Source Code Translation to Dataflow Graphs

The patent introduces an intermediate representation (IR) as a mediator between the source procedural language and the target dataflow language. The IR captures the semantic meaning and data relationships of the source code in a language-agnostic form, enabling accurate translation while handling the fundamental differences in execution modalities. This intermediary layer allows the translator to preserve the original program's logic without being constrained by the syntactic and semantic differences between procedural and dataflow paradigms.

The translation process is segmented into distinct phases: parsing the source code into an abstract syntax tree, converting the AST to an intermediate representation that models data relationships, and then generating the target dataflow language code from the IR. This segmentation allows each phase to focus on specific aspects of the translation, improving both accuracy and manageability of the complex translation task.

The patent extracts the data access and processing logic from sequential procedural operations and reformulates it as a dataflow graph where data moves continuously through processing stages. By taking out the intermediate dataset concept and replacing it with direct dataflow connections between computational nodes, the system eliminates the need for persistent storage of intermediate results while maintaining efficient data access through the dataflow execution model.

The dataflow model enables continuous processing where data flows continuously through the computational graph without being written to and read from storage at each stage. This continuity of useful action eliminates the stop-start nature of sequential processing with intermediate datasets, allowing data to be transformed and passed along the pipeline without interruption or storage overhead.