river-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Mark Brouwer <mark.brou...@cheiron.org>
Subject Re: Order in which unicast discovery providers are consulted
Date Wed, 28 Mar 2007 07:32:04 GMT
Bob Scheifler wrote:
>> That order seems to be determined mainly by
>> the way the service providers at the client side are found through
>> com.sun.jini.resource.Service, which boils down to the semantics for a
>> TreeSet.
>>
>> My question is whether there is a strong rationale behind the TreeSet,
> 
> com.sun.jini.resource.Service is essentially a clone of sun.misc.Service
> from the JDK, and we retained its implementation search order, apparently
> not recognizing that it doesn't match what we have spec'd in various
> places as an ordered list to be maintained.

Hi Bob,

I did a quick scan (again) but I couldn't find any language from which I
could conclude I was dealing with a bug, I saw some implementation code
that gave me a hunch but then again
com.sun.jini.resource.Service.providers explicitly mentions that the
iterator returns the providers in an arbitrary order.

>                                             As of JDK 6, a public version
> of this facility has been exposed as java.util.ServiceLoader, which does
> preserve order.  Another bug to be fixed.

I'll file a bug against River and fix it in my code.
-- 
Mark


Mime
View raw message