design

Status: accepted (2026-10-17, Daniel): Rust only. Proposed by Claude the same day; Daniel chose option 3. Related: Requirements SR-16, SR-30; Reading of the Earlier Vault item 1.

Context

Daniel’s Prompt3: “Julia for rapid development (also because I have no idea what we may do with such code graphs, and having all the numerical/scientific computing libraries available seems beneficial). Run Rust in production.” The earlier Repository Layout read this as a split by concern: a Rust kernel plus a Julia frontend over a C ABI. Moonkale wants a Rust crate (SR-30).

Options

  1. Julia first, then port to Rust. Fastest to explore, but Moonkale cannot link it until the port.
  2. Rust for the store, hashing, syntax and CLI; Julia for experiments. Julia reads and writes the same SQLite file (SQLite.jl), or calls the Rust library later. What Moonkale links is the production part.
  3. Rust only.

Proposal

Option 2. The store format (SQLite tables plus the canonical encoding) is the contract, so either language can produce definitions. A Julia prototype of a checker or a query can run against a real store before anyone ports it.

Decision (Daniel, 2026-10-17)

Option 3: implement in Rust only. In Daniel’s words, roughly:

  • The idea behind Julia was that, once the core stands, one could run machine learning on the code graph: graph kernels, GNNs or other analytics, because Julia’s ecosystem is good for that and Rust’s is not.
  • Some experimental work might be done in Julia, if there is a good reason, “maybe way later, but now not”. Perhaps Sophia can introspect itself by then.
  • The other old idea, writing a Julia skeleton for an agent to port to Rust, is “now entirely unnecessary”.

Consequences

  • Every v0 crate is Rust. The Julia package skeleton (Project.toml, src/*.jl, comment-only) is not part of v0. Whether to delete it or keep it is still open.
  • SQLite as the store format (S-ADR-0005) keeps the later Julia path open without any work now. A Julia analytics tool can read the store, or a later export, directly.
  • Reading of the Earlier Vault item 1 is settled: the earlier split by concern (Rust kernel plus Julia frontend) is dropped.