ternlang-core 0.2.0

Compiler and VM for Ternlang — balanced ternary language with affirm/tend/reject trit semantics, @sparseskip codegen, and BET bytecode execution.
Documentation

ternlang-core

The compiler and virtual machine for Ternlang — a balanced ternary language for AI and ML systems.

Every value is a trit: reject (−1), tend (0), or affirm (+1). The tend state is a first-class computational directive — not null, not false, but "hold until evidence crosses threshold."

What's in this crate

Component Description
Lexer + Parser Tokenises and parses .tern source into an AST
Semantic analyser Type-checks, validates exhaustive 3-arm match
Bytecode emitter Compiles AST → BET (Balanced Ternary Execution) bytecode
BET VM Stack-based VM — 51 opcodes, 27 registers, tensor heap, agent runtime
@sparseskip Codegen directive: emits TSPARSE_MATMUL to skip zero-state weights
StdlibLoader Resolves use std::trit etc. at compile time with zero runtime I/O

Quick start

cargo install ternlang-cli
ternlang run my_program.tern
ternlang repl

Or use the API directly:

use ternlang_core::{Parser, codegen::betbc::BytecodeEmitter, vm::BetVm};

let mut parser  = Parser::new("let x: trit = affirm;");
let mut emitter = BytecodeEmitter::new();
while let Ok(stmt) = parser.parse_stmt() { emitter.emit_stmt(&stmt); }
let mut vm = BetVm::new(emitter.finalize());
vm.run().unwrap();

Language snapshot

fn triage(conscious: trit, vitals: trit) -> trit {
    match conscious {
        reject => { return reject; }   // hard gate — skip all further checks
        tend   => { return tend;   }   // wait for more data
        affirm => { return consensus(conscious, vitals); }
    }
}

License

LGPL-3.0 — open core. See LICENSE-LGPL.