[infinispan-dev] Retrieval operations with the IGNORE_RETURN_VALUES flag

[Mircea Markus]

On 10 Jun 2013, at 10:33, Dan Berindei <dan.berindei at gmail.com> wrote:

> On Thu, Jun 6, 2013 at 9:08 PM, Ray Tsang <saturnism at gmail.com> wrote:
> On Jun 6, 2013, at 13:26, Mircea Markus <mmarkus at redhat.com> wrote:
> 
> >
> > On 4 Jun 2013, at 13:55, Dan Berindei <dan.berindei at gmail.com> wrote:
> >
> >>>> CacheLoaderInterceptor and DistributionInterceptor both honour the IGNORE_RETURN_VALUES flag for get commands, but I think it would be more useful if they ignored it - just like they ignore it for conditional commands.
> >>>>
> >>>> That would make it possible for users to only keep a reference to a cache.getAdvancedCache().withFlags(IGNORE_RETURN_VALUES) and use it for both read and write operations.
> >>>>
> >>>> What do you think?
> >>>
> >>> If I was to take the role of a colleague of the person who's written the Infinispan code, it'd be very confused to see a cache reference created with IGNORE_RETURN_VALUES being used for a get() operation… I can see myself thinking: "Why on earth do you call get with IGNORE_RETURN_VALUES?"
> >>
> >> Isn't Galder's point not to allow invoking get with IGNORE_RETURN_VALUES? As both of you pointed out, Get + IGNORE_RETURN_VALUES doesn't make any sense :-)
> >>
> >>
> >> You'd think conditional operations with IGNORE_RETURN_VALUES don't make sense either, yet we have a special case to handle those as if the flag wasn't present :)
> >
> > I guess you're referring to ISPN-3141?
> 
> Exactly. Does it make sense to call cache.withFlags(IGNORE_RETURN_VALUES).putIfAbsent(k, v)? What should it return?
It does and it will return null as for PutIfAbsent the IGNORE_RETURN_VALUES flag is honoured.
ISPN-3141 affects conditional remove and replace, both returning a *boolean* value, as in this case the cost of not returning the value(null) is the same with the cost of returning the correct value(boolean). 
> 
>  
> > Still I think Get + IGNORE_RETURN_VALUES doesn't make any sense :-)
> 
> +1. It definitely threw me off...
> 
> 
> Ok, maybe IGNORE_RETURN_VALUES wouldn't be the best flag name for what I had in mind... I was thinking of a scenario where the application needs to do both reads and writes, but for writes it never needs to know the previous value. In that scenario it would make sense to call something like 
> 
>     cache = cacheManager.getCache().getAdvancedCache().withFlags(IGNORE_RETURN_VALUES_ON_WRITES)
> 
> at the beginning and only ever use that reference in the application. I agree that using the existing IGNORE_RETURN_VALUES flag for that would be a bit misleading, though.
> 
> Should we change anything about the IGNORE_RETURN_VALUES, then? I guess it would be relatively simple to make it so that get() operations with the flag throw an exception
what about not even throw and exception but (as you suggested previously) ignore the flag itself. That would allow the valid use case you've mentioned to work nicely.
> and (optionally) put() operations always return null. Should I create an issue in JIRA for that?
+1

Cheers,
-- 
Mircea Markus
Infinispan lead (www.infinispan.org)