Somehow the previous custom deleter for xercesc::DOMDocument fails invokable assertions planet a compiler update. A lambda function with the same signature wors for some reason…
* i.e. InputXSLT now requires a C++14 supporting compiler / standard library implementation * this was done while enabling the StreamInputSource to handle all kinds of streams ** this in turn is required to enable e.g. stdin as transformation source while preserving the correct filesystem context
* previous createDocument method used default name "content" ** this led to problems when using xalan::XalanNode as transformation input ** the parameter-less createDocument overload now creates a unnamed xercesc::DOMDocument * changed external function implementations accordingly
* the common use case of ResultNodeFacade inside external function implementations is managing the result root node ** this root node can be fetched through the xercesc::DOMDocument instance ** the new alias overload automatically fetches the root node pointer and as such doesn't require a root node parameter * changed external function implementations accordingly
* pointers to xercesc::DOMDocument were manually released ** this is now solved using a custom deleter for the appropriate std::unqiue_ptr template specialization * added matching factory method to DomDocumentCache * updated external function implementations accordingly ** "constructDocument" is now expected to return a DomDocumentCache::document_ptr instance * updated TransformerFacade accordingly * this change was implemented to get rid of the manual memory management required by xalan / xerces
* using xalan::FormatterToXercesDOM internally ** this required changes to the TransformationFacade::generate member method overloads * TransformationFacade::generate now accepts references to xalan::FormatterListener instances ** "generate(std::basic_ostream<char>&..." instantiates a xalan::FormatterToXML and passes it to the actual generate member method * changed ResultNodeFacade's "getNode" method into a "getResultElement" method ** xalan::FormatterToXercesDOM requires a xercesc::DOMElement instance instead of the previously available xercesc::DOMNode instance ** changed FunctionReadDirectory accordingly * adapted FunctionTransform test case accordingly
* the root node of the result tree of each function is a domain element ** i.e. the root node of "read-xml-file" is "file", the root node of "read-directory" is "directory" * the root node contains the result state of the function call encoded in a "result" attribute ** possible values are "success" and "error" ** the root node may contain additional attributes such as the target path of a called transformation * the actual function result is contained within the child nodes of the function root node ** i.e. the XML file tree returned by "read-xml-file" is a child of the function root node ** if specific errors occured they are also returned as child nodes of the function root node *** this is currently only the case for "transform" where transformation errors are returned as "error" value node childs of the function root node * updated test cases accordingly
* wraps result node construction and appends it to root node on destruction * offers a simpler interface for common node construction patterns * simplifies result node construction in all external function implementations ** most noticeable in FunctionReadDirectory * expanded FunctionReadDirectory result nodes by name, extension, and full-path nodes
* using std::unordered_map for named cache lookups was a nice idea but is not useful when one keeps in mind that: ** we don't want cached external function responses in a situation where we are able to execute transformations inside transformations ** cached responses mean that we can not read a xml file, overwrite it using another transformation and then read it again * the currently available InputXSLT frontend only processes a single transformation anyway, so increased memory requirements should not pose a problem ** if a future frontend should offer batch processing of transformation tasks, one could implement a cache reset method to free memory after a transformation has been processed
* as the xerces document is not instantiated by DomDocumentCache::item and the class only exports a pointer to the converted xalan document, it makes no sense to expose the class to the outside in the first place
* DomDocumentCache::item class now is _finalized_ by default and doesn't perform document instantiation, just lifetime management ** xercesc::DOMDocument is instantiated inside the external function implementations and committed to the document cache for conversion to a xalan document * added mutex with scoped lock to prevent concurrent write access to the std::unordered_map contained withing DomDocumentCache * functionality of the _get_ member method was split into _get_ and _create_ ** added typedef for std::pair specialization type "optional_item" that functions as the return value of _create_ and _get_ * "locator->getSystemId()" was leaking memory as xerces doesn't manage the lifetime of the returned heap-allocated char array ** analog to XMLCh* strings ** transformed XercesStringGuard into template class to be instantiated on either XMLCh or char
* throwing std::out_of_range exception from DomDocumentCache::get instead of returning nullptr in case of a problem * moved xalanToString method back into FilesystemContext compilation unit as is only needed there
* A global DomDocumentCache instance would require key prefixing * switched internal data structure to std::unordered_map for average constant time access
* the plan to return XML-nodes from each external function requires a better way to manage the lifetime of many xerces DOM document instances and their support class instances ** this is why DomDocumentCache and DomDocumentCache::item were implemented ** based on std::map so we can easily access the result of old function calls * changed external read-directory function to return the children of the document node instead of the document node itself ** removes unnecessary cruft in function calls ** will make returning status codes alongside the function result more pleasing to the eye * updated test transformation to reflect new features