[Springframework-developer] Fw: Getting ready for 1.0 final

[<jue...@we...>]

Peter has a point here. Obviously it's easy to address the init-method =
issue by looking for a no-arg method via plain reflection. Note that we =
don't need BeanWrapper's "invoke" method within the framework anymore =
then: Thus, I suggest to drop it completely rather than rework it to =
support overloaded methods. Its current javadoc already discourages =
people to use it anyway, so I don't think that anyone will miss that =
method much. IMO, BeanWrapper is not about invoking arbitrary methods in =
the first place...
=20
Juergen
=20

________________________________

Von: j=FCrgen h=F6ller [werk3AT]
Gesendet: Do 18.03.2004 19:02
An: Peter den Haan
Betreff: Re: Getting ready for 1.0 final


Peter,
=20
I've actually changed the determination of the destroy method today, to =
make it work with "shutdown(boolean force)"-style methods (passing in =
true as argument in this case). A side effect of this change is that it =
looks for a no-arg method first now, then for one with a single boolean =
parameter.
=20
I guess I'll rewrite the init-method determination with plain reflection =
too, to always look for a no-arg method. I'm not fond of the "invoke" =
method in the BeanWrapper interface anyway; I might suggest to drop it =
completely. I'll move this to the list.
=20
Juergen=20
=20

________________________________

Von: Peter den Haan [mailto:pde...@ob...]
Gesendet: Do 18.03.2004 18:13
An: j=FCrgen h=F6ller [werk3AT]
Betreff: Re: Getting ready for 1.0 final


Juergen,
=20
I was intending to write a test case and fix for this, but it looks like =
I may not come around to it in time. I found a limitation in =
BeanWrapperImpl that leads to what is arguably a bug in init-method and =
destroy-method.
=20
BeanWrapperImpl.invoke() looks up the method to invoke by method name, =
not by method signature. In a class with overloaded methods, the =
behaviour of the code (in CachedIntrospectionResults) is essentially =
undefined.
=20
This has repercussions in a couple of places, among others the =
init-method and destroy-method bean tag attributes. If a user specifies =
destroy-method=3D"close", it is clear that the intention is for the =
no-arg form of this method to be called even if there are other, =
overloaded versions. This doesn't work at present.
=20
The solution I wanted to implement is to modify =
CachedIntrospectionResults to key its method information on method =
signature rather than just method name (the current =
getMethodDescriptor(String methodName) can remain supported for =
backwards compatibility, although I would want to better define its =
semantics). A new getMethodDescriptor(String methodName, Object[] args) =
would implement logic to find the best matching method given the =
argument count and types, using the algorithm specified in JLS =
15.12.2.2.
=20
I can have a stab at implementing this over the weekend, or enter it in =
Jira as a bug (or enhancement). Please let me know what you prefer.
=20
Regards,
=20
Peter den Haan
Principal Systems Engineer
OBJECTIVITY
Tel:      024 7642 0000=20
Fax:     024 7642 0001=20
Email:  pde...@ob...
           www.objectivity.co.uk <http://www.objectivity.co.uk/> =20

The contents of this email and any attachments are strictly confidential =
and intended solely for the attention of the named addressee(s).  If you =
are not the addressee(s) you are not authorised to, and must not =
disclose, copy, distribute, or retain this message without prior written =
authority. This footnote also confirms that this message has been swept =
for the presence of computer viruses but this does not guarantee that it =
is virus free.

=20

________________________________________________________________________
This e-mail has been scanned for all viruses by Star Internet. The
service is powered by MessageLabs. For more information on a proactive
anti-virus service working around the clock, around the globe, visit:
http://www.star.net.uk
________________________________________________________________________