harmony-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Aleksey Shipilev" <aleksey.shipi...@gmail.com>
Subject Re: [classlib][luni][performance] IdentityHashMap implementation
Date Tue, 22 Apr 2008 11:23:40 GMT
Tim,

I support your proposal and try to make distinction between
proof-of-concept and clean patches.

Performance improvements could be notorious work, requiring a lot of
resources. The bad thing is, they could give nothing after the work is
done, that's why we are doing thought experiments and proof-of-concept
patches before actually implementing something. Proof-of-concept may
also give a hint on what maximum improvement we can get and set the
expectations for clean patch.

So we are not looking for hiding the problems, rather we're exploring
what could it worth.

F.ex. the experiments on IdentityHashMap performance finally led to
clean, non-breaking, consistent implementation, ready to be committed
patch [1]. Can we focus on reviewing it?

Thanks,
Aleksey.

[1] https://issues.apache.org/jira/browse/HARMONY-5771

On Tue, Apr 22, 2008 at 12:09 PM, Tim Ellison <t.p.ellison@gmail.com> wrote:
> Aleksey Shipilev wrote:
>
> > Right, Tim. It was just the proof-of-concept "Does IdentityHashMap
> > tuning worth it?".
> >
>
>  I think it is unhelpful to mix actual, concrete suggestions for performance
> improvements, with proof-of-concept ideas that break the specification or
> security.
>
>  Please can people mark clearly in their mail whether they are suggesting
> the idea only as "a thought experiment", otherwise somebody may apply the
> change without properly understanding the consequences.
>
>  I realize the onus is on the committer to ensure it is safe but hiding such
> problems is, as I said, unhelpful.
>
>  Thanks,
>  Tim
>
>
>
>
>
> > On Mon, Apr 21, 2008 at 4:50 PM, Tim Ellison <t.p.ellison@gmail.com>
> wrote:
> >
> > > Sergey Salishev wrote:
> > >
> > >
> > > > It also should be noticed that replacing the IdentityHashMap with
> HashMap
> > > >
> > > in
> > >
> > > > Thread impl gives 4.5x boost on ThreadLocalBench.
> > > >
> > > >
> > >  But it would be wrong since it opens a security vulnerability.
> > >
> > >  See:
> > >  http://markmail.org/message/rwq3kif5n33wp65m
> > >  http://svn.apache.org/viewvc?view=rev&revision=477355
> > >
> > >
> > >  Regards,
> > >  Tim
> > >
> > >
> >
> >
>

Mime
View raw message