Evaluation techniques for fast access to structured, semi-structured and unstructured data using a virtual machine that provides support for dynamic code generation

摘要

Computer-implemented techniques described herein provide fast access to structured, semi-structured, and unstructured data using a virtual machine that provides support for dynamic code generation. In an embodiment, a virtual machine allows for the efficient encoding of data objects into hierarchically marked up data. A virtual machine is configured to dynamically create attributes of objects referenced by a body of code and hidden classes used to store attributes of objects referenced by the body of code. The virtual machine will determine whether a hidden class of an object satisfies one or more encoding criteria for encoding instances of objects with the hidden class. Encoding machine code is generated and executed to encode instances of objects with the hidden class. In another embodiment, direct structure decoding enables the generation of efficient machine code that is specialized for accessing only a subset of the input data that is used by an application. A parse operation is received to parse a hierarchical data object and a proxy object is created in a runtime memory space that references the hierarchical data object. It is then determined that the hierarchical data object is syntactically valid, and in response to receiving a request to read a property of the hierarchical data object, a subset of hierarchical data from the hierarchical data object is parsed to materialize a value of the property in the proxy object.