On 2/2/07, Jean-Sebastien Delfino <[EMAIL PROTECTED]> wrote:
Simon Laws wrote: > On 2/1/07, Jean-Sebastien Delfino <[EMAIL PROTECTED]> wrote: >> >> Simon Laws wrote: >> > I notice in implementing the PHP extension (yes - believe it or not >> I'm >> > nearly ready to make a patch for the next version ;-) that, given the >> way >> > that I have implemented the PHP extension there is insufficient >> > information >> > available in the SCA runtime in order to do correct type conversions >> when >> > passing messages between components. I imagine this has been raised >> > before >> > but I looked at the archive and couldn't find a relevant thread. >> > >> > Imagine the scenario: >> > >> > C++ Component (ComponentA) ---WireA---> PHP Component (ComponentB) >> > ---WireB----> C++ Component (ComponentC) >> > >> > Currently the build process looks at the header files described in the >> > component type files and generates wrappers and proxies for the C++ >> > components. I have currently implemented the PHP Extension to use >> generic >> > wrappers and proxies, i.e. it doesn't use those generated based on >> the >> > interface descriptions, so they need to dynamically manage the type >> > conversions for data coming in and going out of a PHP component. >> > >> > WireA. >> > >> > This is OK because the C++ SCA operation object that appears at >> > Component B >> > has a set of data/types based on the generated proxy. The PHP >> > extension can >> > look at this and effect the right type conversions. >> > >> > WireB >> > >> > This is problematic. The dynamic PHP proxy has to generate an >> operation >> > object to pass to the the wrapper of ComponentC. The issue is that >> > there is, >> > as far as I can tell, no dynamic way of getting at the list of types >> that >> > are expected for any particular operation. There is of course a static >> > C++ >> > proxy/wrapper combination that has been generated but I can't inspect >> > it at >> > runtime. >> > >> > I'm not keen on generating PHP specific interface classes. PHP is a >> > dynamic >> > environment and it's a whole stack of extra files we could do without >> > having >> > to manage particularly if we have to adapt the generator for every >> > extension >> > that's constructed. Can we extend the wrapper/proxy mechanism to >> > encapsulate >> > a runtime list of required types alongside the static method >> descriptions >> > that are already generated? We would need this to work for script to >> > script >> > calls as well as for the script/C++ combination so maybe we need >> > something >> > that hangs off the interface description part of the model. I'm not >> that >> > familiar with how that part of the model is used so a little >> > investigation >> > is required. >> > >> > Thoughts? >> > >> > Simon >> > >> >> Simon, >> >> It's an interesting issue. To explore it let's walk through the wiring >> scenario you describe and assume the following: >> - ComponentA (C++) -> WireA -> ComponentB (PHP) -> WireB -> ComponentC >> (C++) >> - ComponentA (C++) passes a short int to ComponentB (PHP) >> - ComponentB executes a PHP script which in turn passes a number to >> ComponentC (C++) >> - ComponentC expects that number to be given as a long int. >> >> Here's what I think should happen in the runtime: >> 1. At the source of WireA, a generated C++ CPPServiceProxy adds to our >> Operation object a Parameter of a type decided by the C++ client code: a >> C++ short int, with type == ParameterType::SHORT. >> 2. At the end of WireA, a PHPServiceWrapper converts that parameter to >> what the PHP script expects, for the sake of simplicity now I am going >> to assume that it needs to convert it to a C++ std::string. >> 3. The PHP script executes, now passes an std::string containing a >> number to the PHPServiceProxy at the source of WireB. >> 4. The PHPServiceProxy does not have much type info about that >> std::string parameter and can only add it to the Operation object as a >> std::string with type == ParameterType::STRING. >> 5. The CPPServiceWrapper at the end of WireB (actually the C++ >> ServiceWrapper generated for ComponentC) receives the std::string and >> should convert it to what ComponentC expects: a long int. >> >> The general idea is that a ServiceProxy sends what it is given (or picks >> the most natural type out of the ParameterTypes that we have defined and >> converts the data to it). A ServiceWrapper converts what it receives to >> the type expected by the code it wraps. I think that the >> CPPServiceWrapper code and the generated C++ ServiceWrappers are simply >> missing the logic to convert data to what the target expects. >> >> At the moment this limitation also prevents a C++ method >> getCustomer(long customerID) to be exposed as a REST service for >> example, as the