Julia in Sophia: the language that matters most, because Moonkale was built as Lenticulum.jl’s editor (SR-114). Status by level (Languages), with the tests behind each claim. 2026-10-20.
Sources: SPEC-008 Module-Level Julia Resolution; Real Code Corpora §5; Moonkale’s
moonkale-content-id(revdacef2e6, normaliser v7);crates/sophia-ingest.
Pages
| Page | Topic |
|---|---|
| Julia Parsing and Normalization | tree-sitter, WP-66 v7, what is refused |
| Julia Name Resolution | module scope, includes, imports, generic functions as (module, name) |
| Julia Multiple Dispatch and World Age | generic identity vs method table vs dispatch result |
| Julia Macros and Generated Functions | what Sophia sees of code that writes code |
| Julia Types | the type system as far as identity and dispatch need it |
| Julia AD and Compiler Backends | Zygote, Enzyme, Mooncake, Reactant: derived programs and artifacts |
| Julia Conformance | the conformance suite: Sophia against Julia itself |
| Julia SophiaLang Interoperability | how Julia definitions relate to sophialang |
Status by level
| Level | Feature | Status | Evidence |
|---|---|---|---|
| L1 | top-level definitions: functions (long, short, where), structs, abstract and primitive types, constants, macros, modules | yes | tree-sitter-julia via arborium; LinearAlgebra: 6 of 3 552 definitions are malformed syntax |
| L2 | canonical form and content id | yes, identical to Moonkale | tests/module.rs reproduces every row of Moonkale’s module.tsv; tests/code.rs the per-file fixtures |
| L2 | refusals | 136 of 3 552 in LinearAlgebra (27 unsupported syntax, 6 malformed, 103 depending on a refused one) | examples/refusals.rs |
| L3 | module-level references, across included files | yes, as Moonkale | 2 469 references in LinearAlgebra, 948 across files |
| L3 | generic-function ownership: which function a method belongs to | 95.1 % agree with Julia on LinearAlgebra (2 777 of 2 920); the rest are re-exported bindings | Julia Conformance |
| L3 | uses of generic functions (calls) | scope-aware since v6; 2 237 in LinearAlgebra | examples/generic_calls.rs |
| L3 | using bindings, re-exports (Base.Array is Core.Array) | approximated (not followed) | all 86 remaining ownership disagreements |
| L3 | macros, @eval, generated functions | not resolved: hashed as written; since v7 nothing inside a call is a unit | Julia Macros and Generated Functions |
| L3 | dispatch: which method a call reaches | not attempted | Julia Multiple Dispatch and World Age |
| L4 | lowering to sophialang or to Julia’s lowered IR | no | — |
| L5 | execution | no; Julia runs Julia | — |
| L6 | verification | no | — |
What Sophia stores for a package
sophia ingest-code julia PKG DIR in module mode (sophia_ingest):
- records: one per identified definition, with Moonkale’s id;
- generic functions: nominal entities
jl.<module path>.<name>with methods as members; methods of other modules’ functions go to those (jl.Base.show); - types and constants: named
jl.<module path>.<name>; - references: content references (in the id) and generic uses (
nominal_refs, not in the id).
LinearAlgebra (Julia 1.13 stdlib, v7): 3 416 definitions with ids (identical to Moonkale), 3 284 distinct records, 798 generic functions (123 of them other modules’ functions that LinearAlgebra extends) with 2 907 methods.
The honest summary
Sophia can identify Julia definitions exactly as Moonkale does, and mostly knows which function a method belongs to. It does not know which method a call reaches, what a macro expands to, or what code runs. Those are the next semantic boundaries (Semantic Boundary Experiments), not more packages.