OASIS Mailing List ArchivesView the OASIS mailing list archive below
or browse/search using MarkMail.

 


Help: OASIS Mailing Lists Help | MarkMail Help

sca-j message

[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]


Subject: Re: [sca-j] NEW ISSUE: @Callback injection could be NULL


Jim Marino wrote:
> 
> On Feb 13, 2009, at 3:34 PM, Simon Nash wrote:
> 
>> David Booz wrote:
>>> I'm going to try to state this again without using the word injection 
>>> because it seems to have too many connotations.....and the email 
>>> thread is wandering all over the place.
>>> The point I'm worried about is a setter method that is annotated with 
>>> @callback (or even a 0..1 @reference as pointed out earlier).
>>> I'm concerned that we provide a consistent model to component 
>>> developers, and therefore the runtime MUST NOT skip invocation of the 
>>> setter method. It may be very important for the logic in the setter 
>>> method to understand that there is no reference target or callback. 
>>> It's the 'skipping of the setter' that is the problem.
>> I don't think we need to drive all these setters with null arguments
>> just in case they have some important initialization to do.  My reason
>> for saying this is that we also have the @Init method, which is called
>> after all the setters have (or have not) run.
>>
>> So the consistent model is that all setters with something to set
>> would run, and then the @Init method would run.  This combination
>> allows the implementation code to handle all possible cases.
>> For example, the setters could each set a flag to say "OK, done my
>> thing", and the @Init method could check these flags to find out
>> if further processing is needed because some setter did not run.
>>
> FWIW, I prefer the approach of using the @Init method. A strong argument 
> can be made that setters should not be performing initialization logic - 
> a setter should just set a value and @Init should be used for 
> initialization Sticking to this rule has the nice effect of keeping 
> setters simple and clean (i.e. no need for null checks) and 
> consolidating initialization in one place where it arguably should be. 
> Also, from an implementation perspective, less invocations to set items 
> increases performance. Of course the exception is CDI in the case where 
> a reference is optional, but that is probably a very specific use-case.
> 
For optional references and properties, CDI allows business logic
to be run with NULL passed in.  Currently, this isn't allowed for
callbacks.... but why not?  The only standard scope that allows callback
injection is stateless, because we have prohibited callback injection
for composite-scoped components.  For callback injection into stateless
scoped components, there's no reason to prohibit using constructor
parameters for the callbacks.

If we were to make CDI for callbacks legal, it makes callbacks more
consistent with references and allows the implementation to provide
business logic that detects the case when a callback is missing.

