[Mono-dev] Problems with System.ServiceModel.Web

Atsushi Eno atsushieno at veritas-vos-liberabit.com
Fri Oct 15 06:46:27 EDT 2010


  Hello,

OK, I have checked in your patch in almost identical form in git master.
I found that GET operation does not pass on .NET when
GetRequestClientFormatter() is invoked, so I split it into another contract
and marked the test as NotWorking (it haven't passed earlier anyways).

Thanks a lot for the patch :)

Atsushi Eno

On 2010/10/13 23:34, Frank Wilhelm wrote:
>  Hello,
>
> thank you for your review. I tried to use the style of the existing 
> code but I guess I missed some spaces.
>
> The problem with your modifications is that it defeats the purpose of 
> my patch. In .NET I can host a method with a definition like:
>
> [OperationContract]
> [WebInvoke(UriTemplate="set?name={name}&value={value}")]
> void Set(string name, string value);
>
> In Mono I couldn't. This is why I created the patch. With your 
> modifications the problem still exists, but now in a different place.
>
> ValidateStyleForMessageFormatter shouldn't reject a method like the 
> one above. Like the modified Validate method it should consider the 
> UriTemplate. In the method above there is no need to wrap because 
> everything is passed in the URI. That works for .NET and also works 
> for Mono if the validation is less restrictive. I think the major 
> problem with my patch is that it doesn't work if the UriTemplate is 
> null. With only that fixed the test RejectTwoParametersWhenNotWrapped 
> will not break and it also works for my purpose.
>
> I also checked the .NET behavior for WrappedRequest and 
> WrappedResponse. In .NET the BodyStyle WrappedRequest resolves 
> problems with multiple inputs and WrappedResponse with multiple 
> outputs. That is what I expected and it is the opposite to what is 
> done in ValidateStyleForMessageFormatter.
>
> So I created a new version that hopefully resolves all issues.
>
> On 12.10.2010 13:30, Atsushi Eno wrote:
>>  Hello,
>>
>> Thanks for the patch, I very much appreciate it. Though there were 
>> some problems in your patch.
>>
>>     - your implementation had an assumptiopn that WebGetAttribute and
>>       WebInvokeAttribute has non-null UriTemplate. They can indeed be 
>> null.
>>     - Your change broke RejectTwoParametersWhenNotWrapped() in
>>       WebInvokeAttributeTest. It is because, in your 
>> WebHttpBehavior.Validate()
>>       change you just removed existing code that checked what this 
>> test exactly
>>       verifies i.e. invalid multiple arguments. You might think it is 
>> "very strange"
>>       and "wrong" but that's what .NET indeed does (looks like it is 
>> done in
>>       GetClientMessageFormatter() in current version of .NET though).
>>     - There was a couple of coding style fixage needed. You had right 
>> and wrong
>>       style, so I assume you might have already known the style, but 
>> in case you
>>       haven't, have a look at http://mono-project.com/Coding_Guidelines
>>       (I found this page was not linked from our "Contributing" page, 
>> which was
>>       our bad, just fixed it.)
>>
>> I'm attaching the modified fix here. It includes some additional 
>> tests. I'll commit the changes unless you have further comments.
>>
>> (As a cosmetic excuse, WebMessageBodyStyle.Bare was new in 3.5 SP1 
>> AFAIR, so it was left ignorant in our implementation from 3.5 era.)
>>
>> Atsushi Eno
>>
>> On 2010/10/09 18:27, Frank Wilhelm wrote:
>>>  Hello Mono devs,
>>>
>>> I tried running my web service on Mono and ran into several issues. 
>>> I use the WebHttpBinding to create REST web services by hosting that 
>>> with the ServiceHost class. This works fine on .NET but the Mono 
>>> implementation of System.ServiceModel.Web shows very different, and 
>>> very limiting, behavior. So I decided instead of waiting for a fix I 
>>> try to track down the issues. Here is the first I discovered.
>>>
>>> The Validation of WebHttpBehavior is very limiting. If a POST 
>>> operation has several parameters it will only host them if you 
>>> define 'wrapped' body format. But this isn't required if the 
>>> parameters are passed in the UriTemplate. Also the requirements for 
>>> WrappedRequest and WrappedResponse look very strange. I have to wrap 
>>> my response if I have multiple input parameters? That looks wrong.
>>>
>>> I looked up in the MSDN what the Validate method does for .NET, and 
>>> then I tried to make the .NET implementation fail. It just doesn't 
>>> behave like specified in the documentation, it never fails no matter 
>>> what.
>>>
>>> I attach a patch that will make the Validate method less picky. It 
>>> will look for placeholders in the URI template and only after that 
>>> decides whether wrapping is needed or not. That is still more 
>>> restrictive than the .NET implementation but for me it looks like 
>>> the right thing to do. I also added a test that reproduces the problem.
>>>
>>> Please accept the patch and give me some feedback on what I can 
>>> improve for further contributions.
>>>
>>>
>>> _______________________________________________
>>> Mono-devel-list mailing list
>>> Mono-devel-list at lists.ximian.com
>>> http://lists.ximian.com/mailman/listinfo/mono-devel-list



More information about the Mono-devel-list mailing list