DSM

What is DSM?

DSM (Digital Substrate Model) is a small declarative DSL for describing data models — namespaces, concepts, clubs, keys, documents, attachments, and the function pools that operate on them. It is the modeling layer of the DevKit ecosystem and the entry point of the code-generation pipeline.

DSM is standalone. Authoring a .dsm file does not require Kibo, the Viper C++ runtime, or dsviper. You can use a DSM model as a typed contract between teams, or feed it into code generation when typed code is needed.

The DSM repository carries the ANTLR4 grammar (DSM.g4) and the canonical JSON wire format — producer-neutral, implemented independently by Viper C++, Kibo, the VS Code extension, and the JetBrains plugin.

namespace Graph {27c49329-...} {

concept Vertex;
concept Edge;

struct Position {
    float x;
    float y;
};

attachment<Vertex, Position> position;

};

Five Fundamental Notions

DSM rests on five notions ; the example above exercises four of them (namespace, concept, struct as document, attachment).

Function Pools

DSM also describes function pools — typed surfaces of operations over the model. Two flavours :

• function_pool — pure, stateless utility functions.
• attachment_function_pool — stateful operations on the model.

Either flavour is exposed both locally (Python/C++ calls) and over the wire by a Service — the surface is the same.

Type System

DSM includes a complete type system :

• Scope : namespace — UUID-scoped, isolating every type defined within.
• Primitives : bool, int8-64, uint8-64, float, double, string, blob
• Math : vec<T, n> for fixed-size numeric arrays (XYZ, UV…), mat<T, cols, rows> for matrices (e.g. 4×4 transforms)
• Containers : vector, xarray, set, map, optional
• Structures : struct, tuple, enum
• References : key<Concept> for type-safe foreign keys

Quickstart

From a .dsm source file, three operations cover almost everything :

# Validate the syntax
python3 tools/dsm_util.py check model.dsm

# Encode to the canonical JSON wire format — what Kibo (and every other producer) consumes
python3 tools/dsm_util.py encode model.dsm model.dsm.json

# Convenience: validate + encode + invoke Kibo with the viper template (Python package)
python3 tools/dsm_util.py create_python_package model.dsm

.dsm is the textual form developers write ; .dsm.json is the form Kibo (and every other producer) reads.

For real-world DSM models, the dsm-samples repository ships three reference models : Tuto (minimal), Ge (Graph Editor — recommended multi-attachment pattern), and Re (Raptor Editor — 20-file production-scale model translated from C++).