perl bless/overload performance problem on RHEL

[Nicholas Clark]

On Mon, Sep 01, 2008 at 07:52:51AM +0100, Dave Cross wrote:
> Elliot Moore wrote:
> > 
> > May interest anyone running stock perl on Red Hat 5, Centos 5, or Fedora 9
> > 
> > http://blog.vipul.net/2008/08/24/redhat-perl-what-a-tragedy/
> > 
> >  (https://bugzilla.redhat.com/show_bug.cgi?id=379791)
> 
> AIUI, the problem is with their builds of 5.8.8 - which is in current
> versions of RHEL and Centos. And was in Fedora 8.
> 
> Fedora 9 contains a build of 5.10.0 which doesn't seem to exhibit this
> problem.

No, the code is different. Full explanation here:

  http://use.perl.org/~nicholas/journal/37274

Wishful thinking fix suggestion here:

  http://use.perl.org/~nicholas/journal/37278

Slashdot covers it here:

  http://linux.slashdot.org/article.pl?sid=08/08/29/1423201



What RedHat did, and why - your guess is as good as mine. But anyone
commenting about the source, cause, or intent of code changes to the
Perl core itself that is in disagreement with what I typed in
http://use.perl.org/~nicholas/journal/37274 is wrong.

I wrote *all* the code in question. I should know.

(Both the code in the trunk (that became 5.10.0) and the code in the 5.8.x.
The code that is now in 5.10 is different, and is the conceptually correct
fix to the underlying problem. It can be different, because it involves an
incompatible change in the semantics of SV flag bits. It doesn't actually
involve changing the values of flag bits, so at first glance it doesn't look
binary incompatible. I don't have access to anything running Fedora 9 (or
Fedora anything else) but my understanding is that they have a very clean
build of 5.10.0, so it follows that it cannot have this problem.)

Nicholas Clark