Aquesta és una versió anterior. La versió actual és la v2.10.0. KernRift v2.10.0 · Data de publicació:
Fitxers d’instal·lació v2.8.0
| Arquitectura | Tipus | Àmbit | Mida del fitxer | Baixa | SHA256 |
|---|---|---|---|---|---|
| 64 bits | Portable | — | — | github.com | VirusTotal |
| ARM64 | Portable | — | — | github.com | VirusTotal |
L’enllaç i l’SHA256 provenen del manifest comunitari del Gestor de paquets del Windows. Convé comparar el hash del fitxer baixat abans d’executar-lo.
Notes de la versió
SSA IR backend is now the default codegen
The new SSA-based IR backend with graph coloring register allocator is now the default for x86_64 code generation. The old stack-based codegen is available via --legacy.
What's new
- SSA IR backend — full language coverage (237/237 IR tests, 310/310 total)
- Graph coloring register allocator — interference-based with liveness analysis
- Bootstrap convergent — the compiler self-compiles through IR at 1.33 MB
- --legacy flag — opt into old codegen if needed
- --emit=ir dump — human-readable IR output for debugging
IR backend features
- All arithmetic, comparisons, logical ops (signed + unsigned)
- Control flow: if/else, while, break/continue, match, early return
- Functions: calls (1-8+ args), recursion, function pointers, indirect calls
- Types: structs (field access, literals, pass-by-value, return), arrays, tuples, enums, type aliases, sizeof
- Memory: alloc/dealloc, load/store (8/16/32/64-bit), memcpy/memset, unsafe pointer ops
- Floats: f64/f32/f16 arithmetic, comparisons, conversions, sqrt, FMA
- Builtins: print/println, str_len/str_eq, file I/O, time_ns, atomics, volatile, ASM blocks
- Statics: cross-function mutable state with proper fixup resolution
Bug fixes
- Fix x86_64 PE single-target hang (1MB → 4MB codegen buffer)
- Fix CI bootstrap (pin to v2.7.1, add src/ir.kr to source concat)
- 16 register allocator bugs fixed for self-compile convergence
Platform verification
All 10 targets compile. Bootstrap convergent. 310/310 tests pass.
Full Changelog: v2.7.1...v2.8.0
