-public

Aaron Patterson <aaron.patterson@gmail.com> wrote:
> > On Jul 15, 2020, at 4:35 PM, Eric Wong <normalperson@yhbt.net> wrote:
> > Btw, does rb_gc_register_mark_object pin?  A quick glance at
> > gc.c tells me it doesn't, and I'll need to revert commit
> > 2a6cb76d5010cb763ef5a2c305728465d15eb7c9 in unicorn:
> > https://yhbt.net/unicorn-public/20181226050857.6413-1-e@80x24.org/
> 
> Yes, it does pin.  I’m not super proud of this code, but here
> is where objects passed to rb_gc_register_mark_object get
> pinned:
> 
>   https://github.com/ruby/ruby/blob/c2a6295ec04a191c689d22254ac1ad5d665e27ad/vm.c#L2307-L2320

Ah, missed that, thanks.  So all is well in unicorn (for now).

> I don’t know why the mark object array is an array of arrays
> (I assume so as not to waste space in the array buffer?).
> Maybe this could be a more friendly data structure.

It seems so, since rb_ary_push can double the internal buffer.
Anyways, ko1 made commit
a617afc643e17cea5c1960a60adcd5ffb25fdfe2 r46714
Maybe it can be flattened nowadays and rely on GC.compact
to reclaim unused space.

/me goes back to hacking in Perl 5...

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, back to index

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
     [not found] <redmine.issue-17023.20200710175402.5550@ruby-lang.org>
     [not found] ` <redmine.journal-86563.20200715200039.5550@ruby-lang.org>
2020-07-15 23:35   ` [ruby-core:99184] [Ruby master Bug#17023] How to prevent String memory to be relocated in ruby-ffi Eric Wong
2020-07-15 23:49     ` [ruby-core:99185] " Aaron Patterson
2020-07-16  0:04       ` Eric Wong

unicorn Ruby/Rack server user+dev discussion/patches/pulls/bugs/help

Archives are clonable:
	git clone --mirror https://yhbt.net/unicorn-public
	git clone --mirror http://ou63pmih66umazou.onion/unicorn-public

Example config snippet for mirrors

Newsgroups are available over NNTP:
	nntp://news.public-inbox.org/inbox.comp.lang.ruby.unicorn
	nntp://ou63pmih66umazou.onion/inbox.comp.lang.ruby.unicorn

 note: .onion URLs require Tor: https://www.torproject.org/

AGPL code for this site: git clone https://public-inbox.org/public-inbox.git