Hi Stevo,
Stevo Slavić wrote:
> Hello castor users,
>
> When generating source code from multiple XSDs using castor-maven-plugin
> v1.5, one may experience random failure. Process itself will not fail, but
> its outputs are not generated well. Problem occurs if one configures plugin
> using packaging, schemaDirectory, without providing binding file, and when
> generateImportedSchemas is set to false, in case of multiple XSDs being
> processed referencing each other. In this case source code generation
> success will be dependent on order the XSDs are processed. E.g. if one xsd
> (1) references the other (2) and 1 becomes processed before 2, types
> generated from 1st XSD with attributes of type generated from 2nd XSD might
> not compile, because attributes will not have correct package reference
> (only types.%TypeName% without packaging prefix) - likely a bug in Castor
> CodeGen module.
Not really a bug, but in my view something that needs to be driven by a
human being, as requirements seem to differ for everybody. But then
again, this is me being biased here.
>
> In my case generateImportedSchemas has to be false, since some imported
> schemas don't have to be processed, only ones from a given directory.
> Correct me if I'm wrong, but I think currently one can not influence order
> XSDs will be processed. After checking the castor-maven-plugin source it
> seems schemas will be processed in order that is more dependent on alignment
> of stars than anything else (java.io.File.listFiles(), java.util.Set, ... no
> guarantees of order). Same goes for the processing success.
Yes, correct.
> Solution was to use binding.xml file and configure XML namespace to package
> name mapping. Then it seems, regardless of the order XSDs are processed,
> source code generator knows to resolve package for given XSD
> type/group/element/... based on its namespace and provided mapping.
Yes, that's where much what we advise users to do. If you run into
problems during code generation, use one of two options:
a) set generateImportedSchemas to 'true', or
b) use a binding file.
> Maybe castor-maven-plugin could be improved, to first analyze all the XSDs
> before they are processed, check who imports whom, and based on that order
> the XSDs for processing, so do something similar maven reactor does at the
> start of the build by ordering modules to be built.
Yes, definitely. That would make for some nice goals. As a first step,
one could provide a new goal that behaves similar to
castor:analyze-schema-set ....
>
> Regards,
> Stevo.
>
---------------------------------------------------------------------
To unsubscribe from this list, please visit:
http://xircles.codehaus.org/manage_email