   Simon

> 
>> I was looking to see what Spring does.  AFAICT it does not drive
>> dependency injection setters when the dependency is missing (though
>> it does pass NULL for missing dependencies when calling constructors).
>> See this article:
>> http://blog.springsource.com/2007/07/11/setter-injection-versus-constructor-injection-and-the-use-of-required/ 
>>
>> The article also mentions the @Init-style approach that I described
>> above as one way to check for missing dependencies.
>>
>>  Simon
>>
>>> The spec text you point out does not clearly say what to do. The text 
>>> is clearly written assuming everything is wired up (I'm guilty of 
>>> thinking that way myself).
>>> Dave Booz
>>> STSM, BPM and SCA Architecture
>>> Co-Chair OASIS SCA-Policy TC and SCA-J TC
>>> "Distributed objects first, then world hunger"
>>> Poughkeepsie, NY (845)-435-6093 or 8-295-6093
>>> e-mail:booz@us.ibm.com
>>> Inactive hide details for Simon Nash ---02/13/2009 09:41:55 
>>> AM---David Booz wrote: > I'm glad we're doing this in email, and 
>>> noSimon Nash ---02/13/2009 09:41:55 AM---David Booz wrote: > I'm glad 
>>> we're doing this in email, and not on the call. We need to do
>>> From:   
>>> Simon Nash <oasis@cjnash.com>
>>> To:   
>>> sca-j@lists.oasis-open.org
>>> Date:   
>>> 02/13/2009 09:41 AM
>>> Subject:   
>>> Re: [sca-j] NEW ISSUE: @Callback injection could be NULL
>>> ------------------------------------------------------------------------
>>> David Booz wrote:
>>> > I'm glad we're doing this in email, and not on the call. We need to do
>>> > more of this.
>>> >
>>> > We need to avoid uninitialized fields and so am asserting that NULL
>>> > would in fact be injected, ensuring that a test for NULL in the app 
>>> was
>>> > always going to give the correct answer. This makes the two cases
>>> > parallel (which they should be). It doesn't do the app any good to 
>>> leave
>>> > uninitialized fields laying around.
>>> >
>>> > Good catch on the "should".
>>> >
>>> > I knew about the 76 overlap in the example and chose to ignore it 
>>> given
>>> > that it hasn't been accepted/applied yet. In any case, I am fine if we
>>> > omit the example updates from the proposal.
>>> >
>>> > This then changes the proposal to:
>>> >
>>> > In section 6.2.4, at the end of the first paragraph (line 395 [1] ) 
>>> add
>>> > the following:
>>> >
>>> > A field or setter method annotated with @callback is injected or 
>>> set to
>>> > NULL if the implementation is invoked through a non-bidirectional
>>> > service or when invoked as a callback. It is RECOMMENDED that the
>>> > component implementation checks for null before using a callback
>>> > reference in this situation (See also the getCallbackReference() 
>>> method
>>> > in section 8.2 "RequestContext").
>>> >
>>> Apart from the question of whether or not NULL gets injected, I don't
>>> think we should make the component implementation logic into an RFC2119
>>> conformance target.  I'd like to proposed alternative wording that
>>> addresses these two points:
>>> A field or setter method annotated with @Callback will not be injected
>>> if the implementation is invoked through a non-bidirectional service or
>>> when invoked as a callback. To allow for this, the component 
>>> implementation
>>> can check for null before using a callback reference in this situation
>>> (See also the getCallbackReference() method in section 8.2 
>>> "RequestContext").
>>>  Simon
>>> >
>>> > Dave Booz
>>> > STSM, BPM and SCA Architecture
>>> > Co-Chair OASIS SCA-Policy TC and SCA-J TC
>>> > "Distributed objects first, then world hunger"
>>> > Poughkeepsie, NY (845)-435-6093 or 8-295-6093
>>> > e-mail:booz@us.ibm.com
>>> >
>>> > Inactive hide details for Simon Nash ---02/10/2009 11:18:56 AM---David
>>> > Booz wrote: > TARGET: Java CAA, section 6.2.4 "AccessingSimon Nash
>>> > ---02/10/2009 11:18:56 AM---David Booz wrote: > TARGET: Java CAA,
>>> > section 6.2.4 "Accessing Callbacks" [1]
>>> >
>>> >
>>> > From:
>>> > Simon Nash <oasis@cjnash.com>
>>> >
>>> > To:
>>> > sca-j@lists.oasis-open.org
>>> >
>>> > Date:
>>> > 02/10/2009 11:18 AM
>>> >
>>> > Subject:
>>> > Re: [sca-j] NEW ISSUE: @Callback injection could be NULL
>>> >
>>> > 
>>> ------------------------------------------------------------------------
>>> >
>>> >
>>> >
>>> > David Booz wrote:
>>> >  > TARGET: Java CAA, section 6.2.4 "Accessing Callbacks" [1]
>>> >  >
>>> >  > DESCRIPTION: There are situations where the field or setter with an
>>> >  > @callback annotation (when used to obtain a reference to a 
>>> callaback
