Snapshot of previously-uncommitted local work so nothing is lost after the
power outage. NOT reviewed for correctness — a WIP checkpoint, not a feature:
- multi-language hachimoji encoders (c/cpp/fortran/julia/octave/r/scala/go/rust/coq)
- formal Lean WIP (BraidTree, Eisenstein, HachimojiCapture, MathlibConnect,
ModularFormBridge, ClusterManifold) + lakefile + E8Sidon edit
- docs/, experiments/ (epyc oisc benches), deploy/, scripts, test scaffolding
- .gitignore: exclude **/target/ and Coq build artifacts
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- floorDiv returns int64 instead of int32 to match avmClamp signature
- instr.Arg wrapped in int64() for avmClamp/avmQ0Clamp calls
- Test expected value cast to int32
fix(julia): add Q16_SCALE export to Q16_16 module
Julia port:
- Extended type tracking from Bool (is_q16) to Int8 tri-state (TYPE_BOOL/Q16/Q0)
- Added ADD_SAT_Q0 / SUB_SAT_Q0 primitive handling
- Added avm_q0_clamp for symmetric Q0_16 range [-32767, 32767]
R port:
- Added PRIM_ADD_Q0 / PRIM_SUB_Q0 primitives
- Added push_q0 instruction, make_q0 / is_q0 helpers
- Added avm_q0_clamp for symmetric Q0_16 range
- Fixed LOAD to preserve type tag on restore
All three ports now support the full AVM ISA type universe (Q0_16, Q16_16, Bool).
Fixes applied to Rust, Julia, and R AVM ISA ports:
1. Division rounding: use floor division (matching Lean Int.ediv)
2. Clamp range: symmetric [-2147483647, 2147483647] for Q16_16,
[-32767, 32767] for Q0_16 (preserves negation involution)
3. V6 sign-decomposition comparison for ltQ16
4. Stack depth limit: maxStackDepth = 1024 with StackOverflow error
5. Added AVM-specific constants and helpers to each port
All three ports now match the Lean reference specification.
