Goals:

• speed up parsing to >1GB/s turtle digestion including populating rdf4cpp node storage and serializing result in turtle to /dev/null.
• eliminate serd dependency
• full unicode support
• cover all major formats (Turtle-like, JSON-based, Binary, XML)
• make it easy to support additional formats

Design suggestions:

• zero-copy (use span/string_view)
• 2-stage parsing:
  1. non-branching, SIMD-friendly structural indexing; chunk-wise
  2. (parallelizable) actions, e.g.: instantiate RDF nodes, push a subject to the stack, etc. (e.g., prefix definitions are barriers for parallelism)
• multi-source (mmap, c-stream, c++ stream, byte-buffer)

Library suggestions:

• Use taskflow and highway for thread/data parallelism.