>>> >  > instance) can be injected with NULL. As noted in section 8.2
>>> >  > RequestContext, the getCallback() method might return NULL in 
>>> certain
>>> >  > situations, and those situations might also occur in combination 
>>> with
>>> >  > the use of @callback. For example, suppose a component 
>>> implements two
>>> >  > services, one has a bidirectional interface and the other does 
>>> not. The
>>> >  > component impl could use @callback to get a callback instance. When
>>> >  > invoked over the bidirectional service, the @callback will 
>>> non-null, but
>>> >  > will be null when the non-bidirectional service is invoked. 
>>> AFAIK, the
>>> >  > specs don't prohibit the example (nor do I think they should).
>>> >  >
>>> > This case isn't quite the same as RequestContext.getCallback().
>>> > In that case, the API can return NULL.  In this case, the callback
>>> > would not be injected, and so the field would be uninitialized.
>>> >
>>> >  > PROPOSAL:
>>> >  > There is no pdf for [1] so bear with me:
>>> >  > In section 6.2.4, at the end of the first paragraph (line 395) 
>>> add the
>>> >  > following:
>>> >  >
>>> >  > A field or setter method annotated with @callback might be 
>>> injected or
>>> >  > set to NULL if the implementation is invoked through a 
>>> non-bidirectional
>>> >  > service or when invoked as a callback.
>>> >  >
>>> > As I said above, it's not correct to say "...might be injected or
>>> > set to NULL...".  It's really "...might not be injected...".
>>> >
>>> >  >                                        The component implementation
>>> >  > should always check for null before using a callback reference 
>>> (See also
>>> >  >
>>> > You've got a non-RFC2119 "should" here.  Even if we found another
>>> > word, I think it's going too far to say that this should *always*
>>> > be done.  IMO it should *only* be done if the component logic
>>> > allows this case to arise, which I believe will be a minority of
>>> > all the cases where callbacks are used.
>>> >
>>> >  > the getCallbackReference() method in section 8.2 "RequestContext").
>>> >  >
>>> >  > In section 6.2.4, the first example (line 400) replace the 
>>> example with
>>> >  > the following:
>>> >  >
>>> >  > @Callback
>>> >  > protected ServiceReference<MyCallback> callback;
>>> >  >
>>> >  >       *public void *someMethod() {
>>> >  >
>>> >  >       *if* (callback) {
>>> >  >       MyCallback myCallback = callback.getCallback(); …
>>> >  >
>>> >  >       myCallback.receiveResult(theResult);
>>> >  >       }
>>> >  >       }
>>> >  >
>>> > These changes overlap with the proposed changes for JAVA-76,
>>> > and they carry forward one of the problems that is the subject
>>> > of JAVA-76 (incorrectly using getCallback() where getService()
>>> > should be used).  Also, bearing in mind my comment above, I'm
>>> > not convinced that this example needs to show the "if".  I think
>>> > it's enough to describe the possibility in added text.
>>> >
>>> >   Simon
>>> >  >
>>> >  >
>>> >  > [1]
>>> >  >
>>> > 
>>> http://www.oasis-open.org/committees/download.php/31121/sca-javacaa-1.1-spec-cd02-rev2.doc 
>>>
>>> >  >
>>> >  > Dave Booz
>>> >  > STSM, BPM and SCA Architecture
>>> >  > Co-Chair OASIS SCA-Policy TC and SCA-J TC
>>> >  > "Distributed objects first, then world hunger"
>>> >  > Poughkeepsie, NY (845)-435-6093 or 8-295-6093
>>> >  > e-mail:booz@us.ibm.com
>>> >
>>> >
>>> >
>>> > ---------------------------------------------------------------------
>>> > To unsubscribe from this mail list, you must leave the OASIS TC that
>>> > generates this mail.  Follow this link to all your TCs in OASIS at:
>>> > 
>>> https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php  
>>> >
>>> >
>>> >
>>> ---------------------------------------------------------------------
>>> To unsubscribe from this mail list, you must leave the OASIS TC that
>>> generates this mail.  Follow this link to all your TCs in OASIS at:
>>> https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php
>>
>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe from this mail list, you must leave the OASIS TC that
>> generates this mail.  Follow this link to all your TCs in OASIS at:
>> https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php
> 
> 
> ---------------------------------------------------------------------
> To unsubscribe from this mail list, you must leave the OASIS TC that
> generates this mail.  Follow this link to all your TCs in OASIS at:
> https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php
> 
> 




[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]