On 8/10/07, Catalin Hritcu <catalin.hritcu@gmail.com> wrote:

Hi and sorry for the delayed response,

On 7/24/07, Asiri Rathnayake <asiri.rathnayake@gmail.com> wrote:
> Hi Vincent,
>
> On 7/24/07, Vincent Massol < vincent@massol.net> wrote:
> >
> > Hi Asiri/Tharindu,
> >
> >
> >
> > On Jul 24, 2007, at 4:33 PM, Asiri Rathnayake wrote:
> >
> >
> >
> >
> > On 7/24/07, Vincent Massol <vincent@massol.net> wrote:
> > >
> > > Hi,
> > >
> > >
> > > If our goal is to implement the Confluence XMLRPC interface (
> http://confluence.atlassian.com/display/DOC/Remote+API+Specification )
> then it seems to me we shouldn't throw XWikiException in any method in our
> ConfluenceRpcInterface interface. Several methods instead should return a
> boolean to indicate success or failure.
> > >
> > >
> > > Does anyone know why we're throwing them?
> >
> >
> > I believe the reason is that we should be able to indicate to the user
> about what went wrong on the server. This is very useful since there may be
> situations where the user needs some help with. For an example if the user
> enters a wrong password when logging in, the exception says that something
> is wrong with logging details (or so it should). Similar situations apply
> for other operations as well. But having said that, the current
> implementation does not provide much help with error messages, we need to
> improve it.
> >
> >
> > >
> > >
> > > Are we ok to remove all exception throwing?
> >
> >
> > I don't think this is a good idea. But we'll wait for other opinions.
> >
> >
> >
> > Thanks for your POV but I think we're not talking about the same thing at
> all.... :)
> >
> >
> > I know exceptions are useful in general and I agree with that of course...
> I said "IF our goal is to implement the Confluence XMLRPC interface". I
> think this is our goal. Why are we doing this? For one reason I can see:
> that anyone who has code that interacts with Confluence can use that code to
> interact with XWiki. So 2 advantages:
> > - we don't create yet a new API
> > - confluence tools can work seamlessly with XWiki
>
> aha... Now I get it. I think we need to verify whether actual Confluence
> implementation throws exceptions or not (i beleive they don't since the API
> doesn't indicate of such a thing). If that is so, it's better we remove the
> Exceptions (so we have more integration). But shouldn't there be any
> mechanism to report server errors and such (i'm still scratching my head) ?
>
I checked and confluence does throw exceptions for all erroneous
situations. Also all methods in swizzle-confluence (a client-side
proxy for the confluence remote api) have a "throws Exception" clause.
This means it makes a lot of sense to always throw an exception when
something goes topsy-turvy on the server (now the behavior depends on
mood of the writer of each method, and we don't check for many error
conditions).

I think we should also throw the generic java.lang.Exception type exceptions along with some error code and a meaningful error message and avoid XWiki specific exceptions (i think it's obvious).

For exceptions to be sent back to the client we need to enable them on
the server using the "enabledForExtensions" parameter (currently it's
turned off).

From http://ws.apache.org/xmlrpc/advanced.html:
"It is recommended to use the property enabledForExceptions. If this
property is set, then the server will attempt to convert the exception
into a byte array, which is transmitted to the client. The client will
be able to convert the byte array back into an exception and throw
that, as if it came right out of the client. Note, that this approach
may cause security and privacy issues, because the serialized
exception may, in theory, contain arbitrary objects."

> >
> > Of course that'll work well only if we implement the real confluence
> API... and that confluence API doesn't throw any exception AFAIK.
> >
> >
> > BTW it's funny that you want to keep the exceptions since it's because of
> your patch that I'm proposing to remove them... ;) You created a patch that
> returns a boolean in addition to throwing an exception. This is obviously
> wrong as they serve the same purpose.
>
> Yes, but there the problem was, when the return type is specified as void in
> the RPC, the request doesn't even reach the implementation (i couldn't find
> out why). I changed it to boolean as a workaround, but still it's not the
> right thing.
>
This is true. The confluence API specifies boolean return types for
some methods just because of bugs in some xml-rpc implementations.
For example:

boolean logout(String token) - remove this token from the list of
logged in tokens. Returns true if the user was logged out, false if
they were not logged in in the first place (we don't really need this
return, but void seems to kill XML-RPC for me)

Anyway, we should stick to what the API specification says, and
sometimes that means returning meaningless booleans.

Voila...

> >
> > In conclusion we need to choose:
> > 1) either we want to implement the Confluence API and we should drop the
> exceptions
>
> Should we do something about void return types then ? The API has few
> methods that return void, but we have a problem with void return types in
> our implementation, think we need to find out why void return types cause a
> problem.
>
[snip]

Please name the methods which caused you trouble and I'll look into it.

Well, everything mentioned in XMLRPC API with a void return type will cause a problem. But I think i have replaced the only one such method (delete page, i think) currently implemented in XWiki. Anyway, as we go on implementing more of the API, we'll need to have a boolean (or something suitable) as the return type for those conflicting methods in API.

Thanks a lot.

- Asiri

Regards,
Catalin

--
You receive this message as a subscriber of the xwiki-dev@objectweb.org mailing list.
To unsubscribe: mailto:xwiki-dev-unsubscribe@objectweb.org
For general help: mailto: sympa@objectweb.org?subject=help
ObjectWeb mailing lists service home page: http://www.objectweb.org/wws