Streaming Language Rewrite Processing (SLRP)

Streaming Language Rewrite Processing (SLRP) is a low-level computing model that is essentially characterized by limited-memory processors rewriting a much larger input stream. Computation under this condition requires multiple passes, but is highly amenable to pipelining and partitioning, and thus parallelism. SLRP processors may have side channels to external devices that we drive through ‘effectful’ rewrite rules.

TLDR: SLRP is a very promising alternative to the von Neumann architecture. SLRP offers advantages for security, concurrency, distribution, and scalability. However, SLRP performs poorly for data plumbing and loops because there is no pass-by-reference within the stream. This weakness can be mitigated by effects, shunting a region of the stream to a remote device and keeping a small placeholder – a reference.

The article develops a Turing complete foundation for a SLRP machine code based on concatenative combinatory logic, and sketches several extensions. The challenge is that the operands for a combinator may be larger than a processor's memory, e.g. for a copy operator `c[X] == [X][X]` the value `[X]` might be larger than what our processor can remember. Thus, a processor cannot simply copy and paste. To solve this involves divide-and-conquer tactics, and a notion of open-ended blocks.