generated C++ ServiceWrapper is missing the logic to >> convert the customerID received in string form from the REST query >> string to the expected C++ long int. >> >> So we just need to add the missing type conversion logic to the C++ >> ServiceWrappers :) >> >> Thoughts? >> >> -- >> Jean-Sebastien >> >> >> --------------------------------------------------------------------- >> To unsubscribe, e-mail: [EMAIL PROTECTED] >> For additional commands, e-mail: [EMAIL PROTECTED] >> >> Hi Jean-Sebastien > > It could well be simpler if, as you suggest, the conversion happens in > the > generated wrapper in this case. So we need such type information, as is > required to effect the type conversion, to be generated into the wrapper. > Yes, and I think that there are three cases to consider here: - The wrapper is generated and the type info can be burned into it (from the C++ class/interface of the service for which the wrapper was generated). - The wrapper is generic and introspects the component implementation (incl. the associated .componentType metadata and the service interfaces it declares, if any) to determine the expected data type. - The wrapper wraps a script written in a language that supports dynamic typing (e.g. Ruby) and best is to not convert and present the data to the script as-is (e.g. a string, an integer). Independent of the type handling, the wrapper is responsible for presenting the data to the target component implementation in a suitable form, for example with Ruby, this means turn the input data into a Ruby VALUE object. Again this is independent of the type, as a Ruby VALUE can represent a int, a string etc. > However I think we should try and arrange it so that we are not trying > to do > any more conversions than are necessary so I'm not convinced that the > proxy > should be guessing what type to produce only for it to be coerced > again at > the wrapper unless this is a natural effect of the transport/protocol > involved, for example, In your REST example. I.e. we want the proxy to > have > the best information available to it in marshaling the data it has. > I agree. I was not clear before and by conversions in the proxy I meant "turn the data into a form that can be carried by our Operation + Operation::Parameter infrastructure". In the case of a C++ proxy, no conversion should be necessary. In the case of Ruby the string, int etc. data needs to be extracted of the Ruby VALUE object that wraps it. I'm not sure about PHP but you probably have a similar mechanism to carry data and track its usage so you may have to extract the raw data out of its PHP representation. So in most cases there is no conversion at all in the proxy or as you said just a natural effect of the transport/protocol involved (for example with REST we are receiving a query string and producing XML for the response). > I also need to address complex types in the PHP extension. Where a remote > interface is used you can imagine the WSDL, SMD or whatever being > parsed at > the client component to construct a proxy capable of suitable type > checking > and of providing convenience functions such as > "giveMeAnSDOWithTheRightModelLoaded()". Looking at the code and > samples at > the moment it seems that the C++ infrastructure loads available XSDs and > provides some handy helper methods for creating SDOs based on the types > available. I don't how they are related to particular interfaces > though. I > expect I'm overlooking something. > We also load WSDL files. A Proxy is created from a Reference, and the Reference can carry an Interface definition, a WSDL portType for example. Composite::findWSDLDefinition(namespace) will give you a WSDLDefinition object from which you can lookup a WSDLPortType and WSDLOperation and the operation input and output types. You should be able to use these types to validate the data flowing through the proxy. > Regards > > Simon > -- Jean-Sebastien --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] OK Jean-Sebastien, I agree with what you have said. So the long and short
of all this is that 1/ the C++ generated wrapper has to be a bit smarter about the type conversions it makes, i.e. it can't assume the values arriving in the Operation object are of the correct type 2/ proxy code should only take responsibility of populating the Operation object and for ensuring there is no loss of data/precision in doing this 3/ Look at how we provided WSDL defined type information to script developers. Currently in PHP SCA you can define references using annotations and include WSDL information there so there is a natural bridge to finding the correct type information for a given reference. We haven't however done the piece of work to share meta data between script and runtime so there is work to do here! Simon
